U.S. Semiconductors Primer
AMERICAS SEMICONDUCTORS EQUITY RESEARCH
December 11, 2013 Chip-by-Chip: Semiconductor ABCs Sector view Remains Neutral Sector Primer
In this detailed industry report, we provide a framework for longer- Research analysts term analysis by examining the market size, competitive landscape, and growth drivers for the semiconductor industry. Separately, we Americas Semiconductors have published an outlook piece, titled “Comfortably Numb,” with Romit Shah - NSI themes, catalysts, and best ideas for 2014. [email protected] +1 212 298 4326 Sidney Ho, CFA, CPA - NSI This report offers historical analysis, a discussion of key themes, and an [email protected] outlook for each of the major sub-segments including Microprocessors, +1 212 298 4329 Wireless, Logic, Memory, Analog, and Graphics. We also detail the Sanjay Chaurasia - NSI [email protected] semiconductor manufacturing processes, Moore’s Law, and new +1 212 298 4305 production technologies such as FinFET, double and quadruple patterning, and 450mm.
We discuss in the Microprocessor section: new product roadmaps, Intel’s leadership position in servers, and ARM’s 64-bit architecture. We provide a PC model forecast with analysis of the weakening correlation between unit growth and GDP. We also examine integrated CPU graphics, PC gaming, and applications that will support GPU in cloud (GRID).
In addition, we look at the evolution of cellular standards, economics of
baseband processors, and Qualcomm’s leadership position in LTE. Emerging market smartphone adoption along the S-curve, increasing complexity of RF, and the number 2 player in LTE are key themes we discuss in the Wireless section.
We also examine the communications infrastructure market with a focus on PLD content, carrier capex trends, and the impact from densification technologies such as Small Cells.
Furthermore, we share our perspective on key technology transitions in DRAM and NAND, as well as a discussion on capacity, pricing, and competition from Samsung.
See Appendix A-1 for analyst certification, important disclosures and the status of non-US analysts.
Nomura | U.S. Semiconductors Primer December 11, 2013
Relative ranking for companies in our coverage Here is our view on relative rankings for companies within our coverage.
Fig. 1: Relative ranking for our coverage
Rating Company Comments Buy Qualcomm China LTE Drives $6.00 EPS Power Xilinx Highest Growth in 2014 Micron Technology Multiple Expansion Drives Next Leg Avago Technologies TD-LTE and Cisco Drive Outsized Growth Broadcom Value is Deep Nvidia Market not Assigning Value to Tegra Cypress Semiconductor Passing the Trough Neutral Maxim Integrated Recent Win at Apple Is Interesting Atmel Show Me the Leverage Marvell Technology Skeptical of Wireless Traction Intel Uncertainty and Safety Advanced Micro Devices PCs Under Significant Pressure Analog Devices Weak Track Record of Growth Linear Technology Great Company, Bad Price Reduce Altera 28nm Share Loss not Discounted Texas Instruments Unjustifiable PE SanDisk Cautious on NAND; Samsung Competition Cavium Great Expectations; Trades at 30x
Source: Nomura research
2
Nomura | U.S. Semiconductors Primer December 11, 2013
Contents
7 Industry Overview Research analysts
9 Ranking Semiconductor Companies (by Revenue) Americas Semiconductors
Romit Shah - NSI 10 Semiconductor Industry by Device Type [email protected]
+1 212 298 4326 12 Semiconductor Industry by End Market Sidney Ho, CFA, CPA - NSI [email protected] 13 Computing (Data Processing) +1 212 298 4329
Sanjay Chaurasia - NSI 18 Wireless Communications [email protected] +1 212 298 4305
23 Communications Infrastructure and Networking
25 Consumer Electronics
27 Automotive
29 Industrial
30 Military and Aerospace
31 Best Growth Opportunities in Semiconductors
32 Semiconductor Represents 20% of Electronics Systems Value
33 The Semiconductor Industry Is Cyclical
35 Key Metrics to Watch
39 Other Metrics to Watch
41 Semiconductor Industry Is Capital Intensive
44 Capital Intensity Is Increasing
48 Intel’s Process Lead May Result in Cost Advantage versus TSMC in the Next Few Years
49 Intel Will Broaden Foundry Customer Base
50 Valuation Multiples Shrink as Market Matures
51 PHLX Semiconductor Sector (SOX)
53 Semiconductor Key Milestones
54 Semiconductor Manufacturing
58 Production Processes and SPE markets
61 Key Trends and Developments in SPE
78 Microprocessors (MPUs)
78 Overview
3 Nomura | U.S. Semiconductors Primer December 11, 2013
78 Brief History of Microprocessors
80 Types of Processors
82 Microprocessor Market Size
83 x86 Architecture – De Facto Standard in PCs
84 Intel’s Tick-Tock Model
85 High-K Metal Gate (HKMG) – Becomes an Important Breakthrough
87 3D or Tri-gate Transistors: 10 Years in the Making
90 Capital Intensity Needed to Follow Moore’s Law Is Increasing
92 End Markets
114 ASPs
117 Client Roadmap
122 Graphics Processing Unit (GPU)
122 GPU processing pipeline
123 Types of GPUs
123 Integrated graphics – Will likely keep gaining share from low-end discrete GPU
125 Discrete graphics – Will continue to dominate the high-end and gaming markets
127 Discrete Graphics Overview and Landscape
131 Market share trends – Nvidia’s share reaching record high
132 Wireless
132 Market Size
133 Key Components in the Wireless Industry
148 Handsets remain the key driver for wireless semi growth
150 Evolution of cellular standards
154 Drivers for LTE
155 Qualcomm sustaining leadership in LTE
157 In the long run, we see Intel and MediaTek vying for the No. 2 spot in LTE
160 MediaTek, over time, could become a meaningful LTE supplier in China
162 LTE adoption could drive a multiyear growth in RF components
165 RF360 – A disruptive technology?
166 We think RF360 is largely a power amplifier play, but will take time to gain traction
4 Nomura | U.S. Semiconductors Primer December 11, 2013
167 Apple’s 64-bit support will likely increase competition among ARM SoC vendors
169 3G ASPs will prove more resilient than investors fear
175 Programmable Logic Devices (PLDs)
175 PLD market
176 Problems with fixed logic
176 PLDs offer fast time-to-market
177 Types of PLDs
178 PLDs have potential to displace ASIC and ASSP
179 PLDs benefit from an expanding addressable market
182 However, PLDs have not outgrown the semiconductor market
184 The biggest driver for PLDs is carrier equipment spending
185 4G/LTE spending should improve, but relative to 3G may be more drawn out
187 Small Cells will play a bigger role in 4G networks; PLD content in small cells is very low
189 Competitive landscape
194 Analog
194 What Is Analog IC?
195 Characteristics of the Analog Market
198 Analog Market Size
200 Analog market segments
215 Analog Follows GDP Growth
219 The Analog Market Works Like a Pendulum
221 Investing For The Future
222 Focus on Shareholder Returns
224 Valuation Is Key to Investing in Analog
226 Outlook for 2014E
228 Memory
231 DRAM Market
233 Consolidation Should Bring More Stability to the Market
234 Technology Migration Drives Price Decline
5 Nomura | U.S. Semiconductors Primer December 11, 2013
236 Capital Spending Is Key to DRAM Cycles
238 The Importance of PC Is Diminishing
240 Mobile DRAM to Exceed PC DRAM in 2014
245 NAND Market
246 Multiple Applications Drive Demand Growth
248 Moving Faster than Moore’s Law
250 More Bits per Cell
252 NAND Capex Set to Rise
253 Bargaining Power of the NAND Industry
254 SSD is the Next Big Growth Driver
257 Scaling Challenges Ahead: 3D NAND and Other Options
263 Company Models
318 Appendix A-1
6 Nomura | U.S. Semiconductors Primer December 11, 2013
Industry Overview Semiconductors serve as the foundational component for modern-day electronic equipment. While the majority of semiconductors are used in personal computing and communications devices, they can also be found in a wide variety of end market applications such as consumer, industrial, medical, military, and aerospace. In 2013, we estimate the semiconductor industry will surpass $300bn in sales for the first time. According to Worldwide Semiconductor Trade Statistics (WSTS) published by the Semiconductor Industry Association (SIA), semiconductor industry revenue reached the $1bn milestone in 1964, followed by $10bn in 1979 and $50bn in 1990. 1990 to 2000: Revenue quadrupled Between 1990 and 2000, the industry grew fourfold to $204bn, representing a CAGR of 15%. There were two notable peaks during this period: the 1995 peak of $144bn (up 42% YoY) was driven by the growth in personal computing, and the peak in 2000 of $204bn (up 37% YoY) was driven by the emergence of the Internet. 2000 to 2010: Revenue decelerated following the Internet bubble Between 2000 and 2010, the industry growth rate declined to only 4% CAGR, largely because of a significant correction of 32% decline in 2001. If we exclude 2000, the growth between 2001 and 2010 was 9% CAGR. Also notable is that industry revenue declined 9% YoY in 2009 during the financial crisis, only to see a strong rebound of 32% YoY in the following year. 2010 to 2013: No growth, excluding memory Semiconductor industry growth continues to decelerate this decade. Between 2010 and 2013, we estimate industry growth slows to only 1% CAGR. As the industry matures, semiconductor revenue growth is more closely tied to global GDP growth. Macro uncertainties particularly in Europe and the U.S. had been a headwind to semiconductor growth in the past two years. Furthermore, in 2012, the emergence of smartphones and tablets drove a decline in worldwide PC sales for the first time in 10 years, and the trend continues in 2013. And although the industry showed 4% year-over-year growth through the first three quarters of 2013, the majority of the growth came from the memory industry. Excluding memory, semiconductor revenue would have been flat year over year.
Fig. 2: Semiconductor revenue YoY growth
400 1990-2000 2000-2010 2010-13 350 15% CAGR 4% CAGR 1% CAGR
300
250
200
150
100 Industry Revenue (in $bn) (in Revenue Industry
50
0
Source: SIA, Nomura estimates
7
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 3: Semiconductor industry revenue milestone
1947 Invention of transistors 1949 Semiconductor industry surpassed $100mn in sales. 1964 Semiconductor industry surpassed $1bn in sales. 1979 Semiconductor industry surpassed $10bn in sales. 1994 Semiconductor industry surpassed $100bn in sales 2000 Semiconductor industry surpassed $200bn in sales. 2011 Semiconductor industry sales reached a record high of $300bn.
Source: Global Semiconductor Alliance (GSA), SIA, Nomura research
Units track revenue Similar to industry revenue, the growth of semiconductor unit shipments also decelerated in the past two decades, declining from 11% CAGR in the 1990s to 6% CAGR in the 2000s and only 2% CAGR so far in the 2010s. Discrete components account for nearly 70% of the industry shipments, while the more complicated integrated circuits, or ICs, account for 30%. Stable ASP Average selling prices (ASP) of semiconductor components have been relatively steady, declining at an average of 1-2% per year since 2000. In 2012, discrete components had an ASP of about $0.10, while integrated circuits had an ASP of about $1.30. Even within integrated circuits, there is also a wide range of ASPs. For example, a microprocessor had an ASP of about $90, while an analog chip had an ASP of only $0.40.
Fig. 4: Semiconductor unit shipments Fig. 5: Semiconductor average selling price (ASP)
900 0.80 1990-2000 2000-2010 2010-13 1990-2000 2000-2010 2010-13
800 11% CAGR 6% CAGR 2% CAGR 0.70 +4% CAGR -2% CAGR -1% CAGR
700 0.60
600 0.50
500 0.40 400
0.30
Industry Units (in bn) (in Units Industry 300 Average Selling Price ($) 0.20 200
100 0.10
0 0.00
Source: SIA, Nomura estimates Source: SIA, Nomura estimates
8
Nomura | U.S. Semiconductors Primer December 11, 2013
Ranking Semiconductor Companies (by Revenue) In 2012, Intel had the highest revenue in the semiconductor industry at $49bn, accounting for 16% of total industry revenue. Samsung was second at $29bn and 10% share. Qualcomm (4% share), Texas Instruments (4%) and Toshiba (4%) round out the top 5 companies. The top 10 companies accounted for about 50% of the industry revenue.
Fig. 6: Semiconductor revenue and market share, 2012
Rank Company Revenue ($mn) Share (%) 1 Intel 49,089 16% 2 Samsung Electronics 28,622 10% 3 Qualcomm 13,177 4% 4 Texas Instruments 11,111 4% 5 Toshiba 10,610 4% 6 Renesas Electronics 9,152 3% 7 SK Hynix 8,965 3% 8 STMicroelectronics 8,415 3% 9 Broadcom 7,846 3% 10 Micron Technology 6,917 2% 11 AMD 5,294 2% 12 Infineon Technologies 4,797 2% 13 NXP 4,114 1% 14 Sony 3,967 1% 15 SanDisk 3,945 1% 16 Freescale Semiconductor 3,738 1% 17 Elpida Memory 3,328 1% 18 MediaTek 3,326 1% 19 Nvidia 3,244 1% 20 Marvell Technology Group 3,157 1% 21 ON Semiconductor 2,895 1% 22 Rohm 2,889 1% 23 Analog Devices 2,691 1% 24 Panasonic 2,680 1% 25 Sharp 2,537 1% Others 93,406 31% Total Market 299,912 100%
Source: Company data, Nomura estimates
9
Nomura | U.S. Semiconductors Primer December 11, 2013
Semiconductor Industry by Device Type The semiconductor industry is generally divided into seven categories. In 2012, logic accounted 28% of total sales, followed by microcomponents (21%), memory (19%), analog (13%), optoelectronics (9%), discrete (7%), and sensors (3%). The first four device types – logic, microcomponents, memory, and analog – are generally considered integrated circuits (ICs).
In this report, we will focus on the four integrated circuits segments. • Logic (28% of sales): there are two sub-segments within logic, which are standard logic (6%) and application specific logic (22%). Standard logic includes general purpose logic, programmable logic devices (PLD), and display drivers. Within application specific logic, the largest segments are communications and computing. In the segment, we will focus on the PLD segment. • Microcomponents (21% of sales): within this segment, the main sub-segments are microprocessors (14%), microcontrollers (5%), and digital signal processors (1%). The primary end market for microprocessors is computing (personal computers and servers). The key end markets for microcontrollers are automotive, consumers, and industrial. DSPs are commonly used in communications equipment. • Memory (19% of sales): this segment is driven by two sub-segments: DRAM (9%) and NAND flash memory (9%), while other memory types such as NOR and SRAM together accounted for about 1% of total 2012 semi sales. The composition of this segment changed significantly in the past decade. In 2000, DRAM was the dominant memory type, followed by NOR flash, and SRAM. • Analog (13% of sales): the analog segment can be by divided standard analog and application specific analog. As the names suggest, application specific analog is designed for one specific end market, while standard analog is a device designed for more than one application and often used by more than one customer.
Fig. 7: Semiconductor industry by device type, 2012
Sensors Optoelectronics 3% Analog 9% 13%
Discretes 7%
Memory Logic 19% 28%
Microcomponents 21%
Source: SIA, Nomura research
10
Nomura | U.S. Semiconductors Primer December 11, 2013
Over the past 5 years, the overall semiconductor market grew at a CAGR of 3%, while the growth rates of IC segments ranged from 0% to 4% CAGR. That said, there could be substantial difference within each major segment. For example, within the memory segment, DRAM declined at a 4% CAGR over the past five years, while NAND flash grew at a 12% CAGR over the same period.
Fig. 8: Semiconductor market by device type, 2008-2013E
(in $bn) 2008 2009 2010 2011 2012 2013E 5-yr CAGR Analog 36 32 42 42 39 41 2% Logic 74 65 77 79 82 85 4% Microcomponents 53 48 61 65 60 62 1% Memory 46 45 70 61 57 58 0% Discretes 17 14 20 21 19 20 3% Optoelectronics 181722232628 10% Sensors 557889 9% Total Semiconductor 249 226 298 300 292 303 3%
Source: Company data, Nomura estimates
The following table shows the major semiconductor vendors in each major segment.
Fig. 9: Major semiconductor vendors
Segment Major Semiconductor Vendors Analog Standard Analog TI, Analog Devices, Maxim, Linear, Intersil, Semtech, Micrel, ON Semi, Fairchild ASSP Analog Consumer ASSP NXP, STMicroelectronics, Broadcom PC ASSP Intersil, Semtech, International Rectifier Comm ASSP TI, STMicroelectronics, Freescale, Infineon, Broadcom, PMC, NXP Logic PLD Xilinx, Altera, Lattice, Microsemi Special Purpose Logic Intel, Samsung, Qualcomm, Braodcom, Marvell, AMD, Nvidia, Infineon Other logic Samsung, IBM, NEC, Fujitsu, Toshiba, Renasas, Magnachip Microcomponents Microprocessors Intel, AMD, Freescale, Oracle, IBM, Renasas Digital Signal Processors TI, Analog Devices, Freescale, Toshiba, NXP Microcontrollers Renasas, Freescale, Infineon, Microchip, STMicro, TI, Atmel, NXP, Fujitsu Memory DRAM Samsung, SK Hynix, Micron, Elpida, Nanya NAND Flash Samsung, Hynix, Toshiba/SanDisk, Micron/Intel NOR Flash Micron, Spansion, Macronix, Samsung, Winbond Other memory Cypress, STMicroelectronics, Renasas, Atmel, Microchip Discretes Infineon, Toshiba, ON Semi, Mitsubishi, Renasas, NXP, Vishay Optoelectronics Sony, Samsung, Sharp, Nichia, Omnivision, Aptina Sensors Robert Bosch, STMicroelectronics, Infineon, Asahi Kasei, Denso
Source: Gartner, Nomura research
11
Nomura | U.S. Semiconductors Primer December 11, 2013
Semiconductor Industry by End Market The semiconductor market serves a number of end markets. In 2012, data processing (or computing) was the largest end market, consuming 39% of the semiconductor market revenue, followed by communications at 29%, consumer electronics at 14%, industrial at 9%, automotive at 8%, and military and aerospace at 1%. Over the past 10 years, the end market split has been relatively stable. The notable exception is communications, which grew from 23% of total sales in 2002 to 29% in 2012. Specifically, wireless communications has been the fastest growing segment, accounting for 22% of total sales and up from 14% in 2002. The growth was driven by the growth in mobile handsets, and more recently the popularity of smartphones. In contrast, the computer electronics segment (including PC) within the data processing market saw two consecutive years of decline as PC growth rate stalled. This is partially offset by the increased adoption of tablets, as well as servers that power traditional enterprises and internet data centers. The consumer segment also saw a steady decline in the past three years, as it appears that consumer spending has switched from PCs and home electronics to smartphones and tablets. We expect these trends to continue over the next 5 years. In the following section, we provide more details of the dynamics of each end market.
Fig. 10: Semiconductor revenue by end market, 2012
Military and Industrial Aerospace 9% 1% Automotive 8%
Data Processing 39%
Consumer 14%
Communications 29%
Source: Gartner, Nomura research
12
Nomura | U.S. Semiconductors Primer December 11, 2013
Computing (Data Processing) Computing is the largest end market in semiconductors. Gartner estimates that computing semiconductors will reach $123bn in 2013, accounting for about 39% of total semiconductor revenue. Gartner further estimates that this segment will grow at a 5% CAGR to $151bn in 2017. The key end products in this segment are desktops, notebooks, tablets and servers. Gartner estimates that these four product types account for roughly 65% of the segment revenue. Semiconductor companies that have the highest exposure to the computing market include processor suppliers (Intel, AMD and Nvidia), hard disk drive system-on-a-chip suppliers (Marvell and LSI) and memory suppliers (Samsung, Hynix and Micron).
Fig. 11: Computing (data processing) semi revenue, 2008-2014E
160 42% 140 41% 120 40% 100 80 39% 60 38%
Market Size ($bn) Size Market 40 37% 20 0 36% 2008 2009 2010 2011 2012 2013E 2014E
Computing Revenue % of Total Semi
Source: Gartner, Nomura research
Desktops, notebooks, and tablets Traditional PC (desktop and notebook) was the primary driver for the semiconductor industry, growing at a CAGR of 10% in the last decade. Notebooks surpassed desktops in 2009, and accounted for 58% of traditional PC shipments in 2012. Geographically, the majority of the growth in the past 10 years came from the emerging markets, which accounted for around 60% of worldwide PC shipments in 2012. The primary reason for the growth in emerging markets is the increasing affordability of PCs, as ASPs declined from $1,300 in 2000 to $600 in 2012.
13
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 12: PC price declined over the past 10 years Fig. 13: PC shipments by geography, 2012
1,400 U.S. Res t of Wor ld 21% 1,200 27%
1,000
800
600 Western Europe 19% 400 PC AveragePC Price Selling
200 Asia Pacific 28% Japan 5% 0 U.S. Western Europe Japan Asia Pacific Rest of World 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Source: IDC, Nomura research Source: IDC, Nomura estimates
The emergence of tablets in 2010 significantly altered the growth trajectory of traditional PCs. In 2012, PC unit shipments declined for the first time in ten years, declining 3% year-over-year. Notably, notebook shipments declined for the first time ever, down 5% YoY. In 2013, traditional PCs and notebooks are on track to decline 10% and 15% YoY, respectively. Consumers delayed their PC replacements and purchases in favor of alternatives that include tablets and smartphones.
Fig. 14: Tablets surpassed 100mn units in 3 years Fig. 15: Notebooks declined for the first time in 2012
200 250
180
160 200
140 150 120
100 100 80
60 50 Notebook Unit Shipment (mn) Shipment Unit Notebook Tablet Shipment Unit (mn) 40
20 0 0 2010 2011 2012 2013E
Source: Gartner, Nomura estimates Source: Gartner, Nomura estimates
We believe the trend of tablets cannibalizing notebooks will continue. Tablets are proving to be cheaper than, and more portable alternatives to, notebooks for common tasks such as web browsing, email, videos, and photos. We think this could particularly affect notebook growth in emerging markets where consumers are more price-sensitive. We expect notebook shipment growth to remain muted for the next few years and grow in the mid- single digits. By contrast, we expect tablet shipments to grow 30%-plus per year for the next few years. By 2015, we expect tablet shipments to exceed traditional PC shipments.
14
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 16: Computing units by end market, 2004-2015E
700
600
500 Units (in mn) (in Units 400
300
200
100
0 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E
Server Desktop Notebook Tablet
Source: IDC, Nomura estimates
Servers Servers account for 12% of computing semiconductor revenue in 2012. While unit volume is small at about 8mn units per year, the average selling price and semiconductor content are significantly higher. Server demand is driven by traditional enterprise spending. More recently, cloud computing and data centers are playing a bigger role. Cloud computing is a service that provides scalable computing capacity to remote customers using the Internet. Cloud computing enables enterprises to have their applications up and running faster, and could be set up as public cloud or private cloud. In addition, because of the massive increase in web service usage, Web 2.0 companies such as Google, Facebook, Twitter, Baidu and Amazon are a significant growth driver for data centers that host racks of server and storage infrastructure. In 2012 and 2013, server shipment growth was challenged due to worldwide GDP slowdown. Budgetary constraints for traditional enterprises delayed server replacement. In addition, the maturation of server virtualization also impacted growth. The sluggishness in traditional enterprise servers overshadowed the strong growth in cloud computing and high performance, according to Intel. Gartner is forecasting server shipments to grow at 4% per year from 2012 to 2014. We believe the forecast could be understated given that large Internet players such as Google and Facebook are not properly represented, as they are not required to report their budget spent on data centers. Intel has more than 90% market share in server processors. Intel generates about half of its Data Center Group (DCG) revenue from traditional enterprise servers, while the other half comes from other segments including cloud computing, high performance computing, telco, and workstations. Intel recently reiterated that it expects DCG revenue to grow at a 15% CAGR from 2013 to 2017, with non-enterprise markets growing at a 20% CAGR, and traditional enterprises growing at an 8% CAGR.
15
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 17: Server shipments, 2000-2013E Fig. 18: Intel expects DCG revenue to growth at 15% CAGR from 2013 to 2017 9
8
7
6
5
4
3 Unit Shipm e nt (mn) 2
1 0
Source: Gartner, Nomura estimates Source: Intel, Nomura research
16
Nomura | U.S. Semiconductors Primer December 11, 2013
PC unit shipment forecast We forecast that total computing devices will increase 14% YoY from 472mn in 2012 to 539mn in 2013. We estimate that traditional PC shipments will decline 11% YoY from 350mn units in 2012 to 312mn units in 2013, with mobile PCs down 10%, and desk- based PCs down 9%. In contrast, we expect ARM-based tablets to grow nearly 80% YoY in 2013 to 216mn units. In 2014, we expect traditional PC shipments to further decline by 1% to 308mn units, with mobile PCs (notebooks) up 2% and desk-based PCs down 2%. We estimate ARM- based tablets will rise 27% to 274mn units. We forecast total computing devices will increase 11% YoY to about 600mn units in 2014.
Fig. 19: PC unit shipment forecast, 2011-2014E Units In millions 1Q13 2Q13E 3Q13E 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2011 2012 2013E 2014E Traditional PCs 77.3 73.4 78.5 82.9 74.1 72.0 78.5 83.4 364.0 350.2 312.1 308.0 % change, year-on-year (13.1%) (14.0%) (10.5%) (6.1%) (4.2%) (1.9%) 0.1% 0.6% 2.0% (3.8%) (10.9%) (1.3%) % change, quarter-on-quarter (12.5%) (5.1%) 7.0% 5.7% (10.6%) (2.8%) 9.1% 6.2% Desk-based PCs 34.0 32.1 32.4 34.8 32.1 31.0 32.6 34.3 155.7 146.5 133.2 130.0 % change, year-on-year (10.4%) (11.1%) (10.3%) (4.4%) (5.6%) (3.2%) 0.8% (1.5%) 1.2% (5.9%) (9.1%) (2.4%) % change, quarter-on-quarter (6.6%) (5.7%) 0.9% 7.5% (7.7%) (3.3%) 5.1% 5.0% % mix of PC shipments 43% 43% 40% 41% 42% 42% 40% 40% 43% 42% 42% 41%
Mobile PCs 44.3 42.4 47.6 50.3 44.0 43.1 48.7 52.4 209.7 204.7 184.6 188.3 % change, year-on-year (13.4%) (14.3%) (8.0%) (4.0%) (0.7%) 1.8% 2.3% 4.3% 2.5% (2.4%) (9.8%) 2.0% % change, quarter-on-quarter (15.3%) (4.4%) 12.4% 5.5% (12.4%) (2.0%) 12.9% 7.6% % mix of PC shipments 57% 57% 60% 59% 58% 58% 60% 60% 57% 58% 58% 59%
Traditional Notebooks 39.6 36.8 40.9 42.8 35.9 34.5 38.5 40.2 181.8 180.2 160.1 149.2 % change, year-on-year (12.5%) (13.3%) (11.0%) (8.1%) (9.3%) (6.1%) (5.7%) (6.2%) 14.9% (0.9%) (11.2%) (6.8%) % change, quarter-on-quarter (15.1%) (7.1%) 11.2% 4.8% (16.2%) (3.8%) 11.6% 4.3% % mix of Mobile PC shipments 89% 87% 86% 85% 82% 80% 79% 77% 87% 88% 87% 79%
Netbooks 1.5 1.2 0.8 0.7 0.5 0.4 0.3 0.3 24.5 14.8 4.2 1.4 % change, year-on-year (64.4%) (75.6%) (74.6%) (71.4%) (68.7%) (68.8%) (62.8%) (63.3%) (28.8%) (39.8%) (71.5%) (66.7%) % change, quarter-on-quarter (37.3%) (21.1%) (31.0%) (16.2%) (31.4%) (21.3%) (17.7%) (17.3%) % mix of Mobile PC shipments 3% 3% 2% 1% 1% 1% 1% 0% 12% 7% 2% 1%
Ultramobile NBs 2.2 3.3 4.4 4.6 5.6 6.0 7.1 8.7 2.1 8.8 14.6 27.4 % change, year-on-year 42.6% 72.9% 84.5% 57.6% NM 80.6% 60.1% 89.0% NM NM 65.6% 87.6% % change, quarter-on-quarter (22.9%) 48.9% 31.6% 4.3% 21.8% 8.0% 16.7% 23.1% % mix of Mobile PC shipments 5% 8% 9% 9% 13% 14% 14% 17% 1% 4% 8% 15%
x86-based Tablet PCs 1.0 1.1 1.5 2.1 2.0 2.2 2.8 3.3 1.3 1.0 5.7 10.3 % change, year-on-year NM NM NM NM NM NM 84.9% 54.0% 4.1% (22.0%) NM 80.4% % change, quarter-on-quarter NM 7.3% 40.7% 40.9% (4.0%) 7.3% 27.4% 17.4% % mix of Mobile PC shipments4%4%5%7%7%7%7%9%1%0%3%5%
ARM-based Tablet PCs 47.6 43.3 55.0 69.7 59.9 61.9 69.8 82.0 59.9 120.2 215.6 273.6 % change, year-on-year NM 54.6% 99.3% 56.6% 25.8% 43.0% 26.9% 17.7% NM NM 79.4% 26.9% % change, quarter-on-quarter 7.0% (9.0%) 27.1% 26.7% (14.1%) 3.4% 12.7% 17.5% % mix of shipments incl. ARM tablets 38% 37% 41% 45% 44% 45% 46% 49% 14% 25% 40% 46%
ARM-based Windows RT Shipments 0.91.31.31.71.11.21.51.9- 1.0 5.2 5.7 % change, year-on-year - - - 63.2% 14.5% (5.8%) 21.9% 13.6% - - NM 10.9% % change, quarter-on-quarter (7.3%) 37.2% (1.5%) 30.3% (35.0%) 12.9% 27.5% 21.4% % mix of shipments incl. ARM tablets2%3%2%2%2%2%2%2%0%1%2%2%
Source: Gartner, Nomura estimates
17
Nomura | U.S. Semiconductors Primer December 11, 2013
Wireless Communications Wireless communications is the second largest end market in semiconductors, only behind computing. Gartner estimates that wireless semiconductor revenue will reach $76bn in 2013, accounting for 24% of total semiconductor revenue. Gartner further estimates that this segment will grow at a 6% CAGR to $98bn in 2017. Semiconductor companies that have the highest exposure to the wireless communications market include Qualcomm, MediaTek, Broadcom, Avago Technologies, Skyworks, and RF Micro Devices.
Fig. 20: Wireless semi revenue, 2008-2014E
90 30% 80 25% 70 60 20% 50 15% 40 30 10%
Market Size ($bn) Size Market 20 5% 10 0 0% 2008 2009 2010 2011 2012 2013E 2014E
Wireless Revenue % of Total Semi
Source: Gartner, Nomura research
The wireless communications market has become an increasingly more important segment, growing from around 20% of total semiconductor revenue in 2005 to 24% in 2013E. Increasing content in handsets is the key driver in the past few years, as the market continues to shift from voice-centric features phones to smartphones. Smartphones take advantage of the improvement in data transfer speed from third- generation (3G) technologies such as WCDMA and HSPA, and fourth-generation (4G) technologies such as LTE. For example, EDGE (a second-generation technology) has a peak downlink speed of 0.5Mbps, versus HSDPA (a 3G technology) peak speed of 14.4Mbps and LTE (a 4G technology) peak speed of more than 100Mbps. The shift from 2G technology to 3G and 4G technologies is driving a significant increase in data traffic over cellular network networks.
18
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 21: Smartphone shipments, 2005-2014E Fig. 22: Handset shipments by technology, 2011-2017E
1,200 100% 90% 1,000 80% 70% 800 60% 50% 600 40% 30% 400 20% 10%
Sm ar tphone Unit Shipment200 (mn) 0% 2011 2012 2013E 2014E 2015E 2016E 2017E 0 2G/2.5G 3G 4G 2005 2006 2007 2008 2009 2010 2011 2012E 2013E 2014E
Source: Gartner, Nomura estimates Source: Gartner, Nomura research
In 2013, we estimate worldwide mobile handsets shipments will grow modestly at 2% YoY to 1.77bn, but wireless semiconductor revenue will be up 7% YoY. This is due to 30%-plus growth in smartphones that have higher semiconductor content, which more than offset the 20%-plus decline in voice-centric phones. We estimate smartphones will account for 54% of total shipments at an ASP of $281, and features phones will account for 46% of shipments at an ASP of $41. Because of the significantly higher ASP, we estimate smartphones will represent nearly 90% of the mobile handset revenue opportunity.
Fig. 23: Mobile handset unit shipments, 2000-2015E
2.25 2.00 1.75 1.50 1.25 1.00 0.75 Handset Units (bn) 0.50 0.25 0.00
Smartphones (bn) Voice phones(bn)
Source: Gartner, Nomura estimates
By 2015, we expect smartphones will account for 74% of total mobile handset shipments and more than 90% of handset revenue. We believe emerging markets will drive future growth of smartphones. In 2013, we estimate that the 3G/4G penetration rate in emerging markets is about 25%, but the subscriber base of emerging markets is nearly five times the size of the developed regions. We believe emerging markets are at an inflection point of the S-curve that should see significant acceleration in the next few years. Qualcomm recently forecast that smartphones will grow from 37% of the worldwide installed base in 2013 to 71% in 2017. We note that penetration rate could exceed 100% due to multiple subscriptions per person.
19
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 24: 3G/4G have significant growth potential in emerging regions
130% Developed regions 120% 120% Emerging regions 120%
125% 110% 100% 100%
120% 100% 80% 80% 115% 90% 60% 60% 110% 80% 40% 40% 105% 70%
20% 20% 100% 60%
95% 50% 0% 0% 2009 2011 2013E 2015E 2017E 2009 2011 2013E 2015E 2017E
Total penetration 3G/4G penetration Total penetration 3G/4G penetration
Source: Qualcomm, Nomura research
Fig. 25: 3G/4G penetration in developed regions vs. emerging markets, 2012
1200 3G/4G subs 1000 2G subs 800 600 400 200 0 2012 Subscribers(in mn) USA India Brazil China Korea Japan Russia Thailand ex-Brazil Americas Indonesia W. Europe W. Developed Regions Emerging Markets
Source: WCIS, Nomura research
A key driver for accelerated adoption in emerging markets is the emergence of low-cost smartphones that are less than $150. The growth in low-end smartphone will likely lead to lower average selling price category, but the mix shift from feature phones to smartphones could increase the blended ASP for mobile headsets. In 2013, we estimate smartphone ASP will decline 10% but blended mobile handset ASP will increase 13%. Similarly, the average semiconductor content per mobile phone should increase, as semiconductor content in smartphones is 4 to 10 times the content in feature phones.
20
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 26: Blended handset ASP increases despite emergence Fig. 27: Smartphones have significantly higher silicon of low-cost smartphones content than feature phones
400 400 $200
350 350 $180 $160 300 300 $140 250 250 $120 200 200 $100 150 150 $80
100 100 $60
50 50 $40
0 0 $20 2008 2009 2010 2011 2012E 2013E 2014E $0 Voice ASP Smartphone ASP Blended ASP Voice phone Low-end smartphone Mid-range High-end smartphone smartphone
Source: Gartner, Nomura estimates
Source: Gartner, Nomura estimates
21
Nomura | U.S. Semiconductors Primer December 11, 2013
Handset shipment forecast We estimate global handset shipments will grow 2% from 1.74bn in 2012 to 1.77bn in 2013, with smartphone shipments growing grow 41% to 954mn, offsetting a decline of 23% in voice-centric phones to 816mn. Smartphone shipments will likely exceed voice- centric systems for the first time, accounting for 54% of total shipments. Because of significantly higher ASP, smartphone is expected to account for 89% of the industry’s shipment value In 2014, we expect to again show low-single-digit growth at 4%, bringing total handset shipments to 1.84bn units. We expect the mix shift toward smartphones to continue. We estimate smartphones will grow 27% in units to 1.22bn, accounting for 66% of handset unit shipments and 93% of handset shipment value. While we expect smartphone ASP to decline 9% YoY, the blended ASP should improve 7% due to the mix shift.
Fig. 28: Nomura handset model, 2006-2015E
Global handset market 2006 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E 3YR CAGR Handset sales (m) 991 1,153 1,222 1,214 1,597 1,775 1,741 1,770 1,840 1,920 3.3% - yoy growth 21% 16% 6% -1% 32% 11% -2% 2% 4% 4%
Voice phone units (m) 909 1,031 1,084 1,043 1,298 1,303 1,063 816 625 501 -22.2% - voice centric growth 19.2% 13.4% 5.1% -3.7% 24.5% 0.4% -18.4% -23.2% -23.4% -19.9% Smartphone units (m) 82 122 139 174 299 472 678 954 1,215 1,420 28.0% - smartphone share of annual sales 8% 11% 11% 14% 19% 27% 39% 54% 66% 74% - smartphone penetration 2.1% 3.1% 3.9% 4.6% 6.9% 11.1% 16.3% 22.9% 30.1% 36.1% - smartphone growth 52% 49% 14% 25% 71% 58% 44% 41% 27% 17%
Voice ASP (USD) 148 133 116 106 77 58 47 41 37 34 -10.3% - yoy change -7% -10% -13% -9% -27% -25% -19% -14% -9% -8% Smartphone ASP (USD) 378 358 359 334 309 329 312 281 256 235 -9.0% - yoy change -8% -5% 0% -7% -7% 6% -5% -10% -9% -8% Total ASP (USD) 167 157 144 139 121 130 150 170 182 182 6.7% - yoy change -5% -6% -8% -3% -13% 8% 16% 13% 7% 0%
Voice revenue (USDbn) 135 137 126 111 100 75 50 33 23 17 -30.2% Smartphone revenue (USDbn) 31 44 50 58 92 155 211 268 311 333 16.4% Total global handset revenue (USDbn) 165 181 176 169 193 231 261 301 334 350 10.2% - market growth 15% 9% -3% -4% 14% 20% 13% 15% 11% 5% - voice growth 11% 2% -8% -12% -9% -25% -34% -34% -30% -27% - smartphone growth 40% 42% 14% 16% 59% 68% 36% 27% 16% 7%
Source: Gartner, Nomura estimates
22
Nomura | U.S. Semiconductors Primer December 11, 2013
Communications Infrastructure and Networking Gartner estimates that the wired communications market will reach $17bn in 2013, accounting for 6% of total semiconductor revenue. Including infrastructure revenue that is within wireless communications, the total communications infrastructure is estimated to be $22bn in 2013, or 7% of total semiconductor revenue. Gartner further estimates that this market will grow at a 3% CAGR to $27bn in 2017, led by service provider and mobile infrastructure spending. Semiconductor companies that have the highest exposure to the communications infrastructure and networking include Altera, Xilinx, Broadcom, Freescale, PMC, Analog Devices, and Linear Technology.
Fig. 29: Communications infrastructure and networking revenue, 2008-2014E
25 9.0% 8.0% 20 7.0% 6.0% 15 5.0% 4.0% 10 3.0% Market Size ($bn) 5 2.0% 1.0% 0 0.0% 2008 2009 2010 2011 2012 2013E 2014E
Comm and Networking Revenue % of Total Semi
Source: Gartner, Nomura research
By application, local area network (LAN and WLAN) accounts for 25% of total segment revenue, followed by DSL/cable/fiber (21%), enterprise (12%), mobile infrastructure (12%), public infrastructure (10%) and service provider (6%).
Fig. 30: Communications infrastructure and networking by application, 2013E
Other 14% DSL/Cable/FTTx 21%
Mobile Infrastructure 12%
Enterprise Service Provider 12% 6%
Public Infrastructure 10% LAN 25%
Source: Gartner, Nomura research
23
Nomura | U.S. Semiconductors Primer December 11, 2013
The exponential growth in data traffic through the Internet, cellular networks, enterprise networks, and home networks is driving the need for more communications infrastructure spending. Cisco forecasts that global Internet Protocol (IP) traffic will grow at a CAGR of 30% over the next 5 years, and that data traffic from mobile devices will grow at a CAGR of nearly 80% during the same period, with the strongest growth coming from mobile video traffic. To support the data traffic growth, Gartner estimates that carrier infrastructure spending will increase at a CAGR of 5% from $77bn in 2012 to $98nm in 2017, while enterprise networking spending will increase at a CAGR of 3% from $41bn in 2012 to $48bn in 2017.
Fig. 31: Carrier infrastructure spending, 2010-2017E Fig. 32: Enterprise networking spending, 2010-2017E In $bn In $bn $100 $50 $90 $45 $80 $40 $70 $35 $60 $30 $50 $25 $40 $20 $30 $15 $20 $10 $10 $0 $5 2010 2011 2012 2013E 2014E 2015E 2016E 2017E $0 Broadband Access Mobile Infrastructure 2010 2011 2012 2013E 2014E 2015E 2016E 2017E Optical Transport Service Provider Traditional Switching Voice Switching Switches WLAN ADC WAN Optimization Routers Security
Source: Gartner, Nomura research Source: Gartner, Nomura research
Over the next few years, we expect the global build-out of 4G LTE networks to drive carrier spending growth in mobile infrastructure and related wireline infrastructure. Gartner estimates that semiconductor revenue from mobile infrastructure and service providers will increase at a 6% CAGR from $3.9bn in 2013 to $5.4bn in 2017. Currently, LTE infrastructure deployment is mostly happening at China Mobile (200k LTE base stations), but we expect more spending to come through in 2014 and beyond. Next year, we expect new LTE deployments at China Telecom followed by China Unicom. In the U.S., we expect to see more spending from T-Mobile and Sprint. We believe that over time other emerging regions (Brazil, India) will start to deploy 4G wireless networks as well. Furthermore, Gartner is forecasting around 1.5mn LTE base-station shipments by 2015. While Europe has lagged infrastructure spending in LTE, we think Vodafone’s recent capex update could spur better spending next year.
Fig. 33: Worldwide service provider capex forecast
$370 7%
$360 6%
$350 5% $340 4% $330 3% $320 2% $310 1% $300 0% $290
$280 -1%
$270 -2% 2010 2011 2012 2013E 2014E 2015E 2016E 2017E
Capex ($ in bn) Growth
Source: Infonetics Research, Nomura research
24
Nomura | U.S. Semiconductors Primer December 11, 2013
Consumer Electronics Gartner estimates that the consumer semiconductor market will reach $43bn in 2013, accounting 14% of total semiconductor revenue. Gartner’s definition of consumer electronics excludes mobile phones and computing devices. Gartner further estimates that this segment will be flattish at $44bn in 2017. Some of the key products in the consumer electronics products include LCD TV, set top boxes, and game consoles. Semiconductor companies that have the highest exposure to the consumer market include Samsung, Broadcom, STMicro, Marvell, Maxim, AMD, and Silicon Image.
Fig. 34: Consumer semi revenue, 2008-2014E
60 20% 18% 50 16% 14% 40 12% 30 10% 8% 20 6% Market Size ($bn) Size Market 4% 10 2% 0 0% 2008 2009 2010 2011 2012 2013E 2014E
Consumer Revenue % of Total Semi
Source: Gartner, Nomura research
The advent of digital content is driving the growth of consumer electronics. In addition to smartphones and tablets, consumers are demanding for home electronics such as TVs, set top boxes, blue ray players, and video game consoles, as well as handheld devices such as digital camcorders, digital still camera, portable media players, and handheld game consoles. Many devices have the capability to connect to the Internet either through cable, fiber, or cellular services In 2013E, LCD TVs account for the largest consumption of consumer semiconductor with 27% share, followed by digital set top boxes (13%), digital still cameras (9%), and video game consoles (8%). The highest average semiconductor content comes from digital camcorders ($105), followed by video game consoles ($91), handheld game consoles ($65) and LCD TVs (57%).
Fig. 35: Consumer semiconductor by application, 2013E Fig. 36: Average semi content by application, 2013E
Digital Set Top $120 Box LCD TV 13% $100 27%
$80 Digital Still Camera $60 9% Digital Camcorder $40 4% $20 Video Game Consoles $0 8%
Video Game Handhelds Other 3% 27% Portable Media Blue Laser DV D Players Players 6% 3%
Source: Gartner, Nomura research Source: Gartner, Nomura research
25
Nomura | U.S. Semiconductors Primer December 11, 2013
Gartner expects consumer semiconductors to be flat from 2013 to 2017 and consumer semiconductor as a percentage of total semiconductors to decline from 14% in 2013 to only 11% 2017. We attribute the decline to some of these functions being integrated into other devices. Mobile phones have already replaced the need for MP3 players. Today, every smartphone is equipped with a high-quality digital camera, and most of them have the capability to take videos, thus reducing the need for standalone portable media players, digital cameras and camcorders. In addition, there are now many devices such as smartphones and tablets that can support casual gaming, which reduces the need for a standalone handheld game console. Gartner forecasts handheld game console shipments to decline at a 10% CAGR between 2013 and 2017.
Fig. 37: Digital still camera shipments Fig. 38: Digital camcorder shipments
140,000 18,000
120,000 16,000
100,000 14,000
80,000 12,000
60,000 10,000 2011 2012 2013E 2014E 2015E 2016E 2017E 2011 2012 2013E 2014E 2015E 2016E 2017E
Digital Still Camera (in 000) Digital Camcorder Shipments (in 000)
Source: Gartner, Nomura research Source: Gartner, Nomura research
One area that we expect will garner a lot attention is video game consoles. In late 2013, Sony and Microsoft refreshed their video game consoles for the first time in seven years. The key hardware change is that both new game consoles will use a customized version of AMD’s processor. With much improved hardware performance and connectivity to the cloud, the new game consoles promise to deliver significantly better gaming experience. In addition, China’s recent end of a 13-year ban on game consoles in the country could open up a potentially huge market for game console makers. Game console shipments typically peak in the second or third year after a generation of game consoles is launched.
Fig. 39: Video game console growth driven by next-gen game console launches
60,000
50,000
40,000
30,000
20,000
10,000
0 2011 2012 2013E 2014E 2015E 2016E 2017E
Game console sales (in 000)
Source: Gartner, Nomura research
26
Nomura | U.S. Semiconductors Primer December 11, 2013
Automotive Gartner estimates that the automotive semiconductor market will reach $25bn in 2013, accounting 8% of total semiconductor revenue. Gartner further estimates that this segment will grow at a 8% CAGR to $35bn in 2017. Automotive electronics enable advanced safety features, new information and entertainment services, and greater energy efficiency. Semiconductor companies that have the highest exposure to the automotive market include ON Semiconductor, Microchip, Linear Technology and Analog Devices.
Fig. 40: Automotive semi revenue, 2008-2014E
30 8.5%
25 8.0%
20 7.5% 15 7.0% 10
Market Size ($bn) Size Market 6.5% 5
0 6.0% 2008 2009 2010 2011 2012 2013E 2014E
Automotive Revenue % of Total Semi
Source: Gartner, Nomura research
Although worldwide automotive production is projected to grow only low-single-digit percentage per year, Gartner forecasts automotive semi revenue will grow twice as fast at an 8% CAGR between 2013 and 2017. This is because the amount of electronics used in automotive continues to increase. For example, hybrid and electric cars are gaining popularity, while in-vehicle communications and entertainment systems are getting increasingly more sophisticated. The key applications within automotive can be classified in a few categories, including safety-related equipment (air bag systems, antilock braking systems), powertrain-related equipment (engine control units), infotainment-related equipment (car radio head units, GPS navigation), body, security, and comfort-related equipment (dashboard instrument cluster, remote keyless entry, and climate control units).
Fig. 41: Worldwide auto sales increases low single digit per year (in mn) 120
100
80
60
40
20
0 2011 2012 2013E 2014E 2015E 2016E 2017E
Source: Gartner, Nomura research
27
Nomura | U.S. Semiconductors Primer December 11, 2013
Gartner estimates that the average semiconductor content per vehicle will increase from $310 in 2012 to more than $350 in 2017E. The amount of semiconductor content is significantly higher in electric and hybrid vehicles than in traditional, internal combustion engine cars. The areas that are likely to see the fastest growth include safety, advanced drivers assistance systems, the connected car, and fuel efficiency and emission reductions.
Fig. 42: Automotive semis grow twice as fast due to increasing content per vehicle
40 $360
35 $350
30 $340
25 $330
20 $320
15 $310
10 $300 Revenue Opportunity (in $bn) (in Opportunity Revenue
5 $290 Content Semiconductor Average
0 $280 2011 2012 2013E 2014E 2015E 2016E 2017E
Automotive Semiconductor Revenue Average semiconductor content
Source: Gartner, Nomura research
28
Nomura | U.S. Semiconductors Primer December 11, 2013
Industrial Gartner estimates that the industrial semiconductor market will reach $26bn in 2013, accounting 8% of total semiconductor revenue. Gartner further estimates that this segment will grow at an 8% CAGR to $36bn in 2017. Key segments include manufacturing systems, medical equipment, security and energy management, and test and measurement. The market is very fragmented in terms of customer concentration. This is a core market for analog and mixed-signal semiconductors, microcontrollers, and discrete semiconductor components. Semiconductor companies that have the highest exposure to the industrial market include Analog Devices, Linear Technology, Maxim Integrated and Avago Technologies.
Fig. 43: Industrial semi revenue, 2008-2014E
35 10% 9% 30 8% 25 7% 20 6% 5% 15 4% 10 3% Market Size ($bn) Size Market 2% 5 1% 0 0% 2008 2009 2010 2011 2012 2013E 2014E
Industrial Revenue % of Total Semi
Source: Gartner, Nomura research
The industrial semiconductor market is highly correlated with worldwide economic growth, but tends to grow at 1 to 2 times the rate of GDP growth. Some of the fastest growing segments in the industrial market include smart energy (smart meters, grids, and appliances) and medical electronics. Over a longer period of time, industrial semis track the growth of industrial OEMs, such as General Electric, Honeywell, and Emerson.
Fig. 44: Industrial market driven by many sub-segments
40
35
30
25
20
15
10
Revenue Opportunity ($bn) Opportunity Revenue 5
0 2011 2012 2013E 2014E 2015E 2016E 2017E
Manufacturing systems Medical Equipment Security/Energy Management Test and Measurement Other Industrial
Source: Gartner, Nomura research
29
Nomura | U.S. Semiconductors Primer December 11, 2013
Military and Aerospace The military and aerospace market is the smallest segment in the semiconductor industry, accounting for $3.6bn in 2013, or about 1% of total semiconductor revenue. Gartner estimates that this segment will grow at 4% CAGR to $4.2bn in 2017. The military and aerospace segment includes subsystems for commercial and military aircrafts, missile guiding systems, radar communications warfare systems such as software-defined radio, electronic warfare systems, unmanned vehicles, secure communications equipment, and intelligence gathering systems. Semiconductor companies that have the highest exposure to the military aerospace market include Microsemi and M/A COM. This segment has become increasingly important for PLD suppliers, including Altera and Xilinx, as their exposure to this market has increased to 10-15% of their total sales. This market is considerably dependent on government spending and policy making. The long-term trend for this segment is a focus on the ever-tightening government budget constraints and declining investments in R&D, which should lead to stronger demand for commercial off-the-shelf solutions. For the commercial aerospace industry, manufacturers are focused on building production capabilities in order to meet the boom in demand for commercial airliners.
Fig. 45: Military and aerospace semi revenue, 2008-2014E
4.5 1.4% 4.0 1.2% 3.5 1.0% 3.0 2.5 0.8%
2.0 0.6% 1.5 0.4% Market Size ($bn) 1.0 0.2% 0.5 0.0 0.0% 2008 2009 2010 2011 2012 2013E 2014E
Military/Aerospace Revenue % of Total Semis
Source: Gartner, Nomura research
3 0
Nomura | U.S. Semiconductors Primer December 11, 2013
Best Growth Opportunities in Semiconductors While revenue growth in the semiconductor industry looks poised to decelerate as the market matures, there are a number of opportunities that we believe will see above- average growth in the next few years. Migration to higher tier phones Nomura communications equipment analyst, Stuart Jeffrey, estimates that global smartphone shipments will increase from 954mn in 2013E to 1.4bn in 2015E, representing a 2-year CAGR of 22%. The growth is driven by the adoption of 3G/4G technologies in emerging markets, which stands at 25% in 2013E. In contrast, we estimate that the adoption rate in developed regions is above 75%. If emerging markets follow the adoption curve of developed markets, the penetration rate in emerging markets should accelerate in the next few years. While smartphone ASPs in emerging markets are likely lower than in developed markets, they are significantly higher than 2G phones that carry ASPs of about $25. Semiconductor companies that we think are well positioned to benefit from this trend are Qualcomm, MediaTek and Avago Technologies. Qualcomm is the largest baseband supplier in the world and collects royalty revenue based on end device sales. MediaTek’s baseband business should benefit from its strong relationships with Asia smartphone OEMs. Avago’s average content increases to address the increased number of frequency bands used in 4G/LTE. Global build-out of 4G wireless infrastructure Since the first commercial 4G-LTE network was launched in 2009, the majority of the investments have been focused in Japan and North America. We expect the 4G build- out to expand to the rest of the world in the next few years, with China and Europe being the biggest opportunities. China Mobile is planning to build 200,000 TD-LTE base stations in 100 cities over this and next year. China Telecom and China Unicom are conducting 4G trials and we expect larger scale deployments next year. In Europe, Vodafone and Deutsche Telekom recently stepped up investments in 4G networks. GSMA Intelligence estimates that in 2017 nearly 500 LTE networks will be in service in 128 countries, nearly double the number of live networks in 2013. Semiconductor companies that we think will benefit from the global build-out include Altera, Xilinx, and Analog Devices. Programmable logic device suppliers, Altera and Xilinx, derive nearly half of their revenue from the communications end market. Analog Devices supplies a wide range of data converters, clocks, and amplifiers to the communications market, and derive about 20% of total sales from this market. Increasing electronic content in automobiles Gartner estimates that global automobile unit sales will grow at a CAGR of 4% from 2012 to 2017. Semiconductor industry revenues should grow considerably faster. This is because the average semiconductor content per vehicle is expected to increase from $310 in 2012 to $350 in 2017. The amount of semiconductor content is significantly higher in electric and hybrid vehicles than in traditional, internal combustion engine cars. Automotive electronics enable advanced safety features, new information and entertainment services, and greater energy efficiency. Some of the key sub-systems include anti-lock braking systems, infotainment systems, and remote keyless entry. Semiconductor companies that we think will benefit from this trend include Linear Technology, ON Semiconductor, and Microchip. We estimate that each of these companies derives 15-30% of its revenue from the automotive market.
31
Nomura | U.S. Semiconductors Primer December 11, 2013
Semiconductor Represents 20% of Electronics Systems Value Semiconductor demand is a function of the demand of the electronics systems that the semiconductors are shipped into. Over the past 10 years, the average semiconductor content averaged 20% of the value of the electronics systems, with a range of 17% to 22%.
Fig. 46: Semiconductor content as a percentage of electronics systems production
25%
20%
15%
10%
5%
0% 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Source: Gartner, Nomura research
In 2012, data processing electronics had the highest percentage of semiconductor content at 28%, followed by communications (23%), automotive (23%), consumer (16%), industrial (13%), and military/aerospace (4%).
Fig. 47: Semiconductor content as percentage of electronics systems production, 2012
Total
Military / Aerospace
Industrial
Automotive
Consumer
Communications
Data Processing
0% 5% 10% 15% 20% 25% 30%
Source: Gartner, Nomura research
32
Nomura | U.S. Semiconductors Primer December 11, 2013
The Semiconductor Industry Is Cyclical Cyclicality driven by macro events and supply chain The semiconductor industry grew at a CAGR of 8% from 1990 to 2012. However, the industry is highly cyclical with an annual growth rate that ranged from 32% decline in 2001 to 42% growth in 1995. On a quarterly basis, the year-over-year growth rates ranged from 4% decline in 3Q 2001 to 60% growth in 1Q 2010. A large part of the cyclicality was tied to specific historic events, most of which were related to macroeconomic factors given that the industry is highly correlated to GDP growth. For example, the oil crisis in 1973 and again in early 1980s, the PC bubble in mid-1980s, the Gulf War in early 1990s, the Asian financial crisis in 1997, the Internet bubble in 2001, and most recently the global recession in 2009. The industry had generally bounced back from the cyclical troughs in one to two years following these macro events, driven by adjustments in the supply chain There are also many inventory driven mini-cycles between macro events. Inventory cycles exist because of the long supply chain, which started from semiconductor equipment to semiconductor, to distribution and, ultimately, sell-through to end customers. Inventory-driven cycles tend to be shorter in duration, as the supply chain usually self- corrects within several quarters.
Volatility will likely decline in the future We expect volatility in industry revenue to lessen in the future, as the semiconductor market matures and the supply-chain gets more integrated. For example, many large OEMs, such as Cisco, have requested semiconductor suppliers to place consigned inventory (or vendor managed inventory) at warehouses near the customer premises, rather than building inventory internally. This is achieved through more sophisticated inventory systems that require deeper collaboration between suppliers and customers. The improved visibility to real-time customer demand should allow semiconductor companies to better manage their own inventory levels, and hence reduce the overall cyclicality of the business.
Fig. 48: Semiconductor industry revenue, YoY % change, 1990-2012 Based on quarterly revenue 80%
60%
40%
20%
0%
Gulf War -20% Asian Financial -40% Crisis Global Internet Recession Bubble -60% 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12
Source: SIA, Nomura research
33
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 49: : Four-Stage semiconductor cycle
Strong Weak Excess Capacity Markets Markets Capacity Constrained Capital Limited Price Price Equipment Equipment Softening Firming Expansion Additions
Source: Nomura research
In general, there are four stages for every semiconductor cycle, but the length and magnitude of each cycle and stage varies significantly: Stage 1: Strong markets – this phase is characterized by strong profits by semiconductor suppliers with prices firming up, while economy continues to recover. Increased profits encourage semiconductor suppliers to invest in new capacity in order to meet forecast demand. Stage 2: Excess capacity – more often than not, semiconductor suppliers overestimate the strength of end demand, or the timing of production output misses the demand due to long equipment lead times, resulting in higher inventory than needed in the supply chain. Stage 3: Weak markets – in order to get rid of excess inventory (many products have short product lives), semiconductor suppliers are willing to cut prices while at the same time cutting back on factory utilization to limit further inventory build. The result is lower profits for the semiconductor industry. Stage 4: Capacity constrained – with lower profits, semiconductor suppliers become very conservative on capital spending in order to preserve their cash balance. As a result, there could be supply shortages in the early stage of demand recovery, which in turn drives better pricing.
Fig. 50: Semiconductor Industry ASPs, YoY % change, 1991-2012 Based on quarterly data 25%
20%
15%
10%
5%
0%
-5%
-10%
-15%
-20% 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12
Source: SIA, Nomura research
34
Nomura | U.S. Semiconductors Primer December 11, 2013
Key Metrics to Watch Semiconductor is a cyclical industry, and factors affecting the industry tend to revert to the mean over time. While turning points (peaks and troughs) are important, the rate at which business moves through the cycle will dictate the magnitude and the length of the business cycles. In general, the industry works like a pendulum, in the sense that the market moves around an equilibrium, and the higher the velocity with which it enters equilibrium, the higher the velocity with which it exits.
Profit margins Semiconductor companies’ profit margins are good indicators of margin leverage, especially for companies with high fixed costs. For example, Texas Instruments’ incremental gross margin averaged 75% over the past few cycles. In addition, corporate profits could indicate the willingness of future capacity spending.
Fig. 51: Semiconductor industry gross margin
65%
60%
55%
50%
45%
40%
Source: Company data, Nomura research
Fig. 52: Semiconductor industry operating margin
35%
30%
25%
20%
15%
10%
5%
0%
Source: Company data, Nomura research
35
Nomura | U.S. Semiconductors Primer December 11, 2013
Inventory Supply chain inventory impacts factory utilization at semiconductor companies or their manufacturing partners. Because of the long supply chain, business fluctuations of semiconductor companies are usually magnified when compared to their end market customers. Semiconductor companies and the supply chain need to balance between serving anticipated demand and the risk of having excess inventory that could become obsolete over time.
Fig. 53: Semiconductor inventory
$20 90
$18 85
$16 80
$14 75 Days of Inventory Inventory Days of
$12 70
$10 65 Inventory ($ billion)
$8 60
$6 55
$4 50 Jun-03 Jun-04 Jun-05 Jun-06 Jun-07 Jun-08 Jun-09 Jun-10 Jun-11 Jun-12 Mar-03 Mar-04 Mar-05 Mar-06 Mar-07 Mar-08 Mar-09 Mar-10 Mar-11 Mar-12 Sep-03 Dec-03 Sep-04 Dec-04 Sep-05 Dec-05 Sep-06 Dec-06 Sep-07 Dec-07 Sep-08 Dec-08 Sep-09 Dec-09 Sep-10 Dec-10 Sep-11 Dec-11 Sep-12 Dec-12
Semi Inventory Semi DOI
Source: Company data, Nomura research
Fig. 54: Supply chain inventory
$32 45.0
$30
40.0 $28
$26 35.0 $24
$22 30.0 $20
$18 25.0
$16
$14 20.0 Inventory ($bn)
$12 15.0 $10
$8 10.0 $6
$4 5.0
$2
$0 0.0 Jun-03 Jun-04 Jun-05 Jun-06 Jun-07 Jun-08 Jun-09 Jun-10 Jun-11 Jun-12 Mar-03 Mar-04 Mar-05 Mar-06 Mar-07 Mar-08 Mar-09 Mar-10 Mar-11 Mar-12 Sep-03 Dec-03 Sep-04 Dec-04 Sep-05 Dec-05 Sep-06 Dec-06 Sep-07 Dec-07 Sep-08 Dec-08 Sep-09 Dec-09 Sep-10 Dec-10 Sep-11 Dec-11 Sep-12 Dec-12
PC OEM Inventory Wireless OEM Inventory Telecom OEM Inventory Total OEM DOI
Source: Company data, Nomura research
36
Nomura | U.S. Semiconductors Primer December 11, 2013
As supply chain companies work closer together, it appears semiconductor companies are carrying a higher percentage of the total supply chain inventory. Inventory at semiconductor companies grew as a percentage of total supply chain from 25% in 2008 to 31% at the end of 2012.
Fig. 55: Semiconductor represents about 30% of total supply chain inventory
100%
90%
80%
70%
60%
50%
Inventory (in $ billion) (in $ Inventory 40%
30%
20%
10%
0% Jun-03 Jun-04 Jun-05 Jun-06 Jun-07 Jun-08 Jun-09 Jun-10 Jun-11 Jun-12 Mar-03 Mar-04 Mar-05 Mar-06 Mar-07 Mar-08 Mar-09 Mar-10 Mar-11 Mar-12 Sep-03 Dec-03 Sep-04 Dec-04 Sep-05 Dec-05 Sep-06 Dec-06 Sep-07 Dec-07 Sep-08 Dec-08 Sep-09 Dec-09 Sep-10 Dec-10 Sep-11 Dec-11 Sep-12 Dec-12
Semi Inventory Supply Chain Inventory
Source: Company data, Nomura research
37
Nomura | U.S. Semiconductors Primer December 11, 2013
Semiconductors vs. end market shipments Semiconductor revenue growth relative to end market growth indicates potential inventory build or depletion. Any over- or under-shipment tends to self-correct in 2 to 4 quarters, and a period of significant over-shipment is generally followed by a period of significant under-shipment. Over a long period of time, semiconductors and end markets grow at a similar rate.
Fig. 56: Semiconductor under-shipped end markets since 1Q 2011
40%
30%
20%
10%
0%
-10%
-20%
Sem i YoY Growth-30% vs. End Market 1Q06 2Q06 3Q06 4Q06 1Q07 2Q07 3Q07 4Q07 1Q08 2Q08 3Q08 4Q08 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13E
Source: Company data, Nomura research
Factory utilization Factory utilization provides information about the potential margin leverage that semiconductor companies can achieve. SICAS discontinued this report in 2012 due to insufficient information from member companies.
Fig. 57: Worldwide fab utilization
2,500 100
2,300
90 2,100
1,900 80
1,700
1,500 70
1,300 Fab Utilization (in %)
WafersPerWeek (in 000) 60 1,100
900 50
700
500 40 1Q97 1Q98 1Q99 1Q00 1Q01 1Q02 1Q03 1Q04 1Q05 1Q06 1Q07 1Q08 1Q09 1Q10 1Q11 Wafer starts per week Fab Utilization
Source: SICAS, Nomura estimates
38
Nomura | U.S. Semiconductors Primer December 11, 2013
Silicon shipment Silicon shipment indicates the amount of wafers that is being used in semiconductor manufacturing. It is considered a leading indicator, as manufacturing throughput times are generally 4 to 8 weeks.
Fig. 58: Worldwide silicon shipment statistics
3000
2500
2000
1500
1000
Silicon Materials (mn sq inches) 500
0 1Q00 1Q01 1Q02 1Q03 1Q04 1Q05 1Q06 1Q07 1Q08 1Q09 1Q10 1Q11 1Q12
Source: SEMI, Nomura research
Other Metrics to Watch Deferred income on semiconductor companies’ balance sheets is an indicator of the amount of inventory stored at their distributors’ warehouses. This applies to companies that recognize revenue on a sell-through basis, Order lead time is an indicator of demand; semiconductor companies tend to quote long order lead times to customers when supply is tight. This could lead to customers placing more orders than they need, which is sometimes referred to as “double order.” There are no official statistics for order lead times. Book-to-bill ratio is a forward indicator of revenue growth; it is more meaningful for semiconductor equipment companies than semiconductor companies given less stringent cancellation policies. Durable goods shipment is an indicator of end market demand strength. Electronics good shipments have a high correlation with semiconductor shipments.
39
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 59: Semiconductor equipment book-to-bill, 2007-2013
1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4
Source: SEMI, Nomura research
Fig. 60: Semiconductor industry growth tracks durable goods shipment, 2004-2013
60% 30%
50%
40% 20%
30%
20% 10%
10%
0% 0%
Sem-10% i YoY
-20% -10% Durable Goods (Electronics) YoY (Electronics) Goods Durable -30%
-40% -20% Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09 Jan-10 Jan-11 Jan-12 Jan-13
Semi YoY Durable Goods (Electronics) YoY
Source: U.S. Census Bureau, Nomura research
40
Nomura | U.S. Semiconductors Primer December 11, 2013
Semiconductor Industry Is Capital Intensive Smaller process size and resulting cost reduction has driven semi growth Smaller process size has enhanced the performance of semiconductor devices by increasing chip speeds and transistor density. Meanwhile, the ability to extract more number of chips from one wafer has lowered semiconductor manufacturing costs. This improvement in semiconductor cost performance led to the expansion of applications for semiconductors. At first, when performance was more important than cost, semiconductors were used almost exclusively for military applications, but in the 1970s, they started being used in supercomputers and eventually office computers. In the 1980s, expansion of the PC market was the driver of semiconductor market growth, but since the 1990s, applications for semiconductors have expanded to telecom equipment, consumer electronics, and automobiles. As a result, the semiconductor market underwent 64-fold growth in the 30 years from 1970. This phenomenal growth was surpassed by the growth of the market for semiconductor production equipment (SPE) over the same period. Smaller process or line width requires significant investments in capital equipment. Over the past 10 years, the semiconductor industry on average spent 20% of its revenue on capital expenditures. In 2013, Intel and TSMC combined are on track to spend 30% of their revenues for capital equipment. Ten years ago, this number was 13% for Intel and TSMC combined. We note that memory companies historically have had very high capex-to-sales ratio. Over the past 10 years, the average capex-to-sales ratio for the DRAM segment was 40%, with a range of 19% in 2009 to 68% in 2007. Fluctuation in capital spending contributed directly to industry supply-demand dynamics in DRAM in the subsequent 1-2 years. Capital spending can be broadly divided into three categories: logic (including mixed- signal), memory, and other (optoelectronics, power discrete and compound). Logic averaged about 50% of the industry’s capex over the past 10 years, followed by memory at 40% and other at 10%. In 2012, capex was skewed more toward logic (67%), as memory suppliers cut back on capacity expansion for both DRAM and NAND. Over the past 10 years, integrated device manufacturers (IDMs) accounted for an average of 75% of the industry capex, while foundries accounted for 19%. In 2012, the split was 67% IDMs and 26% foundries. Not every semiconductor company owns its own fabs. Companies that own fabs, such as Intel and Samsung, are IDMs. Companies that design products but outsource the manufacturing function are called fabless companies; e.g., Qualcomm, Broadcom, NVIDIA, and Altera are fabless companies. Companies that specialize in outsourced manufacturing are called foundries. Examples of foundries are Taiwan Semiconductor Manufacturing (TSMC), United Microelectronics (UMC), and GlobalFoundries.
41
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 61: Growth of the semiconductor and SPE markets
(Market size: $mn) 1,000,000 '00 –'10 '85 –'00 CAGR 4% CAGR 13%
100,000 '70 –'85 CAGR 16% 10,000 '00 –'10 '85 –'00 CAGR -2% CAGR 15%
1,000 '70 –'85 CAGR 23% Semiconductor production equipment Semiconductors 100 1970 1975 1980 1985 1990 1995 2000 2005 2010 (CY)
Source: Nomura, based on WSTS, SEMI data
Fig. 62: Semiconductor industry capital spending
70 30%
60 25%
50 20% 40 15% 30 10% 20
10 5%
0 0% 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Total spending ($bn) Capex as % of semi sales
Source: Gartner, Nomura research
42
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 63: Capital spending by device type, 2002-2014E
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E2014E
Logic (Including Mixed Signal) Memory Others
Source: Gartner, Nomura research
43
Nomura | U.S. Semiconductors Primer December 11, 2013
Capital Intensity Is Increasing Since mid-2000, semiconductor performance enhancements have run up against a technological wall, as process shrink on its own has no longer been able to produce faster computing speeds. The use of new materials and device structures has enabled semiconductor makers to continue raising semiconductors' performance levels, but wafer and production costs are also increasing due to increasing complexities in advanced processes. However, despite moderating improvements from small process size, capital intensity seems to be increasing. Moore's Law stipulates that the number of transistors in a given area doubles every 18-24 months. To achieve that, more advanced process technology is required. For example, Intel transitioned from 45-nanometer (nm) process node to 32-nm in 2010, and again to 22-nm in 2011, and is expected to transition to 14-nm in early 2014. Each technology transition requires significant amount of capital spending, and state-of-the-art fabrication facilities (fabs), and could cost between $3bn and $5bn. We think very few companies will be able to follow Moore’s Law In 1996, the top five companies accounted for 21% of total semiconductor capital investments. This rose to 33% in 2000 and to 41% in 2006. Average capex value among the top 20 companies increased by around 50% from 1996 to 2000. From 2000 to 2006, average capex value at the top five companies increased, but there was a substantial decline among companies ranked sixth and lower. 2012 saw even higher capex concentration for the top three semi companies. Since 2010, only three semiconductor manufacturers have made major capital outlays. When compiling forecast for the overall SPE market, changes in capex by companies other than Intel, TSMC, and Samsung Electronics are now less meaningful. We think the shift to 300mm wafers was a factor in capex concentration. The construction of a single 300mm wafer facility required a minimum investment of $2bn. Compared with 200mm facilities, this represented a doubling of the minimum investment. Only a limited number of companies were able to construct new plants and those that were not able to build 300mm facilities found themselves lagging significantly in terms of cost competitiveness. The semiconductor industry thus came to be dominated increasingly by the majors in all regions of the globe. In 2012, Samsung topped the capex list at $12bn, followed by Intel at $12bn and TSMC at $8.3bn. The three companies combined accounted for 53% of the industry capex. The rest of the top 10 spenders were mostly foundries (GlobalFoundries and UMC) and memory suppliers (SK Hynix, Micron, Toshiba, and SanDisk). In 2013, we expect a similar pattern with the top 3 companies accounting for 50-55% of the industry’s total capex.
44
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 64: Top 10 Spenders on capital expenditures, 2012 and 2013E
14,000
12,000
10,000
8,000
6,000
4,000 Capital Spending ($bn) Spending Capital 2,000
0
2012 2013E
Source: Gartner, Nomura research
More revenue is needed to justify capital spending for each successive node For 200mm fabs, semiconductor companies typically required $3-5bn in revenue to justify the capex investment. According to Intel, this revenue threshold increases to $9-12bn for 300mm fabs and to $15bn-plus for leading edge nodes beyond 2015. Due to increasing capital intensity, we think there will be just a few companies that will be able to deliver Moore’s Law in the long run.
Fig. 65: Beyond 2015, likely only a few companies will be able to deliver Moore’s Law
Intel Samsung TSMC Toshiba Texas Instruments Renesas ST Micro Qualcomm Hynix Micron Broadcom AMD Sony Infeneon Fujitsu Freescale NXP Elpida Nvidia UMC 0 5 10 15 20 25 30 35 40 45 50 Revenue Required ($Bn)
Source: Intel, Nomura research
45
Nomura | U.S. Semiconductors Primer December 11, 2013
Intel’s capex has increased significantly in the recent years Until 2010, Intel spent around $5bn per year on average in capex. This trend changed dramatically starting 2011 when Intel’s capex increased more than 2x its average capex spending. The increased capex was to support 22nm transition. Intel recently guided 2014 capex to be flat year over year. Although capex is not increasing in 2014, investments are shifting away from PC (5% decrease) and Phones (20% decrease) to tablets (75% increase), Internet of Things (20% increase), Multi-comms (15% increase), and data center (10% increase). Intel expects to spend an all-time-high capex of $11bn in 2013, which is primarily driven by 14nm costs and includes around $1bn in spending for 450mm. Intel expects to use 450mm wafers in the later part of this decade to lower transistor costs. In addition to increased capex, Intel is investing in other semiconductor production equipment companies, such as ASML, to drive transistor costs further down. If we look at cumulative capex, Intel is on track to spend $33bn in the three years from 2011 to 2013, which is equivalent to the combined capex that Intel spent in the last seven years prior to 2011. That said, this increased capex has served Intel well. Intel started shipping FinFET transistors in 2011, which now appears to be around four years of lead versus TSMC. We expect TSMC to start building its first FinFET chips in 2015. In addition, Intel is also set to start shipping industry’s first 14nm transistor in 1Q14, well ahead of its nearest competitors in process technology—TSMC and Samsung.
Fig. 66: Intel’s capex has been significantly higher in recent years
$14,000
$12,000
$10,000
$8,000
$6,000
$4,000
$2,000
$0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Capex ($mn)
Source: Intel, Nomura research
46
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 67: Intel’s MPU unit shipments and capex trends, 2003-2014E
3.5x
3.0x
2.5x
2.0x
1.5x
1.0x
0.5x
0.0x 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Intel Capex (indexed to year 2003) Intel MPU units (indexed to year 2003)
Source: Intel, Mercury Research, Nomura research
Intel’s lead and entry into foundry business is also pressuring TSMC TSMC recently increased its capex and now expects its 2013 capex to be in the $9.5-10.0bn range versus around $8bn in 2012. The company aims to accelerate its FinFET roadmap and to bridge the gap with Intel in process technology. TSMC also expects to achieve commercial production of its 16nm FinFET transistors in 2015. Even with 16nm FinFET, we think TSMC’s customers will likely see a limited extension of Moore’s Law. It seems TSMC will be just adding FinFET structure to its 20nm transistors, thus providing minimal scaling of transistors per area of die. While Intel has always been ahead of TSMC in process technology, Intel never directly competed against TSMC. This situation has changed, as Intel recently indicated that it will act as a foundry for any customer, including its competitors. This is a change in strategy from last year when Intel indicated that it would not enable any of its competitors. As a result, we see Intel as a direct competitor to TSMC on this leading edge of the process nodes. In the wake of this lead, Altera announced last year that it would work with Intel for the 14nm FPGA to take advantage of Intel’s higher performance and lower transistor cost in 14nm versus TSMC. We think these dynamics are adding further pressure on TSMC to bridge its gap with Intel. Over the past five years, TSMC’s capex has increased fivefold from under $2bn in 2009 to around $10bn in 2013.
47
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 68: TSMC’s capital spending is up significantly too
$12,000
$10,000
$8,000
$6,000
$4,000
$2,000
$0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Capex ($mn)
Source: Company data, Nomura research
Intel’s Process Lead May Result in Cost Advantage versus TSMC in the Next Few Years TSMC recently indicated that the cost per transistor in the transition to 20nm from 28nm will not fall by as much as the transition to 28nm from 40nm. We also expect limited area scaling in the transition to 16nm FinFET from 20nm for TSMC, although performance and power consumption will likely be improved. On the other hand, Intel reiterated its view at 2013 Intel Developers Forum event that Moore’s Law would continue to work well for the 10nm and 7nm process nodes in its sight. As such, we think Intel will likely have relative transistor cost advantage versus TSMC at 22nm/20nm and 14nm/16nm. Intel recently indicated that it could have around 30-40% transistor cost advantage in these nodes versus TSMC. We, however, note that the advantage will not likely entirely translate into actual cost savings. In the past, Intel’s cost per transistor has been consistently higher than that of TSMC’s due to Intel’s heavy depreciation (i.e., Intel has 2-4 years depreciation for its equipment, vs. TSMC with 6 years depreciation—due to TSMC’s support for n-1 process nodes). We think TSMC’s historical cost advantage will erode in the 22nm/20nm and will likely further erode in 14nm/16nm against Intel.
Fig. 69: Cost per transistor – TSMC’s progress Fig. 70: Cost per transistor – Intel’s progress TSMC's cost per transistor decrease will pause at 20/16nm Intel's cost per transistor decrease will continue in 22nm and 14nm
28nm 20nm 16nm FF 10nm 28nm 22nm FF 14nm FF 10nm
Source: Intel, Nomura research Source: Intel, Nomura research
48
Nomura | U.S. Semiconductors Primer December 11, 2013
Intel Will Broaden Foundry Customer Base Intel indicated during its analyst day in 2013 that it would open up its manufacturing expertise to any customer, including its competitors. This is a marked change in strategy from last year, when Intel held the view that it would not enable its competitors through its advanced manufacturing capabilities. We think that Intel believed that it could use its manufacturing lead to gain traction in mobile, but that the company is realizing that a manufacturing edge alone will not translate into mobile traction and share gain against its fabless competitors (Qualcomm, Broadcom, and MediaTek). We think Intel’s manufacturing advantage alone is not sufficient to gain traction against ARM SoC suppliers. We think that in order to be competitive in mobile, Intel needs to develop a better ecosystem for its x86 application processors and develop a competitive integrated LTE modem. Lower power and better performance of the application processor without meeting these two requirements is not likely to help Intel, in our view. We believe that the new leadership under CEO Brian Krzanich is driving this change in strategy in order to get paid for its manufacturing lead. This change has implications both for foundries (TSMC, Samsung) and for fabless companies. We think TSMC could lose customers who are dependent on leading-edge performance nodes. In addition, we think Intel’s lead in manufacturing is likely to push TSMC and Samsung to increase their capex to bridge their manufacturing gap. We think Intel, over time, could get high performance chip manufacturing contracts from other fabless companies for its 14nm and 10nm nodes. Intel is already acting as a foundry for Altera, Microsemi, and a few other smaller FPGA companies. We think that many other fabless companies, such as Xilinx, Broadcom, Marvell, and Avago, would likely, over time, engage in foundry discussions with Intel to benefit from the performance and transistor cost declines in Intel’s 14nm and 10nm process nodes.
49
Nomura | U.S. Semiconductors Primer December 11, 2013
Valuation Multiples Shrink as Market Matures Price-to-earnings (P/E) ratio and enterprise value-to-sales ratio are the two most commonly used metrics when valuing semiconductor stocks. P/E is generally more volatile, and as the semiconductor industry matures, the average semiconductor P/E has been consistently on a decline over the past 20 years. For example, in the 1990s, semiconductor stocks were valued at 25-30x estimated earnings, but by 2013, the P/E ratio dropped to 15-20x. In contrast, Enterprise value-to-sales is a more consistent metric. Given the semiconductor industry is a cyclical business, current valuation metrics are often compared to historical averages (e.g., 3-year, 5-year and 10-year averages). In addition, valuation metrics relative to the industry averages are compared to historical averages. Because of the difference in capital structure between integrated device manufacturers and fabless companies, investors may also look at valuation based on free cash flow. Price-to-book is generally less meaningful in the semiconductor industry and is used for companies that do not have a track record of consistent earnings, e.g., memory companies.
Fig. 71: P/E ratios of semiconductor stocks decline as the market matures
16% P/E: 25-30x
14% P/E: 20-25x 12% P/E: 18-20x P/E: 12-15x 10%
8%
6% P/E: 15-20x
4%
2%
0% 1995-2000 2003-2008 2005-2008 2010-2012 2013E
GDP Growth Semi Growth
Source: FactSet, Nomura research
Historically, the semiconductor industry attracted talent with stock-based compensation, including employee stock options, warrants, and restricted stocks. Previously, employee stock options were not included as compensation expense given that no cash payments from the employers were involved. However, in 2006, the Financial Accounting Standard Board's FAS 123R required companies to record equity-based compensation based on the economic value given to the employees. As a result, many semiconductor companies report earnings on both a Generally Accepted Accounting Principles (GAAP) basis and a pro forma basis. The definition of pro forma basis also varies, with some companies including stock-based compensation, and others not, but both methods exclude one-time charges. To promote apple-to-apple comparison within the sector, we recommend investors focus on pro forma earnings, including stock-based compensation.
50
Nomura | U.S. Semiconductors Primer December 11, 2013
PHLX Semiconductor Sector (SOX) The PHLX Semiconductor Index (SOX) is the most widely followed index for the semiconductor industry. It is composed of companies primarily involved in the design, distribution, manufacture, and sale of semiconductors. As of June 20, 2010, there were 30 semiconductor companies represented in the index. Texas Instruments, Qualcomm, Intel Applied Materials, and Taiwan Semiconductor Manufacturing have the highest weightings and combined for about 35% of the SOX index.
Fig. 72: SOX components
Ticker Company Weight (%) Ticker Company Weight (%) TXN Texas Instruments 8.1 ASML ASML Holding NV 2.9 QCOM QUALCOMM 8.0 NVDA NVIDIA 2.7 INTC Intel 7.9 MCHP Microchip Technology 2.7 AMAT Applied Materials 6.1 LRCX Lam Research 2.6 TSM Taiwan Semiconductor Mfg 5.5 MXIM Maxim Integrated 2.5 MU Micron Technology 5.4 MRVL Marvell Technology 2.2 SNDK SanDisk 4.4 CREE Cree 2.1 ADI Analog Devices 3.9 ARMH ARM Holdings plc 1.8 BRCM Broadcom 3.8 SWKS Skyworks Solutions 1.5 XLNX Xilinx 3.6 LSI LSI 1.4 AVGO Avago Technologies 3.3 FSL Freescale Semiconductor 1.1 KLAC KLA-Tencor 3.2 TER Teradyne 1.0 ALTR Altera 3.2 ATML Atmel 1.0 NXPI NXP Semiconductors NV 3.2 ONNN ON Semiconductor 1.0 LLTC Linear Technology 3.0 AMD Advanced Micro Devices 0.7
Source: FactSet, Nomura research
51
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 73: Stock performance for our coverage universe As of Dec 9, 2013 Name Tickers Price Performance (%) YTD 1w 1m 3m 12m Advanced Micro Devices AMD 51.3 -0.3 9.3 1.7 55.1 Altera Corp ALTR -7.4 -1.2 -3.8 -15.2 2.2 Analog Devices ADI 16.5 1.6 -1.8 4.2 20.3 Atmel ATML 13.0 -3.3 4.7 -2.0 35.0 Avago Technologies AVGO 49.7 5.9 4.4 22.8 34.9 Broadcom BRCM -16.0 4.5 4.3 5.2 -16.4 Cavium Inc CAVM 14.5 -1.3 -8.0 -8.5 3.6 Cypress Semi CY -10.4 0.3 5.0 -13.5 -4.6 Intel INTC 20.9 4.6 2.8 10.0 23.7 Linear Technology LLTC 28.2 3.4 6.9 12.4 31.2 Marvell Technology MRVL 84.8 -5.7 3.2 11.3 51.3 Maxim Integrated MXIM -3.8 -0.7 -3.3 0.2 -4.6 Micron Technology MU 264.7 9.6 32.1 51.5 260.7 NVIDIA NVDA 24.0 -2.5 2.0 2.2 26.9 Qualcomm QCOM 18.6 -0.3 5.2 7.9 14.3 SanDisk Corp SNDK 58.8 1.3 1.4 18.4 67.2 Texas Instruments TXN 41.1 1.3 3.4 11.2 46.2 Xilinx Inc XLNX 23.4 -0.4 -1.9 -4.2 28.3 Average 37.3 0.9 3.7 6.4 37.5 Median 22.1 0.0 3.3 4.7 27.6 Key Indices Dow Jones Industrial DJII-USA 22.3 -0.4 1.8 7.4 22.6 NASDAQ Composite COMP-USA 34.7 0.2 3.5 11.2 36.1 S&P 500 Index SP50 26.8 0.1 2.1 9.3 27.9 Philadelphia Semi Index SOX 34.6 1.3 2.4 8.8 36.5
Source: FactSet, Nomura research
52
Nomura | U.S. Semiconductors Primer December 11, 2013
Semiconductor Key Milestones Innovation is driving the proliferation of electronics devices in our lives. The first transistor was built in 1947 and the first semiconductor integrated circuit was built in 1958. Moore’s Law (1965) has been the guiding principle for the semiconductor industry. As the chart below shows, Intel was involved with many key milestones of the semiconductor industry over the past 50 years.
Fig. 74: Key milestones of the semiconductor industry
Invention of the first transistor at AT&T Bell Telephone Labs – Transistor 1947 replaced vacuum tube amplifier and provided a tiny, reliable, and relatively inexpensive substitute for the relays in electromechanical telephone exchanges.
First semiconductor integrated circuit (IC) – On 12 September 1958, Jack S. Kilby demonstrated the first working integrated circuit to managers at Texas 1958 Instruments. This was the first time electronic components were integrated onto a single substrate. Moore’s Law –Intel co-founder Gordon Moore observed that the number of components in integrated circuits had doubled every year from the invention of 1965 the integrated circuit in 1958 until 1965 and predicted that the trend would continue. It becomes known as “Moore’s Law” , which has been the guiding principle for the semiconductor industry. First Calculator - Texas Instruments introduces the first commercially available 1967 handheld calculator. The 'miniature' calculator was able to add, subtract, divide, and multiply. Intel is Formed - Gordon Moore and Robert Noyce formed Intel with the 1968 intention of using chips in consumer electronics. Moore and Nyce also saw a need to add memory and data storage to chips.
Intel 4004 - The 4-bit Intel 4004 was originally developed for a calculator. Intel 1971 4004 contained only 2,300 transistors. Intel's modern day chip is 4,000 times faster and 50,000 times cheaper per transistor.
Intel 8086/8088 – A16-bit microprocessor designed by Intel, which gave rise to 1978 the x86 architecture. In 1979, Intel 8088 was modified with an external 8-bit data bus, and was used in the original IBM PCs. IBM Introduces First PC - Model 5150 was the predecessor to the present day 1981 PC. IBM's 5150 came with a floppy disk, cassette system and open architecture. Altera invents PLD - Programmable logic devices, unlike logic gates that have 1983 a fixed function, have undefined functions at the time of manufacture.
Qualcomm Introduces Code Division Multiple Access (CDMA) - Unlike 1989 other wireless standards, CDMA system enabled multiple users to share the same frequency channel at the same time. DRAM reaches peak revenue - Industry revenues exceed $40 billion for the 1995 first time. Nvidia markets world's first discrete GPU chip - Nvidia's GeForce 256 was a 1999 GPU, a single-chip processor, that integrated transform and lighting hardware into the GPU itself. AMD introduces 1GHz chip - AMD's 1-GHz Athlon processor beat Intel 1-Ghz 2000 Pentium chip by a few days to the market. High-K Metal Gate (HKMG) is used by Intel - HKMG was an important 2007 breakthrough in manufacturing integrated circuits because it enabled further miniaturization of transistors. Intel unveils the world’s first 3D transistor - Instead of a single planar layer 2011 controlling the source channel, the source is surrounded by multiple surfaces by creating a three-dimensional structure.
Source: IEEE, Global Semiconductor Alliance, Nomura research
53
Nomura | U.S. Semiconductors Primer December 11, 2013
Semiconductor Manufacturing From Sand to Chips The raw material for semiconductor product is silicon—the basic constituent of sand and one of the earth’s most abundant materials. Silicon is a semiconductor, meaning that it can be turned into an excellent conductor or insulator of electricity with minor amounts of impurities added. This characteristic makes it act as an on/off switch, and the basic circuit that does this is a transistor. Transistors are the basic building blocks of electronic circuits, and the on/off action is known as the 1 and 0 of the digital equipment. Semiconductor processing combines millions of transistors on a chip to produce very complex circuits. The first essential process is purifying the silicon, as any flaws or unwanted impurities will make circuits unusable. Purified silicon is then melted and seeded with a crystal. The result is a long rod (ingot) of single-crystal silicon. The ingot is then sliced with a machine equipped with a moving diamond blade that cuts through the cylinder, dividing it into thin wafers.
Fig. 75: Semiconductor manufacturing process
Source: SEMATECH, Nomura research
54
Nomura | U.S. Semiconductors Primer December 11, 2013
Basic steps in making semiconductor chips (sourced from SEMATECH) Semiconductor manufacturers produce many kinds of chips. The precise process followed to make a chip varies according to the chip type and manufacturing company. However all wafer processing involves six basic steps (steps 3-8). Once wafer processing is complete, each chip (or die) on the wafer is tested for electrical performance, cut apart with wafer saws, and put into individual protective packages. Once packaged, chips are tested again to make sure they function properly before being shipped to distributors or placed in electronic products.
• Step 1: To make wafers, polycrystalline silicon, containing elements that can modify its conductivity, is melted. The melted silicon is then used to grow silicon crystals (or ingots) that are sliced into wafers. • Step 2: To remove even the tiniest scratches and impurities, one side of each wafer is polished to a mirror-like surface. Chips are built on this surface. • Step 3: As the first step to making a chip, a layer of silicon dioxide (SiO2) glass is grown or deposited on the wafer. Because it will not conduct electricity, this layer is called dielectric. It will be patterned and etched to mask the silicon. SiO2 may be grown or deposited in later steps in the process as layers of circuits are built into the chips: (a) SiO2 can be grown on the wafer by exposing the wafer to oxygen at very high temperatures, or (b) silicon can be combined with oxygen and then in a process called chemical vapor deposition (CVD) used to coat the wafer surface. • Step 4: Photolithography, lithography for short, is a process used to create multiple layers of circuit patterns on a chip. First, the wafer is coated with a light-sensitive chemical called photo-resist. Then light is shone through a patterned plate called a mask to expose the resist—much the same way film is exposed to light to form a photographic image. • Step 5: Following the lithography process, the wafer goes to the etch area where materials are removed in a series of steps using various manufacturing tools. Exposure to light in lithography causes portions of the resist to “harden” (or become resistant to certain chemicals). This “non-hardened” resist is washed away in the development process. Then the material below it, for example SiO2, is etched away. Finally, the “hardened” resist is stripped off sp that the material underneath forms a three-dimensional pattern on the wafer. The first lithography and etch process will result in a pattern of SiO2. • Step 6: After several lithography and etch steps, subsequent layers of various patterned materials are built up on the wafer to form the multiple layers of circuit patterns on a single chip. • Step 7: To control the follow of electricity through a chip, certain areas of the wafer are exposed to chemicals that change its ability to conduct electricity. Atoms from the chemicals, called doping materials, can be “diffused”, or forced, into areas of the silicon wafer through chemical exposure and heating. Dopant atoms displace some of the wafer’s original silicon atoms to make the wafer either more or less conductive. Another doping process, called ion implantation, bombards or shoots sections of silicon with charged atoms called ions to displace silicon atoms. • Step 8: The portions of a chip that conduct electricity form the chip’s interconnections. A conducting metal (usually copper) is electro-plated on the entire wafer surface. Unwanted metal is chemically polished off to leave microscopically thin lines of metal interconnects. Today’s more complex chips require layers of metal separated by layers of glass (SiO2) or low-l dielectric insulator. All the millions of individual conductive pathways must be connected in order for the chip to function. This includes vertical interconnections between the layers as well as horizontal interconnections across each layer of the chip. • Step 9: In a process called wafer sort, each chip on a completed wafer is tested for electrical performance. Any failing chips are marked so they can be discarded when they are separated with wafer saws. The chips are put into individual packages which
55
Nomura | U.S. Semiconductors Primer December 11, 2013
will protect the chips and provide connections from the chips to the products for which they are designed. For example, chips destined for computers are placed in packaging that can be plugged into computer circuit boards. Once packaged, chips are tested again to make sure they function properly before being shipped to distributors or used in electronic products.
The semiconductor equipment market is made up of more than 100 suppliers selling equipment to semiconductor manufacturers. The market is split into 3 segments: wafer fab, packaging and assembly, and automated test applications. The top 5 suppliers account for 50% of total market revenue. The following figures show the market size, market segments, and top suppliers in each major segment.
Fig. 76: Semiconductor equipment market, 2007-2012
$50
$40
$30
$20
Revenue (in $bn) $10
$0 2007 2008 2009 2010 2011 2012 Wafer Fab Equipment Packaging and Assembly Equipment Automated Test Equipment
Source: Gartner, Nomura research
Fig. 77: Semiconductor equipment market share, 2012
14%
13%
48%
11%
7% 7%
Applied Materials ASML Tokyo Electron Lam Research KLA-Tencor Others
Source: Gartner, Nomura research
56
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 78: Key segments in semi equipment and top 3 suppliers, 2012
Lithography ($6.6bn) Deposition ($6.0bn) ASML 74% Applied Materials 47% Nikon 15% Tokyo Electron 12% NuFlare Technology 7% Lam Research 11%
Etch, Clean and Planarization ($7.4bn) Process Control ($4.4bn) Lam Research 29% KLA-Tencor 54% Tokyo Electron 22% Hitachi High-Technologies 14% Dainippon Screen Mfg 18% Applied Materials 11%
Source: Gartner, Nomura research
57
Nomura | U.S. Semiconductors Primer December 11, 2013
Production Processes and SPE markets Front-end and back-end processes Semiconductor production processes are usually defined as either a front-end or a back- end process. The former includes all processes related to patterning, while the latter refers to all processes that take place from the probe and test stages. Front-end processes are those involved in the creation of semiconductor circuitry on the surface of the silicon wafer. Back-end processes start with the electrical trials, called probes and tests, conducted on the semiconductor device created on the silicon wafer and continue with slicing/dicing the wafer, placing the chip on the lead frame (a thin metal plate used as the internal wiring of the semiconductor package), and then connecting the electrodes on the lead frame with those on the chip via thin gold wire (bonding). Back-end processing continues with molding, during which the chip is encapsulated in a mold resin for protection, and imprinting the product and/or model name. Devices that successfully get through these processes are then exposed to heat during the burn-in process that accelerates chip aging in order to weed out devices with low reliability and durability. Those that make it this far are then subject to a final electrical test and those devices deemed to be lacking any defects are then shipped out as products. FEOL and BEOL Among the front-end processes, those that involve making transistors (gates) and condensers (capacitors) on the silicon substrate are called FEOL (front-end of line) processes and those involved in embedding the circuitry are called BEOL (back-end of line) processes.
Fig. 79: Semiconductor structure and areas subject to FEOL and BEOL processes
Wire Aluminum & copper wiring
Via Interlayer dielectric film
Wiring layer (BEOL process)
Substrate (FEOL process)
Source: Nomura research
FEOL ratio is high for memory, BEOL ratio is high for logic Semiconductor devices that perform arithmetic operations, such as MPUs and microcontroller units (MCUs), are called logic devices. Such devices have more wiring layers than do memory devices and their manufacture therefore includes a relatively large number of BEOL processes. Memory devices used primarily for data storage have fewer layers of wiring but include transistors (gates) with special structures and require the formation of condensers (capacitors) that do not exist on logic devices. As a result, memory devices are subject to more FEOL processes.
58
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 80: DRAM structure (top) and logic chip structure (bottom)
Capacitor electrode Metal wiring Interlayer dielectric film
Stack electrode
Metal
Bit line (WSi2)
Word line Element isolation film Gate insulation film
Metal wiring Passivation film (aluminum/copper film) (Si3N4 film)
Via hole plug Interlayer dielectric film Titanium nitride/ (W film) titanium film
Aluminum/copper Interlayer dielectric film film below metal wiring
Titanium/ilicide film N well Gate insulation film Element isolation film Pwell Gate insulation film
Source: Nomura research
Memory and logic alternate as the main drivers of SPE sales From the end of the DRAM bubble in1996 until 2001, investment in SPE was led by logic makers, such as Intel. However, in 2002–07, the driving force in the semiconductor market switched from logic devices, which had reached the limits for shrinking chip and process size, to NAND flash memory. As a result, memory device makers’ share of investment in SPE began to rise from 2002, providing a tailwind for the FEOL equipment market. Batch-type and single-wafer systems Front-end process SPE can also be divided into batch-type systems that process dozens of wafers simultaneously and single-wafer processing systems. Figure 80 below summarizes the merits and demerits of these two types of processing equipment. Batch processing equipment provides better cost performance per wafer and higher throughput, but single- wafer equipment is more suitable for small-lot production of multiple device types.
59
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 81: Merits and demerits of batch-type and single-wafer SPE
Single wafer Batch Process uniformity ∆ Reproducibility x Waiting time x Small-lot processing x Throughput ∆ Footprint x Temperature cycling rate ∆ Contamination ∆ Cost performance x
Note: = excellent; = good; Δ = average; x = bad Source: Nomura research
Single-wafer SPE’s share is on the rise The overall trend in SPE shows a rise in the use of single-wafer equipment in line with the increase in multilayer wiring processes. This reflects the greater use of plasma- based CVD, reactive ion etching (RIE), and CMP, all of which are single-wafer systems.
60
Nomura | U.S. Semiconductors Primer December 11, 2013
Key Trends and Developments in SPE Market trends and technological developments Technological impediments to finer process sizes have become even more apparent for semiconductors and this has had a major impact on technological development within the industry. We detect seven main trends: (1) attempts to reduce process sizes by means of next-generation lithography technology, despite the inherent difficulties; (2) use of high-k metal gates and FinFET transistors to reduce process sizes; (3) use of 3D memory (rather than reduced process sizes) to advance flash memory; (4) development of nonvolatile memory as a successor to flash memory or DRAM; (5) use of larger wafers to achieve cost savings for semiconductors without reducing process sizes; (6) development of more sophisticated packaging techniques to improve semiconductor performance without reducing process sizes; and (7) development of a new earnings model. Continuing efforts to reduce process sizes The industry as a whole continues to search for patterning processes of 16nm or better for flash memory and 20nm or better for micro ICs. The most advanced type of patterning in current use is double patterning. There are two types of double patterning: double (or multiple) lithography and the use of a spacer process (self-aligned double patterning). As we expected, extreme ultraviolet lithography (EUVL), the technology seen as best suited to volume production, is unlikely to be commercialized before 2018 as a result of delays in developing light sources. This gap is likely to be filled by either extending the life of double patterning or adopting a completely new technology. Double (multiple) Lithography Aim is to avoid interference by adjoining patterns One factor that can reduce semiconductor resolution is pattern blurring caused by light interference between adjoining patterns. Double lithography seeks to avoid this by thinning out patterns and repeating the processes of exposure, applying and developing the resist, and etching.
Fig. 82: Typical double lithography process
Resist Resist application (1) Exposure/development (1) Etching (1)
Hard mask
Target layer
Resist application (2) Exposure/development (2) Etching (2) Etching (3)
Source: Nomura research
Differences from a spacer process Double lithography was first proposed in 2008. In order to distinguish it from use of a spacer process (self-aligned double patterning), we define "double lithography" as (1) splitting the patterning process and (2) carrying out a number of critical exposures. Requires more systems but not new systems or new light sources Double lithography has the advantage that, although it involves more steps in the production process and more systems than previous exposure methods, it does not require the development of new light sources or resists and lends itself to the use of systems that continue to perform at the same level.
61
Nomura | U.S. Semiconductors Primer December 11, 2013
Disadvantages of use of pattern splitting and pattern overlay as well as increased production costs One disadvantage of double lithography is increased production costs. However, this is not a major problem, partly because there are few alternatives. The main technological problem is how to split the patterning process. A further complication is the difficulty of ensuring the accuracy of the pattern overlay of the two exposures. This technique is in commercial use by Intel and a number of other companies. Sidewall Process (self-aligned double patterning) The complex sidewall process The main approach at the moment is the sidewall process, also known as self-aligned double patterning. The sidewall process is complicated, involving the following steps: (1) spacer formation, (2) thinning resist via plasma trimming, (3) spacer etching, (4) forming sidewalls, (5) removal of spacers via etching, and (6) etching of the underlying layer using sidewalls as a hard mask. Sidewall process has been developed utilizing existing systems used in mass production Development progressed because it was possible to use lithography systems and film deposition equipment available in 2008, the year in which this approach was proposed. NAND flash makers decided to start using the technique in mass production processes from 2009 and it has steadily been adopted in mass production since then. Because this approach makes greater use of CVD and etching systems, companies such as Lam Research, Tokyo Electron, and Hitachi Kokusai Electric have developed processes and systems to match that have proved to be a major opportunity. Low yields and a limitation to only simply patterns have been shortcomings However, an inability to draw other than simple patterns such as line/space patterns, a need for a large number of deposition and etching systems, a greater cleanroom space requirement, and low yields have proved to be shortcomings.
Fig. 83: Sidewall process
(1) Lithographic exposure (2) Trimming (3) Etching (1)
Photoresist Photoresist Hard mask Hard mask Hard mask
Spacer film Spacer film Spacer film
Hard mask Hard mask Hard mask Antireflective Antireflective film film Antireflective film
Polysilicon Polysilicon Polysilicon gate gate gate
(4) Orientation film formation (5) Etching (2) (6) Etching (3) Sidewall
Spacer film Sidewall
Hard mask Hard mask Antireflective Antireflective Antireflective film film film
Polysilicon Polysilicon Polysilicon gate gate gate
Source: Nomura, based on Tokyo Electron materials from Semicon Japan, 5 December 2007
62
Nomura | U.S. Semiconductors Primer December 11, 2013
EUVL (extreme ultraviolet lithography) Widely seen as the main candidate EUVL is widely seen as the main candidate for next-generation lithography technology. Current R&D on EUVL is aimed at the 16nm node and beyond. The technology uses a light source with a wavelength of 13.5nm, an order of magnitude shorter than that of light produced by the current state-of-the-art ArF excimer laser. Up until around 2000, the technology was generally known as X-ray projection lithography. Spurred on by developments in the US, Japanese companies have resumed R&D on EUVL At one time, Japanese semiconductor manufacturers were active in EUVL R&D. However, these efforts slowed as the Japanese SPE industry's position deteriorated and companies faced pressure to cut R&D expenditure. Spurred on by the 1997 formation of EUV LLC (an Intel-led consortium), the Association of Super-Advanced Electronics Technologies (ASET) in Japan embarked on R&D on EUVL systems in 1998. Elsewhere, research was started under the semiconductor Millennium Research for Advanced Information Technology (MIRAI) project and on 4 July 2002, the Extreme Ultraviolet Lithography System Development Association (EUVA) was formed with Ushio chairman Jiro Ushio at the helm and Nikon and Canon among the participants. Canon and Nikon both discontinued development However, Canon (because of concerns about returns on investment) and Nikon (because of the deteriorating profitability of its lithography business) have both, to all intents and purposes, discontinued R&D on EUVL. EUVA was disbanded as originally planned and a consortium of SPE manufacturers (EIDEC) has been carrying out R&D on products such as mirrors, masks, and resists.
63
Nomura | U.S. Semiconductors Primer December 11, 2013
EUVL development delayed by light sources The development of EUVL has also been delayed in other countries. This has been largely ascribed to the difficulties associated with light sources, but steady progress is being made in the development of light sources, regarded by many as the greatest stumbling block. Two types have been developed in parallel: discharge-produced plasma (DPP) and laser-produced plasma (LPP).
Fig. 84: EUVL structure
EUVL characteristics and technological issues
Projection system Illumination system High-precision flat mirror system Mask Design featuring small number of mirrors Ultra-low defect reflective mask Polishing manufacturing technology, mirror tube Development of Koehler illumination mirrors technology Defect detection and correction
Reflectivity of multilayer mirror <70% Lens-based optical systems cannot be used Transparent mask cannot be used
Illumination system EUV light Reticle stage (Wavelength 13.5nm)
:13.5nm
High resolution applicable at 1Xnm nodes Projection optics system Materials absorption is extremely high Light source Wafer stage
New light sources need to be developed Exposure in a vacuum is necessary Thick-film resist cannot be used
EUV light source Exposure system Resist High power Low debris Vacuum stage Alignment Surface-layer imaging technology Low etendue Chuck Temperature control Outgas control
Source: EUVA, Nomura research
Ushio has effectively withdrawn Cymer, Ushio, Xtreme, and Gigaphoton have been involved in the development of light sources. Xtreme became a subsidiary of Ushio in 2008. Gigaphoton was a 50:50 joint venture between Ushio and Komatsu but became a wholly owned subsidiary of Komatsu in May 2011 in order to sever its ties with Ushio. Ushio effectively discontinued R&D on EUVL systems in 2013 and now is focusing on light sources for development purposes. Main features of the DPP method: simple in design, but has failed to gain support DPP is a method by which EUV light is generated by injecting a tin-bearing gas between a pair of charged electrodes and causing them to discharge. The approach is highly efficient in terms of light conversion but has conventionally been prone to electrode damage, which limits its output. As a result of a technological breakthrough―specifically, the use of rotary electrodes―Ushio made considerable progress with development. However, a widespread view among lithography system and SPE manufacturers that output would be limited made them reluctant to use this system and eventually led Ushio to withdraw.
64
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 85: DPP and LPP technologies
Power supply Debris shield
Debris shield
Droplet
Charged gas Intermediate collection point Power Power supply Collection
Collection Insulator mirror mirror
Intermediate collection Laser point
Source: Nomura, based on data in Nikkei Tech-On! article and Toshiba data
Main features of the LPP method LPP is a method by which EUV light is generated by shining a laser beam on droplets of molten tin. This method is compact but somewhat lacking in terms of light conversion efficiency. A second issue is cooling: energy not transformed into light instead becomes heat, which can be difficult to extract. A third problem is that tin within the plasma tends to precipitate on the main condenser mirror. High power is required EUVL uses reflective optics. A beta version has six reflective mirrors and a production version has eight. Each mirror reflects about 70% of the light. With six mirrors, power drops to an eighth by the time it reaches the wafer, and with eight mirrors it drops one sixteenth. Thus, to maintain 100 wafers per hour (WPH) with a six-mirror setup, the output of the light source has to be 115W; likewise, with an eight-mirror setup, it has to be 250W, although this varies from manufacturer to manufacturer. Intel requires 1,000W for full- volume production. However, mass production could probably be supported at outputs of 100W and below if applications were limited to cutting lithography and contact holes. Cymer's problems and takeover by ASML Although Cymer led the development of light sources, it found itself unable to generate a steady output of more than 10W and was hampered by heavy EUVL development costs. Faced with the prospect of having to discontinue development, it was taken over by ASML in 2012. Gigaphoton got a much later start than Cymer in the area of LPP, but following Cymer’s difficulties, it may be able to catch up in performance terms by the time EUVL is ready for mass production.
65
Nomura | U.S. Semiconductors Primer December 11, 2013
Mask-related problems look small in comparison While the lack of mask inspection and repair equipment is also an issue, this is not a major problem in contact hole and other simple pattern applications. Moreover, with Intel having announced that it had made a zero-defect mask, we think that problems associated with masks are relatively minor compared with light source issues. EUVL mask business opportunities Hoya and Asahi Glass make EUVL mask blanks. Hoya's EUVL mask blank sales come to about ¥1bn a year. Developers of masks include Hoya, Dai Nippon Printing, and Toppan Printing. Refractive optics cannot be used EUV radiation at the 13.5nm wavelength is absorbed by all materials and has a refractive index of about 1, as a result of which the lithography system uses mirror-based reflective optics instead of lens-based refractive optics. Masks also need to be reflective, instead of being filled with holes or transparencies, as is the case now. EUVL masks comprise a buffer layer and an absorber layer patterned on a reflective multilayer coating formed on so-called zero expansion glass, a glass substrate with a very low coefficient of thermal expansion. Complexity of EUVL masks The reflective multilayer coating, with a thickness of about 6.9nm, consists of about 40 alternating layers of molybdenum (Mo) and silicon (Si), designed for high reflectivity using multiple reflections in the area of the wavelength of exposing radiation. Chrome and tantalum nitride, which absorb EUV light at a high rate, are being considered as materials for the EUV radiation-blocking absorber layer. The buffer layer is designed to prevent damage from occurring to the multilayer coating during etching of the absorber material and to serve as a protective layer during repair of defective patterns on the absorber material. Silicon dioxide is primarily used for the buffer layer. Development issues When making 40 Mo/Si layers, issues include detecting and repairing defects caused by the introduction of foreign particles, as well as testing and repairing mask patterns. Blank manufacturers are expected to make zero-defect blanks. Lasertec makes blank inspection equipment and has a record of supplying such equipment to a research organization. We do not expect Lasertec to have any intention of booking large returns during the development phase, but we think its inspection equipment costs well in excess of ¥1bn per unit because they are developed individually.
66
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 86: EUVL mask structure and manufacturing process (case involving phase defect)
Substrate 1. Cleaning of substrate
Multilayer Si
2. Formation of Si/Mo Mask blanks multilayer coat Mo
Absorber Buffer 3. Formation of buffer and absorber layers
Mask pattern
4. Pattern formation Mask
Source: Nomura, based on EUVA website
ASML has announced the development of a 43WPH EUVL system The latest development is that ASML, which continues to develop EUVL systems, announced that it has developed a system with a light source output of 55W and a throughput of 43WPH. However, these figures may only apply in limited operating conditions (with, for example, extended maintenance required after a few hours of operation). The company aims to develop an 80W/58WPH system by this year and a 250W/125WPH by 2015. EUVL may be used for limited processes if performance continues to improve Provided a certain minimum throughput can be achieved, there may not be a major replacement of ArF immersion lithography systems in 2015 and Samsung Electronics may use them with contact holes while Intel uses them with cutting lithography. Samsung Electronics still has high hopes of EUVL ASML currently has orders for 11 EUVL systems for mass production. However, if it can increase throughput to 70WPH by 2014, it could receive orders from Samsung Electronics for another seven machines. Nikon has made the right decision in our view ASML has continued to develop EUVL systems. We think that if EUVL had been adopted for use in mass production, Nikon might have discontinued production of lithography systems. However, it appears to have made the right decision because we currently see no sign of EUVL systems being used for that purpose.
67
Nomura | U.S. Semiconductors Primer December 11, 2013
Quadruple patterning + DSA Difficult choice because of delays with EUVL As a result of the delays in the development of EUVL technology, the industry sees the sidewall process and a combination of double (or multiple) exposure and directed self- assembly (DSA) as the most promising options. Under consideration for single-layer multi-patterning are a quadruple patterning technology developed by Toshiba and a number of other companies that involves a single repetition of the spacer process and a triple patterning technology developed by TSMC and a number of other companies that involves two repetitions of the exposure/etching process.
Fig. 87: Quadruple patterning
Normal exposure and development Plas ma tr imming Etc hing Sidew all film formation
Plasma trimming Spacer film etching Sidew all film etching w ith HM Sidew all film formation
Spacer film etching Sidew all film etching w ith HM
Source: Nomura research
Major implications for etching systems Because quadruple patterning significantly increases the number of times etching equipment is used, we expect its uptake to benefit Tokyo Electron and Lam Research considerably. In order to ensure overlay precision, we expect demand to increase for CD-SEMs, which should benefit Hitachi High-Technologies. Increased costs of quadruple patterning Quadruple patterning, which requires sidewall patterning to be done twice and patterns only one-quarter of line and space dimensions with each exposure, causes declines in yields and process stability, so an increase in costs would be a problem. There has been concern that if flash memory were fabricated using quadruple patterning for 16nm processes, the bit cost would be higher than for 19nm processes and that there would therefore be no point in this process migration. However, there are indications that these costs can be reduced. Seen by some as "worst option" A combination of multi-patterning and DSA has been seen by the industry as the "worst option." This is because process costs rise sharply unless it is combined with other measures to increase productivity (such as increasing throughput). Until now, the use of finer processes has failed to achieve lower device costs.
68
Nomura | U.S. Semiconductors Primer December 11, 2013
Next-generation “cutting” or “complementary” lithography In MPUs, a market in which increases in process costs can to some extent be passed on in device prices, Intel is developing next-generation lithography solutions known as "cutting lithography" and "complementary lithography." These involve using quadruple exposure for patterning of very fine lines and spaces before using cutting lithography to create very fine patterns on specific areas.
Fig. 88: Complementary lithography
ArF immersion lithography Multiple patterning Cutting lithography Fine patterning
Source: Nomura research
EUV light and e-beams could be used in cutting lithography EUV light and e-beams are under consideration as light sources for cutting lithography. This would mean designing devices using the grid design rule, which allows only vertical and horizontal, but not diagonal, patterns. The throughput required of cutting lithography is around 50WPH, much lower than the current 200WPH, so there are good prospects for volume production using EUV light and, for limited applications, e-beams.
Fig. 89: DSA reduces line width roughness
Line/space via regular lithography Coating with DSA materials
Heat treatment Selective etching
Source: Nomura research
69
Nomura | U.S. Semiconductors Primer December 11, 2013
FinFET FinFET is key for miniaturization FinFET is one type of transistor with a double-gate construction and the application of equal voltage from both the source and drain sides enables significant improvement in transistor switching performance. Efficiently control current by implanting impurities under the gate For a transistor on a wafer, the current from source to drain is turned on and off by controlling the gate voltage. Implanting impurities on the silicon substrate under the gate and using opposing p-type or n-type semiconductors for the source and drain enables more efficient control of current in the chosen direction.
Fig. 90: FinFET structure
Conventional transistor FinFET
Drain (back side)
Gate
Gate Gate Source Drain
Source Impurity addition channel Impurity atoms
Source: Nomura, based on Advanced Industrial Science and Technology (AIST) data
Impurity variance creates gate performance variance However, the numerical variance of impurities under the gate increases with miniaturization, resulting in wider variance in gate performance. The voltage required to turn current on and off is called the threshold voltage, but threshold voltage needs to be set according to the worst-performing gate.
7 0
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 91: Increase in threshold voltage
Vth variance problem VVdd reduction difficulty Vdd
Over-Drive
Supply voltage (V) Vth variance
Vth design value
Technology generation (nm)
Source: Nomura, based on AIST data
Threshold voltage increases due to variance in gate performance Threshold voltage previously decreased with miniaturization. Going forward, although we expect the median required threshold voltage to decline with miniaturization, threshold voltage based on the lowest gate performance could increase due to impurity variance. Leakage current issues are also a bottleneck for reducing power consumption Leakage current flowing from the source to the drain via the silicon layer below the region of implanted impurities is also a problem. Higher leakage current contributes to higher power consumption and reducing leakage current is an especially large issue for semiconductors used in mobile electronics, for which battery usage time is a key factor.
Fig. 92: Leakage current issues
Gate
Source Source
Leakage current
Source: Nomura research
Solve the foregoing issue through the introduction of FinFET FinFET increases current controllability by inserting the source and drain from both sides with a gate. This limits the increase in threshold voltage caused by impurity variance because impurities are no longer required between the source and the drain. The insertion of source and drain from both sides also limits leakage current.
71
Nomura | U.S. Semiconductors Primer December 11, 2013
Etching and ALD equipment to benefit FinFET started to be introduced for DRAM and logic devices from 2012. Transistor complexity has increased demand for silicon etching equipment and ALD equipment. Hitachi High-Technologies, Lam Research, and Applied Materials are main suppliers of silicon etching equipment to DRAM and logic makers. ASM International has a large share in ALD equipment.
Fig. 93: FinFET production process
1. Start (SOI) 5. Sidewall formation Insulating film sidewall
Silicon
Embedded oxide film
Silicon
2. Fin formation Insulating film sidewall Fin 6. HDD-I/I
Source Drain
Gate metal/ 3. GS formation gate oxide film
7. Silicide Metal
4. Ext-I/I Impurities
8. Wiring Metal wiring
Via
Source: Nomura, based on AIST data
72
Nomura | U.S. Semiconductors Primer December 11, 2013
Shift to 3D NAND Flash Technology
3D memory Toshiba and Samsung Electronics are looking to 3D memory as the technology to enable greater transistor density and lower costs. They are developing technologies called BiCS and TCAT, respectively. BiCS offers the potential for much higher levels of integration and thereby lower costs because it can create memory cells with several dozen vertical layers. Toshiba has already produced samples of 16-layer memory ICs at the 60nm node. Samsung Electronics has announced that it will begin pilot production at a plant in Xian in China in 2014 and has made numerous inquiries with SPE manufacturers.
Fig. 94: 3D memory chip structure
The structure of the new memory cell
Bit line
NA ND
Upper selection gate
Control gate
Low er selection gate
Vertical Source line SONOS FET
Vertical FET
Source: Nomura, based on Toshiba data
Fig. 95: Various types of 3D memory
Toshiba
P-BiC: vertical
Toshiba Samsung Electronics
BiCS: vertical TCAT: vertical
Samsung Electronics
TCAT: horizontal
2007 2008 2009
Source: Nomura, based on company press releases
73
Nomura | U.S. Semiconductors Primer December 11, 2013
Straightforward production process The production method involves creating a stack of insulator and circuit layers, creating holes using etching systems, and filling those holes with doped silicon oxide. The process itself is relatively straightforward.
Fig. 96: 3D memory production process
Source: Toshiba website, Nomura research
Many issues to deal with However, based on our discussions with SPE manufacturers, we think that BiCS and TCAT are unlikely to be used in mass production in the near term. The thickness of the barrier films required means that through holes cannot be made that much smaller or placed too close together and, while a sharp increase in layer count is required for greater transistor density, increasing layer count also raises the ratio of hole length to hole diameter, making it difficult to get electrode materials right to the bottom of the holes. Many methods also exist for 3D memory. We think the method used will determine whether there will be problems in terms of achieving higher densities, the number of rewrites possible, and intercell interference. We expect manufacturers to examine the feasibility of mass production by checking yields, process stability, and the potential for cost reductions while operating pilot lines. For all their similarities, BiCS and 3D NAND are different Based on our discussions, we think that the 3D NAND memory chips being developed by Korean memory makers are likely to have a simpler structure and be easier to manufacture than the BiCS chips being developed by Toshiba. Also, we think that instead of creating a stack with eight or more layers and then using an etching system to form through holes, as Toshiba does with its BiCS chips, Korean memory makers may build up a stack gradually, two or three layers at a time, repeating the etching process each time. Opinion on the optimum number of layers is divided Opinion on the optimum number of layers is divided, with one Japanese semiconductor manufacturer saying that 3D NAND would have to have at least 100 layers in order to gain a cost advantage over planar NAND, while two US processing equipment manufacturers gave 64 and 32 layers, respectively, as the optimum number and a lithography system manufacturer saying 16 layers. Korean companies are counting on 3D NAND Flash memory manufacturers other than Toshiba that have dropped out of the race to achieve finer processes seem to be fully counting on the development of 3D NAND.
74
Nomura | U.S. Semiconductors Primer December 11, 2013
450mm Could Bring Productivity Improvements Shift to 450mm wafers Reducing semiconductor manufacturing costs has been one objective of shrinking process geometries. Lowering costs via improvements in capex efficiency has similarly been one objective of creating larger-diameter wafers. Because it looks as though the move to finer process sizes has hit a wall, the major semiconductor manufacturers are hoping to increase the size of the silicon wafers they use in the semiconductor production process from the current standard diameter of 300mm (12 inches) to 450mm (18 inches). The shift to 450mm wafers was first mapped out by the International Technology Roadmap for Semiconductors (ITRS) in 2003. At that time, the start of mass production was targeted for 2012. Intel in driver seat From around 2006, Intel started to express its desire to shift to 450mm wafers and began exerting strong pressure on industry associations to initiate development. So far, Samsung Electronics and TSMC also appear to have expressed their approval. However, several chipmakers are reluctant to shift to 450mm wafers because their use of 300mm wafers in high-mix, low-volume device production is still not that efficient. SPE and material manufacturers have been less than enthusiastic The response from the SPE and semiconductor material sectors has also been muted. Some manufacturers with small market shares are enthusiastic about the 450mm shift because they spy an opportunity to raise their profiles within the industry, but leading manufacturers have generally reacted negatively to the shift. We think their stance reflects the likelihood of very large expenditure being required to establish 450mm wafer technology without any guarantee of recouping it. We note more differences than similarities regarding the impact of reductions in process size and increases in wafer size. These differences account for SPE manufacturers’ dislike of increases in wafer size. Who pays the development costs? First, the distribution of costs is different. Most development costs for technologies aimed at reducing process sizes are borne by semiconductor manufacturers, so the cost burden has tended to be limited for SPE manufacturers. Depending on the node, there may be no change in SPE specifications, so previous-generation SPE can continue to be used. The cost of developing technologies for increasing wafer size, however, is borne largely by SPE manufacturers. The improvement in processing uniformity resulting from an increase in chamber size and the development of new SPE platforms, for example, have been the responsibility of SPE manufacturers. Sustained development program for reducing process sizes but one-time programs for increase in wafer size Second is the continuity of technology development. There has been continuity in process size reduction with changes occurring over three-year cycles. Both chipmakers and SPE manufacturers have ongoing technology development programs for reducing process sizes and production lines must be upgraded for each new generation, creating considerable demand for SPE. Each time a milestone is reached in reducing process sizes, the amount of value-added in SPE increases and leading manufacturers are able to expand their market shares. For both chipmakers and SPE manufacturers, reducing process sizes is an ongoing process that serves as both a sustainable growth driver and a business opportunity. Difference between ongoing operations that are clearly needed and one-time efforts of unclear utility Increases in wafer size tend to occur once every 10 years, resulting in extended periods during which there is no technology development program under way. For the SPE industry, development of process size reduction technologies is an ongoing process with well-recognized benefits. The benefits of increases in wafer size are less evident and, because technology development is periodic rather than ongoing, SPE manufacturers are reluctant to embark on new development programs without a push to support change.
75
Nomura | U.S. Semiconductors Primer December 11, 2013
Materials costs rise Third is the increase in material costs. The price of silicon wafers rose more than 5x during the shift to 300mm wafers. However, process size reductions do not, in essence, cause increases in silicon wafer prices. The big increase in the price of 300mm silicon wafers affected the rest of the value chain and thus, in order to achieve sizeable cost savings from the increase in wafer size, chip manufacturers attempted to hold down SPE prices to compensate for the rise in material costs. Smaller increases in SPE prices Fourth is the modest increase in SPE prices. Each new generation of finer process sizes comes around once every three years, doubling the number of devices that can be obtained from each wafer. While the amount by which SPE prices rise has not been consistent, the general trend when it comes to reductions in process sizes has been that SPE has risen in price by around 20% whenever value-added has doubled. The shift to 300mm wafers enabled chip manufacturers to produce 2.4x the number of devices from each wafer. Initial guidelines set down a 30% increase in SPE prices, but the final hike was around 10% as a result of pricing pressure from chip manufacturers. Ultimately, therefore, when considering the increase in the number of devices obtained from each wafer, the rise in SPE prices stemming from an increase in wafer size was somewhat smaller than that stemming from a reduction in process sizes, although the difference was not large. However, one problem was that progress in reducing process sizes came at the same time as the introduction of SPE for 300mm wafers. SPE price rises accompanying the increase in wafer size were combined with those for process size reduction, leaving the ultimate mix unclear. Increase in the minimum necessary investment caused chip manufacturers to withdraw from the market Fifth, the shift to 300mm wafers forced many companies out of the business because of an increase in the minimum necessary investment in semiconductor production lines. The minimum investment required in 2002 to ensure the lowest necessary level of productivity for a 300mm fab was ¥160bn for front-end processes alone. This was 2.0–2.5x the level for 200mm wafers. Smaller chipmakers were forced out of the market as a result. The minimum investment for a 450mm fab is estimated at around ¥400bn. It is only natural that chip manufacturers other than Intel, Samsung Electronics, and TSMC would hesitate to invest. In the 1990s, the increase in the minimum investment required for each new generation of process size was just over 15%. Unlike process size reduction, however, increases in wafer size have the disadvantage of giving buyers very much the upper hand in the SPE market. Chips from large wafers have no more value-added Sixth, chips from larger wafers have no more value-added than those from smaller wafers. In addition to bringing down costs, finer process sizes create value-added by boosting device processing speeds and levels of integration. Increases in wafer size, however, do not boost a device's value-added. Increases in the value-added of devices broaden the semiconductor market, thus driving market growth for both devices and SPE.
What Would Trigger Progress in 450mm Development? At present, however, SPE makers have become less rigid about the shift to 450mm and are beginning to see it as a given. The two main problems for which we see no solutions are points (2) and (6) above, but these two points are advantages of process size reductions, without being disadvantages of wafer size increases. Development of 450mm technologies would proceed if other advantages are found, one of which would be upgrade demand for SPE. Increase in the minimum required investment is less of a problem due to increasing oligopolization Point (5)—the increase in the minimum investment required—is less of a problem due to the increasing oligopolization of the semiconductor industry. Only TSMC and Intel have embarked on major semiconductor capex programs at present. The semiconductor industry is already increasingly oligopolized, even before the shift to 450mm wafers.
76
Nomura | U.S. Semiconductors Primer December 11, 2013
Outlook unclear for increases in wafer prices The point above can be seen as a problem of balance within the value chain, in our opinion. The shift to 300mm wafers undermined profit growth in the SPE market but enabled silicon wafer manufacturers to book substantial profits. It will be necessary to carefully watch price rises for parts and materials during the shift to 450mm wafers. We believe that both the semiconductor and SPE industries are aware of this potential problem, and although wafer makers appear superficially reluctant to move to 450mm, discussions are ongoing and semiconductor makers could make greater concessions on prices than during the shift to 300mm wafers in order to encourage wafer makers to move to the larger size. At this point, no particular guidelines for wafer and FOUP or other SPE prices have yet been established. Development costs ought to be borne by chip manufacturers or consortiums established by chip manufacturers Once solutions acceptable to SPE manufacturers have been found for the problems of who bears the cost burden for technology development and the amount by which SPE prices rise, we think that the SPE industry will be much more willing to cooperate in the shift to 450mm wafers. Intel decided to take a stake in ASML and support its R&D activities The advance toward 450mm wafers picked up suddenly on 10 July 2012, when Intel announced that it had decided to take a stake in ASML and provide financial support for its R&D activities. Intel currently has roughly 15% stake in ASML. There will be no dilution because Intel will be purchasing treasury stock. Intel will mainly be supporting the development of production systems based on 450mm wafers, as well as EUVL. Dawn of a new 450mm era ASML's decision to start developing lithography systems based on 450mm wafers has forced rivals reluctant to make the move in the absence of compatible lithography systems to follow suit. Nikon, with financial assistance from Intel and others, is also developing lithography systems. It has received orders for five ArF immersion lithography systems for 450mm wafers, while ASML has received orders for four such systems and two EUVL systems. Boost for Nikon Nikon's decision to concentrate its resources on developing lithography systems for 450mm wafers after having lost its share of the market for advanced lithography systems and discontinuing development of EUVL systems has proved to be the right one. It now has an opportunity to get its own back on ASML in the 450mm lithography system stakes. Schedule for 450mm With the exception of lithography systems, prototypes of all the main SPE systems for 450mm wafers are scheduled to be ready by end-2014. ASML plans to come out with the mass production variant of its lithography system in 2018. In response, Nikon plans to be ready with its prototype in 2015–16 and its mass production model in 2017. As such, we look for demand for SPE to pick up in 2017 once lithography systems have come out and benefits have emerged from the shift to 450mm wafers. However, as of mid-2013, we think some issues (e.g., with production lead times) had already emerged at an early stage.
77
Nomura | U.S. Semiconductors Primer December 11, 2013
Microprocessors (MPUs)
Overview A microprocessor or CPU is an integrated semiconductor circuit that functions as the brain of devices such as PCs, tablets, servers, and others. It is a multipurpose, programmable device that accepts digital data as input, executes instructions stored in its memory, and provides results as output. General-purpose microprocessors in electronic devices are used for computation, running applications, multimedia display, and communication over the Internet. Microprocessors are also used in embedded applications such as cellular phones, networking equipment, automobiles, and industrial process control. A microprocessor consists of an arithmetic logic unit (ALU) that performs calculations and makes logical decisions, registers that store temporary information, a control unit that interprets instruction sets, and interconnect buses that carry data throughout the chip (Fig. 1). More complex microprocessors such as the current generation of x86 chips contain on-chip cache memory to speed up access to the external memory.
Fig. 1: Basic diagram of microprocessor
Source: MIT.edu, Nomura research
Brief History of Microprocessors Intel developed the first commercially available microprocessor, 4004, in 1971. The 4-bit Intel 4004 was originally developed for a calculator. The first 8-bit microprocessor, Intel 8008, was developed in 1972 and was used in computers. The first truly general- purpose 8-bit microprocessor, Intel 8080, was developed in 1974. The more powerful 32-bit microprocessor, with which we are more familiar, was introduced in 1989. Even more powerful 64-bit microprocessors became available in 1992 and are now in mainstream use. Intel 4004 contained only 2,300 transistors. The first general-purpose microprocessor, Intel 8080, contained 4,500 transistors and could execute 2 million instructions per second. By 1989, 32-bit microprocessors contained significantly more transistors (1.2mn) and were capable of executing 20 million instructions per second. By 1990, the number of transistors on microprocessors doubled nearly every 18 months (Fig. 2). The increase in the number of transistors in microprocessors followed an early prediction made by American semiconductor pioneer Gordon Moore, who predicted that the number of transistors on a computer chip would double every two years. This prediction has come to be known as Moore’s Law.
78
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 2: Evolution of microprocessors – transistor count doubling every two years
Source: Intel, Embedded.com, Nomura research
Fig. 3: Moore’s Law in play – comparing Intel 4004 in 1971 to Ivy Bridge in 2012
4004 (1971) Ivy Bridge (2012)
4,000x Faster 5,000x Less Energy / Transistor
50,000x Cheaper / Transistor Source: Intel, Nomura research
79
Nomura | U.S. Semiconductors Primer December 11, 2013
Types of Processors General-purpose processors (GPPs) These processors are typically used in PCs, laptops, servers, and other general-purpose computing systems. Until 2010, processors based on x86 micro-architecture almost completely dominated this category. The introduction of the iPad in 2010 and the subsequent surge in Android tablets have made ARM processors, an alternative architecture to x86, a meaningful contributor to the general-purpose computing space. Almost all tablets shipped in 2013 used an ARM processor. Despite increased adoption of tablets, x86 micro-architecture-based processors continue to dominate this segment at 77% of unit shipments (including tablets). Gartner expects ARM’s share in GPPs to be 21% in 2013. Prior to the introduction of tablets in 2010, x86 accounted for almost all of the processor shipments for general-purpose computing. The x86 architecture was first used for the 16-bit Intel 8086 CPU in 1978. Intel’s 8086 processor allowed assembly language programs written for prior Intel processors to be translated into equivalent 8086 programs. This was significant, as it enabled software migration from older designs into new ones. This, among other reasons such as simpler memory interface, contributed to IBM’s adoption of Intel’s 8088 for its PC instead of 16-bit microprocessors from Motorola, Zilog, and National Semiconductor. The resulting IBM PC became the de facto standard for personal computers, enabling the 8088 and its successors to dominate this large part of the microprocessor market. x86-based processors currently constitute around 95% of the microprocessor market by revenue. The x86 processor market is largely a duopoly between Intel and AMD, with Intel being the dominant player at 82% unit share followed by AMD at 18% share in 2013.
Fig. 4: Vendor unit share in GPPs, 2012
23%
13% 64%
Intel AMD ARM vendors
Source: Mercury Research, Gartner, Nomura research
Fig. 5: Architecture share by unit in GPPs, 2012
23%
77%
PCs, Workstations, Servers (x86) Tablets (ARM)
Source: Mercury Research, Gartner, Nomura research
8 0
Nomura | U.S. Semiconductors Primer December 11, 2013
Embedded processors These processors are used in embedded systems as opposed to general-purpose computer systems. They are usually smaller, use a surface-mount form factor, and consume less power. An embedded system is designed for specific functions within a larger system, often with real-time computing constraints. Embedded systems control many devices in common use today, such as consumer electronics, industrial applications, routers, and switches. The key characteristic of embedded processors is that they handle a dedicated task in a more optimized fashion than a general-purpose microprocessor. In 2012, embedded microprocessor revenue was around $3.5bn. Embedded processors used to be largely non-x86 processors (mostly MIPS and ARM processors); however, over time, x86 processors have gained a significant revenue share of the embedded market. In 2012, x86-based processors held 67% share of the embedded processor market. Key vendors in the non-x86 embedded processors include Freescale, Cavium, Broadcom, Applied Micro, and Renesas.
Fig. 6: Vendor share in embedded CPUs, 2012
1% 1% 1%
3% 4% 4% Intel Freescale Semiconductor 5% Cavium Networks
Broadcom
AMD 50% Applied Micro Circuits
Renesas Electronics
31% Toshiba
IBM Microelectronics
Source: Gartner, Nomura research
Fig. 7: Architecture type in embedded CPUs, 2012
MIPS/Other, 33%
x86, 67%
Source: Gartner, Nomura research
81
Nomura | U.S. Semiconductors Primer December 11, 2013
Microprocessor Market Size SIA estimates total microprocessor revenue in 2013 to be $41bn, with microprocessor (MPU) revenue accounting for 13.5% of the total estimated semiconductor revenue of $304bn in 2013. MPU revenue as percentage of total semi revenue has been in the 14– 16% range in the last 10 years. From 2004 to 2013E, overall microprocessor revenue grew at a 4% CAGR. x86 processor unit volume grew at a 7% CAGR from 202mn to 366mn – units increased each year from 2004 to 2011, but this trend reversed in 2012, when units declined 6%. SIA estimates that microprocessor revenues will be flat in 2013 versus the 10-year YoY average growth rate of 7%. We believe that the decline is largely driven by the cannibalization of PCs from lower-priced and more portable tablets. Poor uptake of Microsoft’s newest operating system, Windows 8, also contributed to the decline in x86 processor units. Most PCs running Windows operating systems are based on Intel’s or AMD’s x86 processors.
Fig. 8: MPUs account for 13% of total Semi revs, 2013E Fig. 9: MPUs as percentage of total Semi revs, 2004–2013E
100% 13.5% 90% 80% 70% 60% 50% 40% 30% 20% 10% 86.5% 0% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
MPU Total Semi Revenue MPU Rev Total Semi Rev
Source: SIA, Mercury Research, Nomura research Source: SIA, Mercury Research, Nomura research
Fig. 10: MPU revenue and growth trends, 2004–2013E Fig. 11: MPU unit shipments and growth trends, 2004–2013E Revenue in $bn Units in billion 50 30% 450 20%
45 25% 400 15% 40 350 20% 35 300 10% 30 15% 250 25 10% 5% 200 20 5% 150 0% 15 0% 10 100 -5% 5 -5% 50 0 -10% 0 -10% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
MPU revenue ($bn) MPU revenue growth (y/y) MPU units (mn) MPU unit growth (y/y)
Source: SIA, Nomura research Source: SIA, Nomura research
82
Nomura | U.S. Semiconductors Primer December 11, 2013 x86 Architecture – De Facto Standard in PCs The 8080 processor is generally considered to be the first true general-purpose microprocessor. In 1978, Intel introduced its first 16-bit processor, the 8086, which gave rise to x86 architecture. Intel was not the only company developing microprocessors during that time. Motorola and Zilog were developing CPUs in parallel. Motorola developed an original design of microprocessors (6800) and Zilog developed the Z-80 CPU, which was compatible with Intel 8080 processor. AMD took a similar route to Zilog and began improving the Intel 8080 design. By the 1980s, Intel’s 8086 had become very popular, and efforts by other companies to derail Intel’s momentum were unsuccessful. Motorola, along with IBM this time, tried to stop x86 from becoming a monopoly in PCs in the 1990s by developing the PowerPC processor architecture. By that time, Intel’s 486 and Pentium processors had become widely popular due to Intel’s relationship with Microsoft. While Apple initially adopted PowerPC processors for its line of computers, this was not enough to curb Intel’s growing dominance. Experts have argued that the PowerPC architecture was built from a clean slate and offered superior performance than the 486/Pentium family in the early 1990s. Although PowerPC had an advantage in performance in the early years, Intel’s architectural improvements and rapidly improving manufacturing capabilities eventually overtook PowerPC in performance. In addition, Intel managed to lower the costs by building a better ecosystem around x86 architecture and by partnering with major OS and OEM vendors (Microsoft, IBM, HP). This shift made x86 very accessible to the Open Source community (Linux, BSD) and further lowered software costs. In 2005, Apple switched to x86, making Intel’s processor architecture a de facto monopoly in PCs.
83
Nomura | U.S. Semiconductors Primer December 11, 2013
Intel’s Tick-Tock Model Intel’s Tick-Tock model is at the heart of the company’s manufacturing strategy and, to a certain extent, the industry’s overall manufacturing technology competitiveness. The Tick-Tock model drives Intel’s architecture and silicon refresh cadence. Every ‘tick’ is a shrinking of process technology of the previous micro-architecture, and every ‘tock’ is a new micro-architecture. The result is that Intel has been able to deliver technological innovation on a reliable and predictable timeline, driving a dependable product roadmap for its microprocessors and platforms. Using this model, Intel has successfully delivered next-generation silicon technology and new processor micro-architecture on alternating years for the past several years. Tick (process technology advancement) Tick has fulfilled the prediction of Intel founder and technology visionary Gordon Moore, Moore’s Law. Intel delivers a new silicon process technology every other year. The tick of the process technology advancement dramatically increases transistor density while enhancing performance and energy efficiency within a smaller, more refined version of the micro-architecture. Tock (new micro-architecture) Tock delivers a new micro-architecture on alternating years. During tock, Intel optimizes the value of the increased number of transistors and technology updates enabled by the new process. These architectural advancements not only improve energy efficiency and performance but also increase functional density with features such as graphics performance, hardware-supported video transcoding, encryption/decryption, and other integrated technologies. For instance, Intel is currently shipping 22nm tock parts (code-named Haswell) and expects to launch 14nm tick parts (code-named Broadwell) in 1H14.
Fig. 12: Intel’s tick-tock manufacturing process
Source: Intel, Nomura research
84
Nomura | U.S. Semiconductors Primer December 11, 2013
High-K Metal Gate (HKMG) – Becomes an Important Breakthrough High-K Metal Gate (HKMG) was first used by Intel in 2007. HKMG was an important breakthrough in manufacturing integrated circuits because it enabled further miniaturization of transistors. Prior to HKMG, silicon dioxide (SiO2) was used as a gate dielectric material. It scaled properly up to 90–60nm process nodes. As transistor geometries scaled to the point where the traditional SiO2 gate dielectric became just a few atomic layers thick (Fig. 13), leakage and power dissipation become critical issues. Beyond 60nm, gate thickness requirements mean that the oxide layer is only a few atoms thick, implying that SiO2 could not be scaled further. That meant that the industry needed to find a new approach to insulate the gate to continue scaling. Without a new dielectric material with increased thickness and high-K (refers to the dielectric constant of a material) value, Moore’s Law would have hit a wall. SiO2 replacement was a challenge for the industry, and efforts to replace SiO2 were undertaken for almost 10 years. While newer materials such as hafnium and zirconium oxides looked promising, they were not as compatible with the polysilicon material used for gate electrodes as SiO2 was. The issues with the combination of new materials with polysilicon gates were defects that caused higher threshold voltages, affecting transistor performance. To solve this problem, the semiconductor industry needed to replace the polysilicon gate electrode with new materials. Intel developed a breakthrough that solved this issue. The company validated a new high-K dielectric material along with a replacement for polysilicon gate electrodes (Fig. 14). The result significantly reduced gate leakage by over 100 times. Intel successfully used this HKMG approach in the 45nm node in 2007. Most foundries today (Intel, TSMC, Samsung, Global Foundries) use some form of HKMG approach along with their own manufacturing recipe. The new high-K material was instrumental in extending Moore’s Law into the next decade. Many experts, however, believe that the newer high-K material will last for another five to six years of scaling, or two to three generations of manufacturing processes.
Fig. 13: Comparison of traditional SiO2 and HKMG-based structure
Gate Gate
1.2nm SiO2 3.0nm high‐dielectrick
Silicon Substrate Silicon Substrate
Existing 90nm Process A potential high‐processK Capacitance = 1x Capacitance = 1.6x Leakage Current = 1x Leakage Current = 0.01x
Source: Intel, Nomura research
85
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 14: Transistor structure explaining changes to high-K dielectric and gate material
Source: Intel, Nomura research
86
Nomura | U.S. Semiconductors Primer December 11, 2013
3D or Tri-gate Transistors: 10 Years in the Making In 2011, Intel unveiled the world’s first 3D transistor, also known as a Tri-gate transistor. It took the company nearly a decade to develop the 3D transistor from concept to commercial production. Intel’s 22nm Ivy Bridge architecture–based CPUs are the first products using 3D transistors. A 3D transistor looks similar to a planar transistor but with one fundamental difference: instead of having a single planar layer controlling the source channel, the source is surrounded by multiple surfaces by creating a three-dimensional structure that looks like a fin (Fig. 16). The three-sided silicon fin that the gate wraps around creates a much larger surface area to control the gate current. This approach alleviates the leakage issues that exist in conventional planar transistors. As the size of individual transistors decreases, planar transistors suffer from undesirable off- state leakage current, which increases the idle power of integrated circuits. In a 3D transistor, the channel is surrounded by all three sides, providing better control over the channel and allowing for a much lower off-state leakage current. Intel indicated that using 3D transistors at 22nm will provide up to 37% higher performance compared to 32nm planar transistors, or it could provide the same performance as 32nm transistors at less than half the power consumption. 3D transistors have better power performance characteristics because 1) the gate now has better control over the flow of current through the transistor; 2) silicon substrate voltage no longer affects current when the transistor is off; 3) the larger inversion layer area enables more current to flow when the transistor is on, 4) transistor density is not affected by this change; and 5) the number of fins can be varied to increase drive strength and performance. In our view, these benefits enable Intel’s 22nm 3D transistors to burn less power than equivalent 22nm planar transistors.
Fig. 15: Traditional 2D planar transistor Fig. 16: Intel’s 22nm 3D tri-gate transistor
Source: Intel, Nomura research Source: Intel, Nomura research
3D transistors will likely extend Moore’s Law for another few years at Intel After crossing the first hurdle to Moore’s Law with HKMG at 32nm, many feared that the scaling would hit a wall again at around 28/20nm. In that regard, Intel’s 3D transistors are an important breakthrough. 3D transistors will enable Intel to continue the path of scaling down to 10nm. Essentially, commercialization of 3D transistors is likely to extend Moore’s Law by another three to four years. Intel’s process technology roadmap suggests a path to 14nm by 2015–16. While this may pave the path for Intel for another two generations of process node transitions (14nm and 10nm), we think other foundries such as TSMC, Samsung, and Global Foundries could take a few years longer to get to the same position as Intel. Indeed, Intel has a two- to three-year lead in manufacturing 3D transistors. We think other foundries have to address two issues simultaneously: 1) develop the 3D transistor backend and library, and 2) improvise on the process to be able to follow Moore’s Law. TSMC is taking a two-step approach to tackle these two challenges, first going to planar the 20nm process and then moving to 16nm by just moving to FinFET transistors without the benefit of scaling. Overall, we think Intel may have cleared its own path to 10nm, but the same is not true for other foundries at this time.
87
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 17: Intel’s process technology roadmap
Source: Intel, Nomura research
First-generation 3D may not provide the same benefit to other foundries TSMC, the second-largest foundry, appears to be encountering bottlenecks in scaling the planar process from 28nm to 20nm. The company is considered the bellwether of the chip sector because it is one of the world’s largest and most advanced foundries of semiconductor chips. Many of TSMC’s customers, such as Nvidia and Broadcom, have indicated that there are no scaling benefits from moving to 20nm because the cost of the transistor is staying the same (Fig. 18). We think the key reason for this is that there are no scaling benefits when moving from planar 28nm to planar 20nm process. In addition, 20nm planar node appears to have higher mask and wafer costs than the 28nm node. Many of TSMC’s key customers have indicated that scaling and cost benefits typical to such transitions may not be realized when moving from 28nm from 20nm. This implication provides an opportunity for Intel to catch up with ARM SoC vendors in power and performance efficiency to a certain in the 20nm generation. This issue highlights the importance of moving to 3D or Tri-gate transistors to continue the path of scaling. We believe that TSMC could achieve commercial production of its 16nm FinFET transistors in 2015. Even with 16nm FinFET, extension of Moore’s Law or doubling transistor density is not guaranteed at TSMC. It appears that TSMC will be adding FinFETs to its 20nm process, providing almost no benefit in scaling transistors per area of die, although it would likely offer lower power and higher performance. However, TSMC sees a path to 10nm by 2016, enabled by FinFET transistors.
Fig. 18: TSMC’s normalized transistor cost based on manufacturing process nodes
1.2
1
0.8
0.6
0.4
Normalized Transistor Cost 0.2
0 2005 2007 2008 2009 2011 2012 2013 2015 2016 2017 80nm 55nm 40nm 28nm 20nm 14nm
Source: Nvidia, Nomura research
88
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 19: Intel’s technology R&D pipeline – another 10 years of Moore’s Law in sight
2011 2013 2015+
22nm 14nm 10nm 7nm 5nm
In Production In Development In Research Lithography, Materials, Interconnect...and more
Source: Intel, Nomura research
89
Nomura | U.S. Semiconductors Primer December 11, 2013
Capital Intensity Needed to Follow Moore’s Law Is Increasing Increasing capital intensity of successive nodes could affect fabless companies Capital spending for each successive node is increasing. For example, 200mm fabs typically require $3–5bn in revenue to justify the capex investment. According to Intel, the revenue threshold increases to $9–12bn for 300mm fabs and to $15bn-plus for leading-edge nodes beyond 2015. While Intel, which has $50bn-plus in revenue in 2012, can support this level of spending, we believe that this could leave many other fabless semiconductor companies (e.g., Qualcomm, Apple, TI, Broadcom, Nvidia, Altera, and Xilinx) at a disadvantage if this issue affects technology transitions of key foundries such as TSMC and Samsung. While we think TSMC has the scale to support the investments needed to make these transitions, the pace of these transitions could slow down from increasing capital intensity of advanced nodes. This could play out in several ways: It could slow down the pace of performance and feature improvements in semiconductor chips. It could also result in fabless companies indirectly or directly supporting the capex of their respective foundry companies, affecting the profitability of fabless companies.
Fig. 20: Beyond 2015, we foresee fewer companies staying on Moore’s law Semiconductor revenue required to support one leading-edge fab
Intel Samsung TSMC Toshiba Texas Instruments Renesas ST Micro Qualcomm Hynix Micron Broadcom AMD Sony Infeneon Fujitsu Freescale NXP Elpida Nvidia UMC 0 5 10 15 20 25 30 35 40 45 50 Revenue Required ($Bn)
Source: IC Insights, Intel, Nomura research
Intel’s capex has increased significantly to maintain its manufacturing lead Until 2010, Intel spent, on average, $5bn per year on capital expenditures. This trend changed dramatically since 2011, when Intel’s capex increased more than two times its historical average spending. The increased capex primarily supported the transition to 22nm. Intel recently guided capex for 2013 to be at an all-time high of $11bn, primarily driven by 14nm costs and by $1bn in spending for 450mm. Intel expects to use 450mm wafers in the later part of this decade to lower transistor costs. In addition to the increase in capex, Intel is investing in other semiconductor production equipment companies such as ASML to drive transistor costs further down. Capex is increasing despite PC Client Group (PCG) revenues declining since 2010, Intel’s capex has more than doubled from $5bn in 2010 to estimated $11bn in 2013 (Figs. 21 and 22). If we look at cumulative capex, Intel is on track to spend $33bn from 2011 to 2013, a figure equivalent to the combined capex that Intel spent in the seven years prior to 2011.
90
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 21: Intel’s capex has increased significantly in the last few years in $mn
$14,000
$12,000
$10,000
$8,000
$6,000
$4,000
$2,000
$0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Source: Intel, Nomura research
Fig. 22: Intel’s MPU unit shipments and capex spend trends, 2004–2013E
3.5x
3.0x
2.5x
2.0x
1.5x
1.0x
0.5x
0.0x 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Intel Capex (indexed to year 2004) Intel MPU units (indexed to year 2004)
Source: Intel, Mercury Research, Nomura research
91
Nomura | U.S. Semiconductors Primer December 11, 2013
End Markets Personal computing shifting to low-power alternatives The key end-market products for microprocessors include servers, desktops, notebooks, and tablets. Notebooks are the largest end market, but Gartner expects tablets to become the largest segment by the end of 2013 (Fig. 23). According to Gartner, tablets and notebooks will account for 42% and 32%, respectively, in 2013, while tablets accounted for only 5% in 2010. In nominal terms, Gartner expects tablets to total 229mn units, while notebooks, desktops, and servers will contribute 179mn, 133mn, and 10mn units, respectively. Since 2010, tablets and servers will be up 129% and 4% YoY, while notebooks and desktops will be down 4% and 5% YoY.
Fig. 23: Units by end market
700
600
500 Units (in mn) (in Units 400
300
200
100
0 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E
Server Desktop Notebook Tablet
Source: Gartner, IDC, Nomura estimates
Fig. 24: Unit growth by end market
40% 350%
300% 30% 250%
200% 20% 150%
10% 100%
50% Y/Y Grow th(Tablet)
Y/Y Grow th (ex. Tablet) 0% 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E 0%
-50% -10% -100%
-20% -150%
Server Desktop Notebook Tablet
Source: Gartner, IDC, Nomura estimates
92
Nomura | U.S. Semiconductors Primer December 11, 2013
MPU unit forecast: unit decline for a second consecutive year The three key end markets for x86 processors are notebooks, desktops, and servers. Among these, notebooks are the largest segment and will account for approximately 53% of unit volume in 2013 (per Mercury Research). Desktops and servers will account for 41% and 5% of unit share in 2013, respectively. For 2013, Mercury Research expects x86 processor shipments to total 366mn, with notebooks, desktops, and servers contributing 195mn, 143mn, and 19mn units, respectively. On a YoY basis, notebook, desktop, and server unit shipments will be down 6%, down 9%, and up 8%, respectively. The notebook segment is also the largest by revenue share. In 2013, the notebook segment will account for 42% ($17bn) of microprocessor share by revenues. Desktop and servers will account for 32% ($13bn) and 26% ($11bn) of revenue share, respectively.
Fig. 25: x86 unit shipment trend by end market, 2004–2013E Fig. 26: x86 unit growth trend by end market, 2004–2013E
450 80%
400 70%
350 60%
300 50%
250 40% 30% 200 20% 150 MPU UnitsMPU (in mn) 10% 100 0% 50 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E -10% - 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E -20%
Server Desktop Notebook Server Desktop Notebook Total
Source: Mercury Research, Nomura research Source: Mercury Research, Nomura research
Fig. 27: x86 processor end-market share by units, 2013E Fig. 28: x86 processor end-market share by revenue, 2013E
5%
26%
42%
41% 54%
32%
Notebooks Desktops Servers Notebooks Desktops Servers
Source: Mercury Research, Nomura research Source: Mercury Research, Nomura research
93
Nomura | U.S. Semiconductors Primer December 11, 2013
Market share: more one-sided than ever The x86-based microprocessor market is essentially a duopoly between Intel and AMD. VIA Technologies, the third company licensed to manufacture x86 processors, has less than 1% market share. Between Intel and AMD, Intel dominates the x86 market in every way. Intel’s x86 processors will account for 82% of total units and approximately 94% of revenue share in 2013, according to Mercury Research. We attribute Intel’s strong market position to its solid track record of product and market execution and to its manufacturing scale advantage, all of which allow for larger investments in R&D and manufacturing process technologies than its competitors. In 2013, AMD’s unit share in processors was at an eight-year low, due to the company’s poor execution on it product roadmap and to its disadvantage on scale and manufacturing process technologies relative to Intel. After reaching a peak unit share of 23% in 2006, AMD has steadily lost share to Intel; Mercury Research estimates that the company will end 2013 with unit share of 18%, a 66bp improvement from 2012. Intel dominates the key end markets of notebooks, desktops, and servers. By the end of 2013, Mercury Research expects Intel to have an 86% share in notebooks, 80% share in desktops, and 97% in servers.
Fig. 29: x86 processor unit market share trend, 2004–2013E
120%
100%
80%
60%
40% M PU Units(in m n)
20%
0% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Intel AMD VIA
Source: Mercury Research, Nomura research
Fig. 30: Market share in notebook, desktop, and server segments, 2013E
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0% Notebook Desktop Serv er
Intel AMD
Source: Mercury Research, Nomura research
9 4
Nomura | U.S. Semiconductors Primer December 11, 2013
ASPs: trends have been favorable According to Mercury Research estimates, Blended ASP for x86-based microprocessors will be $112 in 2013, up $2 from 2012. This follows increases of 11% in 2011 and 4% in 2012. This ASP increase is significant, considering that the 10-year average growth is -1%. ASP increases in the last three years were driven by increases in server (1%) and desktop (4%) processor ASPs. A richer mix of server processors in data centers and improved pricing for desktop processors drove the ASP increase in the last three years. AMD’s lack of competitiveness has also contributed to this ASP increase. We believe that this ASP trend will reverse, as competition from lower-priced ARM processors ($15– 20 in ASP) in tablets and increased competition in x86 server processors from 64-bit ARM processors are likely to pressure x86 processor ASPs.
Fig. 31: MPU ASP growth (YoY) trend, 2004–2013E
15%
10%
5%
0%
-5%
-10%
-15% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
ASP change (y/y)
Source: Mercury Research, Nomura research
Fig. 32: MPU ASP trend, 2004–2013E
$160
$140
$120
$100
$80
$60
$40
$20
$0 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
ASP ($)
Source: Mercury Research, Nomura research
95
Nomura | U.S. Semiconductors Primer December 11, 2013
Notebooks Notebooks are the biggest end market for x86 processors Notebook processors will account for 53% of unit shipments and 42% of total revenue share of x86 processors in 2013, per Mercury Research estimates. In the last 10 years, notebook microprocessors have grown at the fastest rate among the three key categories. Notebook processors grew at a CAGR of 21%, from 36mn units in 2004 to 195mn units in 2013E. Notebook shipments will decelerate to -6% annual growth in 2013E versus the 10-year average growth of 24%, per Mercury Research. We believe that the decline in notebook shipments is driven by tablets cannibalizing notebooks. Tablets are proving to be cheaper than, and more portable alternatives to, notebooks for common tasks such as web browsing, email, videos, and photos. We think this could particularly affect notebook growth in emerging markets where consumers are more price-sensitive. Emerging markets account for more than half of all notebook shipments. We expect notebook shipment growth to remain muted for the next few years and grow in the mid- single digits. By contrast, tablet shipments saw a significant growth of 70%-plus in 2011 and 2012 and, according to Gartner, will grow at a 30%-plus for the next few years.
Fig. 33: Notebook processor unit shipment and growth trend, 2004–2013E in million units
250 80%
70%
200 60%
50% 150 40%
30% 100 20%
10% 50
0%
0 -10% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Notebooks (in mn) Unit growth (y/y)
Source: Mercury Research, Nomura research
Correlation between notebook growth and GDP growth appears to be weakening PC shipment growth and global GDP growth historically have been correlated. The correlation was significant at 66% from 2001 to 2010 (Fig. 34). We are using total PC growth to calculate correlation to account for the fact that notebooks have been gaining share from desktops in the last several years. Since the launch of iPad in Q2 2010, the correlation between PC growth and global GDP growth has been waning. This correlation drops to 45% from 2Q10 to 2Q13 (Fig. 34). Prior to 2010, notebooks were the most portable computational devices. Today, there is a variety of devices, including large-screen smartphones, tablets, and notebook alternatives such as Google’s Chromebook. We are not forecasting this correlation to decline further, but we believe that the fragmented landscape will prevent a return to historically higher correlations.
96
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 34: PC growth (YoY) has largely been correlated with global GDP growth
'01- '10 Correlation: 66% '10 - '13 Correlation: 45% 6% 35%
5% 30%
4% 25%
3% 20%
2% 15%
1% 10%
0% 5%
-1% 0% YoY ChangeYoY in Gobal GDP YoY Change in PC Grow th -2% -5%
-3% -10%
-4% -15%
GDP growth (y/y) PC growth ((y/y)
Source: Gartner, Nomura estimates
Fig. 35: Comparison of desktop, notebook, and tablet shipments, 2004–2014E
350
300
250
200
150 Units in mn
100
50
0 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Desktop Notebook Tablet
Source: Gartner, Nomura estimates
Emerging regions have been the key growth drivers for notebooks, but future growth from these regions is less certain In 3Q13, emerging regions accounted for 57% of total shipments, up from 47% in 2007. Emerging regions have also been the biggest contributors to PC growth since 2000. APAC delivered 12% CAGR in PC shipments between 2000 and 2013. Shipments in Latin America grew 11% during the same period. By contrast, shipments grew at a CAGR of 2% in North America and at 0% in Japan. Emerging markets have been crucial to the growth of PCs, and much of Intel’s expectation for future growth from these regions depends on the vision of at least one PC per household. Low PC penetration in emerging markets (~25%) was a secular growth driver for PCs until the launch of iPad in 2010. Since the introduction of iPad, tablets have dramatically enhanced access to computing in emerging regions by lowering price points. Prior to the introduction of tablets, the average cost of notebooks in emerging regions was $500–700. The introduction of iPad followed by tremendous growth in Android tablets has brought price points to as low as
97
Nomura | U.S. Semiconductors Primer December 11, 2013
$99 in emerging regions. This disruption could be a significant impediment to the growth rate of PCs in emerging markets as tablets achieve functional parity with PCs.
Fig. 36: Regional share of PC shipments, 2004–2013E
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Asia/Pacific EMEA Japan Latin America North America
Source: Gartner, Nomura research
Fig. 37: Split of PC shipments between regions
100
90
80
70
60
50
Units inmn 40
30
20
10
0
Developed Emerging
Source: Gartner, Nomura research
98
Nomura | U.S. Semiconductors Primer December 11, 2013
Tablets are pushing notebook ASPs into the $300–500 price band We believe that Intel’s Bay Trail chipsets (Atom-based) could help the company introduce a number of devices in the $300–500 price band. This segment has been showing some growth and has been outgrowing the overall notebook segment. Last year, total notebook shipments were largely flat, but shipments in the $300–500 category grew 23% YoY (Fig. 41). We believe that this segment is showing growth due to price competition from cheaper alternatives, which are priced in the same band as tablets. In addition to pricing pressures from tablets, x86-based notebooks are seeing pressure from ARM-based notebooks such as Google Chrome ($250 ASP). We believe that these trends are forcing OEMs to focus on the $300–500 notebook segment. During IDF 2013, Intel announced that its Bay Trail products will be priced competitively for products ranging from $99 to $399. We believe that more OEMs plan to focus on tablet-like Windows products in this price range. Intel’s new 22nm Bay Trail would enable OEMs to make notebooks in this price range look more appealing. This was not feasible a year ago with Ivy Bridge (expensive and consumed too much power) or Atom (performance too low). However, the lower price points of these devices would mean that OEMs will likely use more of Bay Trail (Atom) or Celeron (22nm) chips in this segment, and less of Core Haswell chips.
Shipments of high-end notebooks ($800+ ASP) are declining Despite the significant power savings associated with Haswell, Intel has not been successful in driving notebook sales. Gartner data suggest that shipments in the $800- plus segment have been declining since 2009 (Fig. 40). While more power-efficient chips could make this segment (convertibles, hybrids) more appealing, we think Haswell is unlikely to reinvigorate this segment for two reasons: 1) A no-compromise convertible device is still very expensive. Popular convertible devices such as Acer Aspire S7 ($1500), ASUS Taichi ($1200), and Dell XPS 12 ($1200) are priced above $1000. 2) We believe that this price point is high, considering the various alternatives now available in the $300–500 range, and it is debatable whether a Windows-based convertible device could provide an optimal tablet experience. For consumers who use tablet functionality for a majority of their time, we think splitting those functionalities into two separate devices (iOS or Android for tablet and Windows or Mac for notebook) could be more cost- effective and provide a better experience. The $800-plus segment accounted for 23% of notebook shipment and around 40% of notebook revenue in 2012. As such, we view this as a negative trend for x86 suppliers, particularly for Intel, which has more than 80% share of Ultrabooks.
Fig. 38: NB shipments breakdown by ASP Fig. 39: Ultrabook shipment and mix forecast, 1Q11–4Q13E
Source: Gartner, Nomura research Source: Nomura estimates
99
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 40: $800+ notebook segment is declining since 2009 … Fig. 41: …while shipments in $300–500 segment are growing
80 45% 60 30%
70 40% 50 25% 35% 60 40 20% 30% 50 25% 30 15% 40 20% 30 20 10% 15%
20 10% 10 5%
10 5% 0 0% 0 0% 2009 2010 2011 2012 2009201020112012 $300-499 units (mn) $300-500 units as % of total notebooks (RHS) $800+ units (mn) $800+ units as % of total notebooks (RHS) Source: Gartner, Nomura estimates Source: Gartner, Nomura estimates
Low-cost CPUs in tablets will likely remain a long-term competitive threat to notebooks In a traditional notebook, a CPU accounts for approximately 20% of the total bill of materials (BOM) versus only 7% in a typical tablet BOM (Figs. 42 and 43). In tablets, a larger proportion of cost is allocated to NAND flash, touch panel, casing, and polymer battery, one of the reasons why the finish and quality of high-end tablets are superior to many notebooks in a similar price range. We believe that this contributes to Intel’s inability to promote Ultrabooks. A no-compromise Ultrabook is fairly expensive ($899- plus) and is similar to the price of Apple’s MacBook Air. The high price point raises consumers’ expectations, who want a flawless device similar to MacBook Air. Our experience is that none of the Ultrabooks in the market today is perfect in all aspects. Issues range from poor touchpads and cheap wireless chipsets to bigger issues such as uncalibrated displays. We believe that consumers are not comfortable buying a compromised device at these higher price points. Lowering Intel’s CPU ASP over time should help PC OEMs find more margin dollars to fix these issues.
Fig. 42: Typical BOM for a tablet – CPU accounts for 7% Fig. 43: Typical BOM for notebook – CPU accounts for 21%
Source: Nomura estimates Source: Nomura estimates
100
Nomura | U.S. Semiconductors Primer December 11, 2013
Windows 8 and 8.1 failing to spur a PC replacement cycle this year Although there is a lot of optimism around Microsoft’s Windows 8.1 OS, we think key changes are more focused on optimizing Windows OS on small displays (7–10”). While Microsoft has made some UI changes, such as bringing back the Start button, we think these incremental changes are unlikely to boost sales to levels seen in prior OS release cycles (Fig. 44). For example, PC sales rebounded 30%-plus in the quarter following the release of Windows 7. The response to Windows 8 has not been good so far. PC sales declined 10% YoY in the quarter after the release of Windows 8 last year. The key issue, we think, is that tablets filled a void for touch-enabled, low-power, and low-cost computing that now also has a mature application ecosystem. Consequently, we believe that ASPs of touch-enabled notebooks with the new OS (priced at $699–999) will be less attractive than those of tablets at $199–549. In addition, we think increasing interest in 7”–8” tablets could squeeze out consumers’ budgets for PCs. We see improvements in tablets to a point where functionality is bridged between PCs and tablets.
Future corporate PC refreshes will likely have a muted impact on PC growth Corporate or commercial PCs account for approximately 45% of all shipments. Traditionally, IT departments refresh PCs every few years. Since the introduction of the bring-your-own-device trend, enterprises have been more cautious about refreshing PCs en masse. If we analyze enterprise PC shipment growth (Fig. 44), we see that enterprise PC growth is fairly correlated with the growth of PCs in the consumer segment. However, the launch of Windows 7 in 3Q09 had a stronger impact on enterprise growth than on consumer growth because Windows 7 was a large update and generated a strong corporate refresh. We think this correlation made sense in prior periods when consumers and enterprises had fewer options when it came to buying a computing device. We expect enterprise PCs and consumer PCs to be less correlated, considering the weakening consumer interest in PCs due to tablets. However, it appears that both are still correlated (Fig. 44), which could be due to factors such as increasing enterprise interest in alternative devices such as tablets or the increasing trend of bring-your-own- device at workplaces. As a result, corporate workplaces are allowing people to use devices with different operating systems (iOS, Android). In addition, Gartner research indicates that most enterprises are refreshing PCs on a need basis. These trends defy the traditional approach of near-automatic software and hardware refreshes every few years. These trends will make PC refreshes less lumpy and noticeable. For example, Microsoft is set to end enterprise support for Windows XP in April 2014, fueling expectations of a healthier PC refresh cycle in 2013. However, we think that this is not meaningfully affecting PC growth, which is set to decline in the low single digits this year. This view is backed by the fact that Microsoft recently indicated that around 75% of all PCs running Windows XP have been upgraded to Windows 7 or higher, but the refresh did not have a noticeable impact on PC growth this year.
101
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 44: Enterprise PC growth (YoY) and major OS launches
Windows Windows Windows 98 2000 Windows XP Vista Windows 7 Windows 8 60%
50%
40%
30%
20%
10%
0% -97 -04 -11 -99 y y y -10% p Jul-98 Jul-05 Jul-12 Oct-03 Apr-00 Oct-96 Oct-10 Apr-07 Jun-01 Jan-02 Jun-08 Jan-09 Mar-03 Mar-96 Feb-99 Feb-06 Mar-10 Feb-13 Nov-00 Aug-02 Dec-97 Dec-04 Dec-11 Se Sep-06 Nov-07 Aug-09 Ma Ma Ma -20%
-30%
Consumer Enterprise
Source: Gartner, Nomura estimates
Gap between 2H and 1H seasonality appears to be shrinking Notebook shipments historically have been stronger in 3Q and 4Q of the calendar year, largely due to universal back-to-school sales in 3Q and promotions during the holiday shopping period in 4Q across most regions. Consequently, notebook sales have been traditionally back-half loaded. Ten-year seasonal average growth for notebooks shipments in 3Q and 4Q are 16% and 12%, respectively. Notebook sales in the second half of the year, which includes back-to-school and holiday season, have grown at an average of 14% sequentially compared with a mid-single-digit decline in 1H (Fig. 45). Since the introduction of tablets in 2010, we have observed two distinct changes in the historical seasonality pattern: 1) the gap between 1H and 2H seasonality appears to be shrinking (Fig. 47), and 2) back-to-school seasonality appears to be on a decline (Fig. 46). 2H seasonality has been historically higher than 1H seasonality by 15–20ppt. This gap has reduced to less than 10ppt since the iPad was first introduced. We think availability of cheaper alternatives to PCs such as tablets and Google Chromebooks are a factor as well. In addition to tablets’ portability, light weight, and long battery life, they bridge functional parity with traditional notebooks by adding features such as the ability to print, a full keyboard, and remote desktop.
102
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 45: Notebook shipment seasonality, 1Q05 to 2Q13 QoQ growth (%), 4Q growth highlighted 35%
30%
25%
20%
15%
10%
5%
0%
-5%
-10%
-15% Q1-2005 Q2-2005 Q3-2005 Q4-2005 Q1-2006 Q2-2006 Q3-2006 Q4-2006 Q1-2007 Q2-2007 Q3-2007 Q4-2007 Q1-2008 Q2-2008 Q3-2008 Q4-2008 Q1-2009 Q2-2009 Q3-2009 Q4-2009 Q1-2010 Q2-2010 Q3-2010 Q4-2010 Q1-2011 Q2-2011 Q3-2011 Q4-2011 Q1-2012 Q2-2012 Q3-2012 Q4-2012 Q1-2013 Q2-2013
Source: Gartner, Nomura estimates
Fig. 46: Back-to-school seasonality is moderating Fig. 47: Gap between 1H and 2H seasonality is shrinking
20% 50%
15% 40% (Total
10% 30%
5% 20% Growth
Growth)
YoY
0% 10%
Unit 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Sequential Growth
School ‐5% 0%
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 To
‐10% -10% Back
‐15% -20%
Back to School Quarter 1H of the Year 2H of the Year
Source: Gartner, Nomura research Source: Gartner, Nomura research
103
Nomura | U.S. Semiconductors Primer December 11, 2013
Market share: highly skewed According to Mercury Research, Intel had 86% share of notebook processor units in 3Q 2013. AMD’s notebook processor share of 14% was toward the low end of the 10-year range of 10–19%. Since the beginning of last year, AMD lost 280bp share to Intel. The key reason for Intel’s significantly higher notebook share is better performance and battery life of Intel’s processors versus AMD’s. Intel processors’ superior battery life is largely a result of the company’s more advanced manufacturing technology. Intel has been shipping 22nm processors since 2012, while AMD has yet to make the 28nm transition at Global Foundries, AMD’s primary foundry for its mainstream processors. Despite being one full process node behind Intel, AMD has been able to maintain its traction, due to its better graphics performance. Both companies have GPUs (graphics processing units) integrated in their CPUs. In addition, AMD historically has been more exposed to consumers and less to enterprise. As consumers continue to adopt ARM- based tablets, we think that this will increase pressure on AMD. Furthermore, Intel’s expected transition to 22nm Atom processors in 2H 2013 could make Intel more competitive at the low-end market against AMD. Intel’s new Atom chip is likely to have three times improvement in graphics performance over the previous generation of Atom chips. In the mainstream notebook processor segment, we see the process gap continuing to increase between Intel and AMD. We expect Intel to move to 14nm process in 1H14. In contrast, we expect AMD to transition to 28nm process node in 1H 2014 at Global Foundries. We believe that AMD could see further share loss in the notebook segment in upcoming periods.
Fig. 48: Notebook microprocessor unit share, 2004–2013E
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0% 1Q04 2Q04 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05 1Q06 2Q06 3Q06 4Q06 1Q07 2Q07 3Q07 4Q07 1Q08 2Q08 3Q08 4Q08 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13E
Intel AMD Transmeta
Source: Mercury Research, Nomura research
104
Nomura | U.S. Semiconductors Primer December 11, 2013
Tablets While Intel has made progress, ARM stays dominant in mobile devices ARM chips dominate the mobile device landscape. According to Gartner, ARM powers more than 95% of the smartphones and has around 90% share in mobile computing. The low-power feature of ARM chips and its licensing model have made ARM a dominant architecture in mobile devices. The chip’s architecture is licensed to more than 200 companies, including Qualcomm, Nvidia, Broadcom, Marvell, and Texas Instruments. Until 2011, Intel did not have a good offering for mobile devices. Intel’s Atom chip in 2011 consumed 5–10 times more power than ARM’s SoC. Intel’s first full-fledged effort in mobile was a 32nm Medfield SoC, which the company launched in 2012. While Medfield was used in Motorola Razr I and Lenovo K100 smartphones, it had limited success because it was a discrete application processor and did not have the performance or power efficiency to match Qualcomm and other providers. Intel’s follow-up to Medfield, dual-core Clover Trail and Clover Trail + SoCs, were also 32nm chips with improved performance and battery life. Despite these improvements, Intel’s traction in smartphones was limited to tier-2 and tier-3 OEMs (Lenovo, Lava, Megafone). However, Intel made better progress with tablets by winning sockets in Samsung’s Galaxy Tab 10.1. Despite these efforts, Intel’s current share in mobile device is less than 5%.
Manufacturing leadership favors Intel, but it is not enough Intel announced 22nm quad-core Bay Trail SoCs to target tablets and smartphones in 2013. Intel significantly improved performance of chips that target smartphones and tablets. Benchmarks suggest that quad-core Bay Trail has better CPU performance than the current-generation 28nm ARM SoCs from Qualcomm, Nvidia, and Samsung. In addition, Intel claims a much improved battery life. We believe that Intel’s 22nm process technology is helping offset some of the inherent low-power advantages in ARM chips, but that’s just one factor. Most OEMs will likely rely on a multisourced strategy. In addition, iOS and Android operating systems, which have a market share of 90% of mobile devices, are closer to ARM architecture. We forecast 48% share for iOS, 41% share for Android, and a mere 10% share for Microsoft in tablets in 2013 (Fig. 49). x86 CPUs are mostly tied to Microsoft’s Windows platform, which has a very low share in tablets. While Intel’s recent Android tablet (Tab 3 10.1”) design win at Samsung is positive, we think it still has a long way to go to gain meaningful share in Android tablets. Apple, which has dominant share (40–45%) of the tablet market, is vertically integrated, and Samsung, while also vertically integrated, uses multiple chipset providers (Qualcomm, Marvell). The remaining tablet market is dominated by ARM processor companies such as Qualcomm, Nvidia, Samsung, MediaTek, Marvell, and Broadcom.
Fig. 49: Tablet mix by OS, 2013E
1% 10%
48%
41%
iOS Android Microsoft Others
Source: Nomura estimates
105
Nomura | U.S. Semiconductors Primer December 11, 2013
ARM processors are likely to maintain lead in mobile devices at least through 2015 While Intel is likely to bridge the power consumption gap with ARM processors with its 22nm Bay Trail in the second half of 2013 and with its 14nm Broadwell chips next year, we think ARM will continue to dominate mobile devices. A much broader ecosystem of ARM suppliers, IPs, and peripheral support makes ARM chips for smartphones a lower- risk proposition for most handset OEMs. In addition, both Android and Windows Phones are native to ARM architecture, which makes it easier for OEMs to stay with ARM chips to support the latest applications and releases in the respective operating systems. That said, we think Intel will gain share in tablets, where 22nm Bay Trail looks very competitive to equivalent ARM chips. Due to Intel’s advantage in 14nm, we believe that Intel could have an edge on 14nm SoCs, although 14nm SoCs will likely not be available until 2H15. Intel’s mobile SoCs currently lag the leading edge by more than a year. As such, during the next six to eight quarters, we do not see the scales in mobile tipping in favor of Intel.
Fig. 50: Power efficiency comparison between ARM and Intel chips, 2012-17E Relative performance/watt for ARM vs. Intel processors
Source: ARM, Intel, AnandTech, Nomura research
106
Nomura | U.S. Semiconductors Primer December 11, 2013
Desktops Desktops appear to be driven by demand from emerging markets and by replacement trends in developed markets Gartner expects desktop shipments to account for 33% of total PC shipments in 2013, down from 83% in 2000. Desktops have grown at a 2% CAGR since 2000 versus notebooks at 22%. Desktop shipments are projected to decline 7% in 2013, well below the 10-year average of 2% growth. We believe the decline is driven by increased uptake of notebooks and tablets and by consumers and businesses either delaying upgrades or purchases of desktops. Desktops have been declining against notebooks in the last several years. Furthermore, increasing adoption of cloud-hosted services such as online storage for pictures and videos is decreasing the utility of desktops for consumers who already have an Internet-connected notebook or tablet. Desktop volume is primarily driven by developing economies, due to the significant price differences between desktops and notebooks. Notebooks in emerging markets cost, on average, $250–300 more than desktops. In 2013, emerging markets will account for almost 61% of all desktop shipments, according to Gartner. We think that the availability of cheaper tablets in emerging markets could drive a multiyear decline in desktop shipments. In enterprises, PC desktop refreshes are being affected on two fronts: First, many organizations are buying low-cost tablets for employees for front-end content consumption. Second, many IT departments are using desktop virtualization instead of upgrading the desktop hardware. Virtualized desktops are much easier to manage, upgrade, and secure than physical desktops. On the other hand, desk-based PCs are still a popular form factor choice for many organizations. Desktops are also a preferred form factor in the professional market, which has experienced consistent growth. While we do not see many growth drivers for desktops, we think the replacement cycle of four to five years in the installed base of desktops could sustain flat growth for the next several years.
Fig. 51: Desktop and notebook shipments, 2000–2014E In thousands
250,000
200,000
150,000
100,000
50,000
0 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Desktops Notebooks
Source: Gartner, Nomura estimates
107
Nomura | U.S. Semiconductors Primer December 11, 2013
Market share: AMD has bigger slice of desktops At the end of 3Q13, according to Mercury Research, Intel had 79% share of desktop processor units and AMD’s shipments accounted for 19% of units. AMD’s current desktop processor share of 19% is toward the low end of the 10-year range of 16–29%. Two key factors that contributed to AMD’s recent loss in desktop share are lower performance than Intel’s CPUs and poor execution in the 32nm node at Global Foundries. In 2012, AMD was severely supply-constrained at 32nm, which contributed 330bp share loss in desktops to Intel. Historically, AMD has had a larger share, as battery life is less relevant in desktops and lower price is usually the key driver of sales. The 24% YoY decline in desktop volumes in 2013 affected AMD more than it did Intel. In addition, due to widening manufacturing process disadvantage with Intel, AMD’s desktop processor performance has started to lag more noticeably to equivalent Intel parts. Intel’s next- generation mainstream processor, Haswell, shipped in 2H13 and is likely to increase pressure on AMD’s desktop business. Haswell is 20–30% more power efficient than Intel’s last-generation processors (Ivy Bridge) and has significant (two to three times) improvement in graphics performance. We expect AMD to offer 28nm desktop chips for a large part of 2014 against Intel’s 22/14nm chips. In conclusion, we believe that AMD could experience more share loses in desktops, as Intel’s competitive lead could widen when Intel begins 14nm production in 2014.
Fig. 52: Desktop microprocessor unit share, 1Q04–3Q13
100%
80%
60%
Total Desktop MPU share MPU Desktop Total 40%
20%
0% 1Q04 2Q04 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05 1Q06 2Q06 3Q06 4Q06 1Q07 2Q07 3Q07 4Q07 1Q08 2Q08 3Q08 4Q08 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13
Intel AMD VIA
Source: Mercury Research, Nomura research
108
Nomura | U.S. Semiconductors Primer December 11, 2013
Servers Servers bring a disproportionate amount of revenue relative to unit volume According to Mercury Research, server processors will account for 5% of unit shipments and 27% of total revenue share of x86 processors in 2013. x86 processors for servers have grown at a 10% CAGR in the last 10 years and have more than doubled in volume from 8mn units in 2004 to 19mn in 2013E. Mercury expects server processor shipments to increase 8% in 2013, below the 10-year average of 16%. Servers remain a secular growth story for x86 processors. Revenues from x86 server chips will be up 7%, totaling $11bn, per Mercury Research estimates. We believe that, in a weak macro environment, share shifted to more cost-efficient x86 platforms, explaining the delta between revenue growth and unit growth. Mercury’s outlook for 2014 suggests 5% unit growth, but Intel appears to be more bullish for its server business and is guiding for high-single-digit growth in 2014. In addition, Intel expects to double its server business in the next four to five years from approximately $12bn in annual revenue currently.
Fig. 53: Server processor unit shipment and growth trend, 2004–2013E Units in millions
25 50%
40% 20
30%
15 20%
10% 10
0%
5 -10%
0 -20% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Servers (in mn) Unit growth (y/y)
Source: Mercury Research, Nomura research
Fig. 54: Server shipment and growth trends, 2010–2016E Units in millions
14 20%
18% 12 16%
10 14%
12% 8 10% 6 8%
4 6%
4% 2 2%
0 0% 2010 2011 2012 2013E 2014E 2015E 2016E
Serv er Shipments (mn) Serv er Growth (y /y )
Source: Gartner, Nomura research
109
Nomura | U.S. Semiconductors Primer December 11, 2013
Favorable trends in servers driven by cloud and data center to continue over the next several years Cloud computing is a service that provides scalable computing capacity to remote customers using the Internet. Public cloud computing offers cloud services to users who are external to the provider’s organization. Cloud resources are usually shared by multiple users and are dynamically provisioned on demand. Cloud computing enables enterprises to have their applications up and running faster, and it provides improved manageability and less maintenance and cost, enabling IT departments to allocate resources properly to meet fluctuating and unpredictable business demands. For these reasons, the first users of these services were cash-strapped organizations that adopted cloud computing to host their applications from providers such as Amazon. Now, bigger enterprises are following suit and embracing cloud computing. Despite concerns around security, the majority of Fortune 500 companies are either using cloud (private or shared) or are considering using it. Prominent companies such as Netflix and Twitter primarily use cloud-based servers from Amazon. Top cloud computing providers include Amazon, Microsoft, Google, VMware, and IBM. Intel recently indicated that its server shipments to cloud providers have been growing 20–40% YoY. In addition, recent negative trends in traditional hardware server companies such as IBM indicate that workloads are shifting from traditional to cloud-based servers. Increasing cloud-based offerings and a high-level of interest in cloud adoption are the key drivers for data center growth. According to Gartner, total data center spending will grow at a CAGR of 4% between 2013 and 2016. Asia Pacific is expected to experience the strongest growth in spending, with a 6% CAGR during this period. The Cisco Global Cloud Index, which is a measure for the growth of global data center and cloud-based IP traffic, forecasts annual global data center IP traffic and global cloud IP traffic to triple over the next five years. The number of workloads per installed cloud server is expected to increase from 4.2 in 2011 to 8.5 in 2016, with nearly two-thirds of all workloads processed in the cloud by 2017.
Data center growth will likely continue to be driven by cloud adoption and increasing use to web services Gartner is forecasting a 4% CAGR in server and storage units and revenue in data centers through 2017 (Fig. 55). While this growth may look modest, we believe that data center growth of large Internet players such as Google and Facebook is not properly reported. A recent third-party (Uptime) survey with responses from 1,000 data center facilities operators, IT managers, and senior executives from around the globe indicates that data center operators are expecting healthy budgets, with nearly one-third in the United States and Europe expecting increases of 10% or more. Most of the increase is driven by third-party data center operators. Key data center operators such as Google, Amazon, Facebook, Microsoft, and Apple do not report their budgets. Due to the massive increase in web service usage (Facebook users reaching 1bn users), the big Internet players are a significant driver of server infrastructure growth.
110
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 55: Data center server and storage unit forecasts, 2010–2017E In mn 16 7%
14 6%
12 5% 10 4% 8 3% 6 2% 4
2 1%
- 0% 2010 2011 2012 2013E 2014E 2015E 2016E 2017E
Server and storage units Unit Growth
Source: Gartner, Nomura research
Competition from ARM-based servers will likely increase, but it is not a big threat to Intel in the next few years The explosion of smartphones and tablets and the increasing use of cloud-based services require servers that can host these applications. The current server landscape is dominated by Intel’s x86 chips with more than 90% market share. The problem with traditional x86-based servers is that they consume too much power and are only optimized for certain types of applications. Micro-servers, a new category of servers, are designed differently and make use of low-power, multiple-parallel processors to run workloads in a distributed fashion. Recent acquisitions and product announcements indicate increasing momentum toward micro-servers. AMD recently bought SeaMicro, a pioneer in micro-servers. In addition, HP announced project Moonshot, a new breed of low-power server systems offering customers a choice of chips to handle different computing workloads. The company expects the new servers to cut power consumption by 89% relative to existing HP servers, to consume 80% less space, and to reduce costs by more than 75%. Micro-servers such as Moonshot largely rely on chips designed for mobile devices, which have begun to offer more competitive performance while using significantly less power than the traditional x86 server chip. Next year, we will see ARM chips with 64-bit support, which we think would the first serious push of ARM into servers. To counter this, Intel is preemptively creating low-power version of server chips. This appears to have some success in fending off the competition from ARM chips. For example, HP’s new breed of servers, Moonshot, will be initially powered by Intel’s server- class Atom chips (Centerton). Centerton adds features previously unavailable in Atom processors, including Intel’s virtualization technology and support for error correction in the memory interfaces. Intel recently announced a 22nm successor to Centerton, codenamed Avoton. While this could cannibalize Intel’s high-end Xeon server chip sales, Intel seems to be more proactive in the server market to tackle competition from ARM than it has been in the client market. In addition, there are some benefits to using low- power x86 chips in micro-servers, including compatibility with legacy software code.
111
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 56: HP’s Moonshot micro-server
Source: HP, Nomura research
Fig. 57: ARM-based server chip cartridge
Source: Calxeda, Nomura research
Open Compute Project is disruptive, but x86 server chips will likely continue to dominate the server space Gartner’s data show strong growth for vendors lumped under “Others,” which includes Quanta and other companies (Hyve, ZT Systems, WiWynn) building custom servers for large cloud companies. We believe that the Open Compute Project (OCP)initiative is the key driver for this category. OCP is an initiative by Facebook to improve the efficiency of data centers. In Open Compute, member companies such as AMD, Calxeda, Applied, HP, Dell, and Intel are pushing toward an agnostic hardware architecture that will enable server motherboards to be vendor-neutral. Intel, AMD, and ARM vendors such as Applied Micro and Calxeda will support the new hardware architecture. We believe this new hardware architecture will have a significant impact on the data center server market in the long term. Data centers using this architecture can mix CPUs from different suppliers of x86 and ARM-based SoCs. This should add flexibility in data centers by enabling companies to use the most efficient CPU vendor/architecture suited for a specific workload. We believe that variations of OCP’s approach will be eventually adopted by other companies that operate data centers. As a result, there could be more competition in the data center space, where Intel has a dominant position (97% share). That said, we do not expect ARM to gain much traction in servers before 2016, as software availability and ecosystem support take a long time to establish. Gartner is forecasting a stable share of x86 server revenue from data centers through 2016.
112
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 58: Gartner expects a stable share of x86 server revenue from data centers
100%
90%
80%
70%
60%
50%
40%
30% 45% 45% 45% 45% 20% 41% 42% 44%
10%
0% 2010 2011 2012 2013E 2014E 2015E 2016E
x86 Serv er Rev enue non-x8 Serv er Rev enue Storage and Networking Rev enue
Source: Gartner, Nomura research
Market share: Intel dominates servers Intel dominates the server market with 97% unit share. At the end of 3Q13, AMD had only 3% share in servers, down from its peak of 26% in 2006. The server segment is very sensitive to performance and Intel has executed solidly. In addition, AMD has not been able to support cloud and virtualized workloads in data centers. AMD’s most recent Bulldozer architecture has not been able to match Intel’s strength. AMD recently changed its strategy. The company is looking for ways to customize its offerings to gain share in specific verticals and OEMs. It recently started to work on developing an ARM-based server chip. We are positive on this initiative; according to Gartner, ARM’s server market could be 10–15% of the total server opportunity by 2015/2016.
Fig. 59: Server microprocessor unit share, 2004–2013E
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0% 1Q04 2Q04 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05 1Q06 2Q06 3Q06 4Q06 1Q07 2Q07 3Q07 4Q07 1Q08 2Q08 3Q08 4Q08 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13E
Intel AMD
Source: Mercury Research, Nomura research
113
Nomura | U.S. Semiconductors Primer December 11, 2013
ASPs Notebook: uptrend in notebook ASPs in the last few years is not likely to continue Intel’s notebook ASPs have been up in the last three years. After reaching a trough at around $70 in 2009, Intel’s notebook ASPs have been up 40% to around $100 in 2013 (Fig. 60). AMD has not been able to capitalize on this upward ASP trend; its notebook ASPs are down 3% since 2009. Intel’s notebook ASP increase in the last few years was primarily driven by a premium on its higher performance, longer battery life, and lower competition from AMD. The increase in notebook ASPs has also been driven by higher consumer preference of the mid-range Core i5 parts in notebooks, as opposed to entry- level Core i3 chips. However, if we analyze notebook ASP trends over a longer period, we find that notebook ASPs have declined at a CAGR of 5% since the peak of approximately $180 in 2004 to around $100 in 2013. We think increasing competition from ARM-based low-priced tablets and lower notebook system prices will likely pressure notebook ASPs. We believe AMD’s notebook ASP will be under more severe pressure due to Intel’s widening competitive lead in manufacturing process and low-power architectural changes. We think notebook ASPs will continue their long-term decline, perhaps at a slightly faster pace than the long-term -5% CAGR.
Fig. 60: Notebook ASP trend for Intel and AMD, 2004–2013E
$200
$180
$160
$140
$120
$100
$80
$60
$40
$20
$0 1Q04 2Q04 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05 1Q06 2Q06 3Q06 4Q06 1Q07 2Q07 3Q07 4Q07 1Q08 2Q08 3Q08 4Q08 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13E
Intel Notebook ASP AMD Notebook ASP
Source: Mercury Research, Nomura research
114
Nomura | U.S. Semiconductors Primer December 11, 2013
Desktop: ASPs have been more stable than notebook ASPs, a trend we think could persist over the next few years Desktop ASPs peaked in early 2005, when Intel’s desktop ASP exceeded $120 and AMD’s $80. Since 2005, desktop ASPs have steadily declined through 2010, reaching around $90 for Intel and around $50 for AMD. From the peak ASPs in 2005, desktop ASPs have declined at a 3% CAGR through 2013. By contrast, notebooks ASPs have declined at a 5% CAGR. Although Intel’s desktop ASPs have bounced from the trough in 2010 of $80 to the recent $103, we do not think this growth will continue. The increase in desktop ASPs is likely due to increased mix of enterprise and professional desktops, as consumers’ share of desktops has decreased. AMD’s desktop ASPs have been flat, after reaching a trough of around $49 in 2010. While we do not anticipate significant growth in desktop ASPs in upcoming periods, they will likely be more stable than notebook ASPs. This is due to lower competition from ARM and a richer mix of enterprise and professional desktops than in notebooks.
Fig. 61: Desktop ASP trend for Intel and AMD, 2004–2013E
$140
$120
$100
$80
$60
$40
$20
$0 1Q04 2Q04 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05 1Q06 2Q06 3Q06 4Q06 1Q07 2Q07 3Q07 4Q07 1Q08 2Q08 3Q08 4Q08 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13E
Intel Desktop ASP AMD Desktop ASP
Source: Mercury Research, Nomura research
115
Nomura | U.S. Semiconductors Primer December 11, 2013
Server: ASPs have grown double digits YoY in the last few years; we see server ASPs plateauing at current levels in the near term Unlike notebook and desktop ASPs, which have declined in the last 10 years, server ASPs have increased at a CAGR of 3% from around $440 in 2004 to $560 in 2013. After hitting the trough in 2006, Intel’s server ASPs have increased sharply from $330 to $580 in 3Q13. AMD’s server ASPs have remained largely flat at $310. Currently, Intel’s server chips account for 97% of the server market share. While there is virtually no competition for Intel’s server chips, further ASP gains of this magnitude are unlikely. There are a few reasons for that: 1) We are seeing plateauing in terms of new features in server chips. Intel’s latest server chipset, Romley, which has 30% better performance than the last generation, failed to trigger an upgrade from enterprise customers. That said, data centers continue to be a secular driver for higher server ASPs. 2) Intel is cautious about exercising its market pricing power due to emerging competition from ARM’s 64-bit server chips. While high-performance applications will continue to use x86-based server chips, there are certain server loads that can utilize ARM’s more power-efficient server chips. In addition, we will likely see an offsetting impact from Intel’s own offerings of low- power server chips based on Atom (Avaton), which are 8–10 times lower in ASP than Intel’s Xeon server chips. We believe the 64-bit ARM server chips will be available for deployment in 2H14. The 64-bit ARM server chips are likely to be 8–10 times cheaper than Intel’s or at a similar price to Intel’s Avaton chips.
Fig. 62: Server ASP trend for Intel and AMD, 2004–2013E
$700
$600
$500
$400
$300
$200
$100
$0 1Q04 2Q04 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05 1Q06 2Q06 3Q06 4Q06 1Q07 2Q07 3Q07 4Q07 1Q08 2Q08 3Q08 4Q08 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13E
Intel Server ASP AMD Serv er ASP
Source: Mercury Research, Nomura research
116
Nomura | U.S. Semiconductors Primer December 11, 2013
Client Roadmap Intel Intel launched the next-generation core Haswell processor on 22nm in the second half of 2013. The company expects Haswell to reduce the power consumption from 17W in Ivy Bridge to 7-10W. We expect Intel to move Atom chips to 22nm process node with improved cores (4) and performance (2 times) versus the previous generation. The 22nm Haswell will push Intel two generations ahead of AMD on 22nm. AMD is shipping 32nm core CPUs and expects to transition to 28nm in 2H13. Intel is pushing OEMs to make touch a mandatory feature for Haswell-based Ultrabooks that are were launched in 2H13. Intel refreshed its existing Ivy Bridge chips in the server segment this year. The new Ivy Bridge-E chips feature better graphics and performance and lower power.
Fig. 63: Intel notebook roadmap 1H12 2H12 1H13 2H13 1H14 2H14 Notebook / Desktops:
High Performance Ivy bridge Haswell Broadwell 35-55 watts 25-35 watts 17-25 watts 22nm 22nm 14nm Single Core
Medium Performance Ivy bridge Ivy bridge Haswell Broadwell 13-35 watts 10-25 watts 10-17 watts 7-10 watts 22nm 22nm 22nm 14nm Single Core Quad Core
Low Performance Ivy bridge Ivy bridge Haswell Broadwell 17-35 watts 10-25 watts 10-17 watts 7-10 watts 22nm 22nm Bay Trail 14nm Single Core Quad Core 5-10 watts 22nm Source: Intel, Nomura research
Fig. 64: Intel Mobile Roadmap 1H12 2H12 1H13 2H13 1H14 2H14 2015 Core Saltwell (32nm) Saltwell (32nm) Saltwell (32nm) Silvemont (22nm 3D) Silvemont (14nm 3D) Silvemont (14nm 3D) Goldmont (14nm 3D)
Tablet Performance Medfield Clover Trail Bay Trail Merrifield Moorefield Broxton Atom Z24xx Atom Z27xx Atom Z23xx 64-bit 64-bit 64-bit MP in 3Q12 MP in 4Q12 Intel's 1st quad core in tablet 14nm Quad core Quad core 2.5-3.5 watts 2.75-3 watts MP in 4Q13 14nm 14nm 32nm 32nm 2.5-3 watts 22nm LTE Advanced Cherry Trail TD-LTE & TD-SCDMA 64-bit 17 FDD bands, 5 TDD bands 14nm Atom Airmont
Medium / Low Performance
Smartphone Medfield Clover Trail+ Merrifield Moorefield Broxton Atom Z24xx Atom Z25xx 64-bit 64-bit 64-bit MP in 3Q12 Intel's 1st dual core in SP 14nm 14nm Quad core 2.5-3.5 watts 3.5 watts Quad core 14nm 32nm 32nm LTE Advanced TD-LTE & TD-SCDMA Lexington 17 FDD bands, 5 TDD bands Atom Z24xx MP in 1Q13 Medium / Low 32nm SoFIA SoFIA LTE Performance Integrated Global 3G Integrated Global LTE Atom based Atom based external foundry external foundry
Source: Intel, Nomura research
117
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 65: Intel Server Roadmap 1H12 2H12 1H13 2H13 1H14 2H14 Servers:
High Performance Sandy Bridge-EX Ivy Bridge Ivy Bridge-EX-A Kittson 95-150 watts 80-130 watts ------Boxboro MC/EX Boxboro MC / Brickland
Medium Performance Sandy Bridge-EP Ivy Bridge Ivy Bridge-EN 60-150 watts 17-130 watts ------Romley
Low Performance Ivy Bridge Ivy Bridge Ivy Bridge 10-95 watts 17-55 watts Bromolow 32 nm 32nm Denlow Bordenville Edisonville Source: Intel, Nomura research
118
Nomura | U.S. Semiconductors Primer December 11, 2013
AMD AMD is shipping 32nm Richland APUs with Piledriver cores. We expect AMD to refresh its core high-performance APUs with Steamroller cores (Kaveri) and transition to 28nm Global Foundries process in 1H14. After a disappointing performance improvement in Piledriver cores, expectations are high from AMD’s next-generation Steamroller cores. Steamroller cores are expected to support Heterogeneous Computing (HSA). In the medium-performance segment, AMD is expected to launch Kabini APUs using Jaguar cores on TSMC’s 28nm process node in late 1H14. Considering process node disadvantage, it would be crucial for AMD to optimize its architecture to improve performance at lower power consumption to stay competitive with Intel. In the low end, AMD will be refreshing its Brazos platform with Jaguar core-based 28nm APUs in 1H14.
Fig. 66: AMD Notebook Roadmap
1H12 2H12 1H13 2H13 1H14 2H14 Notebook / Desktops: High Performance Vishera Richland Kavaeri -- 35 watts -- 32nm 32nm 28nm Vishera 2/4 Steamroller cores -- Medium Performance Trinity Trinity Richland -- -- 17-25 watts Kaveri 32nm 32nm 32nm -- 28nm 2/4 Steamroller cores Low Performance Brazos 2.0 Brazos 2.0 Kabini 9-18 watts 9-18 watts 9-15 watts 40nm 40nm 28nm 2/4 Jaguar cores Source: AMD, Nomura research
Fig. 67: AMD Mobile Roadmap
1H12 2H12 1H13 2H13 1H14 2H14 Tablets:
Hondo Temash 4,5 watts <4 watts 40nm 28nm Jaguar Cores Source: AMD, Nomura research
Fig. 68: AMD Server Roadmap
1H12 2H12 1H13 2H13 1H14 2H14 Servers:
High Performance Abu Dhabi Opteron 6300 / 4300 Warsaw -- 35-140 watts -- 32nm 32nm 32nm 4/6/8/12/16 Piledriver cores 4/6/8/12/16 Piledriver cores 12/16 Piledriver cores
Medium Performance Seoul Opteron 6300 / 3300 Berlin -- 25-65 watts -- 32nm 32nm 28nm 6/8 Piledriver cores 4/8 Piledriver cores 4 Steamroller cores
Low Performance Delhi Opteron X1150/X2150 Seattle -- 9-22 watts -- 32nm 28nm 28nm 4/8 Piledriver cores 4 Jaguar cores ARM Cortex A57 Source: AMD, Nomura research
119
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 69: AMD Server Roadmap
1H12 2H12 1H13 2H13 1H14 2H14 Embedded Solutions: High Performance R-Series APU/CPU Bald Eagle APU/CPU 17 - 35 watts 17 - 35 watts Hierofalcon CPU 32nm 28nm 17 - 35 watts 2/4 Piledriver cores 2/4 Steamroller cores 28nm 2/4 Steamroller cores Low Performance G Series APU SoC Steppe Eagle 6 - 25 watts 5 - 25 watts 32nm 28nm 2/4 Jaguar Cores 2/4 Jaguar cores
Discrete Graphics Radeon E6160 / E6760 Adelaar GPU -- -- 40nm 28 nm
Source: AMD, Nomura research
120
Nomura | U.S. Semiconductors Primer December 11, 2013
ARM ARM released the Cortex A15 core in 2013, which has been adopted by Samsung and Nvidia. ARM’s A15 cores have two times the performance of A9 cores and are more power-efficient. Samsung is using A15-based processor in Chromebooks, Galaxy Note, and Galaxy S4 smartphones. Nvidia’s Tegra 4 is expected to be used in smartphones and tablets. In 2014, ARM is expected to add 64-bit support for its upcoming A53 and A57 cores, which are to be used in more compute-intensive/server applications.
Fig. 70: ARM Processor Roadmap
1H12 2H12 1H13 2H13 1H14 2H14
High Performance Cortex-A15 Cortex-A15 Cortex-A57 4 watts 4-7 watts -- Dual-Core 1.75GHz 2.5 GHz Quad-Core 2.5GHz 32 nm 28nm 20nm
Medium Performance Cortex A9 Cortex A9 Cortex A12 4-7.5 watts 4-5 watts -- < 2GHz < 2GHz 32nm 28nm
Low Performance Cortex-A5 Cortex-A7 Cortex-A53 -- 0.5 - 1 watts -- Dual Core Single Core, >1 GHz Quad-Core 2.GHz 32nm 28nm 20nm
Source: Gartner, ARM, Nomura research
Fig. 71: Specifications of Apple chips using ARM
App Processor Architecture 64-bit Support Lithography Clock Speed (GHz) # of Cores Manufacturer Application Launch Date iPad, iPhone 4, iPod A4 ARM v7 Cortex-A8 No 45nm 1 1-4 Samsung Touch, Apple TV 2010 iPad 2, iPhone 4S, iPod A5 ARM v7 Cortex-A9 No 45 - 32 nm 0.8 - 1 1 - 2 Samsung Touch, Apple TV 1Q 2011
A5x ARM v7 Cortex-A9 No 45nm 1 2 Samsung iPad 3 1Q 2011
A6 Swift ARM v7 Cortex-A7s No 32 nm Up to 1.3 2 Samsung iPhone 5 3Q 2012
A6x Swift ARM v7 Cortex-A7 No 32 nm Up to 1.4 2 Samsung iPad 4 4Q 2012
A7 Cyclone ARM v8 Yes 28 - 22 nm Up to 1.3 2 Samsung iPhone 5s, iPad Air 2013 Source: Apple, AnandTech, Nomura research
121
Nomura | U.S. Semiconductors Primer December 11, 2013
Graphics Processing Unit (GPU) A graphics processing unit (GPU) is a specialized chip designed to accelerate the image output in a frame buffer intended for output to a display. GPUs are very efficient at manipulating computer graphics and are generally more effective than general-purpose CPUs for processing large blocks of data in parallel. GPUs are used in embedded systems, mobile phones, personal computers, workstations, and game consoles. In a personal computer a GPU can be present on a video card, it can be on the motherboard or on the CPU die. The term GPU was popularized in 1999 by Nvidia that marketed the world's first GPU (GeForce 256).
Fig. 1: Graphics processing pipeline
Source: MIT.edu, Nvidia, Nomura research
GPU processing pipeline • 3D API: OpenGL and Direct3D are the two most commonly used API (application programming interface) exposed to the graphics drivers. Graphics drivers translate OpenGL/D3D commands into GPU-specific commands and send them to the actual GPU for processing. Direct3D is similar to OpenGL API for Microsoft Windows operating systems. • Vertex Processor: A vertex processor, or shader, is a graphics processing function used to add special effects to objects in a 3D environment by performing mathematical operations on objects' vertex data. In addition to the location’s coordinates, each vertex can be defined by other variables such as color. • Rasterization: Rasterization is the task of taking an image described in a vector graphics format (shapes) and converting it into a raster image (pixels or dots) for output on a video display or printer, or for storage in a bitmap file format. In normal usage, the term refers to the popular rendering algorithm for displaying three-dimensional shapes on a display. Rasterization is currently the most popular technique for producing real-
122
Nomura | U.S. Semiconductors Primer December 11, 2013
time 3D computer graphics. Compared with other rendering techniques, such as ray tracing, rasterization is faster. However, rasterization is simply the process of computing the mapping from scene geometry to pixels and does not prescribe a particular way to compute the color of those pixels. Hardware acceleration of rasterization algorithms is now fairly common in consumer PCs that have a dedicated GPU. • Frame Buffer: A Frame buffer is a memory buffer containing a complete frame of data, which is used to drive a video output device. The memory buffer typically consists of color values for every pixel to be displayed. Color values are commonly stored in 1-bit binary (monochrome), 4-bit, 8-bit, 16-bit high-color, and 24-bit true-color formats. An additional alpha channel is sometimes used to retain information about pixel transparency. The total amount of the memory required to drive the frame buffer depends on the resolution of the output signal and on the color depth and palette size. Frame buffer sizes of 1GB are common in the entry-level graphics cards and can exceed 8GB in high-end cards.
Types of GPUs
Integrated graphics – Will likely keep gaining share from low- end discrete GPU The definition of integrated graphics has changed over time. Earlier references to integrated graphics used to be for ‘integrated chipset graphics’ that was integrated as part of the CPU chipset. Integrated graphics in today’s PCs and notebooks refers to CPU-integrated graphics. In CPU-integrated graphics, graphics processing units are integrated in the CPU die. Intel’s Sandy Bridge and AMD’s Llano processors were the first CPUs to integrate GPUs on the same die as the CPU. According to Mercury Research, integrated graphics will account for 72% of total shipments in 2013. Integrated CPU graphics solutions are less costly than dedicated graphics solutions, but also have lower performance. Historically, integrated solutions were incapable of playing 3D games or running graphically intensive programs. Intel and AMD have recently improved the performance of their integrated CPU graphics solutions. AMD’s Accelerated Processing Unit (APU) and Intel’s HD Graphics are now more than capable in handling 3D graphics, but are still not on par with the current generation of discrete GPUs. Integrated graphics’ share has increased over time from 38% of all shipments in 2000 to 73% in 2012. A large part of the increase in integrated graphics share is due to integration of graphics cores in CPUs in the last few years. Integrated CPU graphics has much better performance than the integrated chipset graphics (figure 2), a factor that has also helped in increasing the share of integrated graphics.
123
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 2: Block diagram of integrated graphics (IGP)-based CPU sub-system
CPU
FSB
PCI x16 Slots North Bridge Memory IGP
DMI
South Bridge SATA BIOS USB
Source: Nomura research
Fig. 3: Block diagram of integrated CPU graphics in a CPU sub-system
CPU
Integrated CPU Graphics
FSB
PCI x16 Slots North Bridge Memory
DMI
South Bridge SATA BIOS USB
Source: Nomura research
124
Nomura | U.S. Semiconductors Primer December 11, 2013
Discrete graphics – Will continue to dominate the high-end and gaming markets These are standalone GPUs and are more powerful than the integrated graphics. Discrete GPUs typically interface with the motherboard by means of an expansion slot such as PCI Express (PCIe) or Accelerated Graphics Port (AGP). Discrete graphics is typically used in gaming PCs, high-end desktops and notebooks, and in gaming consoles. GPUs are also used in High Performance Computing. Due to GPU’s highly parallel architecture, many compute-intensive programs can run faster in GPUs than in CPUs. This concept utilizes the computational power of a modern graphics accelerator's shader pipeline into general-purpose computing power. In certain applications, GPUs can yield several orders of magnitude higher performance than a conventional CPU. GPU-based high performance computers are starting to take on a significant role in large-scale modeling. Most of the new generation supercomputers take advantage of GPU acceleration. Discrete graphics share in the last 3–4 years has declined a bit but has been relatively stable. In 2013, discrete graphics will account for 28% of all PC graphics shipments. Performance of discrete graphics remains superior to that of integrated CPU’s. Most recent benchmarks suggest that integrated CPU graphics are now similar to the performance of entry-level discrete graphics.
Fig. 4: Block diagram of discrete graphics in a CPU sub-system
CPU
FSB
Discrete PCI x16 Slots North Bridge Memory Graphics
DMI
South Bridge SATA BIOS USB
Source: Nomura research
125
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 5: Graphics performance comparison, Integrated chipset, Integrated CPU (Intel and AMD), and Discrete
40.0x
35.0x
30.0x
25.0x
20.0x
15.0x
10.0x
5.0x
0.0x Integraed Chipset Iv y Bridge HD4000 AMD A8 Radeon 6550D Entry -lev el discrete
Source: Toms Hardware, Nomura research
Fig. 6: Integrated and discrete graphics share trend, 2000–2014E
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
Total Discrete - % Total Total Integrated - % Total
Source: Mercury Research, Nomura research
Fig. 7: Graphics shipments trends by type, 2000–2014E Fig. 8: Share of graphics shipment by type, 2000–2014E in thousands units
600,000 100%
90% 500,000 80%
400,000 70%
60% 300,000 50% 200,000 40%
100,000 30%
20% 0 10%
0% 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E Total Integrated Chipset Total Integrated CPU Total Discrete Total Desktop Discrete - % Total Desktop Integrated - % Total Portable Discrete - % Total Portable Integrated - % Total Source: Mercury Research, Nomura research Source: Mercury Research, Nomura research
126
Nomura | U.S. Semiconductors Primer December 11, 2013
Discrete Graphics Overview and Landscape NVIDIA and AMD are the two primary companies that develop and sell discrete graphics chips. Discrete graphics chips are primarily used in three key end-markets: 1) PC gaming, 2) high-end PCs and workstations, and 3) high-performance computing (for compute intensive applications). We estimate the total size of the discrete graphics market will be $4.6bn in 2013. GPU revenue split among gaming, high-end PCs & workstations, and High Performance Computing was 72%, 24%, and 4%, respectively. According to Mercury Research, discrete graphics shipments totaled 128mn in 2012 and will decline 7% to 119mn in 2013E. Third party estimates suggest that GPU shipments should modestly rebound in 2014.
Fig. 9: Discrete Graphics Market and Segments ($mn), 2013E
$175 , 4%
$1,104 , 24%
$3,307 , 72%
Gaming, desktops, and notebooks Professional Graphics HPC
Source: Nvidia, AMD, Nomura research
127
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 10: Discrete graphics shipment and growth trend, 2000–2014E in thousand units
160,000 20%
140,000 15%
120,000 10% 100,000 5% 80,000 0% 60,000 -5% 40,000
20,000 -10%
0 -15%
Total discrete shipments (in K) Total Discrete GPU growth (y/y)
Source: Mercury Research, Nomura research
Integrated CPU graphics will likely drive discrete GPU attach rates lower over time Improving performance of integrated CPU graphics could be a long-term headwind for discrete graphics shipments at the low end. Apple recently switched its 13” and 15” MacBooks to integrated CPU graphics from Intel. Earlier, Apple was using discrete GPU for the 15” model. That said, we think discrete graphics will maintain its lead in performance at the high end for the next several years. While the performance of integrated graphics CPU has increased, both Nvidia and AMD have also improved performance for their high-end GPUs. We see the gap in integrated and discrete graphics in the high end to persist. In addition, we think the revenue impact from the loss of low-end discrete GPUs to integrated graphics is modest. To back our view, we note that Nvidia’s GPU revenue is expected to increase 1% in CY13, while Intel’s PC Client revenue is expected to decline 5% in 2013. The reason for this is that most of the discrete graphics revenue is locked in the high-end applications, such as professional, HPC, and gaming, that are not under threat from integrated graphics.
Fig. 11: Discrete graphics attach rate trend, 2002–2014E in thousand units
450,000 70%
400,000 65%
350,000 60%
300,000 55% 250,000 50% 200,000 45% 150,000
40% 100,000
50,000 35%
0 30% 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
MPU units Discrete Graphics units Attach rate
Source: Mercury Research, Nomura research
128
Nomura | U.S. Semiconductors Primer December 11, 2013
PC gaming is the biggest driver of discrete GPU revenue We estimate that more than half of the industry’s discrete graphics revenue come from sales to PC gamers. PC gaming accounts for around two-thirds of Nvidia’s total GPU revenue. According to Gartner, spending in PC gaming software will exceed $2bn in 2013. While discrete revenues are expected to be correlated with PC gaming software sales, we find that in the last few years, discrete graphics revenue has outpaced the low- single-digit decline in PC gaming software sales. We believe that this is because of the rise of free online PC games. We expect PC gaming to remain relevant. We believe ‘massively multiplayer online role-playing games,’ such as World of Warcraft, and real- time strategy games, such as StarCraft, have sizable user bases and steady revenue streams. These games are also difficult to port to consoles. Furthermore, PC games can offer better graphics as modern PCs' capabilities are far better than those of current game consoles, including the newly released Xbox One and Sony PS4. Currently, the total number of PC gamers is estimated to be around 54mn worldwide. That said, we think other factors, such as gamers switching to mobile devices and media tablets, could cause a decline in the PC gaming market over time.
Fig. 12: Discrete graphics and PC gaming software sales trend, 2011–2014E In $mn
6,000
5,000
4,000
3,000
2,000
1,000
0 2011 2012 2013E 2014E
PC Gaming revenue Discrete Graphics revenue
Source: Gartner, Nomura estimates
Adoption of GPU as a cloud service could become a new revenue stream Earlier this year, Nvidia unveiled a new service called GRID. The idea behind GRID technology is that games and graphics-compute dependent applications can have a device-agnostic, location-independent future. In enterprises, GPUs don’t need to be confined physically near the users of graphics-heavy applications. For enterprise customers, Nvidia offers the ability to virtualize GPUs and thus allow the data center/enterprise IT manager provision GPUs to work efficiently and on demand. Nvidia is working with partners such as Citrix, Microsoft, and VMware. For GRID, the GPUs can be clustered in a server rack and allocated to users via network. This approach is similar to CPU/server virtualization, which is now a mainstream approach in most enterprises and data centers. We think, over time, GRID in enterprise could become a meaningful revenue driver. Nvidia recently indicated that GRID is under trial at more than 200 enterprise customers and all major server OEMs (Cisco, HP, IBM, Dell) are supporting the technology. VMware recently announced that it is adding vDGA (virtual Dedicated Graphics Acceleration) using Nvidia’s GRID technology. Amazon also recently started offering GRID in its EC2 platform services. While it is difficult to assess the adoption rate of this service, we believe that virtualizing GPUs would make it easier to provision and deploy GPUs in an organization. Nvidia believes that GPU streaming is a $10bn opportunity over time. As such, we think increased usage of virtualized GPUs could prove to be an offset over time to the headwinds that discrete GPUs face from share loss to integrated CPU graphics and from increasing
129
Nomura | U.S. Semiconductors Primer December 11, 2013 adoption of mobile devices (tablets) that do not use discrete GPUs. We expect the company to start seeing GRID revenues in the back half of next year (2014). Similarly, for gaming application, Nvidia is enabling streaming of games via a dedicated set of GPUs. A tablet or an iPad may not have a capable GPU on board, but using GRID, they can now stream graphics power. NVIDIA is utilizing its low-latency remote display technology to improve the user experience by reducing the lag when going back-and-forth with the rendering server. Nvidia is using this technology in its SHIELD game console. We believe Sony is also using Nvidia’s GRID technology in its PS4 console platform.
130
Nomura | U.S. Semiconductors Primer December 11, 2013
Market share trends – Nvidia’s share reaching record high Nvidia’s current share is approaching its historic high market share of 67% in discrete graphics. After losing share to AMD in 2010, Nvidia has been steadily gaining market share with its more power efficient Kepler architecture. Nvidia’s current Kepler-based GPUs are more power efficient and have won most of the design wins in the first generation of Ultrabooks. In 1Q13, Nvidia had around 67% of overall discrete market share and AMD had the remaining 33%. Within the segment, Nvidia had 68% in desktops and 64% in notebooks. We believe Nvidia’s share gain will likely moderate in 2014.
Fig. 13: Overall discrete market share, 1Q09–1Q13 Fig. 14: Desktop discrete market share, 1Q09–1Q13
100% 100%
90% 90%
80% 80%
70% 70%
60% 60%
50% 50%
40% 40% Total standalone market standalone Total
30% 30%
20% Market Standalone Desktop 20%
10% 10%
0% 0% 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 NVIDIA AMD/ATI Others NVIDIA AMD/ATI Others
Source: Mercury Research, Nomura research Source: Mercury Research, Nomura research
Fig. 15: Portable discrete graphics market share, 1Q09– Fig. 16: Discrete graphics – portable vs. desktop, 1Q09– 1Q13 1Q13
100% 40,000
90% 35,000
80% 30,000 70% 25,000 60%
50% 20,000
40% 15,000
30% 10,000 20% Portable Standalone Market Standalone Portable 5,000 10%
0% 0 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 1Q09 2Q09 3Q09 4Q09 1Q10 2Q10 3Q10 4Q10 1Q11 2Q11 3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 NVIDIA AMD/ATI Others Total Desktop Discrete Total Portable Discrete
Source: Mercury Research, Nomura research Source: Mercury Research, Nomura research
131
Nomura | U.S. Semiconductors Primer December 11, 2013
Wireless
Market Size According to Gartner, total wireless semiconductor revenue will be $73bn in 2013. The majority of the revenue will be driven by increasing content in mobile phones, which will account for 87% of total wireless semiconductor revenue. Wireless revenue is expected to account for 24% of the total semiconductor revenue of $304bn in 2013, up from 23% in 2012. Wireless semi revenue as percentage of total semi revenue has increased from around 20% in 2005 to 24% in 2013E. We think increasing semiconductor content in handsets is the key driver for the growth in wireless semi growth in the last three to four years. Although overall handset units are expected to grow modestly this year at 2% YoY, wireless semi revenue is expected to be up 7% YoY. We believe that this is due to 30%-plus growth in smartphones this year, which have higher semiconductor content.
Fig. 1: Wireless accounts for 24% of total Semi revs, 2013E Fig. 2: Wireless as percentage of total Semi revs, 2004-2013E
100% 90% 24% 80% 70% 60% 50% 40% 30% 20%
76% 10% 0% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
Wireless Semi Revenue Total Semi Revenue Wireless Semi Revenue Total Semi Revenue
Source: SIA, Gartner, Nomura research Source: SIA, Gartner,, Nomura research
Fig. 3: Revenue growth trends, 2004–2013E Fig. 4: Wireless revenue and growth trend, 2004–2013E $ in millions
50% $80,000 50%
40% $70,000 40%
30% $60,000 30%
20% $50,000 20% 10% $40,000 10% $30,000 0% 0% $20,000 -10% $10,000 -10% -20% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E $0 -20% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E Wireless Semi Revenue (y/y) Total Semi Revenue (y/y)
Total Semi Rev enue Ex . Memor y ( y /y ) Wireless Semi Revenue Wireless Semi Revenue (y/y)
Source: SIA, Gartner, Nomura research Source: SIA, Nomura research
132
Nomura | U.S. Semiconductors Primer December 11, 2013
Key Components in the Wireless Industry Overview The wireless semiconductor industry compromises companies that produce cellular baseband processors, application and multimedia processors, RF components, connectivity chips, memory, and others. This section of our report focuses on application- specific subcomponents, which include baseband and applications processors, connectivity, and RF. As cellular networks become more complex, basebands ICs are constantly upgraded to support new wireless protocols. The emergence of 2G, 3G, and 4G standards has increased the content costs of basebands, and the increasing computing horsepower of smartphones has pushed up applications processor content costs. The chart below shows the market share of the components within the wireless semiconductor segment in 2012.
Fig. 5: Market share of components in the wireless semiconductor segment, 2012
23%
38%
20%
4% 8% 8%
Baseband Memory RF Devices Application/Mulimedia Processors Connectivity Others
Source: Gartner, Nomura research
The total BOM (bill of materials) of mobile phones has increased dramatically since the introduction of smartphones and new cellular standards. For example, the chart below shows the BOM comparison of Nokia 105s and iPhone 5s. Although Nokia 105 is a relatively new phone, it is designed after Nokia 1100, which was released in 2003 and has sold more than 250 million units over the last 10 years. We use Nokia’s 105 model as the proxy for a voice-only legacy phone and iPhone 5 as the proxy for a modern-day smartphone.
Fig. 6: BOM comparison of voice-only and smartphone Voice-Only Smartphone Nokia 105 iPhone 5S Baseband / RF / Memory $5.25 Baseband $32.00 Processor $19.00 Connectivity / RF $4.20 NAND / DRAM $20.40 Enclosures / Connectors $3.50 $28.00 Charger / Battery $2.50 $10.60 Display $2.25 $41.00 Camera $13.00 Sensors $15.00 Power Management $7.50 Total Content Cost $13.50 $190.70
Source: iSuppli, Nomura research
133
Nomura | U.S. Semiconductors Primer December 11, 2013
The content costs of all the parts inside a phone have increased due to the increasing complexity of a phone’s capabilities. Unlike the original Nokia 105, the new generation of smartphones can play sophisticated games, connect to 3G/4G cellular networks for data usage, play music and movies on a high-definition screen, and make voice calls. Even with the dramatic change in content costs, iSuppli estimates that Apple’s implied gross margin is approximately 70% while Nokia 105’s is 29%. Baseband Processors A baseband processor (or baseband) manages the radio control functions, such as signal modulation, encoding, and radio frequency shifting. It also manages the transmission of signals. Basebands have their own control software that allows changes to the software application and operating system without affecting the operation of the baseband. Basebands require a lengthy certification process at each carrier before they are approved for use. To support multiple standards and bands, the complexity of baseband processors increases, as every new generation of products has to be backward-compatible and be able to support all legacy standards. For example, an LTE-capable device needs to work on LTE bands in addition to being backward-compatible with all 3G and 2G standards. The baseband market is directly tied to mobile devices. Every mobile handset requires one baseband to handle the communications function. However, not every tablet requires a baseband, as some can operate through WiFi networks alone. We estimate that the baseband attach rate in tablets has declined from 50% in 2010 when Apple launched iPad to 20–30% currently. Many tablets used in emerging markets are WiFi-only.
Fig. 7: Top 5 vendors in the Baseband subsegment, 2012 Fig. 8: Baseband revenue and growth trend
18,000 15% 13% 16,000
4% 14,000 10%
5% 12,000 5% 10,000
8,000 9% 0% $ in$ millions 60% 6,000
4,000 -5% 10% 2,000
0 -10% 2007 2008 2009 2010 2011 2012
Qualcomm MediaTek Intel STMicroelectronics Broadcom Others Baseband Processors YoY Growth
Source: Gartner, Nomura research Source: Gartner, Nomura research
According to Gartner, the baseband sub-segment made up 23% of the wireless industry and was 50% of the application-specific category (Baseband, RF, applications processor, connectivity) in 2012. Baseband components are the second-most expensive segment in the BOM above because they are at the heart of the phone’s functionality. At the core of the baseband’s functionality, the baseband chip manages the phone’s antennas and decodes voice and data signals. The price of baseband chips has gone up due to evolving cellular standards and the increasing number of bandwidths and network bands. In addition to supporting 2G, 3G, and 4G networks, a smartphone’s baseband chip needs to seamlessly switch between all networks and minimize latency. Baseband chips can be found standalone or be part of system on chip that includes an applications processor. In 2012, Qualcomm was the leading supplier of baseband revenue with 60% share, followed by MediaTek (10%), Intel (9%), STMicroelectronics (5%), Broadcom (4%), and Spreadtrum (4%). Within the CDMA/UMTS market, Qualcomm had 62% share, followed by Intel at 10%, and STMicroelectronics at 7%. Within the GSM/GPRS market, MediaTek led with 37%, followed by Intel at 19% and Spreadtrum at 16%. The CDMA/UMTS market was about three times as big as the GSM/GPRS market in 2012. In the TD-SCDMA market, Spreadtrum was the leader with 40% share, followed by Marvell at 26% and
134
Nomura | U.S. Semiconductors Primer December 11, 2013
MediaTek at 20%. However, perhaps the most important category going forward is the LTE market, which is expect to grow to four to five times in the next five years. In the LTE market, we estimate Qualcomm is enjoying more than 90% with a very significant price premium. That said, we expect competition in LTE to increase, as most suppliers, including, MediaTek, Intel, Broadcom, Marvell, and NVIDIA, have plans to ramp production of LTE chipset in the first half of 2014.
Fig. 9: Baseband market share by technology (LTE, CDMA, GSM, TD-SCDMA)
LTE Market Share, 2013E CDMA Market Share, 2012
Others, 5% Intel, 10%
STMicro, 7%
Broadcom, 6%
MediaTek, 5%
Qualcomm, 62% Others, 9%
Qualcomm, 95%
Qualcomm Others Qualcomm Intel STMicro Broadcom MediaTek Others
GSM Matker Share, 2012 TD-SCDMA Market Share, 2012
STMicro, 1% Others, 1% Others, 12% Leadcore, 12%
Broadcom, 5%
MediaTek, 37% Spreadtrum, 40% STMicro, 11% MediaTek, 20%
Spreadtrum, 16%
Intel, 19% Marvell, 26%
MediaTek Intel Spreadtrum STMicro Broadcom Others Spreadtrum Marvell MediaTek Leadcore STMicro Others
Source: Gartner, Nomura research
135
Nomura | U.S. Semiconductors Primer December 11, 2013
The baseband market is estimated to be more than $16.5bn in 2013, up 6% from 2012. This revenue growth rate is higher than the unit growth rate as average selling price expands. Gartner estimates baseband revenue will grow at a CAGR of 7% in the next five years to $22bn in 2017 while units will grow at only a CAGR of 3%.
Fig. 10: Baseband processor market, 2012–2017E
25,000
20,000
15,000
10,000
Baseband revenue (in $mn) (in revenue Baseband 5,000
0 2012 2013E 2014E 2015E 2016E 2017E
Source: Gartner, Nomura research
Baseband ASPs are increasing despite growth in emerging markets. The increase in ASP may seem counterintuitive given that the growth will come from emerging markets and that the growth in high-end smartphone market is slowing. However, we believe the following factors will drive higher ASPs:
136
Nomura | U.S. Semiconductors Primer December 11, 2013
Mix shift from 2/2.5G phones to smartphones, as baseband use in entry-level smartphone costs nearly twice as much as 2/2.5G basebands. Over the next five years, Gartner estimates that the entry-level smartphone will grow at a CAGR of 28% and account for 43% total mobile handset shipments in 2017, while traditional 2/2.5G will decline at a CAGR of 21% to only 15% of total mobile handset shipments. The premium smartphone market is expected to grow at a CAGR of 12%. One of the key drivers for smartphone adoption in the next few years is the increasing adoption of turnkey reference solution by companies such as Qualcomm, MediaTek, and Spreadtrum. These turnkey solutions offer high-performance chipsets and improved features, while driving time to market and attractive device ASPs ($150–200) for consumers who would have purchased a low-end 2/2.5G phone.
Fig. 11: Basic smartphones are expected to be the majority of unit shipments in 2017
100% 58% 43% 34% 27% 20% 15% 90%
80% 43% 41% 70% 37% 60% 33% 50% 27% 40% 14% 39% 42% 30% 37% 33% 30% 20% 28% 10% 0% 2012 2013E 2014E 2015E 2016E 2017E
Premium smartphone Basic smartphone Traditional phone
Source: Gartner, Nomura research
Fig. 12: Mix shift to smartphone is driving blended average semiconductor content increase
$70
$60
$50
$36 $38 $39 $35 $40 $32 $40
$30
$20
$10
$- 2012 2013E 2014E 2015E 2016E 2017E
Premium smartphone Basic smartphone Traditional Blended
Source: Gartner, Nomura research
137
Nomura | U.S. Semiconductors Primer December 11, 2013
Mix of integrated baseband processors is increasing. Gartner estimates that integrated basebands will grow from 25% in 2012 to 67% in 2017, and ASPs of integrated basebands could be two to three times higher than discrete basebands. Integrated solutions save $5–7 on BOM relative to discrete basebands, due to the elimination of power management IC and memory, as well as the packaging, test, and assembly costs required for additional chips. In the future, it is likely that other functions, such as wireless connectivity, audio codec, and touch controllers, will be integrated into basebands. That said, we think premium smartphones will continue to use discrete basebands and applications processors for performance and differentiation.
Fig. 13: Growth of integrated baseband processors is driven by smaller form factors, battery life, and low-cost smartphones
100% 75% 64% 54% 44% 37% 33% 90% 80% 70% 60% 67% 63% 50% 56% 40% 46%
30% 36%
20% 25% 10% 0% 2012 2013E 2014E 2015E 2016E 2017E
Integrated % of total unit Discrete % of total unit
Source: Gartner, Nomura research
Fig. 14: Integrated baseband processors drives ASPs higher
$20 $18 $16 $14 $12 $9.09 $9.27 $9.30 $10 $8.51 $7.79 $8.10 $8 $6 $4 $2 $0 2012 2013E 2014E 2015E 2016E 2017E
Discrete baseband - ASP Integrated baseband - ASP Blended ASP
Source: Gartner, Nomura research
138
Nomura | U.S. Semiconductors Primer December 11, 2013
Multi-mode LTE basebands cost two to three times as much as the average 3G baseband. Based on market statistics, LTE’s adoption in the first eight quarters was much faster than 3G’s during the same period. Gartner forecasts LTE (FDD and TD combined) will grow from 5% of total shipments in 2012 to 34% in 2017. Given the significant premium of LTE basebands, the rapid adoption of LTE should lift overall baseband ASPs, even assuming a modest decline in LTE baseband ASPs over time.
Fig. 15: LTE basebands cost 3x as much as 3G basebands
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2012 2013E 2014E 2015E 2016E 2017E
GGE CDMA/EVDO UMTS TD-SCDMA LTE-FDD TD-LTE
Source: Gartner, Nomura research
139
Nomura | U.S. Semiconductors Primer December 11, 2013
Application & Multimedia Processors An applications processor is a central processing unit similar to the one installed in a personal computer. It is often used to manage the software programs on a mobile device. An applications processor is a system-on-a-chip that integrates a variety of systems, including CPU, graphics accelerators, image signal processors, storage interfaces, memory management, and input/output ports. Most of the applications processors today are either dual-core or quad-core. Theoretically, quad-core processors would operate twice as fast as dual core processors and four times as fast as single-core processors. In reality, the performance depends on the programs being run, as well as compatibility with other hardware in the system. In 2012, applications processors made up 8% of the wireless industry and were 17% of the application-specific category. This subsegment grew 56% from 2011 to 2012 because of the growing need for mobile handsets to provide discrete processors to cope with enhanced graphics, gaming, and video capabilities. Application processors metaphorically stand in the middle of a mobile phone because of their central importance. Memory, sensors, power management, touch-screen control, displays, baseband processors and connectivity all plug into the applications processor, where functions are centrally coordinated. Total subsegment revenue grew 56% in 2012, with total revenues $4.8bn and $2.8bn in 2011 and 2012, respectively.
Fig. 16: Block diagram of a mobile phone
Source: Samsung, Nomura research
Although applications processors are only 8% of the total market, they are the fastest- growing subsegment with a 12% CAGR from 2007 to 2012. The applications processor market is largely owned by Samsung and Apple, which accounted for 81% of total revenues in 2012. Both firms largely use these processors for internal consumption to meet tailored demands. Samsung’s ARM cortex-based Exynos processor is used for its flagship galaxy phones and tablets, and in Google’s chromebooks. Apple’s ARM cortex- based A7 processor is used for its flagship tablets and phones.
1 40
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 17: Top 5 in the App Processor subsegment, 2012 Fig. 18: Application & multimedia processor revenue and growth trend
6,000 60% 4% 4% 5% 50% 5,000 7% 40% 4,000 30%
20% 3,000 13% 10% $ in$ millions 2,000 0%
-10% 68% 1,000 -20%
0 -30% 2007 2008 2009 2010 2011 2012
Samsung Apple Texas Instruments Nvidia Qualcomm Others Application/Multimedia Processors
Source: Gartner, Nomura research Source: Gartner, Nomura research
Gartner defined market share differently from other sources, as Gartner includes Samsung’s foundry business for Apple to be part of Samsung’s market share. Hence, Samsung’s share appears to be significantly higher than Apple’s. This should change in 2014 if Apple changes its foundry partner to TSMC, which Gartner will include as “Apple foundry.” In the integrated processor market, Qualcomm and MediaTek are the leading suppliers, and we expect more competition from competitors such as Marvell, Broadcom, and Intel in 2014. Overall, we expect Texas Instruments’ share to decline in 2014, as the company announced its exit from this market in 2012. With the exception of Intel, most suppliers license their CPU technology from ARM Holdings. Gartner estimates that mobile phones with integrated solutions will grow from 25% in 2012 to 67% in 2017. As the mid-range and low-end smartphones continue to grow, there is a need to lower cost and improve time to market delivery. We estimate the cost of an integrated baseband is half the cost of the discrete solution (discrete baseband processors plus discrete applications processor), due to the elimination of power management IC and memory, as well as the packaging, test, and assembly costs required for additional chips. We believe premium smartphones will continue to use discrete basebands and applications processors for performance and differentiation. The competitive landscape is different in the tablet market, given that most tablets are not equipped with a baseband processor. As such, the qualification requirements for tablets are significantly lower than the applications processors used in mobile handsets. As a result, there are more suppliers for discrete applications processors, especially in the low-end white-box market. Examples of these suppliers include Allwinner, Rockchip, Amlogic, and Wondermedia.
141
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 19: Growth of integrated baseband processors is driven by smaller form factors, battery life, and low-cost smartphones
100% 75% 64% 54% 44% 37% 33% 90% 80% 70% 60% 67% 63% 50% 56% 40% 46%
30% 36%
20% 25% 10% 0% 2012 2013E 2014E 2015E 2016E 2017E
Integrated % of total unit Discrete % of total unit
Source: Gartner, Nomura research
According to Strategy Analytics, the applications processor market, including discrete and integrated processors, grew from $2.5bn in 2009 to $12.9bn in 2012, representing a three-year CAGR of 70%. Based on data from the first half of 2013, the market is on track to grow another 40–50% in 2013. In 2012, the discrete application market was approximately $4.8bn, representing about 37% of the total applications processor market. The growth in the discrete application market is primary driven by flagship products from Apple and Samsung.
Fig. 20: Application Processor revenue, 2009–2013E
$20
$18
$16
$14
$12
$10
$8
$6
$4 Apps Processor($bn) Revenue
$2
$0 2009 2010 2011 2012 2013E
Source: Gartner, Nomura research
142
Nomura | U.S. Semiconductors Primer December 11, 2013
Connectivity – 802.11ac should provide a boost next year Wireless connectivity generally refers to the means of connecting to the Internet or to other devices through radio waves rather than physical wires. In mobile devices, wireless connectivity refers to protocols such as Bluetooth, wireless LAN (WLAN), global positioning system (GPS), and near-field communications (NFC) but excludes cellular protocols such as WCDMA and LTE that are generally handled by baseband processors. Wireless connectivity solutions can be offered as single-function solutions, but the majority is offered as combination solutions (or wireless combo) to save space and cost. Most of the wireless combo chips now include a least three functions: Bluetooth, WLAN, and GPS. We expect a quad combo chip that includes NFC functionality (e.g., Broadcom’s BCM43341) to see increasing rates of adoption in the future. Furthermore, new standards are being introduced within each protocol. For example, the most common WLAN standard today is 802.11n, but we expect more mobile devices to adopt the latest standard of WiFi, 802.11ac, which has throughput speed that is two to three times higher than 802.11n’s. While the rate of adoption of 802.11ac was less than expected this year, we think next year we could see a broader adoption of 802.11ac in enterprise LAN and mobile devices. We believe that Apple, which did not use 802.11ac WiFi this year, is likely to adopt the new standard in its devices next year. According to Gartner, the connectivity subsegment made up 4% of the wireless industry and was 8% of the application-specific category (Baseband, RF, applications processor, connectivity) in 2012. Connectivity components have a low ASP relative to the rest of the mobile phone semiconductor content. In the BOM diagram above, connectivity coupled with RF represents only 2% of the total cost. Total subsegment revenue grew only 2% in 2012, with total revenues at $2.3bn and $2.5bn in 2011 and 2012, respectively. Broadcom derived $1.2bn in revenue from connectivity in 2012, producing a market share of approximately 47%. The next-closest competitor was Qualcomm, which had $0.3bn in revenue and a 10% market share. The rest of the companies accounted for 42% of the market share or $1.1bn in revenue. Connectivity subsegment has grown at a CAGR of 12% over the last five years. The largest spike in revenue growth happens in 2010, when smartphones grew by 71% on a unit basis.
Fig. 21: Top 5 in the Connectivity subsegment, 2012 Fig. 22: Connectivity revenue and growth trend
3,000 50% 16% 2,500 40%
30% 8% 2,000 20% 47% 1,500 10%
9% $ in millions 1,000 0%
500 -10% 10%
0 -20% 10% 2007 2008 2009 2010 2011 2012
Broadcom Qualcomm CSR Texas Instruments MediaTek Others Wireless Connectivity YoY Growth
Source: Gartner, Nomura research Source: Gartner, Nomura research
In 2012, Broadcom led the overall wireless connectivity market for mobile phones with a 47% share and had more than 75% market share in combo chips. Broadcom continues to have the vast majority of high-profile smartphones, including iPhone 5S and Galaxy S4. Broadcom is also the leader in 802.11ac, which has been adopted by a few high-end smartphones, including Samsung Galaxy S4. Further adoption of 802.11ac should help Broadcom’s ASP in the future.
143
Nomura | U.S. Semiconductors Primer December 11, 2013
Qualcomm was second with 10% share. Qualcomm’s share gain partly came from the acquisition of Atheros Communications, but Qualcomm also benefited from increase connectivity attach rates with MSM8960 solution and from its Qualcomm Reference Design (QRD) program for entry-level smartphones in emerging markets. CSR was third with 10% share. CSR sold the handset connectivity business to Samsung in 2012, which completes the final piece that is required for Samsung to produce a fully integrated solution that combines an applications processor, a baseband processor and wireless connectivity. We think Samsung’s integrated solution will likely start with low- end smartphones. TI captured 9% share in 2012, but its share should decline in 2013, as it announced its exit from this market. TI underinvested in this market and was unable to gain traction at tier-1 smartphone OEMs. Meanwhile, TI’s main customers, Nokia and Blackberry, continue to lose share.
144
Nomura | U.S. Semiconductors Primer December 11, 2013
RF – Above-average industry growth trend to continue on increasing mobile data usage and LTE adoption RF components include filters, power amplifiers, low-noise amplifiers, duplexers, and switches. RF component vendors work with gallium arsenide to make components (power amplifiers), rather than silicon. Gallium arsenide has higher electrical conductivity characteristics but is generally more difficult to manufacture than its silicon counterpart. According to Gartner, the RF subsegment made up 8% revenue of the wireless industry and 17% of the application-specific category (Baseband, RF, applications processor, connectivity). The total RF segment revenue grew 4% in 2012, with total revenues of $5.3bn and $5.1bn in 2012 and 2011, respectively. In 2012, the top 5 RF vendors by revenue were Skyworks, Qualcomm, RD Micro, TriQuint, and Avago. Skyworks Solutions derived around $1bn in revenue from RF components in 2012, producing a market share of approximately 20%. The next-closest competitor was Qualcomm, which had $0.8bn in revenue and a 15% market share. The rest of the companies accounted for 65% of the market share and $3.4bn in revenue.
Fig. 23: Top 5 in the RF subsegment, 2012 Fig. 24: RF revenue and growth trend
7,000 20% 20% 15% 6,000 10%
38% 5,000 5% 0% 4,000 -5% 15% 3,000 -10% $ in$ millions 2,000 -15% -20% 1,000 7% 13% -25% 8% 0 -30% 2007 2008 2009 2010 2011 2012 Skyworks Solutions Qualcomm RF Micro Devices TriQuint Semiconductor Avago Technologies Others RF Devices YoY Growth
Source: Gartner, Nomura research Source: Gartner, Nomura research
An increasing number of RF bands in 4G devices are boosting RF silicon content in phones. We believe increasing use of mobile data and adoption of LTE devices will be a long-term driver for growth in RF components. RF content varies from $3 to $5.50 in single-band to quad-band 3G devices. While 3G devices utilize up to five bands, LTE has a higher number of bands across carriers and geographies. The RF content varies from $3 in a single-band phone to around $5.50 in a quad-band phone. 4G networks utilize up to 40 bands, but LTE support in a regional device could mean an additional seven to eight bands, driving RF content in regional LTE-enabled phone in the $7–9 range.
Fig. 25: RF content costs on 3G vs. 4G enabled phones
Source: TriQuint, Nomura research
145
Nomura | U.S. Semiconductors Primer December 11, 2013
While RF silicon content in phones is increasing due to LTE, the adoption of LTE- enabled smartphones is also increasing. Gartner estimates that LTE connections have only a 2% penetration rate for all mobile connections, which is estimated to be 6.5bn devices by 2013. At the same time, smartphone connections to LTE networks will have a penetration rate of 14%, and this number will likely double by 2015. Mitigating this seemingly low penetration rate is that mobile phones that are LTE-enabled but do not have an LTE connection are not counted in the 14% rate. Gartner estimates that LTE- enabled smartphones will be approximately 250mn or 26% of the smartphone market in 2013 and 350–380mn in 2014.
Fig. 26: LTE connections and adoption rates
450 30% 400 25% 350 300 20% 250 15% 200 150 10% Devices in mn 100 5% 50 0 0% 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E
LTE Connections Adoption Rate for All Connected Devices Adoption Rate for Smartphones
Source: Gartner, Nomura research
Connected devices are moving beyond mobile phones. According to Cisco’s VNI global IP traffic forecast, traffic from wireless devices will exceed traffic from wireline devices by 2014. Moreover, mobile data traffic should increase by a magnitude of 18x from 2012 to 2016, mainly driven by Internet users adopting 3G/4G networks on a variety of connected devices. Gartner estimates that mobile connections over the last six years (2007–2013E) will increase at a CAGR of 12%. Not only are total devices increasing, but the share of 3G/4G connections is increasing as a percentage of the total market. Tablets are the prominent example of non-mobile phone devices gaining connections to cellular networks. Gartner estimates that more than 200mn tablets will be cellular-enabled in 2013, rising to more than 500mn by 2017.
146
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 27: Mobile connections by technology, 2007–2015E
8
7
6
5
4
Units in bn 3
2
1
0 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E
2G 3G 4G
Source: Gartner, Nomura research
Fig. 28: Tablets connected to cellular networks, 2010–2017E
700
600
500
400
300 Units in mn
200
100
0 2010 2011 2012 2013E 2014E 2015E 2016E 2017E
Cellular Enabled Tablets
Source: Gartner, Nomura research
We expect wireless connected devices to go beyond mobile phones to smart energy products, automotive sector, medical equipment, and machine-to-machine (M2M) technology. Skyworks Solutions estimates that this broader definition of connected devices will reach 50bn units by 2020. Refrigerators, thermostats, and cars are just a few examples of existing devices that will soon be connected wirelessly. In some cases, these products will require a multitude of RF content to function properly.
147
Nomura | U.S. Semiconductors Primer December 11, 2013
Handsets remain the key driver for wireless semi growth Smartphones have significantly higher silicon content We think handsets are the primary drivers for wireless semiconductor revenue growth. We expect overall handsets to grow at a CAGR of 3% from 2012 to 2015, with smartphones growing at a significantly higher CAGR of 28%. Smartphones have significantly more amount of silicon content than feature phones. We estimate that a voice/feature phone has around $20 of total BOM (bill of material) cost versus BOM cost for low-end, mid-range, and high-end smartphones of roughly $80, $125, and $180, respectively. Currently, based on Nomura estimates, smartphones account for around 50% of total handset shipments. As such, we think increasing penetration of smartphones, which we think will exceed 80% of all handset shipments by 2017, is a secular driver for wireless semiconductor growth over the next several years.
Fig. 29: Smartphones have significantly higher silicon content than feature phones
$200
$180
$160
$140
$120
$100
$80
$60
$40
$20
$0 Voice phone Low-end smartphone Mid-range High-end smartphone smartphone
Source: Nomura estimates
Fig. 30: Handset unit growth
2.25
2.00
1.75
1.50
1.25
1.00
0.75 Handset Units (bn) Units Handset
0.50
0.25
0.00 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E
Smartphones (bn) Voice phones(bn)
Source: Nomura estimates
148
Nomura | U.S. Semiconductors Primer December 11, 2013
In the last five years, the most significant increase in silicon content has come from memory, applications processor, and connectivity. The content of these components combined grew at a double-digit rate over the last five years in handsets while mobile phone revenue grew in the high single digits.
Within handset, cellular connectivity is a secular growth driver for wireless semiconductors Demand for mobile data driven by need to access information and web services from anywhere, anytime is driving growth for higher data speed cellular connectivity technologies. Since 2011, the share of 3G+ phones (includes 4G phones) increased from around 40% to more than 50% in 2013E. We expect the penetration of faster cellular technology to increase to around 90% by 2017. Increasing usage of high-speed cellular modem implies higher silicon content, as a faster modem requires a more expensive modem (baseband), a faster processor, and a higher amount of memory per device. Gartner forecasts baseband revenues to increase at 8% CAGR from 2013 through 2017, which is much higher than the expected overall handset growth of low- to mid-single-digit.
Fig. 31: Handset units by cellular standards
2,500 30%
20% 2,000
10% 1,500 0% 1,000 -10%
500 -20%
0 -30% 2011 2012 2013E 2014E 2015E 2016E 2017E 3G+ units 2G units 3G+ growth (YoY) 2G+ growth (YoY)
Source: Gartner, Nomura research
Fig. 32: Cellular technology penetration rates for key regions, 2Q13 Developed Nations Emerging Markets
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% USA Japan UK Germany Italy India Brazil IndonesiaTurkey
2G 3G/4G
Source: Gartner, Nomura research
149
Nomura | U.S. Semiconductors Primer December 11, 2013
Evolution of cellular standards First-Generation Systems First-generation systems were based on analog modulation and were primarily designed for carrying voice. These services were different from prior systems because they used cellular topology and provided automatic switching and handover of voice calls across cell sites. Japan’s Nippon Telephone implemented the world’s first cellular system in 1979. Nordic Mobile Telephone system, deployed in Europe in 1981, was the first system that supported automatic handover of calls. In the United States, the first generation cellular systems were based on AMPS (Advanced Mobile Phone Service).
Advanced Mobile Phone Service (AMPS) AMPS or Advanced Mobile Phone Service was developed by AT&T Bell Labs in the late 1970s. In addition to the United States, AMPS was deployed in several countries in South America, Asia, and Canada. AMPS systems used a Frequency Modulation (FM) scheme for transmission of voice. Similar to first-generation systems, AMPS used analog transmission. Even after the deployment of second-generation (2G) systems, AMPS continued to be used by operators in North America as a fallback service for providing roaming between different operator networks that had incompatible 2G systems.
2G Systems 2G systems were based on improved hardware processing platforms. Like prior systems, 2G systems were primarily targeted for voice communication, but they used digital modulation instead of analog. The shift from analog to digital modulation enabled several improvements in system performance and bandwidth. Operators could now use spectrally efficient digital codecs and multiplex several users on the same frequency channel, improving performance and capacity of 2G systems. 2G systems also used simple encryption to provide transmission security, which was missing from the earlier analog systems. Besides providing improved voice quality, capacity, and security, 2G systems enabled new applications. SMS (Short Messaging Service) was a key application enabled by 2G systems. SMS application was first deployed in Europe in 1991 and quickly became a popular service among mobile users. In addition to SMS, 2G systems supported wireless data applications. Initial 2G systems supported circuit switched data services and later evolved to support packet data services. Low data rates (10–40 kbps) and limited display space in the handsets required special protocols such as Wireless Access Protocol (WAP) to deliver Internet content to mobile devices.
Fig. 33: Evolution of cellular standards
Evolution of Digital Cellular Standards 2G 2.5G 3G 4G World GSM GPRS EDGE WiMAX (TDMA) Japan PDC WCDMA HSPA (TDMA) (UMTS) (UMTS) U.S. iDen HSPA+ LTE (TDMA) (UMTS) U.S. IS-136 EV-DO LTE-A (TDMA) (CDMA 2000) U.S., Asia IS-95A IS-95B 1x (CDMA) (CDMA) (CDMA 2000)
Source: The Computer Language Co., Nomura research
150
Nomura | U.S. Semiconductors Primer December 11, 2013
GSM (Global System for Mobile communication) In 1982, Global System for Mobile communication (GSM) was formed with a charter to develop a system that could deliver cheaper wireless voice services and to work seamlessly across all of Europe. Prior to GSM, the European cellular market was fragmented with a variety of mutually incompatible systems. By late 1980s, the European Telecommunications Standards Institute (ETSI) took over the development of the GSM standard and the first version, named GSM Phase I, was released in 1990. Shortly thereafter, several operators in Europe deployed GSM. GSM quickly gained acceptance beyond Europe, and the standard was renamed as the Global System for Mobile Communications. Besides voice and SMS, the GSM standard supported circuit-switched data at 9.6kbps. By the mid-1990s, GSM Packet Radio Systems was introduced as an evolutionary step for GSM systems to support higher data rates. Under ideal conditions, GPRS could provide a maximum data rate of 160kbps. Typical implementations of GPRS supported a user data rate of 20–40kbps.
CDMA (Code Division Multiple Access) In 1989, Qualcomm proposed Code Division Multiple Access (CDMA) as a more efficient wireless technology. In 1993, Telecommunication Industry Association adopted Qualcomm’s proposal as an IS-95 standard. Unlike other wireless standards, CDMA system enabled multiple users to share the same frequency channel at the same time. Instead of time-sharing multiple users in a given frequency channel, each user is assigned a different orthogonal code that is used to separate their signals at the receiver. Due to frequency reuse, CDMA systems provided a higher voice capacity than GSM. However, CDMA systems did not succeed in gaining similar adoption rates as GSM did. However, due to better spectral efficiency, CDMA technology was incorporated in the evolution plan for 3G.
CDMA 2000 and EV-DO (Evolution, Data only) The 3G evolution of IS-95 standard is known as the CDMA2000 standard. To achieve higher data rates (up to 2Mbps) and to improve throughput for packet data, the CDMA2000 standard evolved to CDMA2000-EVDO (Evolution, Data Only). This standard is applicable to only data traffic. EV-DO originally was developed by Qualcomm for use in fixed and mobile applications meeting the 2Mbps low mobility requirements and was the first system to provide real broadband-like speeds to mobile users.
UMTS (Universal Mobile Telephone Service) Universal Mobile Telephone Service (UMTS) was originally developed as the 3G system based on the evolution of GSM systems. The 3GPP (3G partnership project), which oversees the standards setting for 3G, was formed with the collaboration of six regional telecommunications standards. 3GPP completed and published the first 3G UMTS standard in 1999, known as UMTS Release 99. UMTS Release 99 is widely deployed around the world. UMTS includes (1) a core network (CN) that provides switching, routing, and subscriber management; (2) the UMTS Terrestrial Radio Access Network (UTRAN); and (3) the User Equipment (UE). The basic architecture is based on backward compatibility with GSM/GPRS architecture, with support for 3G capabilities. While UMTS retains the basic architecture of GSM/GPRS networks, the 3G air interface known as Wide-band CDMA (WCDMA) is significantly different from the 2G interface. The WCDMA design leveraged the IS-95 standard and its features. WCDMA uses a Direct Sequence Spread Spectrum CDMA system, where user data is multiplied with pseudo-random codes that provide channelization, synchronization, and scrambling. WCDMA is specified for both FDD and TDD operations, although FDD is by far the most widely deployed.
HSPA (High-Speed Packet Access) HSPA includes two key enhancements by 3GPP to UMTS-WCDMA: (1) High-Speed Downlink Packet Access (HSDPA) introduced in Release 5 in 2002 and (2) High-Speed
151
Nomura | U.S. Semiconductors Primer December 11, 2013
Uplink Packet Access (HSUPA) introduced in Release 6 in 2004. HSDPA was first deployed by AT&T in late 2005 and quickly became widespread globally. According to Ericsson, population coverage for HSPA will increase from 50% in 2012 to more than 85% in 2017. There are more than 300 operators that have deployed HSPA worldwide. Increasing demand for mobile data download paved the way for HSDPA. HSDPA defined a new downlink channel capable of providing up to 14.4Mbps peak throughput; however, typical user throughputs are in the 500kbps to 2Mbps range.
Fig. 34: Summary of key 3G standards
W-CDMA CDMA 2000 1X EV-DO HSPA 3GPP 3GPP Standard 3GPP2 3GPP2 Release 99 Release 5/6
850 / 900 MHz, 450 / 900 MHz, 450 / 900 MHz, 450 / 900 MHz, Frequency Bands 1.8 / 1.9 / 2.1 GHz 1.7 / 1.9 / 2.1 GHz 1.7 / 1.9 / 2.1 GHz 1.8 / 1.9 / 2.1 GHz
Channel Bandwith 5MHz 1.25MHz 1.25MHz 5MHz
DL: 2.4 - 4.9 mbps DL: 3.6 - 14.4 mbps Peak Data Rate 384 - 2084 kbps 307 kbps UL: 800 - 1800 kbps UL: 2.3 - 5 mbps
Typical User Rate 150 - 300 kbps 120 - 200 kbps 400 - 600 kbps 500 - 700 kbps
User-Plane Latency 100 - 200ms 500 - 600ms 50 - 200ms 70 - 90ms
Multiple Access CDMA CDMA CDMA / TDMA CDMA / TDMA
Duplexing FDD FDD FDD FDD
DS-SS: QPSK, DS-SS: QPSK, Data Modulation DS-SS: QPSK DS-SS: BPSK, QPSK 8PSK and 16QAM 16QAM and 64QAM
Source: Nomura research
HSPA+ 3GPP Release 7 HSPA, also referred to as HSPA+, contains a number of additional features that improve the system capacity, end-user throughput, and latency. The key enhancements included in HSPA+ are 1) higher-order modulation and multiple input multiple output (MIMO) to achieve higher peak rates (LTE further enhances the support for higher order modulation and MIMO) and 2) dual-carrier downlink operation (dual- carrier operations achieve higher data rates when there are multiple carriers available and are deployed in a single cell). This approach doubles the peak data rate from 21Mbps to 42Mbps and substantially increases the overall capacity of a cell.
LTE – Key features LTE was developed to incorporate several key enhancements. 3GPP set several goals for LTE to enhance data rates and to reduce operator’s cost per bit of mobile data. LTE’s design goal was to achieve an average downlink speed that is 3–4 times better than that of HSPA and an average uplink speed that is 2–3 times better. Higher data rates are achieved by improving spectral efficiency by 2–4 times. LTE requirements also called for increased bit rate while maintaining the same site locations as deployed today. LTE makes use of the following key technologies:
152
Nomura | U.S. Semiconductors Primer December 11, 2013
• Orthogonal Frequency Division Multiplexing (OFDM) One of the key differences between the 3G and LTE is the use of Orthogonal Frequency Division Multiplexing (OFDM) as the underlying modulation technology. 3G systems such as UMTS and CDMA2000 are based on Code Division Multiple Access (CDMA) technology. CDMA performs well for low data rate communications such as voice. However, for high-speed applications, CDMA requires a large bandwidth and thus becomes less efficient. OFDM has proven to be a better technology for achieving high data rates. The use of OFDM provides better multipath interference and allows LTE to be deployed in a variety of spectrum allocations and channel bandwidths. • Multi-Antenna Techniques The LTE standard provides support for implementing multi-antenna configurations to improve link performance, capacity, and spectral efficiency. Multi-antenna techniques supported in LTE include the following: 1) Transmit diversity, which is a technique to combat multipath fading in the wireless channel. This technique replicates the same signal and sends it over multiple transmit antennas. Transmit diversity can increase system capacity and cell range. 2) Beam-forming, a technique where multiple antennas are used to transmit the same signal, so that there is more than one spatial path to the receiver. This beam-forming improves the chances that signal reaches the receiver without interference. LTE supports beam-forming in the downlink. 3) Multiplexing, where more than one stream can be combined in parallel by using multiple transmit antennas. These signals can then be separated at the receiver.
153
Nomura | U.S. Semiconductors Primer December 11, 2013
Drivers for LTE Increasing usage of mobile data Increasing adoption of smartphones and tablets is driving an explosion in mobile data usage as more users consume content on the go. Web services such as Facebook and YouTube are driving media-rich applications, which demand high amounts of mobile data usage. According to Strategy Analytics, mobile data traffic is set to rise by 300% by 2017 to about 20 Exabytes from 5 Exabytes in 2012. The biggest driver of this growth is expected to be a rise in streaming video services.
Proliferation of smart devices Convergence of multiple functionalities such as camera, GPS, and video players into smartphones and tablets is driving the demand of these converged devices. This is evidenced by the fact that tablets took only two to three years from launch to hit the cumulative shipment volume of 100mn, while notebooks took more than 10 years to reach the same cumulative shipment volume. In the future, we expect most appliances that are not yet connected to the Internet to be connected and to be accessible via mobile devices. We see “Internet of Things” driving further increase in demand for mobile data in the future. Global adoption of LTE as a single standard will likely accelerate the adoption of these connection devices and services.
Need for lower cost per bit for mobile data LTE offers peak data rates of up to 170 Mbps in the downlink and 58 Mbps in the uplink, faster than today’s HSPA by a factor of 10. In addition, latency is very low at 10–20 ms, considerably boosting end-user experience of services such as gaming and browsing, as well as the performance of true real-time applications such as VoIP and the throughput of a HTTP page access. LTE also has higher spectral efficiency than HSPA, enabling service providers to squeeze more data into their available spectrum. In addition, LTE uses spectrum in widths selectable from 1.4 MHz up to 20 MHz, whereas WCDMA uses a 5 MHz spectrum carrier. Due to the characteristics of OFDM, it has a much better average cell throughput than HSPA does. In addition, LTE simplifies operator network by using a flat architecture and fewer all-IP interfaces, which means less interoperability testing.
1 54
Nomura | U.S. Semiconductors Primer December 11, 2013
Qualcomm sustaining leadership in LTE Qualcomm started shipping LTE-enabled MSM8960 chipset in late 2011. Qualcomm currently has around 90% share of LTE shipments, which is much higher than its overall share in basebands of 55–60%. While we expect solutions from Intel, Broadcom, and MediaTek to start shipping in 2014, we think that the impact to Qualcomm’s share will be initially minimal. We believe Qualcomm is ahead on some of the advanced LTE features such as carrier aggregation, LTE-A, and multiple LTE band support. Despite LTE product announcements from companies such as Intel, Broadcom, and Marvell, we do not believe that there is a meaningful threat to Qualcomm’s lead in the LTE space in the near term. We see the following reasons for this: • Qualcomm has invested more in baseband than all of its competitors combined. In the last three years, Qualcomm spent around $11bn in R&D. We think a majority of this spending went into basebands. In contrast, Qualcomm’s competitors combined have spent less than $10bn in baseband R&D in the last three years. We estimate Intel and Broadcom spent around $3bn in the last three years, while Marvell and MediaTek spent $2bn and $1bn, respectively. We believe this buys Qualcomm a significant lead in terms of integrating and testing advanced LTE features.
Fig. 35: Baseband R&D investments in the last three years $mn 12,000
10,000
8,000
6,000
4,000
2,000
0 QCOM Intel BRCM MRVL MediaTek
Baseband R&D investment in last 3 years
Source: Company data, Nomura research
• OEMs want an integrated global solution. OEMs such as Apple require a single piece of silicon to address all the cellular standards across the regions. Only Qualcomm has the capability to provide this integration. While companies such as Samsung and MediaTek could provide a WCDMA stack, they lack a CDMA stack, which is needed in North America and Latin America. We think Qualcomm’s competitors would likely be able to get sockets for SKUs that are regional, but we think that Qualcomm will remain dominant with OEMs that need a single global SKU for all regions. Even if an OEM has a separate device SKU to support relevant bands in a region, we think the OEMs would like to support all cellular standards in the baseband. • Large number of bands in LTE takes longer to get the optimal solutions ready. LTE has around 40 frequency bands cross various regions, which makes it difficult to test and support these different bands for various handset OEMs. Unlike 3G, which needed four to six bands, LTE basebands need to be tested in a larger number of bands. OEMs that are launching their first-generation product next year will not get all the bands supported in the first iteration. Incomplete or suboptimal band coverage can greatly reduce the total opportunity of the LTE chipset. As several OEMs get their first silicon out, we think it will be difficult for them to support all the needed LTE bands, and
155
Nomura | U.S. Semiconductors Primer December 11, 2013
hence would make the chipsets deployable only in a subset of regions. Qualcomm, due to its head start, is ahead on LTE band support along with other advanced features. We think this would provide more cushion to Qualcomm’s competitive lead.
Need for a competitive integrated baseband and applications processor further increases the complexity. Delivering the LTE baseband alone is not sufficient to gain share in the mainstream LTE market. Basebands need to be paired with a competitive applications processor design. While the competitive situation in applications processors may improve over time at Broadcom and Intel, we think that Qualcomm has a better applications processor and related IPs in its SoC. Qualcomm designs its own CPUs using licenses from ARM. Broadcom, Intel, and Marvell lack a competitive applications processor roadmap. The move to multi-core support in applications processors makes CPUs a competitive dimension to deliver properly on both power and performance.
Fig. 36: Qualcomm mobile roadmap
1H12 2H12 1H13 2H13 1H14 GOBI LTE 9X35, thin modem QCOM S805 (Premium) Adreno 420, 4K QCOM S800 (Premium) APQ8064 8974 8974Pro MTK (mid range) MT6589 MT6582 MT6593 Quad core QCOM S400/600 (mid range) 8x26 8962 SRRD (mid range) Shark Whale MTK (entry level) MT6589m MT6582m QCOM S200 (entry level) 8225Q 8x12 SPRD (entry level) T-Shark
1H12 2H12 1H13 2H13 1H14 QCOM S400/600 (mid range) 8x60 8x30 8x30AB MTK (entry level) MT6577 MT6572 Dual core QCOM S200 (entry level) 8225 8x10 MTK (low cost) MT6572m SPRD
1H12 2H12 1H13 2H13 1H14 Single core MTK MT6575 (ultra low cost) QCOM 7227A SPRD 7710 RDA RDA8850 1. Numbers in red are LTE chips - MT6593 is MediaTek's first LTE smartpohne chip - Whale is SPRD's first LTE smartphone chip - 7710 is SPRD's first WCDMA smartphone chip. - Shark is SPRD's first TD/WCDMA dual mode quad core chip integrating connectivity - RDA8850 is RDA's first WCDMA smartphone chip Source: Qualcomm, Nomura research
156
Nomura | U.S. Semiconductors Primer December 11, 2013
In the long run, we see Intel and MediaTek vying for the No. 2 spot in LTE Intel’s manufacturing advantage will take time to materialize in mobile TSMC recently indicated that the cost per transistor in the transition to 20nm from 28nm will not fall as much as the transition to 28nm from 40nm. While TSMC’s 20nm product is still an attractive node to customers, we think Intel has an advantage at 14nm from its ability to fully scale the benefits of Moore’s Law. For TSMC, we expect minimal transistor scaling in the transition to 16nm FINFET from 20nm. On the other hand, Intel sees Moore’s Law to continue to work through 10nm, providing Intel an edge over its competitors on applications processor cost and performance. The 22nm Bay Trail chipset represents Intel’s revamped Atom product line with major architectural changes, including out-of-order execution engines, four cores up from one or two cores in the prior chip, and significantly improved graphics engine. Bay Trail chips have substantially improved CPU (higher by 3x) and GPU (higher by 5–6x) performance when compared to Atom chips. In addition to these changes, Intel is pushing the Bay Trail chips into the leading edge of manufacturing process technology (22nm). Intel is also offering high- performance processors at competitive price points ($10–40). Intel needed to enhance the value proposition to compete effectively with ARM-based tablets, which are proving to be better in power and performance than the prior-generation Atom chips. While Intel is focusing on Android tablets as well, we think Windows 8.1-based tablets continue to be a differentiated play. We expect to see Bay Trail devices in the market starting in 4Q13. Intel also has a multi-mode LTE data-only modem that is available for commercial shipments. We think Bay Trail chipsets and LTE modems could gain some traction in the tablet space. Although Samsung Tab 3 10.1 is using Intel's data-only LTE modem, we think Intel's traction in smartphones could take longer. Intel's lack of integrated LTE modem (both data and voice) and the fact that x86 is not native to Android are impediments. While the latter issue may not be a constraint with big vendors, we think Intel will have to guide smaller, white-box players to optimize Android apps on its x86 platform to win their business. Overall, Intel has all the necessary pieces to make inroads, manufacturing edge, a solid applications processor in the making, and an LTE roadmap, but we do not see Intel gaining a meaningful share in smartphones in the near term.
Fig. 37: Cost per transistor – TSMC’s progress Fig. 38: Cost per transistor – Intel’s progress TSMC's cost per transistor decrease will pause at 20/16nm Intel's cost per transistor decrease will continue in 22nm and 14nm
28nm 20nm 16nm FF 10nm 28nm 22nm FF 14nm FF 10nm
Source: Intel, Nomura research Source: Intel, Nomura research
157
Nomura | U.S. Semiconductors Primer December 11, 2013
Intel faces other several issues in fine-tuning its mobile strategy, such as BOM cost reduction Our research suggests that the cost of peripheral chips needed with Intel’s solution is high, making the BOM cost of Intel’s solution unattractive relative to ARM’s solution. In contrast, MediaTek and Qualcomm benefit from a larger support ecosystem for ARM peripheral solutions. Intel is planning to resolve this problem with the launch of an economically priced Bay Trail in 2014. In our checks, most OEM customers indicated that it takes a longer design cycle time with x86 chips than with ARM chips. By our estimate, it takes big customers (with Intel’s full support) at least six months to enable Intel’s AP chips (from design in to mass production) and three months longer if customers also need to enable Intel’s LTE. However, it takes even longer if small customers intend to use Intel’s chips, especially without Intel’s full support. This is a problem particularly when low-price mobile device makers (mainly in China) are used to the turnkey solutions with fast time to market. Last, slow refresh cadence and poor segment coverage also appear to be issues for Intel. Intel is launching one chipset per year for each segment. We think this does not align with the needs of most OEMs, which create multiple chipsets optimized for various handset/tablet segments. If Intel intends to take share in the already-competitive smartphone chip market, we think it needs to shorten its product refresh cadence. Intel also has limited product segmentation. The company segments its offerings in different CPU speeds and GPU specs (by downgrading some specs while all chips are based on the same die). However, it does not provide segmentation by display resolution, memory density, camera pixel, etc., making it impossible to expand to low-price mass volume market. Intel's voice and data LTE chip XMM7260 (LTE-advanced) are scheduled to enter mass production in 2H14. However, in the low-price mass market, the company will not have integrated LTE modem and applications processor SoC until the end of 2014.
Fig. 39: Bay Trail roadmap by price segment over the next one year
Intel Platform Guidance Price Platform Attributes HR'13 1H'14 BTS'14 Display Size: 10" 16-32GB storage Quad Core >$249 WIFI/3G&4G Bay Trail T Bay Trail T Bay Trail T 19x12 IPS or higher Options* 2GB RAM Display Size: 8-10" 16-32GB storage Quad Core $199-249 WIFI/3G&4G Bay Trail T Bay Trail T Bay Trail T Up to 19x12 IPS Options* 1-2GB RAM Display Size: 7-8" 8-16GB storage Dual & Quad Core $149-199 WIFI/3G Clover Trail + Bay Trail T Opt.0 Bay Trail CR opt.2 8x12 IPS Options* 1GB RAM Display Size: 7" 512MB-1GB RAM $99-149 Dual & Quad Core 8GB storage Clover Trail + Clover Trail + Bay Trail CR opt.2 6x10 or lower WIFI only
Source: Nomura estimates and research
158
Nomura | U.S. Semiconductors Primer December 11, 2013
Intel’s current applications processor roadmap is competitive on tablets, but lags in smartphones Intel’s manufacturing process execution is the most advanced in the world. The company launched Ivy Bridge platform for PC/NB segments in 2013 using 22nm tri-gate (3D) process for the first time in the industry. In 2014, we expect Intel to migrate to 14nm by 4Q, leading ARM suppliers by up to three quarters. We think Intel’s tablet offerings will migrate to 14nm (Cherry Trail, Airmont core) with mass availability by 4Q14, which would likely be three quarters earlier than ARM SoCs for tablets using 16nm TSMC node. We think 16nm FF TSMC solutions will enter mass production by 3Q15 (e.g., Apple A9X by TSMC). However, in smartphones, we see Intel and ARM migrating to 22nm/20nm by mid-2014. For LTE, Intel is currently offering Merrifield (22nm Silvermont dual core) with its standalone LTE modem. In 2H14, we expect Intel to start offering its first quad-core AP for smartphones (Moorefield), which would still use 22nm Silvermont core. Due to the long design in process of LTE modem, we do not expect smartphones using the chip to enter mass production before 2015. Intel recently announced that it would launch its first integrated modem (3G) and AP SOC chip (28nm SoFIA) for smartphone application by end-2014. Intel will target the low-end smartphone market with this SOC chip, in our view. In 2015, Intel is expected to launch the next-gen tablet chip (Broxton, Goldmont core) using its leading 14nm process technology. We expect Broxton to start shipping in 2H15, while smartphone chips will finally catch up with the tablet chip schedule by using 14nm (14nm SoFIA LTE) toward the end of 2015. However, we do not expect smartphones using 14nm smartphone chips to ramp for volume production before 1H16. In 2016, we expect Intel to enter volume production for 10nm process node, which would be a full year ahead of TSMC’s plan to manufacture 10nm chips in 2017.
Fig. 40: Intel mobile roadmap
1H12 2H12 1H13 2H13 1H14 2H14 2015 Core Saltwell (32nm) Saltwell (32nm) Saltwell (32nm) Silvemont (22nm 3D) Silvemont (14nm 3D) Silvemont (14nm 3D) Goldmont (14nm 3D)
Tablet Performance Medfield Clover Trail Bay Trail Merrifield Moorefield Broxton Atom Z24xx Atom Z27xx Atom Z23xx 64-bit 64-bit 64-bit *MP in 3Q12 *MP in 4Q12 *Intel's 1st quad core in tablet 14nm Quad core Quad core 2.5-3.5 watts 2.75-3 watts *MP in 4Q13 14nm 14nm 32nm 32nm 2.5-3 watts 22nm LTE Advanced Cherry Trail TD-LTE & TD-SCDMA 64-bit 17 FDD bands, 5 TDD bands 14nm Atom Airmont
Medium / Low Performance
Smartphone Medfield Clover Trail+ Merrifield Moorefield Broxton Atom Z24xx Atom Z25xx 64-bit 64-bit 64-bit *MP in 3Q12 *Intel's 1st dual core in SP 14nm 14nm Quad core 2.5-3.5 watts 3.5 watts Quad core 14nm 32nm 32nm LTE Advanced TD-LTE & TD-SCDMA Lexington 17 FDD bands, 5 TDD bands Atom Z24xx *MP in 1Q13 Medium / Low 32nm SoFIA SoFIA LTE Performance Integrated Global 3G Integrated Global LTE Atom based Atom based *external foundry *external foundry
Source: Intel, Nomura research
159
Nomura | U.S. Semiconductors Primer December 11, 2013
MediaTek, over time, could become a meaningful LTE supplier in China MediaTek has benefited from the smartphone growth in China in a much more significant way than Qualcomm has. MediaTek appears to be on track to ship more than 200mn smartphone chipsets in 2013, and it expects to double that number in 2014. In contrast, we think Qualcomm shipped only 50–60mn smartphone chipsets in China in 2013. Despite aggressive price cuts from Qualcomm, MediaTek’s share at China smartphone makers increased to 50%-plus in 2013 from 45% in 2012 and 15–20% in 2011. We think MediaTek’s turnkey business is one of the key reasons behind this strong share gain. In addition, our analysis suggests that MediaTek has been able to optimize chip costs better than Qualcomm has. We attribute MediaTek’s market share gain to its aggressive cost- optimization efforts. Our calculations show that, by moving to 40/45nm, MediaTek’s chip die size decreased by about 45%, while Qualcomm saw a decrease of only about 30%. We believe that MediaTek has a better cost structure at 28nm as well.
Could MediaTek’s success in 3G translate into success in LTE as well? While MediaTek’s LTE chip is a few quarters away and is expected to launch in 2H14, we think there is a reasonable likelihood that MediaTek could translate its success from 3G into LTE. MediaTek recently indicated that it will spend $1bn over the next year on chipsets and basebands to become competitive with Qualcomm. We think this increase is significant, considering that MediaTek spent around $1bn in basebands (Nomura estimate) in the last three years combined. We think this increase in R&D spending could accelerate MediaTek’s competitive positioning over the next few years. Moreover, MediaTek has plenty of time to ready its chips, as LTE deployment in China is still in its infancy. We think it could take another two to three years before Chinese operators have meaningful LTE deployments. If MediaTek can execute just the way it has in 3G, it could have a sizable share of the LTE market in China.
Fig. 41: Market share in China smartphone market
QCOM MTK SPRD RDA Other 100%
80%
60%
40%
20%
0% 2011 2012 2013E 2014E
Source: Nomura estimates and research
160
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 42: Die size – QCOM vs. MediaTek: from 65nm to 40nm and 28nm MediaTek has done a better job in optimizing costs in low-price smartphone segment 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 65nm 40nm 28nm Single-Core Dual-Core Quad-Core
MTK QCOM
Source: Nomura estimates
Fig. 43: 3G WCDMA smartphone chip roadmap – MTK vs. QCOM, 3Q11–1Q14E
3Q11 4Q11 1Q12 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13 1Q14 Quad core QCOM 8225Q QCOM 8x26 QCOM 8x12
MT6589 MT 6582
Dual core QCOM 8225 QCOM 8x10
MT6577 MT6572
Single core QCOM 7227 QCOM 7227A
MT6573 MT6575
65nm 40/45nm 28nm
Source: Company data, Nomura research
161
Nomura | U.S. Semiconductors Primer December 11, 2013
LTE adoption could drive a multiyear growth in RF components Mobile data usage continues to grow rapidly with the explosion of smartphones and tablets. Cisco is forecasting a 66% increase in monthly mobile data traffic from 2.2EB per month in 2013 to 3.6EB per month in 2014. In addition, adoption of cloud services is driving an increase in mobile traffic. The significant increase in mobile data traffic over the last few years has had major implications on RF components that are used in mobile devices. Increasing adoption of smartphones is creating significant demand for mobile data, driving adoption of LTE across the world. The fragmented nature of LTE frequency bands (around 40) across the world implies a significant increase in cellular RF content. Fragmented operator frequency spectrum is driving the need for aggregating these non-contiguous frequency bands to improve capacity. Carrier Aggregation, which is a feature to address this situation, helps by maximizing utilization of available LTE frequencies. In addition, co-existence requirements in increasingly crowded frequency spectrum are driving more RF contents in mobile devices. For example, Band 38 and 40 at China Mobile sit close to the WiFi frequency band and require more stringent filtering to avoid interference. In addition, increasing mobile data rates affect RF components and circuit complexity. Due to these challenges, OEMs and suppliers are placing an emphasis on the overall efficiency in transmitting data to base stations. The migration to higher data rates (LTE), combined with a move to multi-band power amplifiers, is affecting the efficiency of RF circuits. New solutions such as envelop tracking are likely to improve efficiency by decreasing power consumption. Increasing demand for higher download speeds is driving the need for receiver diversity, MIMO, and carrier aggregation technologies.
Most RF components are likely to see growth from increasing adoption of LTE The market for RF components in wireless handsets is poised to make a substantial growth in revenue as LTE gains traction across the globe. Revenue from RF components is expected to increase mid-single digit to low-double digit for the next few years. Handset OEMs need to determine the optimum number of LTE bands to support, impacting how RF suppliers integrate RF components. While integrating RF components to support more RF operational modes will likely reduce the number of discrete components and hence the ASPs, we think that the overall growth in RF component will more than offset the typical price erosion. LTE adoption is expected to drive a higher level of mode and band integration in mobile devices than in 2G/3G devices. Of the major RF component types, power amplifiers including transmit modules and power amplifier duplexer modules are expected to generate the most revenue. Power amplifier content per handset is expected to be driven by LTE and increasing mix of 3G handsets over the next few years. Switches are also expected to grow driven by these drivers. As the number of power amplifiers increases in a handset to provide multimode and multiband support to the growing number of 3G/4G handsets, we expect increasing switch count to increase per device as well. Duplexers are expected to grow driven by an increase in mix of 3G/4G devices. 2G phones are expected to decline to just 20% of all handsets by 2016, compared with 60% in 2010. Growth in 3G/4G devices requires handset OEMs to adopt duplexers for transmit and receive functions. However, not all manufacturers will use discrete duplexers (Apple).
Avago remains well positioned to take advantage of the industry’s adoption of FBAR filter technology 4G smartphones require more RF bands, boosting Avago’s silicon content. We believe Avago is well positioned with its FBAR filter technology. The company’s differentiated FBAR technology is gaining traction, driven by coexistence issues and tighter spectrum allocation. FBAR-type filters provide superior band isolation compared to SAW-type filters in mobile devices. While priced at a premium, OEMs are demanding superior isolation due to increasing coexistence issues related to the inclusion of multiple radios such as WiFi, GPS, NFC, and multi-mode modems. In addition, cellular spectrum is becoming crowded from higher mobile data demand. This is also driving the need for more stringent band isolation requirements. While most of the bands in 2G phones are
162
Nomura | U.S. Semiconductors Primer December 11, 2013 based on SAW-type filters, several bands in 3G and LTE phones use FBAR-based filters. U.S PCS bands 2 and 25, UMTS band 7, and LTE band 17 are using FBAR- based filters. We believe the upcoming band 13 for Verizon LTE is likely to use FBAR- based filters. CDMA phones also use FBAR filters. For the next few years, we think Avago will benefit from FBAR filters in LTE bands, carrier aggregation beginning with AT&T in 2014, and envelop tracking in LTE power amplifiers. Avago is in a strong position for FBAR products, as it currently ships almost all of the FBAR-based filters. While TriQuint and Skyworks are working on their second- and third-generation FBAR- based filters, respectively, Avago is on its tenth generation of optimization in FBAR- based filters. In addition, Avago recently quadrupled its FBAR manufacturing capacity to meet the expected demand from tier-1 customers (Apple and Samsung).
Fig. 44: Global mobile data traffic growth, 2012–2017E
Source: Cisco Systems, Nomura research
163
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 45: Forecasts for band combinations needed to support LTE in different regions
Source: RF Micro, Nomura research
164
Nomura | U.S. Semiconductors Primer December 11, 2013
RF360 – A disruptive technology? In early 2013, Qualcomm introduced the RF360 Front End Solution to address cellular radio frequency band fragmentation, which enables a single SKU for a global 4G LTE design. With more than 40 LTE bands worldwide, designing a handset to support these bands is a complex challenge. For instance, the iPhone 5S comes in four versions, and the Galaxy S4 has multiple versions with separate set of band support. This means that LTE will not be available outside the supported RF bands outside the home market because each carrier uses different frequencies. Qualcomm RF360 solution would allow handset makers to build a single hardware design to address every (or most) LTE markets. RF360 tries to address this issue by combining a family of RF chips that is expected to help OEMs in developing multiband, multimode mobile devices supporting all cellular modes. RF360 solution includes envelope power tracker for 4G LTE mobile devices, a dynamic antenna matching tuner, an integrated power amplifier-antenna switch, and a 3D-RF packaging solution incorporating key front-end components. The RF360 solution is designed to reduce power consumption while reducing the RF front- end footprint inside of a smartphone by up to 50% compared to the available solutions in the market right now. The RF360 front end solution comprises the following components: • Dynamic Antenna Matching Tuner (QFE15xx) is a modem-assisted and configurable antenna-matching tuner that extends antenna range to operate over 2G/3G/4G LTE frequency bands from 700–2700 MHz. • Envelope Power Tracker (QFE11xx) is modem-assisted envelope tracking technology designed for 3G/4G LTE mobile devices. This chip is designed to reduce overall thermal footprint and RF power consumption by up to 30%, depending on the mode of operation. • Integrated Power Amplifier / Antenna Switch (QFE23xx) – The integrated CMOS power amplifier (PA) and antenna switch provide multiband support across 2G, 3G, and 4G LTE cellular modes. This takes a smaller PCB area and has a smaller PA/antenna switch footprint. • RF POP (QFE27xx) is a 3D RF packaging solution that integrates the QFE23xx multimode, multiband power amplifier and antenna switch, with all the associated SAW filters and duplexers in a single package. Designed to be easily interchangeable, the QFE27xx enables OEMs to change the substrate configuration to support global or region-specific frequency band combinations.
Fig. 46: RF360 solution from Qualcomm
Band Band
Band Band
Antenna Envelope Matching Power Tuner Tracker
Power Amp RF & Antenna POP Qualcomm Switch RF360 *Half the size
Source: Qualcomm, Nomura research
165
Nomura | U.S. Semiconductors Primer December 11, 2013
We think RF360 is largely a power amplifier play, but will take time to gain traction We expect traditional GaAs-based amplifiers to maintain their dominance Qualcomm’s RF360 solution includes multiband, multimode CMOS power amplifiers capable of transmitting LTE signals, as well as 3G. This was previously largely achieved by GaAs-based power amplifiers. Shipments of GaAs-based multimode, multiband power amplifiers have grown rapidly over the past three years, primarily in smartphones. Qualcomm’s product appears to be a challenge to the traditional power amplifiers. The company is leveraging the benefits of envelope tracking, a new technology that improves efficiency of power amplifiers. It remains to be seen if envelop tracking can make CMOS power amplifiers successful in smartphones. Last year, CMOS power amplifiers accounted for less than 10% of the power amplifier market. While CMOS power amplifiers could significantly reduce cost, we think GaAs power amplifiers are extremely linear and efficient, and suppliers have driven costs down for more than a decade. In addition, we think solutions similar to RF360 could easily be built using traditional power amplifiers. CMOS power amplifiers had moderate success in the 2G handsets due to performance concerns. We think that a long battery life and a small footprint are critical in mid- to high- end smartphones, where traditional power amplifiers have performed well. Traditional power amplifiers will likely retain their dominance in the mid- to high end of the 3G/4G market. However, we think launch of RF360 could result in a closer partnership of RF vendors and chipset suppliers. For example, we believe Avago is already working with Qualcomm to fine-tune its power amplifiers with Qualcomm’s envelop tracker and baseband chips. Even with the gains from envelop tracking (ET) feature, we think it would take Qualcomm a few years to catch up in performance. That said, we think Qualcomm’s entry in the RF space could bring further adoption of CMOS- based power amplifiers. Another appeal of this solution is a higher level of integration in the form of a single chip that integrates the RF front end with the transceiver. While this integration aspect of CMOS power amplifiers is appealing, we think it would likely come at the cost of performance and efficiency.
166
Nomura | U.S. Semiconductors Primer December 11, 2013
Apple’s 64-bit support will likely increase competition among ARM SoC vendors Apple A7 is the first mobile chip with 64-bit support Apple recently announced its new applications processor, A7, which is used in iPhone 5S and iPad Air. Based on a custom ARMv8 core, A7 is the first mobile processor that supports 64-bit instruction set. Apple A7 has a completely revamped architecture with peak instruction issue width of six instructions, which appears to be more than twice the width in Apple’s prior generation cores and in Qualcomm’s Krait. In addition, it appears that Apple A7 chip has the widest ARM architecture of any shipping mobile chip at this time. As a result, A7 likely has a significantly reduced memory latency and increased bandwidth. Recently, Qualcomm announced the support for 64-bit in its entry-level S410 chipsets. We expect Qualcomm to include 64-bit support in most of its chipset tiers next year. Intel already has 64-bit support in its Silvermont-based chipsets. We believe other mobile chip suppliers such as Nvidia (Project Denver) and Samsung are also working on the 64-bit versions of their mobile chips. We believe next year many new Android devices will likely support 64-bit capability. Why 64-bit? We think 64-bit transition will help in enabling support for desktop- class applications in mobile devices We think support for 64-bit has less to do with the support of more than 4GB of physical memory. This can easily be achieved in the existing ARM chips using address bit extension on Cortex-A15 and Cortex-A7. In server applications, OS and application software are frequently 64-bit, so addressing 64-bit is critical in these applications. We think support for 64-bit in ARMv8 will enable ARM processors to become more broadly deployed in server and desktop applications, and will provide future-proof support for the eventual migration of 64-bit operating systems to mobile applications. Apple recently demonstrated support for iWork applications for iPads, where consumers can use traditionally desktop class 64-bit applications such as photo editing. Eventually, we expect both Apple iOS and Android/Chrome to move up in the application/device ecosystem and provide more support for applications that typically run on traditional PCs. We think Apple’s 64-bit transition with the use of A7 chip will accelerate the industry’s transition to enable desktop class applications in mobile devices. We expect mobile chip competition to increase on 64-bit support While we do not see 64-bit changing the competitive landscape in the smartphone space, we believe support for 64-bit in mobile chips for Android tablets could be a competitive advantage for mobile chip vendors in 2014, as we do not see a lot of advantages of 64-bit in smartphones. While some vendors may choose to use the same 64-bit silicon in smartphones for cost reasons, we think the real utilization of 64-bit would be in tablets. Consequently, we see more adoption of ARMs A50 series chip architecture in the coming periods in tablets and other mobile computing devices. ARM’s A50 series chips are based on ARMv8 instruction set with 64-bit support. In terms of competition, we see Intel and Qualcomm among the first vendors supporting 64-bit for Android tablets next year. We believe Nvidia and Samsung will also launch 64-bit devices next year. That said, we note that Google has not yet announced support for 64-bit for Android. However, we think this support would be available in Android devices in 2014, which should boost the availability of desktop class applications on mobile devices.
167
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 47: ARM v8 architecture will support 64-bit
Source: ARM Holdings, Nomura research
168
Nomura | U.S. Semiconductors Primer December 11, 2013
3G ASPs will prove more resilient than investors fear 3G ASPs are a significant driver of Qualcomm’s earnings Qualcomm Licensing business (QTL) accounts for approximately one-third of total sales but contributes to around two-thirds of the company’s earnings. In 2013, we expect QTL to generate $8bn (31% of sales) in licensing revenue and contribute around $2.72 (61% of total earnings) in EPS. Qualcomm’s licensing business is driven by two key variables: growth in 3G units and growth in 3G device ASPs. We expect Qualcomm’s royalty coverage and royalty rate to be stable over the next several years. While there is less debate on 3G unit growth, which consensus believes will grow in double digits over the next several years, growth (or decline) in 3G device ASPs has been a point of debate. This is due to the fact that Qualcomm ASPs are a significant driver of Qualcomm’s royalty revenue. We estimate that a 5% decline in 3G device ASPs from midpoint of the guidance of $229 for CY13 to $217 will lead to a $0.14 decline in EPS or a 3% decline in the company’s overall EPS.
Fig. 48: Sensitivity table for 3G device ASP vs. QTL EPS
3G Device ASP
2.70 $206 $217 $229 $240 $252
12% $2.34 $2.47 $2.61 $2.74 $2.87
15% $2.40 $2.53 $2.66 $2.80 $2.94
17% $2.45 $2.59 $2.72 $2.86 $3.00
20% $2.50 $2.64 $2.78 $2.92 $3.06
3G Device Growth 3G 22% $2.55 $2.70 $2.84 $2.98 $3.13
Source: Nomura estimates
Investors are concerned that 3G device ASPs could see a meaningful drop The key concern for ASP declines center around the growth in low-cost smartphones (less than $200 ASP). Gartner expects penetration of low-cost smartphones to increase meaningfully in all regions. Low-cost smartphone shipments are expected to double in 2013, driven by APAC region, which alone is expected to contribute more than 100 million low-cost smartphones in 2013. Low-cost smartphones are expected to account for approximately 5% of total handset shipments in 2011 to more than 20% of total shipments in 2013E. Gartner forecasts that low-cost phones will account for around 30% of total smartphone shipments in 2015. Consequently, investors are concerned that increasing penetration of low-cost smartphones will likely affect 3G device ASPs.
Fig. 49: Low-cost smartphone shipments by region Fig. 50: Low-cost smartphone as % of total handset shipments
350 1,250 100%
300 1,000 80% 250
200 750 60% 150
100 (millions) 500 40% Shipments (millions) 50
250 20% Low-Cost Share Smartphone 0 Shipments Smartphone Total 2010 2011 2012 2013E 2014E 2015E 2016E Asia Pacific Latin America Middle East & Africa Eastern Europe 0 0% Western Europe North America 2010 2011 2012 2013E 2014E 2015E 2016E
Source: Gartner, Nomura research Source: Gartner, Nomura research
169
Nomura | U.S. Semiconductors Primer December 11, 2013
3G ASPs can prove to be more resilient than feared Since 2010, Qualcomm’s 3G device ASPs have steadily increased from $176 to $229 in 2013E. Qualcomm’s recent ASP guidance largely backed our view and guided ASPs down 1% for FY14. We believe that the Street is modeling a larger decline of mid-single digits. The key reasons we think 3G ASPs will prove more resilient are as follows:
We see a limited impact from low-cost smartphone growth in China to the overall mix of royalty bearing 3G devices this and next year. The substantial growth in China does not appear to be affecting Qualcomm’s mix. We forecast total smartphone shipments globally increasing 25–30% from 950m units in 2013 to 1,200m in 2014. IDC forecasts shipments in China to increase 100m units to 450m in 2014, accounting for 40% of total incremental volumes globally. However, we do not expect the growth from China to affect Qualcomm’s licensing business, as much of the growth will come from TD-SCDMA devices. TD-SCDMA is a proprietary 3G standard in China, and Qualcomm does not collect royalties on this technology. Gartner forecasts TD-SCDMA phones in China to grow from 65mn units in 2012 to 137mn in 2013. In contrast, royalty bearing 3G phones in China are forecast to increase 21% in 2013.
Fig. 51: 3G device ASP trend for Qualcomm
$250
$200
$150
$100
$50
$0 2009 2010 2011 2012 2013E
Qualcomm ASPs
Source: Company data, Nomura estimates
Fig. 52: China device shipments by technology, 2011–2016E Fig. 53: China device shipments by segment, 2011–2016E
600 120%
500 100%
400 80%
300 60%
200 40%
100 20%
0 0% 2011 2012 2013E 2014E 2015E 2016E 2011 2012 2013E 2014E 2015E 2016E
2G+ TD-SCDMA 3G+ High-end Mid-range Voice phones
Source: Gartner, Nomura research Source: Gartner, Nomura research
170
Nomura | U.S. Semiconductors Primer December 11, 2013
ASP impact from transition of feature phones to mid-/high-end phones is less understood. According to the latest handset report from Gartner, blended ASPs for handsets are expected to be up in six of eight regions. While this may appear to be a contrast to the growth of low-cost smartphones, we think this makes sense because of the transition of feature/voice phones to higher-end devices. In 2012, proportion of high- end phones grew from 21% to in 2011 to 27%, largely at the expense of mid-range phones. It appears that the mix of high-end phones will increase from 27% in 2012 to around 30% in 2013, largely due to decrease in feature phones (down 400bps). While this mix shift looks small, the ASP impact is not because of a significant (three times) price difference between the three handset segments.
Fig. 54: Blended ASPs for handset by region, 2012–2013E
$400
$350
$300
$250
$200
$150
$100
$50
$0 E. Europe Lat. Am. NA W. Europe Mature Greater Emerging Middle East APAC China Asia & Africa
2012 2013E
Source: Gartner, Nomura research
Fig. 55: Worldwide handset segment split, 2011–2016E
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0% 2011 2012 2013E 2014E 2015E 2016E
High-End Mid-Range Voice phones
Source: Gartner, Nomura research
171
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 56: Blended ASPs across handset segments, 2011–2016E
Segment ASPs ($) 2011 2012 2013E 2014E 2015E 2016E High-end ASP 356 424 406 384 371 361 Mid-range ASP 105 97 92 88 85 82 Voice phone ASP 30 27 28 28 27 26
Blended ASP (Overall) 130 162 165 168 173 177 ASP growth (y/y) - 24% 2% 2% 3% 2% Source: Gartner, Nomura research
Fig. 57: Blended ASPs could stay flat to up over the next few years
450 200
400 180
350 160 140 300 120 250 100 200 80 150 60 100 40 50 20 0 0 2011 2012 2013E 2014E 2015E 2016E
Blended ASP (RHS) High-end ASP Mid-range ASP Voice phone ASP
Source: Gartner, Nomura research
172
Nomura | U.S. Semiconductors Primer December 11, 2013
Handset ASP trends for OEMs that pay royalties to Qualcomm support our thesis. We think there is less clarity on royalty coverage issues among China handset OEMs. MediaTek’s recently amended its agreement with Qualcomm, whereby it would no longer provide details of its OEMs customers or shipments to Qualcomm. We think this will make it difficult to understand Qualcomm’s actual royalty coverage in China. However, if we look at the ASP data from key handset OEMs that are known to pay royalties to Qualcomm, we find that the blended ASPs are up for the last four years (2010–13E). Blended handset ASPs of these OEMs is expected to increase from $141 in 2010 to $248 in 2013E. We expect a majority of handset shipments of these OEMs (Apple, Samsung, HTC, LG, Sony, Nokia, and Blackberry) to be royalty-bearing devices. As such, we expect these OEMs to contribute to a majority of the royalty bearing devices for Qualcomm’s licensing revenue. These OEMs are expected to collectively ship more than a billion phones (55% of total handset shipments and almost 100% of MSM shipments of 1081mn in 2013E).
Fig. 58: Total units and blended ASPs of key handset OEMs
1,200 $300
1,000 $250
800 $200
600 $150 Blended ASP Blended 400 $100 Handset Units (mn) Units Handset
200 $50
0 $0 2010 2011 2012 2013E
Units Blended ASP
Source: Nomura estimates
Fig. 59: Units, ASP, and blended ASP of key handset OEMs 2010 2011 2012 2013E Apple Units (mn) 47 89 129 139 ASP $640 $669 $662 $596 HTC Units (mn) 25 43 32 32 ASP $346 $314 $299 $282 LG Units (mn) 114 86 58 64 ASP $97 $115 $152 $219 Nokia Units (mn) 461 422 334 249 ASP $84 $79 $60 $65 Blackberry Units (mn) 47 52 34 31 ASP $307 $289 $238 $244 Samsung Units (mn) 279 314 385 467 ASP $116 $145 $195 $238 Sony Mobile Units (mn) 42 33 31 39 ASP $199 $223 $212 $315 Total Units 1,015 1,039 1,003 1,021 Blended ASP $141 $177 $213 $248
Source: Nomura estimates
173
Nomura | U.S. Semiconductors Primer December 11, 2013
We expect these trends to remain steady over the next one to two years. We continue to expect the proportion of high-end devices to increase as a percentage of total shipments in the next few years. We also see features phones converting into higher-tier phones, providing support for a decline in handset ASPs. Overall, we think 3G device ASPs will stay more resilient than many investors fear for the next one to two years.
Fig. 60: We expect 3G units to grow in double digits, 2013–15E
4,000 30%
3,500 25% 3,000 20% 2,500
2,000 15%
1,500 10% 1,000 5% 500
0 0% 2011 2012 2013E 2014E 2015E 2016E 3G+ units 2G+ units 3G+ growth (YoY)
Source: Gartner, Nomura research
174
Nomura | U.S. Semiconductors Primer December 11, 2013
Programmable Logic Devices (PLDs) Logic devices provide specific functions, including device-to-device interfacing, data communication, signal processing, and almost every other function a system must perform. Logic devices can be classified into two broad categories – fixed and programmable. As the name suggests, the circuits in a fixed logic device are hard-wired, i.e., they perform a specific set of functions. Once fabricated, fixed-function devices cannot be changed. Fixed logic can be further split between application-specific integrated circuits (ASICs) and application-specific standard products (ASSPs). ASICs are devices that are designed for one application for one specific customer, e.g., Apple’s A6 application processor. ASSPs are devices that are designed for one application for multiple customers, e.g., Broadcom’s WiFi combo connectivity solutions. On the other hand, programmable logic devices (PLDs) are standard, off-the-shelf parts that offer customers a wide range of logic capacity, features, speed, and voltage characteristics—and these devices can be reconfigured to perform different set of functions.
PLD market PLD is a relatively small part of the logic market. In 2012, the worldwide market for programmable logic devices was about $4.4bn, as compared to the logic market of $82bn, according to SIA. Over the past 10 years, the PLD market grew at a CAGR of 7%, which is roughly in line with the semiconductor market at a CAGR of 8%. In contrast, the ASIC market grew at a CAGR of 2%, while the ASSP market grew at a CAGR of 8%.
Fig. 1: PLD Revenue and Growth Trends, 2003–2012 in $mn 6,000
5,000
4,000
3,000
2,000 PLD Market $mn) (in Size PLD
1,000
0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Source: Gartner, Nomura research
175
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 2: Over Last 10 Years, PLD Growth Has Been in Line with Overall Semi Growth
9.0%
8.0%
7.0%
6.0%
5.0%
4.0%
3.0%
2.0%
1.0%
0.0% Ov erall Semi ex-Memory ASIC ASSP PLD
10-y r CAGR
Source: Gartner, SIA, Nomura research
Problems with fixed logic Fixed logic is ideal for designs that are in high volume, as fixed costs are spread over a large number of units, or for designs with functions that are well-defined or standardized. While the majority of the logic market is implemented with fixed logic, the time required from design to prototype and then production can take several months to more than a year, depending on the complexity of the design. Implementing fixed function devices involves more risk than implementing the design in a PLD. If an ASIC does not work properly, or if the requirements change, a new chip must be redesigned. The up-front work of designing and verifying fixed logic devices involves substantial "non-recurring engineering" costs, or NRE. NRE represents all the costs a system designer incurs each time a new ASIC is designed and includes expensive mask sets for manufacturing the various metal layers of the chip and the cost of initial prototype devices. These NRE costs can run from a few hundred thousand dollars to several million dollars depending on the process node targeted for the ASIC.
PLDs offer fast time-to-market PLDs provide a faster time-to-market and are more suitable when logic changes frequently or if equivalent logic is less economical to implement in fixed logic (ASIC or ASSP) applications. PLD tools allow designers to quickly develop, test, and realize their designs. The design then can be programmed into a PLD and immediately tested as a live circuit. The PLD used for this prototyping is usually the same PLD used in the final production. There are significantly less NRE costs, and the final design is completed much faster than that of a custom, fixed logic device. Another key benefit of PLDs is that during the design phase customers can change the circuitry as often as they want. That's because PLDs are based on re-writable memory technology—to change the design, the device is simply reprogrammed. Once the design is final, customers can go into immediate production by simply programming as many PLDs as they need with the final software design file. This flexibility, however, comes at a cost. PLDs generally have large die size and therefore are much more expensive than fixed logic devices for high-volume applications.
176
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 3: Difference between ASICs, ASSPs and PLDs
ASIC ASSP PLD Customizable Yes, by fab No Yes, by end user Reprogrammability No No Yes Relative time to market Slow Immediate Fast Relative unit cost Low Moderate High Customers' development cost High Low Moderate Field upgradability No No Yes
Source: Nomura research
Types of PLDs Programmable logic devices can be further divided into two sub-segments: (1) Field Programmable Grid Arrays, or FPGA, and (2) Complex Programmable Logic Devices, or CPLD. Because of product differences, FPGAs and CPLDs do not compete directly. The FPGA market has outgrown the CPLD market in the past few years, and should continue to be the fastest-growing segment in PLD. We estimate 80–90% of PLD market revenue is derived from FPGAs today.
FPGA (Field Programmable Grid Arrays) FPGAs offer the highest amount of logic density, the most features, and the highest performance. FPGAs today provide high densities of more than ten million logic gates and also offer embedded processors like ARM or PowerPC, inbuilt memory, clock management, and support for high-speed interconnect technologies. FPGAs are used in a wide variety of applications ranging from data processing and storage to instrumentation, telecommunications, and digital signal processing.
CPLD (Complex Programmable Logic Device) CPLDs, by contrast, offer much smaller amounts of logic – up to about 10,000 gates. But CPLDs offer very predictable timing characteristics and are therefore ideal for critical control applications. CPLDs require extremely low amounts of power and are very inexpensive, making them ideal for cost-sensitive, battery-operated, portable applications such as handhelds and consumer devices.
Fig. 4: Basic FPGA Architecture
Source: EEWeb.com, Nomura research
1 77
Nomura | U.S. Semiconductors Primer December 11, 2013
PLDs have potential to displace ASIC and ASSP Rising costs of ASICs and ASSPs The cost of each successive generation of process technology is rising exponentially. For example, we estimate that the cost of a 45nm SoC design is $80mn and the cost of 32nm SoC increases to about $130mn. For a 32nm design, we estimate a revenue opportunity of around $500mn is required for the investment to pay off. At the 28nm process node, the breakeven revenue required is even higher. Xilinx estimates that, at 16/14nm, the breakeven revenue will surge to $1.5–3.0bn. It is therefore no surprise that the number of ASIC and ASSP design starts has declined over time. Since 2000, the number of design starts has come down 70% for ASICs and 40% for ASSPs. Gartner forecasts that design starts will continue to decline by 2–3% per year over the next five years.
Fig. 5: ASIC and ASSP Design Starts
9000
8000
7000
6000
5000
4000
3000
2000
1000
0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E 2015E 2016E 2017E
ASIC Design Starts ASSP Design Starts
Source: Gartner, Nomura research
Widening process technology gap Furthermore, there seems to be a widening gap between the technology processes used by PLDs and ASICs. In 2003, the primary design node for both PLD and ASIC suppliers was 130nm. Today, PLDs are two to three generations ahead. Specifically, the primary process node for ASICs is currently 90nm and moving to 65nm in the next 1–2 years, while the primary process node used by PLDs is 40/45nm and moving to 28/32nm. While PLDs generally have larger die size than ASICs given that more logic elements are used, the cost of each logic element is mitigated by this use of a more advanced process node, making PLDs more cost competitive with ASICs.
178
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 6: ASIC/ASSP Design Starts by Process Node, 2012
28/32nm 350nm+ 3% 9% 40/45nm 250nm 13% 3% 180nm 6%
65nm 130nm 22% 19%
90nm 25%
Source: Gartner, Nomura research
PLDs benefit from an expanding addressable market $16bn served addressable market by 2016 Given their high cost relative to ASICs, PLDs historically were used in prototyping activities or applications involving relatively small volumes. Communications infrastructure was the primary target market. With their widening process technology gap and hence cost competitiveness, PLDs are now finding more success in other markets including industrial, automotive and consumer. Xilinx estimates that its served available market will grow to $16bn in 2016, as the core PLD market continues to expand to $6bn, while new opportunities to displace ASIC and ASSP could reach $8bn and the embedded market could be another $2bn of revenue opportunity.
Fig. 7: Large Addressable Market
2016 Xilinx SAM
Embedded $2B SAM ASSP ASIC ASIC/ASSP Displacement $8B SAM
Core PLD $6B SAM PLD >$16B
2016 Serviceable
Source: Xilinx,
179
Nomura | U.S. Semiconductors Primer December 11, 2013
PLD application market led by communications PLDs address a variety of end markets. In 2012, the largest end market for PLDs was communications, which represented 47% of total PLD revenue. Within this segment, wired communications was 25% of total revenue and wireless was 22%. However, wireless has been growing at a much faster rate (5-year CAGR of 15%) than wired communications (5-year CAGR of 1%). Industrial/medical is the second-largest end market for PLDs at 22% (5-year CAGR of 5%), followed by military/aerospace at 13% (5- year CAGR of 7%), and consumer at 9% (5-year CAGR of 5%). Automotive was only 3% of total PLD revenue but grew at a CAGR of 10%, while computing was 5% but declining at a 5-year CAGR of 3%.
Fig. 8: PLD Revenue Trend by End Market, 2008–2012
6,000
5,000 Military/Aerospace 4,000 Industrial/Medical Auotmotive
3,000 Consumer Wir eles s Comms
2,000 Wired Comms
PLD revenuePLD (in $mn) Computing Storage 1,000
0 2008 2009 2010 2011 2012
Source: Gartner, Nomura research
Fig. 9: PLD Share of Revenue by End Market, 2012
Military/ Storage Computing Aerospace 3% 3% 13%
Wired Comms 25%
Industrial/ Medical 22%
Auotmotive 3% Consumer Wireless Comms 9% 22%
Source: Gartner, Nomura research
180
Nomura | U.S. Semiconductors Primer December 11, 2013
The chart below shows the major vertical markets for PLDs and typical applications within each vertical.
Fig. 10: Major Vertical Markets and Applications for PLDs
Vertical Market Applications Communications Transmission, Access, Cellular Infrastructure, WLAN, Microwave Industrial Process control, security, energy Automotive Driver assistance, Infotainment Military Secure communications, Radar Networking Routers, Switches Computer Servers, Mainframes Storage Storage systems, Storage Area Networks Office Automation Copiers, Printers Broadcast Studio, Audio/Video Consumer Set-top decoder boxes, HDTV Medical Diagnostic imaging
Source: Altera, Xilinx, Nomura research
Fig. 11: PLD 5-year CAGR by End Market, 2008–2012
20%
15%
10%
5%
0%
-5%
-10% Wireless Wired Data Processing Data Industrial/Medical Storage Automotive /Aerospace Military Total Computing Consumer
5-y r CAGR
Source: Gartner, Nomura research
181
Nomura | U.S. Semiconductors Primer December 11, 2013
However, PLDs have not outgrown the semiconductor market In theory, the improving cost competitiveness and expanding addressable market should allow PLDs to outgrow the semiconductor market. Both Altera and Xilinx have been talking about a shift to FPGAs for the past decade (see Fig. 12). However, that has not been the case. Over the past 10 years, PLDs grew at a CAGR of 7%, which is comparable to the semiconductor industry (CAGR of 8%). In fact, PLDs only outgrew the semiconductor industry (excluding memory) in three years (2006, 2008 and 2010) and the logic segment in two years (2006 and 2010). While it may see counterintuitive, our discussions with leading networking and infrastructure OEMs suggest that ASICs do not need to be on the most advanced nodes to provide sufficient integration and performance because implementations are more targeted and custom-oriented. In addition, FPGAs generally have lower silicon utilization as compared to ASICs. In other words, there are additional gates that are required to make an FPGA programmable. Therefore, in order to be competitive to ASICs/ASSPs in logic densities, FPGAs need to be on a more advanced node. Today more than 45% of ASIC starts are on 90/130nm. We see ASIC designs moving to 40/45nm over the coming year, minimizing the impact from 28nm.
Fig. 12: Altera and Xilinx over the Years
April 2003, Altera Q1 2003 Earnings Call "With the increase in cost of ASIC design and the decreasing R&D spend of ASSP, it has been clear to us for some time that PLDs would replace ASICs and ASSPs at an accelerating pace… Customers have confirmed for us for some time that their ASIC starts were decreasing and they were transitioning engineers to FPGA design." April 2006, Altera Q1 2006 Earnings Call "The biggest opportunity for growth for programmable logic is in replacing ASICs and ASSPs, I mean we started saying that five years ago and that remains the big opportunity." October 2007, Altera Q2 2008 Earnings Call "In the old days, FPGAs were only used for prototyping or for very low volumes in the high volume market, which still is hundreds thousands of units. But now we really see, thanks to the cost improvements driven by 90-nanometer and further cost reductions in the 90-nanometer product lines, the Spartan-3 product line, we see more and more acceptance of FPGAs even in relatively high volume production, millions of units type of volume production." January 2008, Altera Q4 2007 Earnings Call "PLDs in contrast have aggressively adopted new process generations enabling higher density and performance devices with lower costs to continually replace prior generation ASIC and ASSPs… We estimate today that we are two generations of process technology ahead of the mainstream ASIC industry moving to three generations this year." November 2009, Altera Analyst Meeting "On the technology side, at 40-nanometer where we opened a 3 generation gap over most ASICs. We definitely see the tipping point, the replacement of ASICs and ASSPs is accelerating." February 2010, Xilinx Analyst Meeting "We continue to see evidence to the programmable imperative. That continues to grow quarter-by-quarter. We see more customers shifting to an architecture that’s driven by FPGAs versus the alternatives." April 2011, Altera Q1 2011 Earnings Call "The rising costs of semiconductor design continue to favor programmable products, with the tipping point well underway… We expect to continue to outgrow our customers as we displace ASICs and ASSPs in the infrastructure markets." February 2012, Xilinx Analyst Meeting "So programmable imperative... is driven by an economic phenomenon with each generation's technology developing a product from scratch increases in terms of the cost, the risk and the time. And as a result, ASICs and ASSPs are less and less viable... And this trend... is accelerating." November 2013, Altera Analyst Meeting "First message is that if we look at what's happening from a cost perspective and a cost is both from a cost of development, as well as the increasing cost of transistors, what we find is the ASIC and ASSP models are under pressure, very difficult for companies offering these products to achieve the expected return on investment from an investor. So from many markets we see that these technologies and products are actually retreating; and as such we have a great opportunity to continue to come in and replace what are very large markets much larger than programmable logic, replace them with our products and continue to grow."
Source: Company data, Nomura research
182
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 13: Revenue Growth Comparison
Revenue (in $mn) SIA ex-memory Logic PLD* 2000 155.2 34.6 2.9 2001 114.1 24.3 2.0 2002 113.7 31.3 1.8 2003 133.9 36.9 2.1 2004 165.9 49.3 2.6 2005 179.0 57.7 2.8 2006 189.2 60.2 3.2 2007 197.8 67.3 3.1 2008 202.3 73.5 3.3 2009 181.5 65.2 2.9 2010 228.7 77.4 4.3 2011 238.8 78.8 4.3 2012 234.6 81.7 4.0
Revenue YoY SIA ex-memory Logic PLD 3-year CAGR 9% 8% 11% 5-year CAGR 3% 4% 5% 7-year CAGR 4% 5% 5% 10-year CAGR 8% 10% 8%
* PLD here includes only ALTR and XLNX, and thus the 10-year PLD CAGR (8%) differs from Gartner's calculation (7%) Source: SIA, Gartner, Nomura research
We think ASICs continue to be more cost effective for volume production In 2013, Altera’s customer Huawei converted a high-volume PLD design into an ASIC. While we expect PLDs to outgrow ASICs and ASSPs combined long term, we do not subscribe to the view that the 28nm technology node will cause a tipping point in FPGA adoption. Both Altera and Xilinx have claimed that 28nm FPGAs could cause a tipping point in the industry in favor of FPGAs. However, our research indicates that this will not be the case. While 28nm FPGAs will enable more IP integration and faster data-rate applications, we believe the relative attractiveness of ASICs/ASSPs on power, performance and cost is not changing meaningfully. The companies also claim that increasing NRE costs for ASICs at more advanced nodes will tip the balance in favor of FPGAs. While ASIC costs are increasing, system manufacturers such as Cisco tell us that a similar level of performance is achievable with an ASIC designed at two generation-older nodes (i.e., 65nm, 90nm). As such, we believe the 28nm transition of FPGAs is merely an evolutionary step that keeps pace with increasing integration and the availability of new processes similar to those of an ASIC.
183
Nomura | U.S. Semiconductors Primer December 11, 2013
The biggest driver for PLDs is carrier equipment spending Outperformance in 2010 driven by 3G wireless infrastructure In 2010, PLDs (up 47% yoy) significantly outgrew the semiconductor industry (up 26%) and the logic market (up 19%). We attribute the outperformance to sharp capex cuts (down 16% yoy) by telco operators during the financial crisis in 2009. As the economy recovered, telco operators in China (China Telecom, China Mobile), Europe (Deutsche Telekom, Vodafone), and North America (Verizon, AT&T) raced to build out 3G base stations in 2009 and 2010 in order to manage an increasing number of subscribers and to service an explosion in mobile data usage. Both 3G and 4G base-station spending increased from 25% of mobile infrastructure spending in 2008 to 50% in 2010. A sharp increase in 3G base station capex pressured equipment manufacturers to minimize time-to-market. In addition, 3G standard in its early years was less stable. As such, vendors such as Ericsson and Huawei chose to employ PLDs, which enable faster design cycles than ASICs and ASSPs and are more flexible to accommodate changes in the radio standards. In addition, these deployments were largely first- and second-generation 3G base stations which required flexibility in order to incorporate design changes and operator-specific requirements, dynamics that also favored PLDs over other components. Furthermore, Altera and Xilinx experienced a two-fold increase in PLD silicon dollar content per 3G box compared with 2G, due to higher bandwidth requirements. Over a six-quarter period from June 2009 to September 2010, Xilinx’s communications revenue increased 61% and Altera’s telecom and wireless business increased 77%. Beyond 2013, Gartner is forecasting the carrier capex to grow in low-single digit declining to a flat growth in 2016. Carrier capex is expected to grow 3% in 2014 and 2% in 2015. We think PLD companies should benefit from the moderate growth in the carrier capex over the next few years.
Fig. 14: Carrier Service Provider Capex Trends, 2010–2017E
$370 7%
$360 6%
$350 5% $340 4% $330 3% $320 2% $310 1% $300 0% $290
$280 -1%
$270 -2% 2010 2011 2012 2013E 2014E 2015E 2016E 2017E
Capex ($ in bn) Growth
Source: Gartner, Nomura research However, future rounds of 3G capex unlikely to boost PLD revenue We expect the next round of 3G capex will be less beneficial to PLDs, as equipment vendors reduce PLD content in base-station offerings. We think there was a programmable imperative in first- and second-generation 3G base stations, driven by a need for fast time-to-market and design flexibility. However, 3G base-station deployments today represent third- and fourth- generation rollouts, and within these systems manufacturers are shifting to ASICs and ASSPs in order to reduce cost. We estimate that FPGAs cost $1,000–3,000 per device versus $100–150 for an ASIC/ASSP (after NRE costs). While LTE base station is an opportunity for PLD vendors, we think LTE standard is lot more stable than the 3G standard at this point in adoption. We believe LTE base stations, on a relative basis, will see a larger ASIC/ASSP content than the 3G base stations in the last generation.
184
Nomura | U.S. Semiconductors Primer December 11, 2013
4G/LTE spending should improve, but relative to 3G may be more drawn out 4G rollout is in early innings We believe that LTE infrastructure is in the early innings of deployment. Currently, LTE infrastructure deployment is mostly happening at China Mobile (200k LTE base stations), but we expect more spending to come through in 2014 and beyond. Next year, we expect new LTE deployments at China Telecom followed by China Unicom. In the US, we expect to see more spending from T-Mobile and Sprint. The company believes that over time other emerging regions (Brazil, India) will start to deploy 4G wireless networks as well. Furthermore, Gartner is forecasting around 1.5mn LTE base-station shipments by 2015. While Europe has lagged infrastructure spending in LTE, we think Vodafone’s recent capex update could spur better spending next year. Vodafone recently raised capex guidance by 37% in its H1 FY14 results. Spending increases are driven by increases in mobile infrastructure spending. The wireline investment outlook also remains positive. In addition, Vodafone announced Project Spring earlier this year which featured a 42% increase in capex versus our European telecom services team’s (James Britton et al) prior expectations. We think higher spending by Vodafone could make other operators in Europe also increase their capex spending. In these 4G infrastructure deployments, we see opportunity for PLD companies in wireless macro base stations, radio heads, and backhaul.
Fig. 15: Capex Summary and Forecast for China Mobile China Mobile Group 1H12 2H12 1H13 2H13F FY11 FY12 FY13F FY14F FY15F (CNY bn) Total 73 101 72 150 160 173 222 210 195 Listco 59 69 57 133 129 127 190 181 166 2G/WiFi 29 27 20 18 73 56 38 30 30 TD-LTE - 42 42 53 39 Transmission 15 20 23 36 23 36 59 59 59 Others 15 21 14 37 32 36 51 Parentco 14 32 15 31 46 32 29 28 TD-SCDMA (3G) 3 6 15 12 21 22 19 18 TD-LTE (4G) 5 5 Wireline (Railcom) 1 19 20 10 10 10
Source: Company data, Nomura estimates
Fig. 16: Capex Summary and Forecast for China Unicom
China Unicom Group 1H12 2H12 1H13 2H13F FY11 FY12 FY13F FY14F FY15F Total 38.9 60.9 21.6 58.4 76.7 99.8 80 91 85 Mobile networks (3G) 17.8 19.9 21.4 37.7 25.6 11.7 7.4 4G/FD-LTE 23.8 21.5 Broadband & data 9.6 15.9 19.8 25.5 19.2 20 20 Mobile networks (2G) 2.4 0.8 2.5 3.2 2 2 Transmission 7.5 18 15.4 25.5 19.2 18 18 Others 1.6 6.2 17.6 7.8 16 16 16
Source: Company data, Nomura estimates
Fig. 17: Capex Summary and Forecast for China Telecom China Telecom Group 1H12 2H12 1H13 2H13F FY11 FY12 FY13F FY14F FY15F Total CAPEX 32.6 39.9 33.1 42 71.6 72.5 75 85.7 80 Wireless(CDMA) 7 11.8 22 18.8 25 10 0 4G/LTE 10 10 45.7 45 Listco Fixedline (listco) 25.6 28.1 49.6 53.7 40 30 30 Fixedline breakdown 25.6 28.1 49.6 53.7
Source: Company data, Nomura estimates
185
Nomura | U.S. Semiconductors Primer December 11, 2013
That said, we think PLD content in LTE infrastructure may be lower Specifically, we believe lower PLD content in 4G relative to 3G infrastructure will be driven by the following key issues: • Many newer 3G base stations that were installed in the past few years are software upgradable to 4G. This reduces the need for new hardware installations, resulting in lower PLD demand. In addition, carriers are likely to deploy small cells to address the capacity issues in their 4G infrastructure. We believe there is very little PLD content in small cells. Specific to the first round of TD-LTE base-station rollout, we think the overall PLD content is lower: ~$600–800 per base station versus $800–1000 in 3G base stations. We believe software upgradeability is one of the factors behind this lower PLD content. • LTE standard is more stable than the 3G standard was at a similar point during carrier technology transition. Our research indicates that the current LTE standard is largely in the final stage with only minor changes forthcoming. This is in contrast with the 3G standard, which took quite a long time (3–5 years) to reach a stable state similar to where the LTE standard is today. For this reason, we think more wireless vendors have confidence in developing an ASIC and are willing to forego the flexibility of PLD- based solutions than the last time during the 2G-to-3G transition.
In addition, many telcos in emerging regions are struggling to grow revenue, which is impacting capex commitments Total telco revenue as a percentage of GDP in Asia was 2.8% in 2012 (on average) and this has been flat for the past three years. In this region, South Korea, Japan, and Thailand are the only countries that have seen an increase in the past three years, mainly due to faster total revenue growth vs GDP growth rates. Other countries have seen a decline. Total telco revenue growth rates have been 6–7% pa for the past three years (on average), but nominal GDP growth rates have exceeded this level in most countries.
• Total wireless service revenues are only 1.6% of GDP and there has been hardly any change in the past three years for telcos. Thailand is the only country that has seen an increase, which is mainly due to faster wireless revenue growth vs GDP growth. Wireless revenue growth rates in the region have been 5–7% pa over the past three years, while nominal GDP growth rates have been higher. • South Korea has seen a consistent increase in telcos’ revenues as a % of GDP (2.9% in 2006 to 4.0% in 2012). Telco revenue growth rates in South Korea over the past three years have been 6–14% pa, and GDP per capita has also risen consistently. But in wireless, revenues have seen a slight drop over the past three years to around 1.8% pa. • Among EMs, wireless revenues in Indonesia, India and China are around 1.4% of GDP, and have been flat to declining. GDP growth rates have been a lot stronger, and, at the same time, revenues have been impacted by competition, especially in the case of India. There is a large ARPU differential too – India and Indonesia are $3–4 each while China is around $9 blended.
As such, while we think 4G spending will over time boost PLD growth, we believe the network build-out will be more drawn out due to a good mix of software upgradeable base stations, inclusion of small cells, and many vendors preferring cheaper ASIC solutions due to higher stability of the LTE standard.
186
Nomura | U.S. Semiconductors Primer December 11, 2013
Small Cells will play a bigger role in 4G networks; PLD content in small cells is very low Mobile operators are seeing unprecedented data traffic on their networks due to the strong adoption of smartphones. Mobile data traffic looks set to increase 65% this year and is expected to grow at a 70+% CAGR for the next five years. This massive increase in mobile data traffic is challenging operator’s capabilities, competitiveness and profitability. Operators currently face data capacity challenges in dense urban areas in the face of declining profitability per MB of data. Spectrum availability, high cost of deploying macro base stations, finding locations for new macro cell sites, dealing with regulatory hurdles, and uneven loading in the network are also issues further complicating the picture.
Small cells a key part of operators’ toolkit Operators and equipment vendors view small cells as one of the key tools that could help resolve many of the data capacity challenges that operators face today. Small cells are operator-managed, low-power wireless access points that operate in licensed and unlicensed spectrum. Small cells provide improved cellular coverage and capacity for homes, enterprises, metropolitan areas, and rural public spaces. They include technologies variously described as femtocells, picocells, and microcells which can cover a range of 10m to several hundred meters. In contrast, a typical base station has a range of several tens of kilometers. Small cells (femtocells, picocells and microcells) can address the needs of small to large enterprises, public spaces, and even hotspots, while being part of a single, coordinated operator network.
Adoption looks set to accelerate after initial disappointments Small cells were introduced a couple of years ago with much fanfare, but despite all the hype rapid adoption failed to materialize. However, a few key changes have occurred since the introduction of small cells that, we think, are bringing them closer to wider deployment. First, mobile data capacity issues have become worse for many operators. Second, the cost of capacity for operators is becoming higher than the revenue they bring in. Third, equipment vendors and small cell eco-systems have made significant progress in improving the technology maturity in both software and in hardware. This has resulted in much improved low-power and low-cost small cells. The combination of these factors will, in our view, drive multi-year adoption of small cells in operators’ networks to address data capacity issues in a cost efficient manner. We note that market research firm Infonetics expects unit shipments to grow at a 100+% CAGR over the next five years. In dollars, small cells are expected to be a $2bn-plus opportunity by 2016.
Fig. 18: Small Cells: Units and Revenue, 2011–2016E Fig. 19: Small Cells: Units by Technology, 2011–2016E
Source: Infonetics, Nomura research Source: Infonetics, Nomura research
187
Nomura | U.S. Semiconductors Primer December 11, 2013
We expect PLDs to be negatively impacted by small cells We believe small cells are a negative for PLD companies for two key reasons: 1) despite uncertainties, we believe small cell deployment could impact base-station growth, and 2) SoCs developed for small cells (micro and picocells) have increased in maturity and are ready to move up market, which could reduce PLD content in next-generation macro base stations.
• We believe small cells could impact base-station growth over the next several years. While the dollar value forecast for small cells in 2015 is not significant at around $2bn vs. traditional base-station spending ($34.9bn in 2011), we expect many operators to use small cells to address capacity challenge in their WCDMA networks and later in LTE networks. We believe this could reduce the need for macro base-station cards in existing base stations or could eliminate the need for new base stations altogether in congested areas, both of which are likely to impact PLD consumption. Although our conversations with equipment vendors indicate that it is too early to quantify the impact, we note that each traditional base station card has $1–2k worth of PLD content vs. zero to minimal content in small cells. Most small cell implementations today have PLD content, which is lower by a factor of 5–10 vs. traditional base-station cards.
– Increased maturity of SoC/ASSP solutions for small cells could impact the traditional base-station design. We think that small cell SoC/ASSP solutions available from vendors like Texas Instruments, Freescale, Qualcomm, ip.access and Picochip are likely to move up market and impact traditional base-station designs. We believe this would be driven by following: – SoC solutions are much improved now and provide a significant increase in capacity, cost reduction, and lower power consumption. SoC offerings from vendors like Texas Instruments (Keystone platform), Freescale (QorIQ), Picochip and Qualcomm pack significant processing power by integrating high-performance DSPs and processing cores. Our research indicates that these SoCs are now powerful enough to be used in traditional base-station designs, which typically use a combination of PLDs and discrete processing elements. – The improved SoC solutions and capabilities are enabling vendors to offer multi-standard, multi-service base stations. SoC vendors are combining multiple radios and processing elements in ASICs/ASSPs, which provide a higher level of integration vs. FPGAs. Our research indicates that most newly deployed base stations support multi-standard capability. We expect these SoCs/ASIC solutions to replace the traditional PLD-based architecture in the new base-station designs that support multi-standard (ZTE, Huawei). – We believe improved software architectures from equipment vendors are also helping in this transition. Earlier-generation base-station designs used FPGAs for design flexibility. As operators move to a distributed architecture, a portion of this flexibility is being pulled into software from hardware, which we believe will increase the adoption of low-cost, SoC-based base-station designs.
188
Nomura | U.S. Semiconductors Primer December 11, 2013
Competitive landscape The PLD market is essentially a duopoly between Xilinx and Altera In 2012, Xilinx had an estimated 50% share of the PLD market, followed by Altera at 39%, Lattice Semiconductor at 6% and Microsemi at 5%. The top two suppliers have together held more than an 80% share since 2002, with their combined share reaching a high of 89% in 2012. By major end market, Xilinx had larger a market share in wired communications, consumer, automotive, industrial and military/aerospace, while Altera was stronger in wireless communications, storage and computing. The main competitive factors include performance, densities, system-level functionality, as well as cost and power consumption at each manufacturing process node. On the process technology front, both Altera and Xilinx are shipping 28nm parts.
Fig. 20: PLD Market Share by Revenue, 2012
Altera 39%
Lattice 6%
Microsemi and others 5%
Xilinx 50%
Source: Gartner, Nomura research
Fig. 21: PLD Market Share, 2001–2012
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Xilinx Altera Lattice Microsemi and others
Source: Gartner, Nomura research
189
Nomura | U.S. Semiconductors Primer December 11, 2013
PLDs offer one of the best business models in the semiconductor industry As noted above, Altera and Xilinx address a large served available market (SAM). We believe the total SAM will be greater than $16bn in 2016, of which less than 30% is currently penetrated, we estimate. As PLD devices offer fast time-to-market and design flexibility, end-market acceptance is broad. Altera and Xilinx are entrenched in multiple applications within networking and infrastructure, industrial, automotive and consumer. Furthermore, these PLDs are highly profitable; low capital intensity and a duopolistic competitive landscape support gross margins that are in the mid-60% to low-70% range and EBIT margins in the 30–40% range. And their balance sheets appear very solid. Barriers to entry are high given the large software installed base and use of the most advanced process technology nodes.
Altera’s share gains started to reverse in 2012 After 2002, Altera steadily gained market share from Xilinx and other players. Altera’s share gains against Xilinx accelerated in 2009 due to its head start in 40nm parts. In 2010, Altera’s total revenue was only about 10% below Xilinx’s, versus a 35–40% gap in the early part of the decade. However, that trend began to reverse in 2011 when Xilinx became more aggressive in its 40/45nm product offerings and introduced a mid-range product, Kintex, to fill a product gap between the high-end Virtex and low-end Artix. As a result, we believe Xilinx began to gain share at the 40/45nm node, where we believe Xilinx had only a 35% share, versus its 50–60% shares in the 65nm and 90nm nodes.
Fig. 22: Market Share Trends, 2002–2013E ALTR and XLNX revenue only 65%
60%
55%
50%
45%
40%
35%
30% 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
ALTR XLNX
Source: Company data, Nomura estimates
Fig. 23: Xilinx Market Share by Node, 2Q13
70%
60%
50%
40%
30%
20%
10%
0% 90nm 65nm 40/45nm 28nm
PLD share by node
Source: Company data, Nomura estimates
190
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 24: PLD Segment Comparison between Altera and Xilinx
Source: Company data, Nomura research
Xilinx is taking even more share at the 28nm node We believe Xilinx is winning a majority of 28nm design wins. The value proposition of Xilinx’s Kintex product (mid range) is proving superior to Altera’s high-end offering. Since inception, Xilinx has generated more than $220mn in cumulative revenue from 28nm products through the third quarter of 2013. This compares to around $125mn of 28nm revenue for Altera, and equates to Xilinx having a 60–70% share at the 28nm node. The key drivers behind Xilinx’s performance are its mid-range Kintex family, 3D IC FPGAs, and the programmable system-on-chip Zynq family. Adoption of Kintex FPGAs is driven by wireless customers (80% socket wins in wireless radio designs), 3D IC by wireline, and Zynq by industrial and automotive customers. We believe Xilinx generated 20–30% of its 28nm revenue from 3D ICs. And while Altera often contends that 3D IC revenue is only for prototyping and therefore not scalable, we believe Xilinx will continue to expand its momentum in 28nm design wins due to the strong performance of its mid-range FPGAs, which often match the performance of Altera’s high-end FPGAs. Further, Altera also lags behind with its FPGA SoC offering, whereas Xilinx’s Zynq is seeing strong traction in industrial, automotive, and video processing areas.
Fig. 25: Comparison of 28nm Revenue, 2Q12–4Q13E
100 $90+ 90 $80+ 80 70 60 $50+ 50 $40+ up qoq 40 $33 30 $20+ $25 $20+ $23 $18 $18 20 $10+ $8 28nm Quarterly Revenue (in $mn) 10 0 2Q12 3Q12 4Q12 1Q13 2Q13 3Q13 4Q13E
XLNX ALTR
Source: Company data, Nomura estimates
191
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 26: Comparison of 28nm and 40nm revenue for Altera and Xilinx
28nm Rev ($mn) 3Q12 4Q12 1Q13 2Q13 3Q13 XLNX 20+ 25-30 40+ 50 80 ALTR 18 20-25 20+ 20+ 33
40nm Rev ($mn) 3Q12 4Q12 1Q13 2Q13 3Q13 XLNX 89 100 105 124 136 ALTR 137 149 140 153 163
New Product Rev ($mn) 3Q12 4Q12 1Q13 2Q13 3Q13 XLNX 109 127 145 174 216 ALTR 155 171 160 173 196 % of total FPGA rev 25% 31% 32% 35% 39%
Source: Company data, Nomura estimates
Altera’s tailored approach could prove difficult to scale Altera uses a tailored approach for different segments of FPGAs. While this approach has advantages, it also requires more resources to fine-tune silicon and software for each segment. In contrast, Xilinx utilizes the same approach for silicon and software across the product segment. Our checks indicate that Altera may not be willing to deploy resources needed to fine-tune software and silicon for each segment. As such, we expect Xilinx to scale faster and more cost-efficiently for 28nm FPGAs. We expect Xilinx’s recently introduced Kintex and Zynq families to put more competitive pressure on Altera. Until last year, Xilinx lacked products in the mid-range segment that could compete with Altera’s Arria family of devices. While Altera’s mid-range segment offerings contributed only 1% of its FPGA revenue in 2010, we believe this segment is becoming increasingly important in next-generation design wins. We expect the 28nm version of mid-range FPGAs to be a “sweet spot” in terms of performance and cost for many applications across end markets, in particular wireless. Our checks indicate that Kintex is taking a large share of next-generation wireless base-station design wins. In addition, we believe Altera does not have a competitive offering for Xilinx’s recently launched Zynq FPGAs. Zynq FPGAs are similar to SoCs and include a dual-core Cortex-A9 processor for platform-centric applications. Our checks indicate that Xilinx is getting almost 100% of the design wins from automotive and military/defense end-markets for Zynq.
Altera’s move to Intel’s 14nm is positive, but the impact won’t be seen for several years Altera historically has used TSMC as its sole foundry, but the company recently announced that it is moving its high-end production to Intel’s 14nm foundry. Although Intel is a new entrant to the foundry business, we think Altera’s move is positive because TSMC’s execution in 28nm has been weak, and because progress so far in FinFET technology has been limited. Overall, we think Intel is 1–3 years ahead in 3D transistor technology. As such, we believe, despite execution and potentially higher opex, Altera is likely better positioned in making the 14nm transition versus Xilinx, which is using TSMC. That said, 14nm will likely be immaterial for several years. This gives Xilinx time to strike a deal with Intel. Additionally, we think TSMC may be able to catch up with Intel to mitigate any competitive impact on Xilinx. Altera’s management, in our recent meeting, indicated some sort of exclusivity, but we think this will largely be time-duration exclusivity as opposed to a lock on Intel’s 14nm process.
192
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 27: Expected Technology Transitions for Altera and Xilinx
40nm 20nm Altera 28nm 14nm (Intel)
40nm 20nm Xilinx 28nm 14/6nm
2009 2012 2014E 2015E
Source: Nomura research
193
Nomura | U.S. Semiconductors Primer December 11, 2013
Analog
What Is Analog IC? An analog or linear integrated circuit (IC) is a circuit with a continuous, variable signal, as opposed to a digital circuit where a signal must be one of two discrete levels. Examples of digital ICs are microprocessors and memory. Analog circuits within electrical equipment can convey information through changes in the current, voltage, or frequency. Unlike digital ICs that operate at distinct “on” and “off” levels, most analog function at all inputs levels in a predefined range and provide an output based on that input. Analog circuits interface with real-world inputs in order to measure physical properties such as temperature, speed, pressure, sound, color, and electrical voltage levels. Most electronics today use digital circuits because they are easier to design and less susceptible to noise. However, a device that interfaces with the environment requires at least some analog components to take in this information before converting the signal to digital. Analog ICs are manufactured using bipolar, CMOS, or BiCMOS technologies, including monolithic linear ICs (100% analog) and monolithic mixed-signal ICs (having both digital/numeric and signal/power forms). In contrast, most digital ICs are manufactured using a standard CMOS process. Mixed-signal ICs combine digital and analog functions onto a single device. Typically, a mixed-signal device converts an analog signal into its digital (A-D) equivalent or a digital signal into an analog format (D-A). Examples of mixed-signal ICs are MPEG encoder and decoders, and Ethernet PHYs (physical layer devices).
Fig. 1: Digital Signals Represent Predefined Discrete Values, in this case 0s and 1s
1
0
Source: Nomura research
Fig. 2: Analog Signals Are Variable Across an Infinite of Values
Source: Nomura research
194
Nomura | U.S. Semiconductors Primer December 11, 2013
Characteristics of the Analog Market The analog IC market is highly fragmented, with higher barriers to entry than the digital IC market. The primary reasons for the barriers to entry are that there has been a shortage of skilled analog engineers, and that the design of analog circuits is more of an art than a science. The analog circuits do not require significant capital investments, but rather focus on feature set (performance, functional value), quality, and reliability. As a result, once products are designed into electronics systems, analog companies enjoy relatively long product cycles and strong profitability. Shortage of analog engineers There has been a shortage of skilled analog design engineers worldwide. The majority of college curricula focus on digital design, resulting in a shortage of college curricula to educate students in analog design. Many young engineers seem to be more interested in designing the next microprocessor than in designing the voltage regulator that will make it possible for the microprocessor to turn on and stay on. As a result, most analog engineers learn their skill through on-the-job training, and consequently may take two to three times as long for them to become as productive as their digital engineer counterpart. Most of the best analog design engineers have 20 to 30 years of experience in the analog field. More an art than a science While digital designs focus on improving speed and performance and lowering cost and power using advanced CMOS process technologies, analog products are often designed to provide output signals with very precise specifications. This precision is the result of a careful match between circuit design and the manufacturing process. Because of these special modifications to the manufacturing process, there is no “standard” analog process. For example, some analog IC requires high voltage or high current output to drive other components, but CMOS has poor driving capability. Typical process technologies used for analog circuits are bipolar, BiCMOS, and BCD. Some high frequency circuits require different materials such as GaAs. Lack of specialized design automation software The availability of electronic design automation software is heavily skewed in favor of digital circuit designs. This is because digital functions can be easily expressed in a special branch of mathematics, which can be programmed into software algorithms. In analog circuits, there is no corollary branch of mathematics. In addition, most digital circuits are highly regular and repeated, with predictable performance. The structure in analog circuits is very different because they tend to be highly customized for specific customers and applications. This lack of regularity reduces leverage of software automation in the design process. Complex testing requirements In addition to more difficult design and manufacturing processes, analog circuits are also more complicated to test. In contrast to a digital tester that samples an output at a known time for a high- or low-voltage level, an analog tester must be able to measure intermediate voltage levels and operate at much tighter tolerances for precision analog devices. Analog testing also needs to take into account the noise and distortion produced by resistors and capacitors used in the system. Low capital requirements Digital circuits require leading-edge manufacturing processes in order to reduce die size and increase performance. Moore’s Law suggests that the number of resistors in a given area could double in every 18- to 24-month period. So far Intel has been able to keep up with that pace, but it requires very heavy investments in capital. In contrast, the capital requirements for producing analog circuits are much lower. This is because larger circuit feature sizes are needed to design and manufacture ICs that can deliver high precision and operate at relatively higher voltages than their digital counterparts. Furthermore, lower capital requirement for analog companies implies lower fixed costs. This is important during any industry downturns, when their digital IC counterparts with leading edge fabs will likely need to keep factory utilization high and hence should see more severe price competition.
195
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 3: Capital intensity for Intel and Texas Instruments
25%
20%
15%
10%
5%
0% 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Intel's capex-to-sales TI's capex-to-sales
Source: Company data, Nomura estimates
Fragmented market The analog market is made up of many niche markets. Different end users have different requirements for analog ICs in terms of accuracy, speed, power, linearity, and signal amplitude capabilities. Large analog suppliers have broad standard product portfolio with general purpose products (sometimes referred to as catalog products) that typically target certain segments of the market with low degree of competition. Smaller analog suppliers tend to focus on certain vertical applications with specialized products (application specific standard products, or ASSPs) before larger suppliers address those applications. While ASSP suppliers will likely yield initial success, their business models are less attractive relative to general-purpose product supplier given lower margins, shorter product cycles, high customer concentration and hence higher volatility. Standard products generally have longer product cycles and yield higher margins. Proprietary products Many products in the analog market are designed for one customer for one application and therefore tend to be proprietary products and are sole-sourced. Companies such as Linear and Maxim have built a large product portfolio consisting almost entirely of proprietary products. As a result, competition is not based on price but on feature set, quality, reliability, and service. The proprietary nature allows analog products to enjoy more stable pricing than digital products that tend to compete on cost in addition to performance. Long product life cycles Analog products typically have much longer lives than their digital counterparts. This is due to the relatively long life of many end products in which they are used. For example, some of the products in the industrial market have life cycles exceeding ten years. Even if some of the digital components in these end products may be upgraded during the life cycle, the analog products tend to be unchanged until the next major product refresh. This is because analog products act as an interface between the real world and the digital world, and the interface with the analog world rarely changes. Furthermore, the value of analog components tend to be significantly less the digital components, and the risks of changing out analog components that are working fine far outweigh the benefits the cost savings for a new part. Overlapping annuities Because of the long product life cycles, analog components usually have multiple product revenue streams overlapping each other, which add up to a growing total revenue stream. Relative to digital analog companies where product life cycles are much shorter, this overlapping revenue profile allows analog companies to see relatively more stable revenue and financial results.
196
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 4: Revenue profile of an analog company
Revenue
Product 7 revenue Product 6 revenue Product 5 revenue Product 4 revenue Product 3 revenue Product 2 revenue Product 1 revenue
Time
Source: Nomura research
197
Nomura | U.S. Semiconductors Primer December 11, 2013
Analog Market Size The analog IC market was a $39bn opportunity in 2012, representing about 13% of the total semiconductor market (14% in 2011 and 2010). According to SIA, the analog IC market declined 7% to $39bn in 2012 from $42bn in 2011. The 10-year (2003–2012) revenue CAGR was 5.1%, with unit growth of 8.3% and ASP decline of 3.0%. SIA forecasts that analog IC revenue will grow 5% in 2013 and 6% in 2014, which is about in line with the overall semiconductor market.
Fig. 5: Analog IC Revenue, 2000–2014E
50.0
45.0
40.0
35.0
30.0
25.0
20.0
15.0
Analog IC Revenue ($bn) Revenue IC Analog 10.0
5.0
0.0
Source: SIA, Nomura research
Fig. 6: Analog IC Units and ASPs, 2000–2012
70 $0.60
60 $0.50
50 $0.40 40 $0.30 30 ASP $0.20 20 Standard Analog Units (bn) Units Analog Standard 10 $0.10
0 $0.00 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Unit ASP
ASP (in $, line) Source: SIA, Nomura research
198
Nomura | U.S. Semiconductors Primer December 11, 2013
Analog Market Is Highly Fragmented The analog market is highly fragmented. According to market research firm Databeans, in 2011, Texas Instruments had the highest market share (15%), followed by STMicroelectronics (10%), Analog Devices (6%), Infineon (5%), and Maxim Integrated (5%). The top 5 suppliers accounted for 41% of the analog IC market, the top 10 suppliers accounted for 57%, and the top 20 suppliers accounted for 77%. Historically, market share changes happened very slowly.
Fig. 7: Analog IC Market Share, 2011
Others 43%
Renasas 3% ONNN 3%
LLTC 3%
SWKS 3%
NXPI 4% TXN 15%
MXIM 5%
IFX 5% ADI 6% STM 10%
Source: Databeans, Nomura research
199
Nomura | U.S. Semiconductors Primer December 11, 2013
Analog market segments The analog market is divided into two segments: (1) standard analog (sometimes referred to as general purpose analog or high performance analog) and (2) application specific analog (sometimes referred to as application specific standard products (ASSPs)). In 2012, general purpose analog accounted for 39% of the analog market, while ASSPs accounted for 61% of the analog market. Standard analog ICs are designed for multiple end markets, while ASSPs are designed for specific application. Standard analog tends to have high switching costs, unique analog signal, and power requirements, sole-sourced, and long product life cycles, which drive more stable ASP and better margins than ASSPs. For example, the ASP of standard analog declined by an average of 1.7% per year over the past five years, while the ASP of ASSPs declined by an average of 3.5%
Fig. 8: Standard Analog vs. Application Specific Analog (ASSP)
45.0
40.0
35.0
30.0
25.0
20.0
Revenue ($bn) Revenue 15.0
10.0
5.0
0.0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Standard Analog ASSPs
Source: SIA, Nomura research
Standard Analog According to SIA, the standard analog market declined 9% from $17.1bn in 2011 to $15.5bn in 2012, driven by a 2% decline in units and 8% decline in ASP. The standard analog market can be further divided in four segments: amplifiers, voltage regulators (or power management), data converters, and interface. In 2012, power management accounted for 56% of standard analog, followed by amplifiers (17%), converters (15%), and interface (12%).
2 00
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 9: Standard Analog Market, $17bn in 2012
Converters 15%
Amplifiers 17%
Interface 12% Voltage Regulators 56%
Amplifiers Interface Voltage Regulators Converters
Source: SIA, Nomura research
Fig. 10: Standard Analog Unit and ASP
70 $0.60
60 $0.50
50 $0.40 40 $0.30 30 ASP $0.20 20 Standard Analog Units (bn) 10 $0.10
0 $0.00 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Unit ASP
ASP (in $, line) Source: SIA, Nomura research
1. Voltage regulators Voltage regulators, or commonly known as power management IC, accounted for 56%, or $8.7bn, of the standard analog market in 2012. A voltage regulator is a device that maintains voltage of a power source within acceptable limits. It is important for voltages to stay within a prescribed range that can be tolerated by the equipment using the voltage. Excess variations of voltages could be harmful to the electronic equipment. Voltage regulators can be found a wide variety of applications, ranging from computers and smartphones, to automobiles and power plants.
201
Nomura | U.S. Semiconductors Primer December 11, 2013
Power management IC is likely the fastest growing segment within standard analog. This is driven by the increasing requirement of voltage and current levels with complex electronic systems, and the need to isolate the systems from their power sources. Switch-mode regulators, such as DC-DC converters, AC-DC converters, LED lighting drivers, and lithium-ion battery charge controllers, are expected to have the best growth opportunities. Texas Instruments has the highest market share in power management with 17% share, followed by Maxim Integrated (10%), Linear Technology (7%), ON Semiconductor (5%), and STMicroelectronics (4%). The top 10 suppliers account for 61% of the power management market.
Fig. 11: Market Share in Power Management IC, 2012
TXN 17%
MXIM 10%
Others 57% LLTC 7%
ONNN 5% STM 4%
TXN MXIM LLTC ONNN STM Others
Source: Gartner, Nomura research
Fig. 12: Power Management IC by End Market, 2012
Military/Aero Industrial 1% 17% Data Processing 29%
Automotive 7%
Consumer 20% Communications 26%
Data Processing Communications Consumer Automotive Industrial Military/Aero
Source: Gartner, Nomura research
202
Nomura | U.S. Semiconductors Primer December 11, 2013
2. Data Converters Data converters accounted for 15%, or $2.3bn, of the standard analog market in 2012. Data converters convert analog electrical signals representing real-world phenomenon, such as light, sound, temperature or pressure, for data processing purposes. Digital-to- analog converters (DACs) convert digital signals into analog format, while analog-to- digital converters (ADCs) convert analog signals into digital format. For example, a thermometer detects the temperature and converts the data into a digital format to be displayed on the readout. Analog Devices has the highest market share in data converters with 37% share, followed by Texas Instruments (18%), Maxim Integrated (7%), Cirrus Logic (4%), and Linear Technology (4%). The top 10 suppliers account for 29% of the data converter market.
Fig. 13: Market Share in Data Converters, 2012
Others 29%
ADI 37%
LLTC 4%
CRUS 4%
MXIM 7% TXN 18%
ADI TXN MXIM CRUS LLTC Others
Source: Gartner, Nomura research
Fig. 14: Data Converters by End Market, 2012
Military/Aero Data Processing 2% 10%
Industrial 30%
Communications 27%
Automotive 6%
Consumer 25% Data Processing Communications Consumer Automotive Industrial Military/Aero
Source: Gartner, Nomura research
203
Nomura | U.S. Semiconductors Primer December 11, 2013
3. Amplifiers Amplifiers accounted for 17%, or $2.6bn, of the standard analog market in 2012. An amplifier is an electronic device that amplifies a relatively small input signal into a much larger output signal. There are many types of amplifiers. For example, a power amplifier is used to supply power ranging from a few watts in an audio amplifier to thousands of watts in a radio transmitter. An operational amplifier, or op-amp, produces output voltage that is typically hundreds of thousands of times larger than the voltage difference between its input voltages, and is most widely used electronic device today. Texas Instruments has the highest market share in amplifiers with 22% share, followed by Analog Devices (20%), STMicroelectronics (6%), Linear Technology (6%), and Maxim Integrated (6%). The top 10 suppliers accounted for 82% of the segment.
Fig. 15: Market Share in Amplifiers, 2012
TXN 22%
Others 40%
ADI 20%
MXIM 6% LLTC 6% STM 6%
TXN ADI STM LLTC MXIM Others
Source: Gartner, Nomura research
Fig. 16: Amplifiers by End Market, 2012
Military/Aero Data Processing 5% 11%
Industrial 25%
Communications 32%
Automotive 9%
Consumer 18% Data Processing Communications Consumer Automotive Industrial Military/Aero
Source: Gartner, Nomura research
204
Nomura | U.S. Semiconductors Primer December 11, 2013
4. Interface Interface ICs accounted for 12%, or $1.9bn, of the standard analog market in 2012. Interface ICs are semiconductor chips that are used to control and manage the sharing of information between devices. Examples of common interface protocols include RS- 232, RS-485, and RS-422 in telecommunications and computing, USB, PCI-Express and SATA in computing, HDMI and 1394 in consumer applications. Texas Instruments has the highest market share in interface ICs with 30% share, followed by Maxim Integrated (16%), NXP (10%), Linear Technology (7%), and Analog Devices (6%). The top 10 suppliers accounted for 88% of the interface IC segment.
Fig. 17: Market Share in Interface ICs, 2012
Others 31% TXN 30%
ADI 6%
MXIM 16% LLTC 7% NXP 10%
TXN MXIM NXP LLTC ADI Others
Source: Gartner, Nomura research
Fig. 18: Interface ICs by End Market, 2012
Military/Aero 2% Industrial 21% Data Processing 30%
Automotive 6%
Consumer 15% Communications 26% Data Processing Communications Consumer Automotive Industrial Military/Aero
Source: Gartner, Nomura research
205
Nomura | U.S. Semiconductors Primer December 11, 2013
Application Specific Standard Products (ASSPs) According to SIA, the application specific standard product market declined 6% from $25.2bn in 2011 to $23.8bn in 2012, driven by a 6% decline in units and 8% decline in ASP. The ASSP market can be further divided in sub-segments by end market: consumer, computing, communications (infrastructure and handsets), automotive, and industrial/other. In 2012, communications accounted for 48% of ASSPs, followed by automotive (23%), computing (11%), consumer (10%), and industrial/other (8%).
Fig. 19: Application Specific Standard Product Market, $24bn in 2012
Industrial and Consumer Other 10% 8% Computing 11% Automotive 23%
Communications 48%
Consumer Computing Communications Automotive Industrial and Other
Source: SIA, Nomura research
Fig. 20: Application Specific Standard Product Unit and ASP
40 $1
35 $1 30 $1 25
20 $1 ASP
15
ASSP Units (bn) $0 10 $0 5
0 $0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Unit ASP
ASP (in $, line) Source: SIA, Nomura research
206
Nomura | U.S. Semiconductors Primer December 11, 2013
1. Wireless Handsets Wireless handsets was $8bn in 2012, representing 33% of the ASSP market or 70% of communications ASSPs. Key applications in handsets include battery management, audio, display and lighting, and touchscreen applications. Companies with the highest exposure to handsets are Avago (45% of sales), Maxim Integrated (35%), NXP (25%), Atmel (20%), and ON Semiconductor (15%).
Fig. 21: Wireless handsets as % of Sales, 2012
AVGO
MXIM
NXPI
ATML
ONNN
TXN
FSL
LLTC
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Source: Company data, Nomura research
2. Communications Infrastructure and Networking Communications Infrastructure and Networking was $3.6bn in 2012, representing 15% of the ASSP market, or 30% of communications ASSPs. Communications equipment that use ASSPs includes wireless base stations, backhaul equipment, routers, switches, and customer premise equipment. Companies with the highest exposure to communications infrastructure and networking are Avago (35% of sales), Texas Instruments (25%), and Fairchild Semiconductor (23%). Intersil, NXP, Freescale, Linear Technology, and Analog Devices all have about 20% exposure to this market.
Fig. 22: Communications Infrastructure and Networking as % of Sales, 2012
AVGO
TXN
FCS
ISIL
NXPI
FSL
LLTC
ADI
MCHP
MXIM
0% 5% 10% 15% 20% 25% 30% 35% 40%
Source: Company data, Nomura research
207
Nomura | U.S. Semiconductors Primer December 11, 2013
3. Automotive Automotive was $5.5bn in 2012, representing 23% of the ASSP market. Key ASSPs used in automotive include sensor-amplifier-data converters, networked power actuators, RF transmitter receivers, and bus interface. Examples of automotive applications include antilock brake systems, engine control units, remote/keyless entry, airbag control modules, and dashboard instrument clusters. Companies with the highest exposure to automotive are ON Semiconductor (25%), Freescale (25%), NXP (21%), Analog Devices (18%), and Linear Technology (16%).
Fig. 23: Automotive as % of Sales, 2012
ONNN
FSL
NXPI
ADI
LLTC
ISIL
MCHP
FCS
ATML
TXN
0% 5% 10% 15% 20% 25% 30%
Source: Company data, Nomura research
4. Computing Computing was $2.5bn in 2012, representing 11% of the ASSP market. These devices are used in desktops, notebooks, servers, PC peripherals, and storage systems. Key applications include CPU power management, graphics processors, battery management, hard disk drives, printers, PC peripherals, network interface, and audio. Companies with the highest exposure to computing are Intersil (21%), ON Semiconductor (20%), Texas Instruments (20%), Maxim Integrated (15%), and Fairchild Semiconductor (13%).
208
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 24: Computing as % of Sales, 2012
ISIL
ONNN
TXN
MXIM
FCS
LLTC
MCHP
0% 5% 10% 15% 20% 25%
Source: Company data, Nomura research
5. Consumer Consumer was $2.3bn in 2012, representing 10% of the ASSP market. These devices are designed primarily for use in audio and video electronics. Other applications include home appliances, cameras, camcorders, televisions, digital video recorders, DVD players, set top boxes, home gateways, and game consoles, Companies with the highest exposure to consumer applications are Microchip (20% of sales), ON Semiconductor (23%), Intersil (22%), Fairchild Semiconductor (18%), and Analog Devices (17%).
Fig. 25: Consumer as % of Sales, 2012
MCHP
ONNN
ISIL
FCS
ADI
FSL
TXN
MXIM
CY
LLTC
0% 5% 10% 15% 20% 25% 30% 35%
Source: Company data, Nomura research
209
Nomura | U.S. Semiconductors Primer December 11, 2013
6. Industrial and Other Industrial was $1.9bn in 2012, representing 8% of the ASSP market. These devices are used primarily for the control of analog signals such as temperature, pressure, and velocity. Key applications include factory automation, motor controls, test and measurement equipment, medical equipment, etc. Companies with the highest exposure to industrial applications are Analog Devices (45% of sales), Linear Technology (42%), Fairchild Semiconductor (35%), Microchip (35%), and Atmel (35%).
Fig. 26: Industrial as % of Sales
ADI LLTC FCS MCHP ATML FSL MXIM ISIL AVGO MSCC ONNN NXPI TXN CY 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Source: Company data, Nomura research
210
Nomura | U.S. Semiconductors Primer December 11, 2013
Microcontrollers Microcontrollers (or MCUs) are not analog IC; rather they are classified as part of microcomponents. However, they exhibit many of the same characteristics as analog ICs, such as high sensitivity to global economic growth, fragmented customer base, high barriers to entry, and long product life cycles. A microcontroller is a self-contained system with peripherals, memory, and a processor. Most microcontrollers are embedded in consumer products or machinery such as peripherals, automobiles, and household appliances. Programmable microcontrollers contain general purpose input and output pins that can be configured by software. When a pin is configured as an input, it can be used to read external signals. When a pin is configured as an output, it can be used to drive external devices such as LED displays and motors.
Fig. 27: Examples of Microcontroller Block Diagram
Source: Atmel and Microchip, Nomura research
The microcontroller market can be further divided into 8-bit, 16-bit, and 32-bit, which refers to the number of bits that each processor operation can handle. Generally speaking, 32-bit MCUs are used in more complicated systems such as motor controls, while 8-bit MCUs are used in less complicated systems such as cordless phones. In addition, 32-bit MCUs tend to have higher average selling prices than 16-bit MCUs and 8-bit MCUs given the increased complexity and performance of the chips. In 2012, the three primary end markets for microcontrollers are automotive (35%), industrial (22%), and data processing (20%). However, there are significant variations for the three variants of MCUs. In the 32-bit MCU market, more than 60% of the revenue was driven by automotive, followed by industrial (22%). In the 16-bit MCU market, the primary end markets were data processing (32%), automotive (30%), and industrial (28%). The 8-bit MCU market was more equally split between industrial (23%), data processing (22%), consumer (20%), communications (18%), and automotive (16%).
211
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 28: Microcontrollers by applications in 2012
100%
90%
80%
70% Military and Aerospace 60% Industrial
50% Automotive Consumer 40% Communications 30% Data Processing 20%
10%
0% 8-bit 16-bit 32-bit total
Source: Gartner, Nomura research
The microcontroller market was $14bn in 2012, or 5% of the semiconductor market. The 8-bit market was the largest segment, accounting for 40% of the overall market, followed by 32-bit at 36% of the market and 16-bit at 24%. However, given the increased complexity of electronic systems, we expect the 32-bit segment to grow 5–10% per year, which is 2–3 times faster than the growth rate of other segments of 2–4% per year. As such, we expect the 32-bit segment to surpass the 8-bit market in the next two to three years.
Fig. 29: Microcontroller market
$20 $18 $16 $14 $12 $10 $8 $6 MCU Revenue ($bn) Revenue MCU $4 $2 $0 2011 2012 2013E 2014E 2015E 2016E 2017E
8-Bit MCU 16-bit MCU 32-bit MCU
Source: Gartner, Nomura research
212
Nomura | U.S. Semiconductors Primer December 11, 2013
Like the analog market, the microcontroller market is highly fragmented. In 2012, Renasas was the largest supplier, accounting for 27% of the market revenue. Renasas has a strong presence in the Japanese automotive market. However, no other supplier accounted for more than 10% share. The second largest supplier was Freescale at 10% share, followed by Infineon (7%), Microchip (7%), STMicroelectronics (7%), Texas Instruments (6%), and Atmel (6%).
Fig. 30: Overall Microcontroller Market Share, 2012
Others 18% Renesas 27% Samsung 4%
Fujitsu 4%
NXP 4%
Freescale Atmel 10% 6%
Texas Instruments Infineon 6% 7% STMicro Microchip 7% 7%
Source: Gartner, Nomura research
The competitive landscape also differs in the various segments of the MCU market. In the 8-bit MCU market, the top 3 suppliers are Renasas, Microchip, and Atmel. In the 16-bit MCU market, the top 3 suppliers are Renasas, Infineon, and Texas Instruments. In the 32-bit MCU market, the top 3 suppliers are Renasas, Freescale, and STMicroelectronics.
Fig. 31: 8-bit MCU Market Share, 2012
Others Renesas 19% 17% Panasonic 2% Silicon Labs 2% Microchip Cypress 14% 5% Samsung 6%
Freescale Atmel 7% 11% NXP STMicro 8% 9%
Source: Gartner, Nomura research
213
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 32: 16-bit MCU Market Share, 2012
Intel Inside Secure Others 2% 2% 6% Toshiba Renesas 3% 25% Microchip 4% Samsung 4% Fujitsu 9%
Freescale 10% Infineon 22%
Texas Instruments 13%
Source: Gartner, Nomura research
Fig. 33: 32-bit MCU Market Share, 2012
Fujitsu Others 3% 8% Toshiba 3% NXP 3% Infineon Renesas 5% 39% Denso 5% Atmel 5%
Texas Instruments 8% STMicro Freescale 9% 12%
Source: Gartner, Nomura research
214
Nomura | U.S. Semiconductors Primer December 11, 2013
Analog Follows GDP Growth The analog market has historically grown at a multiple of GDP growth. Since 1990, the analog market grew at a CAGR of 7.6%, as compared to worldwide GDP growth of 3.3% CAGT, which is a factor of 2.0–2.5x. That said, the growth rate of the analog market has decelerated substantially in the last 10 years. In the 1990s, the analog market growth of CAGR of 15% was nearly 5x the growth rate of worldwide GDP (3.1%). In contrast, over the past 10 years from 2003 to 2012, analog growth rate of CAGR of 5.1% was only about 50% higher that the growth rate of worldwide GDP (3.7%). We believe this reflects the maturation of the analog market.
Fig. 34: Analog Growth Decelerates
50 1990-2000: 15% CAGR
40 2002-2012: 5% CAGR
30
20
10
0 YoY growth rate (%) -10
-20
-30
Source: SIA, Nomura research
Over the long term, we believe analog semiconductors market will grow at 1–2x of worldwide GDP growth. The increasing proliferation of electronics in every end market segment is driving the need for both digital and analog semiconductors. Furthermore, we believe semiconductor content per system will likely increase over time. Some of the faster growing markets for analog semiconductors include smartphones, automotive, and smart energy.
215
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 35: Mobile Handset Market
2,500
2,000
1,500
1,000
500 Handset Unit Shipment (mn) Shipment Unit Handset
0
Voice phone units (m) Smartphone units (m)
Source: Nomura estimates
Fig. 36: Automotive Market Fig. 37: Smart Meter Market
40 $360 70 $1,200
35 $350 60 $1,000
30 $340 50 $800 25 $330 40 $600 20 $320 30
Millions of Units $400 15 $310 20 Millions of Dollars US
10 $300 10 $200 Revenue Opportunity (in $bn) (in Opportunity Revenue
5 $290 Average Semiconductor Content 0 $0 2011 2012 2013E 2014E 2015E 2016E 2017E 0 $280 2011 2012 2013E 2014E 2015E 2016E 2017E Smart Meter Shipments Semiconductor Revenue
Automotive Semiconductor Revenue Average semiconductor content Source: IHS iSuppli, Nomura estimates
Source: Gartner, Nomura research
Over the past three years, the fastest growing companies in the analog market were Avago Technologies (15% CAGR), Maxim Integrated (13%), and Linear Technology (12%).
216
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 38: Analog Semiconductor Revenue Growth YoY
2008 2009 2010 2011 2012 2009-12 CAGR SIA Analog -2% -10% 32% 0% -7% 7% ADI -6% -12% 35% 1% -8% 8% ATML -4% -22% 35% 10% -21% 6% AVGO 7% -6% 39% 7% 1% 15% CY 13% -63% 31% 13% -23% 5% FCS -6% -25% 35% -1% -12% 6% ISIL 2% -21% 34% -8% -20% 0% LLTC 7% -23% 61% -8% -4% 12% MXIM -8% -13% 40% 6% -2% 13% TXN (ex-NSM) -1% -13% 42% 2% -5% 11% Average 0% -22% 39% 3% -10% 8% Median -1% -21% 35% 2% -8% 8%
Source: Company data, Nomura research
217
Nomura | U.S. Semiconductors Primer December 11, 2013
As the chart below shows, market share in analog does not change very quickly. As such, market share is not a big driver for growth. What we believe is more important is the companies’ end market exposure. Avago and Maxim both benefited from their exposure in the wireless handset market in the last few years. Linear Technology had an outsized growth in 2010, but its growth in 2011 and 2012 was more subdued. Companies with high exposure to the PC market, such as Intersil, underperformed the analog market.
Fig. 39: Market Share in Analog Tends to Move Slowly
16%
14%
12%
10%
8%
6%
Analog Market Share 4%
2%
0% 2008 2009 2010 2011 2012
ADI AVGO FCS LLTC MXIM TXN (ex-NSM)
Source: Company data, Nomura research
Fig. 40: Company Revenue by End Market
Computing Handset Comm Infra Consumer Industrial Automotive Military & Aero Other ADI 20% 17% 45% 18% ATML 20% 25% 35% 10% 5% 5% AVGO 45% 35% 20% CY 15% 25% 35% 5% 10% 5% 5% FCS 13% 23% 18% 35% 11% FSL 3% 20% 15% 30% 25% 7% ISIL 21% 20% 22% 22% 15% LLTC 11% 1% 20% 3% 42% 16% 7% MCHP 5% 15% 30% 35% 15% MSCC 31% 19% 50% MXIM 15% 35% 15% 10% 25% NXPI 25% 20% 15% 21% 19% ONNN 20% 15% 23% 17% 25% TXN 20% 15% 25% 10% 15% 10% 5%
Source: Company data, Nomura estimates
218
Nomura | U.S. Semiconductors Primer December 11, 2013
The Analog Market Works Like a Pendulum The analog market works like a pendulum, in the sense that the market moves around an equilibrium, and that the higher the velocity goes through this equilibrium, the higher the velocity comes out of the equilibrium. Over time, analog semiconductor growth should be similar to the end market growth. The factors that determine the velocity and magnitude of these cycles include inventory balance, order lead times, and margin leverage. The analog market tends to be more volatile than their end customers due to order lead times and inventory builds. In periods when supply is tight, customers tend to place more orders than they need in hopes that they will get enough supply even if part of the orders cannot be fulfilled. Seeing increasing orders, semi suppliers are likely to lengthen order lead times to signal they have increasing demand, In addition, semi suppliers tend to increase production, with the hopes that they can capture incremental business should demand come in stronger than the customers indicated. The decision to build inventory needs to be made well in advance given production lead times could be as long as 8–10 weeks. This multiplying effect could drive supply chain inventory growth significantly different than end demand growth in a short period of time. Over a long period of time, analog semiconductors growth tend to be similar to end market growth. For example, between 2004 and 2008, the difference between the year-over-year growth rates of semiconductors and end markets ranged from -10% to +14%, but the average growth rate over the 5-year period was almost the same (6% CAGR for semiconductors and 5% CAGR for end markets).
Fig. 41: Semiconductor Growth and End Market Growth Were Similar in 2003–2008
Y/Y growth 2004 2005 2006 2007 2008 5-yr CAGR Networking 16% 4% 25% 21% 3% 13% Wireless 17% 13% 22% 16% -1% 13% Storage 9% 0% 5% 11% 5% 6% Automotive 10% -1% 1% -6% -16% -3% Consumer 6% -1% 8% 14% 7% 7% Industrial 13% 10% 13% 10% 6% 10% PC 13% 10% 6% 11% 5% 9% Total End Markets 11% 3% 7% 7% -2% 5%
Semiconductors 25% 4% 9% -3% -6% 6%
Source: Company data, Nomura research
One of the key factors that determine the magnitude of business cycles is the supply chain inventory level. Inventory replenishment and depletion add to the end demand trajectory that is underlying any business cycle. For example, as business recovers from a cyclical trough, customers tend to also build inventory in anticipation of future demand. This will generally lead to much stronger growth for semiconductor suppliers. Conversely, as demand reaches a cyclical peak, customers tend to reduce inventory levels to maximize cash conversion, resulting in a faster deceleration of businesses for semiconductor suppliers than end market demand. In general, one should expect the faster the decline of business is during an inventory correction, the faster it will snap back when business recovers. Many of the semiconductor cycles in the past 10 years were inventory-driven cycles. Over the past 10 years, analog semi inventory dollars and days have been on the rise. Inventory days ranged from 55 days to 106 days, and the average was 87 days. Analog inventory was at 100 days at the end of 2012.
219
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 42: Analog Inventory Dollars and Days, 2003–2012
$7.0 110
$6.0 100
$5.0 90 Days of Inventory of Days $4.0 80
$3.0 70 Inventory ($ billion)
$2.0 60
$1.0 50
$0.0 40 Jun-03 Jun-04 Jun-05 Jun-06 Jun-07 Jun-08 Jun-09 Jun-10 Jun-11 Jun-12 Mar-03 Mar-04 Mar-05 Mar-06 Mar-07 Mar-08 Mar-09 Mar-10 Mar-11 Mar-12 Sep-03 Dec-03 Sep-04 Dec-04 Sep-05 Dec-05 Sep-06 Dec-06 Sep-07 Dec-07 Sep-08 Dec-08 Sep-09 Dec-09 Sep-10 Dec-10 Sep-11 Dec-11 Sep-12 Dec-12
Analog Inventory Analog DOI
Source: Company data, Nomura research
Furthermore, the variability of inventory will also drive fluctuations in margins for semiconductor companies. Given most analog companies own and operate their own fabs that carry fixed operating costs, the margin fall-through rate tends to be very high. The chart below shows that incremental gross margin for a group of high quality analog companies is 70–80% over a 3-, 5-, and 10-year periods. That said, margin leverage works both ways. For example, Texas Instruments utilized about 50% of its manufacturing base in 4Q 2012, resulting in an underutilization charges that are equivalent to 5–6 points of gross margin impact.
Fig. 43: Example of Incremental Gross Margins Over the Past 10 Years
3-yr average 5-yr average 10-yr average ADI 93% N/M 84% ALTR 75% 82% 74% AVGO 64% 72% N/A LLTC 76% 62% 76% MXIM 75% 66% 56% TXN 57% 93% 72% XLNX 78% 83% 74% Average 74% 76% 73%
Source: Company data, Nomura research
220
Nomura | U.S. Semiconductors Primer December 11, 2013
Investing for the Future In the analog market, research and development is important to drive future revenue growth. This is because analog products have long product life cycles, and multiple product revenue streams overlap each other to drive a growing total revenue stream. Over the past five years, Maxim spent nearly 25% of revenue on R&D, followed by Analog Devices at 19%, Linear Technology at 17%, Avago Technology at 15%, and Texas Instruments at 14%. In contrast, digital companies tend to spend more on R&D expenses as the development costs of each successive generation of process technology increases. For example, Broadcom spent an average of 30% of revenue on R&D over the past 5 years, Marvell at 30%, Nvidia at 25%, Xilinx at 20%, and Altera at 18%. Intel spent only 16% of revenue on R&D primarily due to its scale.
Fig. 44: R&D Expenses as a Percentage of Sales
30%
25%
20%
15%
10% R&D as% of s ale s
5%
0% ADI AVGO LLTC MXIM TXN
2008 2009 2010 2011 2012
Source: Company data, Nomura research
221
Nomura | U.S. Semiconductors Primer December 11, 2013
Focus on Shareholder Returns Long-term investors tend to like the analog semiconductor segment because it offers a stable business model, high barriers to entry, diversified source of income, high cash flow generations, and strong dividend yields. During periods of recovery, analog companies tend to have strong margin leverage. In addition, this segment also has lower headline risks in terms of customer and end market concentration.
Fig. 45: Analog companies offer strong free cash flow and dividend yields As of Nov 15, 2013 FCF Yield (2013E) FCF Yield (2014E) Dividend Yield ADI 5.1% 5.7% 2.7% AVGO 5.3% 6.8% 1.9% LLTC 5.3% 5.7% 2.5% MXIM 7.4% 6.8% 3.3% TXN 5.5% 6.1% 2.6% ALTR 3.4% 4.3% 1.8% XLNX 5.7% 6.4% 2.3% Average 5.4% 6.0% 2.4% Median 5.3% 6.1% 2.5%
Source: Company data, Nomura estimates
Recently, the market has shifted focus to capital allocation. Many analog companies have raised their long-term target to return to shareholders 50% to 100% of their free cash flow generation in the form of share repurchases and dividends. The following chart shows the payout ratio for ADI, AVGO, LLTC, MXIM, and TXN. Year-to-date, the ratio ranges from 51% to 159%.
Fig. 46: Share Repurchases and Dividends
(in $mn) FCF Repurchases Dividends Total Payout % ADI 2011 776 307 287 594 77% 2012 651 120 356 476 73% 2013 YTD 416 5209214 51% AVGO 2011 669 172 98 270 40% 2012 478 44 150 194 41% 2013 YTD 216 49 99 148 69% LLTC 2011 549 64 223 287 52% 2012 529 57 297 354 67% 2013 YTD 401 85 124 209 52% MXIM 2011 682 267 253 520 76% 2012 539 201 269 470 87% 2013 YTD 410 414 214 628 153% TXN 2011 2,442 1,973 644 2,617 107% 2012 2,916 1,800 819 2,619 90% 2013 YTD 1,880 2,134 849 2,983 159%
Source: Company data, Nomura research
222
Nomura | U.S. Semiconductors Primer December 11, 2013
The increased focus on shareholder return has been a major driver for stock performance this year. Year to date, while earnings estimate revisions for analog companies have been flattish to down slightly, the valuation multiples have expanded significantly. The chart below shows the year-to-date stock performance for ADI, AVGO, LLTC, MXIM, and TXN. The average earnings estimates revision for this group is -3% year-to-date, while the average P/E multiple for the group increase from 14x in 2012 to 17x in 2013
Fig. 47: Stock Performance in 2013 YTD As of Nov 12, 2013
Blue percentages represent YTD stock performance 100%
80%
60% 43% 35% 40% 19% 19% 20%
0%
(20%) (1%) (40%)
(60%) ADI AVGO LLTC MXIM TXN YTD Stock Performance Attributed to Multiple Expansion YTD Stock Performance Attributed to Earnings Grow th
Source: FactSet, Nomura research
223
Nomura | U.S. Semiconductors Primer December 11, 2013
Valuation Is Key to Investing in Analog Valuation is a particularly useful tool when investing in analog companies given our view that the business cycle works like a pendulum. There is a high correlation between companies’ valuation multiples to companies’ gross and operating margins. Companies with the highest operating margins tends to trade at a higher EV/S multiple. Companies that are below the trend line are deemed to be undervalued relative to the comp group, while companies above the trend line are deemed to be overvalued. The chart below shows a scatter plot between EV/S multiple and operating margins for a sample size of 24 analog companies, with a very high correlation (r-squared of 0.84). Not surprisingly, the market is assigning a high multiple for companies that have a track record of high profit margins (e.g. LLTC), while companies with low or inconsistent margins are assigned a lower multiple. The correlation between EV/S and gross margin are also high (r-squared of 0.7).
Fig. 48: There is a strong correlation between EV/S and operating margins As of Nov 15, 2013 8.0
7.0 LLTC R² = 0.8412 6.0
5.0 TXN ADI 4.0 AVGO MXIM EV/S (2013E) EV/S 3.0
2.0
1.0
0.0 -10% 0% 10% 20% 30% 40% 50% Ope rating M argin
Source: Company data, Nomura estimates
Given the cyclical nature of the business, it is also useful to compare current valuation to historical periods. Enterprise value-to-sales is a good indicator assuming there are no structural changes to the business. As of November 2013, the analog group traded at 4.4x EV/S, which is higher than the 3-year average for the group. The more commonly used metric is price-to-earnings. As of November 2013, the group trades at 22x 2013 EPS estimates, which is a 3-year high.
224
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 49: EV/S multiples for analog companies As of Nov 15, 2013 EV/S EV/S NTM CY14E 3-yr Max 3-yr Avg 3-yr Min ADI 4.1x4.1x3.0x1.8x ATML 1.9x 3.4x 2.0x 1.0x AVGO 3.3x3.4x2.8x2.1x CY 2.1x 3.5x 2.3x 1.4x FCS 1.0x1.3x1.0x0.6x ISIL 1.9x 2.1x 1.5x 0.9x LLTC 6.3x 6.2x 5.1x 3.9x MXIM 2.9x 3.3x 2.7x 2.0x MCHP 4.2x 4.1x 3.4x 2.7x MSCC 2.6x 2.5x 2.1x 1.7x ONNN 1.3x 1.8x 1.1x 0.8x TXN 3.8x4.0x2.7x1.7x Average 2.9x 3.3x 2.5x 1.7x Median 2.7x 3.4x 2.5x 1.7x
Source: FactSet consensus estimates, Nomura research
Fig. 50: P/E multiples for analog companies As of Nov 15, 2013 P/E P/E NTM CY14E 3-yr Max 3-yr Avg 3-yr Min ADI 19x 20x 16x 11x ATML 13x 22x 14x 8x AVGO 13x 14x 12x 10x CY 17x 21x 15x 10x FCS 17x 23x 15x 7x ISIL 15x N/M N/M 14x LLTC 21x 21x 17x 12x MXIM 15x 18x 15x 12x MCHP 17x 19x 16x 14x MSCC 10x 13x 9x 7x ONNN 10x 16x 11x 6x TXN 20x 20x 16x 10x Average 16x 19x 14x 10x Median 16x 20x 15x 10x
Source: FactSet consensus estimates, Nomura research
225
Nomura | U.S. Semiconductors Primer December 11, 2013
Outlook for 2014E We are cautious on the analog sector heading into 2014. Valuation multiples for analog companies have expanded sharply since the beginning of 2013 likely on hopes that business will turn around quickly as worldwide economy improves. While the economy has improved somewhat, analog companies have not seen a significant pick up in business due to continued cautiousness of end customers, as well as low order lead times and lean inventory in the supply chain. As a result, earnings estimates for analog companies have essentially been unchanged since the beginning of the year. We believe this environment will continue in 2014.
Fig. 51: Since the Beginning of 2013, Multiples Have Expanded but Estimates Are Unchanged
20x 5%
18x 0%
16x -5%
P/E NTM P/E 14x -10% EPS RevisionEPS
12x -15%
10x -20% Jul-12 Jul-13 Oct-12 Oct-13 Jan-12 Jun-12 Jan-13 Jun-13 Apr-12 Apr-13 Feb-12 Mar-12 Feb-13 Mar-13 Sep-12 Nov-12 Dec-12 Sep-13 Nov-13 Aug-12 Aug-13 May - 12 May - 13
Average P/E Average EPS Revision
Source: FactSet, Nomura research
We also note that consensus estimates have already factored a strong recovery in earnings, with EPS growth of about 20% from 4Q13 to 4Q14. Such growth will normally require both revenue acceleration and margin expansion, but we believe those elements are lacking next year. Whereas revenue snapbacks in previous cycles were driven by temporary mismatch of semiconductor shipment and end consumption, we believe the discrepancy is not material (semi growth of 5% yoy in 4Q13E, versus end market growth of 2% yoy) and hence not enough to drive a snapback. Longer term, semiconductors track the end market growth rates.
Fig. 52: Consensus is Modeling Strong YoY EPS Growth
YoY Consensus EPS 1Q13 2Q13 3Q13 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 4Q14 / 4Q13 ADI $0.52 $0.57 $0.62 $0.49 $0.55 $0.60 $0.63 $0.59 21% AVGO $0.61 $0.74 $0.89 $0.79 $0.78 $0.86 $0.94 $0.89 13% LLTC $0.47 $0.43 $0.45 $0.44 $0.47 $0.50 $0.53 $0.52 19% MXIM $0.45 $0.44 $0.41 $0.40 $0.41 $0.47 $0.52 $0.50 27% TXN $0.32 $0.42 $0.61 $0.51 $0.47 $0.57 $0.64 $0.59 16% ALTR $0.37 $0.31 $0.37 $0.30 $0.33 $0.37 $0.40 $0.41 35% XLNX $0.47 $0.56 $0.49 $0.54 $0.57 $0.61 $0.60 $0.60 10%
Source: FactSet consensus estimates, Nomura research
2 26
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 53: No Big Discrepancy Between Semiconductor and End Market Growth Rates
5-yr CAGR 10-yr CAGR Y/Y revenue growth 1Q13 2Q13 3Q13 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2008-2013E 2003-2013E Networking 3% 7% 5% 4% 6% 5% 4% 5% Wireless 2% -3% -3% 0% 1% 6% 8% 5% Storage -3% -6% -3% -1% 1% 1% 3% 2% Automotive 5% 13% 12% 8% 11% 6% 6% 4% Consumer -9% -11% -10% -4% 2% 12% 10% 1% Industrial 1% 2% 4% 5% 5% 5% 6% 5% PC -2% -1% 0% -1% -1% -1% 1% 1% Total End Market*0%1%2%2%4%4%5%3%3%5%
Total Semiconductors** -2% -4% -1% 5% 6% 7% 6% 8% 2% 4%
* includes more than $1 trillion of end market revenue annually ** semiconductors include ADI, LLTC, MXIM, TXN, ISIL, MCHP, ALTR and XLNX
Source: FactSet consensus estimates, Nomura research
Furthermore, we believe margin expansion will be limited in 2014. Consensus is already forecasting gross margins for mixed-signal companies to reach a record-high in 2014. And while consensus’ operating margins for 2014 are still a few percentage points below 2010 record levels, we believe many companies will need to start spending more on operating expenses after several years of tight expense controls. As such, we believe margin leverage will be limited next year.
Fig. 54: Gross Margins for Mixed Signal Companies, 2006–2014E
Gross Margin 2006 2007 2008 2009 2010 2011 2012 2013 2014 ADI 60% 59% 60% 56% 66% 66% 64% 64% 65% AVGO 37% 40% 43% 44% 51% 51% 51% 51% 51% LLTC 78% 77% 77% 75% 78% 77% 75% 75% 76% MXIM 62% 62% 58% 56% 63% 63% 62% 62% 60% TXN 51% 53% 50% 48% 54% 49% 49% 52% 54% ALTR 67% 65% 67% 67% 71% 70% 70% 68% 68% XLNX 61% 62% 64% 63% 65% 65% 66% 68% 69%
Average 59% 60% 60% 58% 64% 63% 62% 63% 63% Median 61% 62% 60% 56% 65% 65% 64% 64% 65%
Source: FactSet consensus estimates, Nomura research
Fig. 55: Operating Margins for Mixed Signal Companies, 2006–2014E
EBIT Margin 2006 2007 2008 2009 2010 2011 2012 2013 2014 ADI 23% 21% 23% 19% 34% 35% 31% 31% 32% AVGO 7% 14% 16% 19% 29% 30% 30% 29% 30% LLTC 51% 48% 47% 41% 52% 49% 45% 45% 46% MXIM 20% 25% 20% 15% 29% 28% 26% 26% 25% TXN 24% 25% 19% 19% 31% 22% 15% 23% 25% ALTR 23% 22% 30% 25% 44% 41% 33% 27% 29% XLNX 21% 21% 25% 21% 33% 29% 26% 30% 33%
Average 24% 25% 26% 23% 36% 33% 30% 30% 31% Median 23% 22% 23% 19% 33% 30% 30% 29% 30%
Source: FactSet consensus estimates, Nomura research
227
Nomura | U.S. Semiconductors Primer December 11, 2013
Memory The memory segment represented a $56bn market in 2012, down from $61bn in 2011. In 2012, memory accounted for 20% of the semiconductor industry. Over the past 10 years, memory revenue grew at a CAGR of 8%, but year-over-year growth rates ranged from 56% growth to 20% decline. Memory represented between 19% and 24% of the total semiconductor industry over the past 10 years. Gartner forecasts that the memory market will grow 25% in 2013 to $69bn and 7% in 2014 to $74bn.
Fig. 1: Memory IC Revenue, 2000–2014E
80,000
70,000
60,000
50,000
40,000
30,000 Revenue (in $mn) (in Revenue 20,000
10,000
0
Source: Gartner, Nomura research
In 2012, Samsung led the memory market with a 34% share, followed by SK Hynix (16%), Micron Technology (12%), Toshiba (10%), and SanDisk (7%). The top 10 suppliers accounted for 92% of the memory market.
Fig. 2: Memory IC Market Share, 2012
Spansion Macronix Others 2% 1% 8% Nanya 2% Intel 2% Elpida Samsung 6% 34%
SanDisk 7%
Toshiba 10%
SK Hynix Micron 16% 12%
Source: Gartner, Nomura research
228
Nomura | U.S. Semiconductors Primer December 11, 2013
In 2012, DRAM and NAND flash accounted for the majority of the memory market, accounting for 47% and 42% of sales, respectively. The next biggest segment was NOR flash, which accounted for only 6% of sales. The composition was significantly different than in 1995, when DRAM accounted for 77% of sales followed by SRAM (11%) and NOR (4%), while NAND flash sales was not meaningful. Over the next five years, we expect NAND to surpass DRAM in terms of revenue. We believe NOR will become a smaller portion of the memory market as mobile phones continue to transition to a DRAM–NAND architecture. In this report, we focus on the two primary markets, DRAM and NAND.
Fig. 3: Memory Market by Segment, 1995–2012 As percentage of memory market 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
DRAM NAND Flash NOR Flash Other
Source: Gartner, Nomura research
Difference between DRAM and NAND Dynamic Random Access Memory (DRAM) is commonly used for random access memory in personal computers and workstation. A typical PC today has about 4 GB of DRAM content. An increasing portion of the DRAM market is being used in smartphones and tablets. Data in DRAM are stored in memory cells in the form of 0s and 1s. DRAM is a volatile memory, which means it loses the data stored in the memory cell once the electricity is cut off. DRAM is dynamic in that it needs to have storage cells refreshed every few milliseconds to maintain its content. Random access means that the microprocessor can access any part of the memory directly, rather than searching the data in a serial manner. NAND flash is a type of non-volatile storage technology that does not require power to retain the data. There are two types of flash memory, NAND and NOR. NAND flash is primarily used for data storage in devices such as memory cards, USB drives, smartphones, tablets, and solid state drives. A typical memory card today comes with 16 to 32 GB of NAND flash, while a solid state drive has 128 to 256 GB of NAND flash. NAND flash is significantly cheaper than DRAM per gigabyte. While NAND can be read fast, writing to the NAND flash is usually very slow relative to other memory technologies. Another problem with NAND flash is that it has a limited number of write-cycles, typically in the range of a few thousand cycles. In contrast, NOR flash is used for code execution and can be found mostly in mobile phones and embedded systems. They are faster but a lot more expensive than NAND.
229
Nomura | U.S. Semiconductors Primer December 11, 2013
Consolidation in the memory market In the past two years, the memory market went through another round of consolidation as Micron successfully acquired Elpida, purchased a majority share in Rexchip and restructured its manufacturing agreements with Inotera and Intel. Today, the DRAM market has three major groups of suppliers – Samsung, Micron, and SK Hynix; while the NAND market has four major groups of suppliers – Samsung, Toshiba/SanDisk, Micron/Intel, and SK Hynix. Samsung and Micron also produce NOR flash, making them the only two companies in the world that supply all three major memory types of products. SK Hynix supplies both DRAM and NAND, while Toshiba and SanDisk only produce NAND flash. Other suppliers (including Spansion, Nanya, Macronix, Powerchip, ProMOS, and Winbond) collectively own less than 10% of the memory market and focus primarily on niche or specialty memory. The most recent round of consolidation has had a profound impact on the competitive landscape, especially in the DRAM market. We believe fewer competitors will lead to more rational behaviors in terms of capital spending and pricing strategy, which in turn should drive more stable and likely more profitable business cycles for the remaining suppliers.
Fig. 4: Competitive Landscape in Memory
Source: Micron
230
Nomura | U.S. Semiconductors Primer December 11, 2013
DRAM Market In 2012, the DRAM segment represented the largest segment within the memory market, accounting for 47% of total memory sales. DRAM revenue of $26bn in 2012 declined by 11% year over year, with unit shipments (in gigabyte) increasing by 32%, offset by ASP (per gigabyte) decline of 33%. ASPs of commodity DRAM (primarily used in PCs) declined by 44%. Given the strong recovery in DRAM pricing since the end of 2012, we forecast that DRAM revenue will increase 33% in 2013, driven by 29% increase in unit shipments and 5% increase in ASPs.
Fig. 5: DRAM Market Revenue, 1990–2014E
45,000
40,000
35,000
30,000
25,000
20,000
15,000
10,000
5,000
0
Source: Gartner, Nomura estimates
In 2012, Samsung was the largest DRAM supplier with 41% market share, followed by SK Hynix (25%), Elpida (13%), Micron Technology (12%), and Nanya (4%). The 2013 acquisition of Elpida by Micron effectively doubled Micron’s market share to 25%.
Fig. 6: Market Share in DRAM, 2012
Others Nanya 5% 4%
Micron 12%
Samsung 41%
Elpida 13%
SK Hynix 25% Samsung SK Hynix Elpida Micron Nanya Others
Source: Gartner, Nomura research
231
Nomura | U.S. Semiconductors Primer December 11, 2013
The DRAM market has gone through many business cycles, with revenue peaks in 1995, 2000, 2006, and 2010. Each peak was followed by a year of 50%+ decline in prices. To demonstrate the volatility of the business, ASPs over the last 10 years declined at a CAGR of 26% per year, but the year-over-year change ranged from 52% decline to 20% growth. In terms of units, bit shipments grew at a CAGR of 51% over this period. In the next few years, we expect industry bit growth to decelerate to 20–30% per year as PC growth declines and memory content growth decelerates. We do not believe new operating systems (e.g., Windows 8) will drive a need for higher memory configurations. However, moderate capacity additions will likely lead to less severe price erosion of 15– 25% per year, leading to revenue CAGR of 0–5% over the next few years. By application, PC has been the workhorse, as DRAM content per PC quadrupled over the past five years. However, DRAM shipments to the PC market recently decelerated. In 2012, PC DRAM bit volume increased only 17%, versus non-PC DRAM growth of 39%. As a result, PC contribution to the DRAM market declined from 47% in 2011 to 42% in 2012. We estimate PC contribution will continue to decline in 2013 to 30–35%. Non-PC DRAM products, also known as specialty DRAM, are shipped for servers, storage, networking, industrial, and consumer applications. Included in specialty DRAM is a variant called mobile DRAM, which is used in mobile applications such as smartphones and tablets due to its lower power characteristics. Mobile DRAM grew from 11% of industry shipment in 2011 to 18% in 2012, and we expect mobile DRAM to grow to 30% of the industry shipments in 2013. Specialty DRAM generally commands a price premium over commodity PC DRAM.
Fig. 7: DRAM End Market Mix
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
PC Mobile Server Graphic Others
Source: Nomura estimates
232
Nomura | U.S. Semiconductors Primer December 11, 2013
Consolidation Should Bring More Stability to the Market The overarching theme of the DRAM industry in 2013 is industry consolidation. The DRAM market has attracted many competitors in the past two decades, but the capital intensity and volatility of the business have driven many suppliers out of the market, including Intel in 1985 and Texas Instruments in 1998. Meanwhile, the market attracted new competition from Korea and Taiwan in the 1990s. While there was no meaningful consolidation in the early part of 2000s, the financial crisis in 2008–2009 forced DRAM suppliers to form five competing groups. The industry lost another supplier when Qimonda filed for bankruptcy in 2009. With the slowdown in PC growth since 2010, many Taiwanese suppliers that were technology laggards and cash-strapped decided to cut back on PC DRAM production. In 2013, Micron completed its acquisition of Elpida, which filed for bankruptcy protection in February 2012. As a result, the market has now consolidated to three DRAM groups—Samsung, SK Hynix, and Micron.
Fig. 8: DRAM Market Competitive Landscape
60000 5 DRAM ~20 DRAM ~10 DRAM groups suppliers suppliers 4 DRAM groups
50000 Qimonda 3 DRAM Taiwan fall out groups Korean suppliers enter suppliers enter Taiwanese suppliers exit 40000 TI and LG drop out MicronMicron acquired buys EplidaElpida (pending)
30000
20000 Industry revenue ($mn)
10000
0
Source: Company data, Gartner, SIA, Nomura estimates
With only three major suppliers remaining, we expect the industry as a whole to act more rationally when it comes to pricing strategy and capital spending. In the past, smaller suppliers would try to gain share by adding new capacity funded by profits they earned in an up-cycle. Oftentimes, the excess investment led to inventory build in the supply chain when demand decelerated, which in turn led to very aggressive pricing. Given its technology leadership, Samsung historically weathered the down-cycles much better than its competitors. While we believe Samsung continues to have a technology lead over SK Hynix and Micron, we think Samsung is less inclined to use pricing as the primarily means to compete given the technology (and hence cost) gap between the remaining competitors is smaller than in the past. Instead, all three suppliers are focusing on profitability by cutting back on capital spending and shifting production to more profitable segments such as mobile DRAM. We thus believe the DRAM market will be more stable and likely more profitable in the future.
233
Nomura | U.S. Semiconductors Primer December 11, 2013
Technology Migration Drives Price Decline DRAM ASP per gigabit (Gb) declined by an average of 28% per year over the past 20 years, ranging from -77% in 2001 to +20% in 2010. The decline was made possible by aggressive process technology migration, which allowed more transistors to be packed in a given die area. In general, DRAM suppliers were able to move from one technology node to another over a period of 6 to 8 quarters, which provided a cost reduction of 40– 50% per Gb over the same period. The primary technology process node in 2012 was 3x-nanomenter, accounting for nearly 50% of total production. Cost savings are generally passed downstream, resulting in similar decline in ASP. As the chart below shows, ASP increase did not happen often, but when it did, it was usually driven by over- correction in the supply chain.
Fig. 9: DRAM ASP Year-over-Year Change
40%
20%
0%
-20%
-40%
-60%
-80%
-100% 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E
ASP Y/Y
Source: Company data, Gartner, Nomura estimates
In conjunction with process technology transitions, the density of a monolithic, or single, die also increases. In 2012, the mainstream density was 2Gb, accounting for 62% of all production. The average density of a monolithic die increased at a 29% CAGR over the past five years, from 450Mb in 2007 to 1.6Gb in 2012.
Fig. 10: Technology Migration in the DRAM Industry, 2008–2013E
100.0% 90.0% 80.0% 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% 2008 2009 2010 2011 2012 2013E 0.10/0.11um 90nm 80nm 7Xnm 6Xnm 5Xnm 4Xnm 3Xnm 2Xnm
Source: Company data, Nomura estimates
234
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 11: Average Capacity of Monolithic DRAM Die Increases
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2010 2011 2012 2013E 2014E 2015E 2016E 2017E
< 1Gbit 1Gbit 2Gbit 4Gbit 8Gbit 16Gbit
Source: Gartner. Company data, Nomura estimates
By supplier, Samsung remains in the lead as it has been progressing its technology migration into 25nm starting in 1Q13 and appears to be ahead of its competition by 6 to 12 months. Following the successful migration into 29nm, SK Hynix launched its technology migration process into 25nm starting from 4Q13. Micron/Inotera fully converted to 30nm in mid-2013 but do not expect the transition to 20nm until mid-2014. Elpida, which is now part of Micron, will proceed to transition to 25nm by the end of 2013, before converting to 20nm by mid-2014 The manufacturing cost dictates how aggressive any given DRAM supplier can be in the marketplace. Given the high fixed cost of the business, suppliers are incentivized to run their factories at full utilization, as long as the average selling price still exceeds cash cost. As such, it is not uncommon for DRAM suppliers to sell components at a loss, hoping that the market would turnaround fast enough before their balance sheets significantly deteriorate. It is when ASP falls below cash cost that suppliers will consider cutting utilization. In the past, large DRAM suppliers have also used their cost leadership to set lower prices in order to gain market share, which essentially squeezed the margins for suppliers that have less competitive cost structure.
Fig. 12: Examples of DRAM Cash Cost and Full Loaded Cost
Total Fabricated Costs (Manufacturing plus Packaging and Test) Process Technology 40 nm 32 nm 25 nm 25 nm Chip Density 2 Gb 2 Gb 2 Gb 4 Gb Cash cost per 2Gb equiv $0.80 - $0.90 $0.60 - $0.70 $0.50 - $0.60 $0.40 - $0.50 Full loaded cost per 2Gb equiv $1.40 - $1.50 $1.10 - $1.20 $0.90 - $1.10 $0.80 - $0.90
Source: Gartner, Nomura estimates
235
Nomura | U.S. Semiconductors Primer December 11, 2013
Capital Spending Is Key to DRAM Cycles The DRAM industry is very capital intensive. It costs $4–5bn for a new fab with a capacity of 100,000 wafers per month, and the cost for technology migration (sometimes referred to as “maintenance capex”) costs about $50mn per year for every 10,000 wafers per month of capacity. Since 2000, the DRAM industry spent about 40% of its cumulative revenue on capital expenditures. However, DRAM capex has seen a decline from a peak of $21bn (68% of sales) in 2007 to $5bn (16% of sales) in 2013E. The estimated capital intensity in 2013 is even lower than the trough of 19% during the financial crisis in 2009. After two years of under-investment, we expect DRAM capex to increase about 40% in 2014, although the implied capex-to-sales ratio of 20% is still significantly below the historical average. We expect the majority of the investment to focus on technology transitions for mobile DRAM.
Fig. 13: DRAM Capital Spending, 2000–2014E
25,000 80%
70% 20,000 60%
15,000 50% 40%
10,000 30%
20% 5,000 10% DRAM Capital Spending Capital inDRAM $mn
0 0%
Total DRAM Capex Capex as % of sales
Source: Company data, Nomura estimates
Capital spending has been the main driver for DRAM business cycles that are oftentimes different from the business cycles for the broader semiconductor industry. To illustrate, we examined the capital spending starting in 2006. The aggressive capital spending in 2006–2007 drove significant oversupply in 2008. Despite 62% bit shipment growth in 2008, ASP declined by 59%, resulting in total DRAM revenue declining 22% year over year. The problem was further exacerbated by the financial crisis in 2009, as bit demand growth sharply decelerated to only 23%. In response, suppliers not only cut back on utilization but also took most of the 200mm wafer capacity offline that were deemed to be not cost competitive. These actions resulted in industry capacity declining nearly 33% between mid-2008 and early 2009. While fab utilization increased following the financial crisis, the slowing of PC growth and further industry consolidation prompted DRAM suppliers to cut back on utilization again in 2011. Meanwhile, DRAM suppliers have been aggressively switching their product mix from PC DRAM to mobile DRAM and NAND, further limiting DRAM supply growth. By the end of 2012, total industry wafer starts were 30% below the 2008 peak. The sharp reduction in industry supply set the stage for strong pricing recovery in 2013.
236
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 14: DRAM Wafer Capacity, 2008–2014E
1,400 Actual 1,300 Forecast
1,200
1,100
1,000
900 Wafer start (300mm x1000 equiv)
800
Source: Company data, Nomura estimates
The cost of scaling also plays a role in determining bit growth. As DRAM scaling becomes increasingly difficult due to physical limitations, the cost of each technology migration increases. At the same time, the amount of bit growth declines for each successive shrink. The chart below shows that the transition from 44nm to 38nm drove about 60% increase in bit growth per wafer, but the transition from 25nm to 20nm yields only a 35% bit increase. Furthermore, the cost for the technology migration jumped from $100mn per 10,000 wafers for the 38nm transition to $200mn per 10,000 wafers for the 20nm transition. We believe the reduced level of capital investment in DRAM and the increasing cost of scaling will drive future industry bit supply growth significantly below the historical average (10-year CAGR of 48%).
Fig. 15: % Bit Growth per Wafer and CAPEX for 10k Capa on Each Tech Migration Phase
(US$mn) % Bit growth per wafer (RHS) CAPEX for 10k capa (LHS) 250 80% 70% 200 60% 150 50% 40% 100 30% 20% 50 10% 0 0% 54nm to 44nm 44nm (8F2) to 38nm to 29nm 29nm to 25nm 25nm to 20nm 38nm (6F2)
Source: Nomura estimates
237
Nomura | U.S. Semiconductors Primer December 11, 2013
The Importance of PC Is Diminishing In the last decade, PC was the primary market for DRAM consumption, accounting for 50–60% of DRAM demand. The DRAM industry growth is a function of PC unit growth and the growth in average DRAM content per PC, both of which are experiencing decelerating growth. As a result, PC reached a peak of 57% of total DRAM shipments in 2009, and has since declined steadily to 42% in 2012. We expect the percentage to further decline to 32% in 2013 and 25% in 2014.
Fig. 16: PC DRAM as a Percentage of Total DRAM Shipments Declines Units in millions 14,000 70%
12,000 60%
10,000 50%
8,000 40%
6,000 30%
4,000 20%
2,000 10%
0 0% 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
PC DRAM Shipments (1-Gb equiv) PC As % of DRAM Bit Shipments
Source: Gartner,Nomura estimates
In 2012, PC units declined for the first time since 2001, declining by 4% year over year. With smartphones and tablets providing other ways to connect to the Internet, we expect PC shipment growth to remain muted. This is in sharp contrast to the historical average growth of 10–15% per year in the last decade. In 2013, sales of notebook PCs continued to be sluggish despite catalysts such as Windows 8 and touch-based notebooks.
Fig. 17: Traditional PC Growth Is Slowing
400 20%
350 15%
300 10% 250 5% 200 0% 150 -5% 100
50 -10%
0 -15% 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
PC units (in mn) PC units YoY
Source: Company data, Gartner, Nomura estimates
238
Nomura | U.S. Semiconductors Primer December 11, 2013
In addition, memory content growth per PC is slowing. Historically, every new Microsoft operating system significantly raised the DRAM requirement of a PC system. For example, Windows 2000 raised the recommended DRAM content from 64 MB to 128 MB, Windows XP increased it to 256 MB, and Windows Vista raised it further to 1 GB. As such, up until 2008, memory content per PC was growing at a CAGR of 40–45%. In 2007 memory content per PC grew 96%.
Fig. 18: PC DRAM Content Growth Decelerates
120%
100%
80%
60%
Bit growth Bit YoY 40%
20%
0% 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Source: Gartner, Nomura estimates
However, since the end of 2008, DRAM content growth has decelerated sharply to only 15% CAGR between 2008 and 2012. Windows 7, which was launched in 2009, had the same minimum DRAM requirement of 1 GB as the Windows Vista for 32-bit systems (2 GB for 64-bit systems), at a time when the average DRAM content was already close to 2 GB. In 2012, Microsoft released Windows 8 with the same minimum DRAM requirements of 1 GB for 32-bit systems and 2GB for 64-bit systems (although Microsoft recommends DRAM content of 4 GB); however, the average DRAM content per system was already above 4 GB in 2012. Although PC performance can still benefit from higher memory content, we expect memory content to increase only gradually in the future.
Fig. 19: DRAM Content Growth in PCs Decelerates
9,000
8,000
7,000
6,000 2007: Window s 7 5,000 Min. DRA M 1 GB 4,000
3,000 2012: Window s 8 Min. DRAM 1 GB
DRAM per MB) (in PC DRAM 2,000
1,000
0
Average DRAM Content per PC Expon. (Average DRAM Content per PC)
Source: Gartner, Nomura estimates
239
Nomura | U.S. Semiconductors Primer December 11, 2013
Mobile DRAM to Exceed PC DRAM in 2014 The emergence of smartphones and tablets are driving strong demand for low power DDR SDRAM (LPDDR), which is also known as mobile DRAM. LPDDR is a variant of DDR SDRAM, with additional features that reduce overall power consumption. Example of power-savings features include temperature-compensated self refresh (TCSR), partial array self refresh (PASR), deep power-down (DPD), and clock stop mode, which are not found in standard DDR SDRAM. The commonly used LPDDR today is LPDDR2, which features operating speed of up to 1066 Mbps at 1.2V power supply. The industry has begun shipping LPDDR3, which features transfer rate up to 1600 Mbps, and is about 50% faster than LPDDR2. We expect mobile DRAM to grow from 18% of total DRAM shipments in 2012 to 30% in 2013 and 36% in 2014. By 2014, mobile DRAM bit shipments should exceed PC DRAM (estimated 25% of total bit shipments) The increase in mobile DRAM bit shipment is driven by an increase in unit shipments of mobile devices, as well as an increase in DRAM content per mobile device.
Fig. 20: Mobile DRAM as a Percentage of Total DRAM Shipments Increases Rapidly
18,000 40%
16,000 35% 14,000 30% 12,000 25% 10,000 20% 8,000 15% 6,000 10% 4,000
2,000 5%
0 0% 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Mobile DRAM Shipments (1-Gb equiv) Mobile As % of DRAM Bit Shipments
Source: Gartner, Nomura estimate
Most recent high-end smartphones and tablets, such as Samsung Galaxy S4, Galaxy Note 10.1, HTC One, and Blackberry Z10, are equipped with 2GB of LPDDR, and some smartphones such as Samsung Galaxy Note III comes with 3GB of LPDDR. The notable exception is Apple, which is still using 1GB of LPDDR in the latest generation of iPhones and iPads. Furthermore, we see an increasing need for mobile devices to use more mobile DRAM as the application processors move from 32-bit to 64-bit over the next few years. For example, Apple is already using a 64-bit application processor in both iPhone 5S and iPad Air. Mobile DRAM is not only used in high end smartphones and tablets. Low-end to mid- range 3G/4G smartphones are mostly shipping with 256MB to 1GB of mobile DRAM, as opposed to 2G feature phones that typically do not use any mobile DRAM (they use NOR flash instead). As such, mobile DRAM should benefit as the market continues to shift from 2G phones to 3G/4G phones in the next few years. Given strong growth potential of mobile DRAM and reduced demand in the PC market, most DRAM suppliers have been shifting capacity from PC DRAM to mobile DRAM. We expect Samsung and Hynix to be the most aggressive in terms of pushing technology migration in the mobile DRAM market.
240
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 21: Mobile DRAM Benefits from Both Unit and Content Increases (in mn) 1400 1,400
1200 1,200
1000 1,000
800 800
600 600
400 400
200 200
0 0 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Smartphone units (in mn) Average DRAM Content (Mb)
Source: Gartner, Nomura estimates
In 2012, Samsung led the mobile DRAM with 53% market share, mostly benefiting from very strong sales to Samsung’s handset division. This is followed by Hynix with 26%, and Elpida with 20%. Elpida benefited from its strong relationship with Apple, but Elpida’s share is low given Apple’s mobile DRAM usage is generally lower than many of its competitors. Micron has very limited share in 2012 given its relationships were mostly with Blackberry and Nokia, which were both losing share. The acquisition of Elpida has significantly strengthened Micron’s position in the mobile DRAM market.
Fig. 22: Mobile DRAM Market Share, 2012
Micron 1% Elpida 20%
Samsung 53%
Hynix 26%
Samsung Hynix Elpida Micron
Source: Company data, Nomura estimates
241
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 23: Nomura DRAM Industry Model
2011 2012 1Q13 2Q13 3Q13 4Q13E 2013E 1Q14E 2Q14E 3Q14E 4Q14E 2014E demand (1Gb eq mn) 21,797 28,043 7,710 8,419 9,376 10,839 36,343 10,036 10,651 12,022 13,921 46,629 y-y 44.1% 28.7% 19.8% 25.0% 32.4% 39.1% 29.6% 30.2% 26.5% 28.2% 28.4% 28.3% q-q -1.1% 9.2% 11.4% 15.6% -7.4% 6.1% 12.9% 15.8% t PC DRAM demand (1Gb eq mn) 10,166 11,876 2,773 2,809 3,020 3,073 11,674 2,831 2,853 3,050 3,157 11,891 y-y 27.0% 16.8% -6.3% -2.3% 0.7% 0.9% -1.7% 2.1% 1.6% 1.0% 2.7% 1.9% q-q -8.9% 1.3% 7.5% 1.8% -7.9% 0.8% 6.9% 3.5% ent PC shipment ('000 units) 363,915 349,383 76,777 75,644 79,923 80,974 313,318 74,223 74,097 78,305 79,790 306,415 % y-y 1.7% -4.0% -13.3% -11.4% -9.5% -7.0% -10.3% -3.3% -2.0% -2.0% -1.5% -2.2% % q-q -11.8% -1.5% 5.7% 1.3% -8.3% -0.2% 5.7% 1.9% ntents/box (MB) 3,576 4,351 4,769 4,967 % y-y 24.9% 21.7% 9.6% 4.2% er DRAM demand (1Gb eq mn) 4,008 5,169 1,542 1,731 1,940 2,149 7,361 2,133 2,376 2,746 3,032 10,286 y-y 101.5% 29.0% 32.7% 42.1% 44.6% 48.4% 42.4% 38.4% 37.3% 41.6% 41.1% 39.7% q-q 6.5% 12.3% 12.1% 10.8% -0.8% 11.4% 15.6% 10.4% RAM demand (1Gb eq mn) 3,450 3,773 839 936 1,083 1,341 4,199 1,111 1,158 1,331 1,719 5,320 y-y 23.8% 9.4% -10.9% -5.8% 26.6% 36.5% 11.3% 32.5% 23.7% 22.9% 28.2% 26.7% q-q -14.7% 11.7% 15.7% 23.8% -17.1% 4.2% 14.9% 29.1% hic DRAM demand (1Gb eq mn) 896 1,036 294 293 358 496 1,441 373 375 483 744 1,976 y-y 3.6% 15.6% 40.1% 29.2% 33.0% 50.1% 39.1% 26.7% 28.2% 34.9% 50.0% 37.1% q-q -10.8% -0.6% 22.4% 38.4% -24.7% 0.6% 28.8% 53.9% e DRAM demand (1Gb eq mn) 2,457 5,091 2,087 2,450 2,763 3,555 10,855 3,350 3,646 4,164 5,019 16,179 y-y 115.3% 107.2% 124.9% 108.1% 91.1% 93.9% 113.2% 60.5% 48.8% 50.7% 41.2% 49.0% q-q 13.8% 17.4% 12.8% 28.7% -5.8% 8.8% 14.2% 20.5% supply (1Gb eq mn) 22,476 28,521 8,362 8,962 9,801 9,268 36,392 9,981 11,027 12,161 13,141 46,310 y-y 48.4% 26.9% 28.9% 25.9% 41.9% 15.7% 27.6% 19.4% 23.0% 24.1% 41.8% 27.3% q-q 4.4% 7.2% 9.4% -5.4% 7.7% 10.5% 10.3% 8.1% and/supply 0.97 0.98 0.92 0.94 0.96 1.17 1.00 1.01 0.97 0.99 1.06 1.01 shipment value ($mn) 29,368 26,143 7,074 8,659 9,678 9,453 34,864 9,782 10,035 10,094 10,250 40,159 y-y -25.1% -11.0% 13.7% 25.3% 51.5% 42.8% 33.4% 38.3% 15.9% 4.3% 8.4% 15.2% q-q 6.8% 22.4% 11.8% -2.3% 3.5% 2.6% 0.6% 1.5% age unit price (1Gb eq $) 1.31 0.92 0.85 0.97 0.99 1.02 0.96 0.98 0.91 0.83 0.78 0.87 y-y -49.5% -29.8% -11.8% -0.5% 6.7% 23.4% 4.5% 15.8% -5.8% -15.9% -23.5% -9.5% q-q 2.4% 14.2% 2.2% 3.3% -3.9% -7.1% -8.8% -6.0%
Source: Company data, Nomura estimates
Fig. 24: DRAM supply forecasts
2011 2012 1Q13 2Q13 3Q13 4Q13E 2013E 1Q14E 2Q14E 3Q14E 4Q14E 2014E Total Supply (1Gb eq. mn) 22,476 28,521 8,362 8,962 9,801 9,268 36,392 9,981 11,027 12,161 13,141 46,310 y/y 48.4% 26.9% 28.9% 25.9% 41.9% 15.7% 27.6% 19.4% 23.0% 24.1% 41.8% 27.3% q/q 4.4% 7.2% 9.4% -5.4% 7.7% 10.5% 10.3% 8.1% Samsung Electronics 8,262 10,550 2,842 2,977 3,443 4,006 13,268 4,126 4,363 4,602 4,853 17,945 y/y 50.1% 27.7% 22.5% 16.7% 28.9% 33.2% 25.8% 45.2% 46.5% 33.7% 21.1% 35.2% q/q -5.5% 4.8% 15.6% 16.4% 3.0% 5.7% 5.5% 5.4% SK Hynix 4,935 7,444 2,282 2,748 2,706 2,311 10,048 2,561 3,174 3,672 4,079 13,487 y/y 51.2% 50.9% 35.6% 51.5% 56.3% 4.3% 35.0% 12.2% 15.5% 35.7% 76.5% 34.2% q/q 3.0% 20.4% -1.5% -14.6% 10.8% 24.0% 15.7% 11.1% MU group (MU+Elpida + Rexchip + Inotera) 6,991 8,239 2,540 2,698 2,770 2,512 10,520 2,638 2,825 3,230 3,553 12,246 y/y 17.8% 37.3% 27.0% 32.4% 15.6% 27.7% 3.9% 4.7% 16.6% 41.4% 16.4% q/q 16.9% 6.2% 2.7% -9.3% 5.0% 7.1% 14.3% 10.0% Nanya own 536 495 70 138 275 311 794 321 328 336 341 1,325 y/y 55.9% -7.8% -63.7% -8.2% 744.7% 161.6% 60.5% 358.5% 136.9% 22.1% 9.6% 66.8% q/q -41.2% 98.0% 98.6% 13.1% 3.1% 2.3% 2.3% 1.5% Powerchip own 667 664 130 180 185 190 685 185 185 170 160 700 y/y 10.0% -0.5% -20.7% 0.0% 2.8% 35.7% 3.2% 42.3% 2.8% -8.1% -15.8% 2.2% q/q -7.1% 38.5% 2.8% 2.7% -2.6% 0.0% -8.1% -5.9% Winbond 99 148 33 36 38 38 145 40 41 41 42 164 y/y -34.1% 49.9% -5.7% 0.0% 0.0% -2.6% -2.0% 21.2% 13.9% 7.9% 10.5% 13.1% q/q -15.4% 9.1% 5.6% 0.0% 5.3% 2.5% 0.0% 2.4% ProMOS 230200000000000 y/y -23.3% -91.3% q/q
Source: WSTS, Nomura estimates
242
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 25: DRAM demand forecasts – PC, server and workstation only Units in Gb Computer Total 14,174 17,045 4,315 4,540 4,959 5,222 19,035 4,964 5,229 5,796 6,189 22,178 y/y 41.8% 20.3% 4.7% 11.0% 14.3% 16.3% 11.7% 15.1% 15.2% 16.9% 18.5% 16.5% q/q -3.9% 5.2% 9.2% 5.3% -4.9% 5.3% 10.8% 6.8% PC Total 12,125 14,178 3,410 3,530 3,844 4,002 14,786 3,717 3,833 4,206 4,417 16,174 y/y 34.8% 16.9% -2.6% 3.6% 7.0% 8.9% 4.3% 9.0% 8.6% 9.4% 10.4% 9.4% q/q -7.2% 3.5% 8.9% 4.1% -7.1% 3.1% 9.7% 5.0% Shipment (units in 000) 371,930 357,392 78,636 77,593 82,014 83,215 321,459 76,244 76,119 80,526 82,112 315,001 y/y 1.7% -3.9% -13.1% -11.2% -9.3% -6.6% -10.1% -3.0% -1.9% -1.8% -1.3% -2.0% q/q -11.8% -1.3% 5.7% 1.5% -8.4% -0.2% 5.8% 2.0% Average contents per box (MB) 4,173 5,078 5,550 5,823 6,000 6,156 5,887 6,241 6,445 6,686 6,886 6,572 y/y 32.5% 21.7% 12.1% 16.7% 17.9% 16.6% 15.9% 12.4% 10.7% 11.4% 11.9% 11.6% q/q 5.1% 4.9% 3.0% 2.6% 1.4% 3.3% 3.7% 3.0% Client PC Total 10,166 11,876 2,773 2,809 3,020 3,073 11,674 2,831 2,853 3,050 3,157 11,891 y/y 27.0% 16.8% -6.3% -2.3% 0.7% 0.9% -1.7% 2.1% 1.6% 1.0% 2.7% 1.9% q/q -8.9% 1.3% 7.5% 1.8% -7.9% 0.8% 6.9% 3.5% Shipment (units in 000) 363,915 349,383 76,777 75,644 79,923 80,974 313,318 74,223 74,097 78,305 79,790 306,415 y/y 1.7% -4.0% -13.3% -11.4% -9.5% -7.0% -10.3% -3.3% -2.0% -2.0% -1.5% -2.2% q/q -11.8% -1.5% 5.7% 1.3% -8.3% -0.2% 5.7% 1.9% Average contents per box (MB) 3,576 4,351 4,623 4,753 4,836 4,857 4,769 4,883 4,929 4,986 5,064 4,967 y/y 24.9% 21.7% 8.1% 10.4% 11.3% 8.5% 9.6% 5.6% 3.7% 3.1% 4.3% 4.2% q/q 3.3% 2.8% 1.8% 0.4% 0.5% 0.9% 1.2% 1.6% Desktops 4,122 4,906 1,166 1,166 1,215 1,247 4,794 1,189 1,191 1,233 1,271 4,884 y/y 12.9% 19.0% -7.7% -5.1% 1.4% 2.5% -2.3% 2.0% 2.1% 1.5% 1.9% 1.9% q/q -4.2% 0.1% 4.2% 2.6% -4.7% 0.2% 3.6% 3.0% Shipment (units in 000) 154,785 148,329 33,909 33,173 34,180 34,693 135,955 33,076 32,780 33,586 34,245 133,686 y/y -1.4% -4.2% -12.2% -9.4% -6.4% -5.1% -8.3% -2.5% -1.2% -1.7% -1.3% -1.7% q/q -7.2% -2.2% 3.0% 1.5% -4.7% -0.9% 2.5% 2.0% Average contents per box (MB) 3,409 4,234 4,400 4,500 4,550 4,600 4,513 4,600 4,650 4,700 4,750 4,676 y/y 14.5% 24.2% 5.2% 4.8% 8.3% 8.0% 6.6% 4.5% 3.3% 3.3% 3.3% 3.6% q/q 3.3% 2.3% 1.1% 1.1% 0.0% 1.1% 1.1% 1.1% Notebooks(ex.Netbook) 5,860 6,836 1,608 1,642 1,805 1,826 6,881 1,643 1,662 1,817 1,886 7,008 y/y 45.0%16.6%-2.8%2.8%2.2%0.5%0.7%2.2%1.2%0.7%3.3%1.8% q/q -11.5% 2.2% 9.9% 1.2% -10.0% 1.2% 9.3% 3.8% Shipment (units in 000) 189,530 187,254 42,868 42,471 45,743 46,281 177,363 41,147 41,317 44,719 45,545 172,729 y/y 16.2% -1.2% -5.7% -3.5% -5.3% -6.5% -5.3% -4.0% -2.7% -2.2% -1.6% -2.6% q/q -13.4% -0.9% 7.7% 1.2% -11.1% 0.4% 8.2% 1.8% Average contents per box (MB) 3,958 4,673 4,800 4,950 5,050 5,050 4,966 5,110 5,150 5,200 5,300 5,193 y/y 24.8% 18.1% 3.1% 6.5% 7.9% 7.4% 6.3% 6.5% 4.0% 3.0% 5.0% 4.6% q/q 2.1% 3.1% 2.0% 0.0% 1.2% 0.8% 1.0% 1.9% Netbooks 183 134 0 0 0 0 0 0 0 0 0 0 y/y -41.7% -26.8% q/q Shipment (units in 000) 19,600 13,800 y/y -48.3% -29.6% q/q Netbook Ratio (%) 5.3% 3.9% Average contents per box (MB) 1,195 1,241 y/y 12.6% 3.9% q/q x86 Servers 1,960 2,302 637 721 825 929 3,111 886 979 1,156 1,261 4,282 y/y 98.3% 17.5% 17.7% 35.1% 38.5% 47.1% 35.2% 39.2% 35.8% 40.2% 35.7% 37.6% q/q 0.8% 13.3% 14.4% 12.7% -4.6% 10.5% 18.0% 9.1% Shipment (units in 000) 8,015 8,008 1,859 1,950 2,091 2,241 8,141 2,021 2,021 2,221 2,321 8,586 y/y 3.7% -0.1% -3.6% -0.1% 1.4% 8.6% 1.7% 8.7% 3.7% 6.2% 3.6% 5.5% q/q -9.9% 4.9% 7.3% 7.2% -9.8% 0.0% 9.9% 4.5% Average contents per box (MB) 31,294 36,792 43,836 47,343 50,468 53,067 48,921 56,110 62,019 66,617 69,517 63,845 y/y 91.2% 17.6% 22.1% 35.3% 36.6% 35.4% 33.0% 28.0% 31.0% 32.0% 31.0% 30.5% q/q 11.9% 8.0% 6.6% 5.2% 5.7% 10.5% 7.4% 4.4% Enterprise Servers 1,350 2,019 658 734 817 901 3,110 919 1,027 1,185 1,328 4,459 y/y 122.4% 49.6% 52.9% 52.5% 55.1% 55.2% 54.0% 39.6% 40.1% 45.1% 47.3% 43.4% q/q 13.3% 11.5% 11.3% 10.4% 1.9% 11.9% 15.3% 12.1% Shipment (units in 000) 1,622 1,865 508 529 552 557 2,146 550 570 610 630 2,360 y/y 14.2% 15.0% 14.2% 15.0% 16.2% 14.8% 15.1% 8.2% 7.8% 10.5% 13.2% 10.0% q/q 4.8% 4.1% 4.3% 0.9% -1.2% 3.6% 7.0% 3.3% Average contents per box (MB) 106,501 138,563 165,715 177,487 189,397 207,222 185,478 213,772 230,716 248,584 269,823 241,825 y/y 94.7% 30.1% 33.9% 32.6% 33.5% 35.2% 33.9% 29.0% 30.0% 31.3% 30.2% 30.4% q/q 8.1% 7.1% 6.7% 9.4% 3.2% 7.9% 7.7% 8.5% Workstations 699 848 247 276 298 318 1,140 328 369 405 443 1,545 y/y 77.3% 21.4% 29.7% 36.1% 35.9% 35.4% 34.4% 32.9% 33.5% 35.7% 39.2% 35.5% q/q 5.0% 11.9% 8.0% 6.8% 3.1% 12.4% 9.7% 9.5% Shipment (units in 000) 3,041 3,134 780 815 828 833 3,256 810 830 850 870 3,360 y/y 4.7%3.1%1.0%4.5%5.2%4.8%3.9%3.8%1.8%2.7%4.5%3.2% q/q -1.8% 4.5% 1.6% 0.6% -2.7% 2.5% 2.4% 2.4% Average contents per box (MB) 29,418 34,647 40,525 43,405 46,122 48,960 44,827 51,872 56,904 60,973 65,215 58,872 y/y 69.4% 17.8% 28.4% 30.2% 29.2% 29.2% 29.4% 28.0% 31.1% 32.2% 33.2% 31.3% q/q 7.0% 7.1% 6.3% 6.2% 5.9% 9.7% 7.2% 7.0%
Source: Company data, Nomura estimates
243
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 26: DRAM demand forecasts – Consumer products only Units in Gb 2011 2012 1Q13 2Q13 3Q13 4Q13E 2013E 1Q14E 2Q14E 3Q14E 4Q14E 2014E Consumer 3,447 6,742 2,401 2,787 3,208 4,221 12,617 3,748 4,077 4,753 5,988 18,566 y/y 75.0% 95.6% 106.2% 94.0% 79.2% 79.6% 87.1% 56.1% 46.3% 48.1% 41.9% 47.2% q/q 2.1% 16.1% 15.1% 31.6% -11.2% 8.8% 16.6% 26.0% Game Console (Main/CPU) 129 215 62 53 75 163 352 110 107 166 387 771 y/y 0.7% 67.0% 101.8% 75.1% 53.0% 54.5% 63.8% 77.8% 103.7% 122.0% 137.8% 118.8% q/q -41.0% -15.5% 42.7% 117.3% -32.2% -3.2% 55.6% 132.7% Shipment (units in 000) 73,120 72,300 15,000 11,200 15,100 31,000 72,300 14,100 10,528 14,194 29,140 67,962 y/y -6.3% -1.1% 3.4% -5.5% -4.4% 2.8% 0.0% -6.0% -6.0% -6.0% -6.0% -6.0% q/q -50.2% -25.3% 34.8% 105.3% -54.5% -25.3% 34.8% 105.3% Average contents per box (MB) 225 381 530 600 635 672 624 1,002 1,300 1,500 1,700 1,452 y/y 7.5% 68.9% 95.1% 85.3% 60.1% 50.3% 63.8% 89.1% 116.7% 136.2% 153.0% 132.7% q/q 18.5% 13.2% 5.8% 5.8% 49.2% 29.7% 15.4% 13.3% Flat Panel TV (LCD) 795 1,075 238 268 351 483 1,339 277 311 407 566 1,560 y/y 24.3% 35.3% 23.7% 25.2% 25.1% 24.2% 24.6% 16.7% 16.1% 15.9% 17.1% 16.5% q/q -39.0% 12.6% 31.1% 37.8% -42.7% 12.1% 30.8% 39.2% Shipment (units in 000) 205,071 210,000 44,095 45,180 54,750 70,775 214,800 44,573 45,859 55,902 73,671 220,005 y/y 7.1%2.4%2.2%2.4%2.2%2.3%2.3%1.1%1.5%2.1%4.1%2.4% q/q -36.3% 2.5% 21.2% 29.3% -37.0% 2.9% 21.9% 31.8% Average contents per box (MB) 496 655 690 758 820 874 798 796 867 931 983 908 y/y 16.1% 32.2% 21.0% 22.3% 22.4% 21.4% 21.8% 15.4% 14.4% 13.5% 12.5% 13.8% q/q -4.2% 9.9% 8.2% 6.6% -8.9% 9.0% 7.3% 5.6% Flat Panel TV (PDP) 676814162020701113161656 y/y 8.3% 1.6% 4.0% 6.8% 23.5% -12.9% 3.6% -19.2% -19.9% -20.6% -19.3% -19.8% q/q -39.8% 18.0% 22.9% -0.2% -44.2% 17.0% 21.9% 1.4% Shipment (units in 000) 17,221 13,307 2,563 2,750 3,125 2,925 11,363 1,794 1,925 2,188 2,099 8,005 y/y -6.6% -22.7% -14.1% -12.7% 0.9% -28.3% -14.6% -30.0% -30.0% -30.0% -28.2% -29.5% q/q -37.2% 7.3% 13.6% -6.4% -38.7% 7.3% 13.6% -4.1% Average contents per box (MB) 495 651 690 758 820 874 790 796 867 931 983 899 y/y 16.0% 31.4% 21.0% 22.3% 22.4% 21.4% 21.3% 15.4% 14.4% 13.5% 12.5% 13.9% q/q -4.2% 9.9% 8.2% 6.6% -8.9% 9.0% 7.3% 5.6% Cell Phone (ex. Smart Phone) 534 632 145 140 136 137 558 116 112 110 111 449 y/y 34.5% 18.3% 3.9% -10.0% -15.8% -22.0% -11.8% -20.2% -19.8% -19.1% -18.8% -19.5% q/q -17.4% -3.5% -2.7% 0.7% -15.6% -3.0% -1.9% 1.0% Shipment (units in 000) 1,302,822 1,063,092 215,776 205,739 197,837 196,929 816,281 164,445 157,836 151,474 151,396 625,151 y/y 0.4% -18.4% -21.5% -22.5% -23.5% -25.5% -23.2% -23.8% -23.3% -23.4% -23.1% -23.4% q/q -18.4% -4.7% -3.8% -0.5% -16.5% -4.0% -4.0% -0.1% Average contents per box (MB) 53 76 86 87 88 89 87 90 91 93 94 92 y/y 34.0% 45.0% 32.3% 16.0% 10.0% 4.7% 14.9% 4.7% 4.6% 5.7% 5.6% 5.1% q/q 1.2% 1.2% 1.1% 1.1% 1.1% 1.1% 2.2% 1.1% Smart Phone 1,483 3,536 1,463 1,833 2,099 2,555 7,951 2,580 2,826 3,115 3,660 12,181 y/y 166.6% 138.4% 131.7% 130.6% 126.9% 115.8% 124.8% 76.3% 54.2% 48.4% 43.2% 53.2% q/q 23.6% 25.3% 14.5% 21.7% 1.0% 9.5% 10.2% 17.5% Shipment (units in 000) 471,743 677,547 210,046 224,341 240,845 278,601 953,833 273,243 288,602 308,351 344,999 1,215,195 y/y 57.9% 43.6% 42.9% 46.0% 42.4% 34.2% 40.8% 30.1% 28.6% 28.0% 23.8% 27.4% q/q 1.1% 6.8% 7.4% 15.7% -1.9% 5.6% 6.8% 11.9% Average contents per box (MB) 402 668 892 1,046 1,116 1,174 1,067 1,209 1,254 1,293 1,358 1,283 y/y 68.9% 66.0% 62.2% 58.0% 59.4% 60.8% 59.7% 35.5% 19.9% 15.9% 15.7% 20.3% q/q 22.2% 17.3% 6.7% 5.2% 3.0% 3.7% 3.1% 5.0% White box Smart Phone 13 185 62 69 85 96 312 98 104 112 117 430 y/y N/A 1331.7% 148.7% 83.8% 55.0% 41.2% 68.4% 57.5% 50.6% 32.7% 21.5% 38.1% q/q -8.7% 11.1% 22.8% 13.3% 1.8% 6.3% 8.3% 3.8% Shipment (units in 000) 8,500 73,500 21,000 22,000 25,500 27,600 96,100 25,000 26,000 27,500 28,000 106,500 y/y N/A 764.7% 75.0% 37.5% 21.4% 12.7% 30.7% 19.0% 18.2% 7.8% 1.4% 10.8% q/q -14.3% 4.8% 15.9% 8.2% -9.4% 4.0% 5.8% 1.8% Average contents per box (MB) 195 322 378 401 425 445 415 500 511 523 533 517 y/y N/A 65.6% 42.1% 33.7% 27.6% 25.4% 28.8% 32.3% 27.4% 23.1% 19.8% 24.7% q/q 6.5% 6.1% 6.0% 4.7% 12.4% 2.2% 2.3% 1.9% Digital Still Camera 12010816172122771517191970 y/y 1.1% -10.0% -31.6% -33.1% -31.4% -21.1% -29.2% -6.8% -3.3% -8.7% -16.8% -9.4% q/q -43.2% 8.3% 21.1% 5.9% -32.9% 12.4% 14.3% -3.4% Shipment (units in 000) 115,534 98,150 13,863 14,615 17,248 17,816 63,542 11,704 12,805 14,260 13,431 52,200 y/y -4.9% -15.0% -36.2% -38.4% -37.5% -29.0% -35.3% -15.6% -12.4% -17.3% -24.6% -17.8% q/q -44.8% 5.4% 18.0% 3.3% -34.3% 9.4% 11.4% -5.8% Average contents per box (MB) 133 141 148 152 156 160 155 164 168 173 177 171 y/y 6.3% 6.0% 7.2% 8.6% 9.8% 11.1% 9.4% 10.4% 10.4% 10.4% 10.4% 10.3% q/q 2.8% 2.7% 2.6% 2.6% 2.1% 2.7% 2.6% 2.6% BRAND Tablet PC 292 595 301 281 300 610 1,492 418 451 658 941 2,469 y/y 557.6% 103.8% 355.5% 210.8% 67.9% 134.9% 150.7% 39.0% 60.9% 119.3% 54.2% 65.5% q/q 15.7% -6.7% 6.9% 103.4% -31.5% 8.0% 45.8% 43.1% Shipment (units in 000) 59,994 77,518 34,200 29,100 30,600 60,100 154,000 39,180 39,990 54,513 75,318 209,000 y/y 226.0% 29.2% 157.1% 78.5% 54.4% 113.9% 98.7% 14.6% 37.4% 78.1% 25.3% 35.7% q/q 21.7% -14.9% 5.2% 96.4% -34.8% 2.1% 36.3% 38.2% Average contents per box (MB) 623 982 1,125 1,234 1,255 1,300 1,240 1,365 1,445 1,545 1,600 1,512 y/y 101.7% 57.7% 77.1% 74.1% 8.8% 9.8% 26.2% 21.3% 17.1% 23.1% 23.1% 22.0% q/q -5.0%9.7%1.7%3.6% 5.0%5.9%6.9%3.6% White box Tablet PC 35 293 91 102 112 122 427 114 126 139 158 536 y/y N/A 741.5% 162.2% 54.7% 30.5% 14.8% 45.9% 25.1% 23.7% 23.8% 29.7% 25.7% q/q -14.2% 11.5% 10.6% 8.6% -6.6% 10.2% 10.7% 13.8% Shipment (units in 000) 7,000 52,000 15,000 16,000 17,000 18,000 66,000 16,000 17,000 18,000 20,000 71,000 y/y N/A 642.9% 114.3% 33.3% 13.3% 0.0% 26.9% 6.7% 6.3% 5.9% 11.1% 7.6% q/q -16.7% 6.7% 6.3% 5.9% -11.1% 6.3% 5.9% 11.1% Average contents per box (MB) 636 720 777 812 845 867 828 911 945 988 1,012 967 y/y N/A 13.3% 22.3% 16.0% 15.1% 14.8% 14.9% 17.2% 16.4% 16.9% 16.7% 16.9% q/q 2.9% 4.5% 4.1% 2.6% 5.1% 3.7% 4.6% 2.4% Portable Music Player 27 35 9 9 10 12 40 10 10 11 13 43 y/y 12.3% 28.2% 16.1% 14.8% 13.5% 12.5% 14.0% 8.6% 8.6% 8.6% 7.8% 8.3% q/q -19.1% 0.6% 9.3% 26.3% -21.9% 0.6% 9.3% 25.4% Shipment (units in 000) 128,406 126,800 28,272 27,094 28,272 34,162 117,800 26,576 25,468 26,576 31,880 110,500 y/y -11.6% -1.3% -7.1% -7.1% -7.1% -7.1% -7.1% -6.0% -6.0% -6.0% -6.7% -6.2% q/q -23.1% -4.2% 4.3% 20.8% -22.2% -4.2% 4.3% 20.0% Average contents per box (MB) 27 35 40 42 44 46 43 46 49 51 53 50 y/y 27.1% 29.8% 25.0% 23.5% 22.2% 21.1% 22.8% 15.5% 15.5% 15.5% 15.5% 15.5% q/q 5.3% 5.0% 4.8% 4.5% 0.4% 5.0% 4.8% 4.5%
Source: Company data, Nomura estimates
244
Nomura | U.S. Semiconductors Primer December 11, 2013
NAND Market In 2012, the NAND segment represented about 42% of the memory IC market. NAND revenue of $24bn in 2012 declined by 3% year over year, with unit shipments (in gigabyte) increasing 62%, offset by ASP (per gigabyte) decline of 40%. Given the supply constraints and pricing strength that started in the second half of 2012, we forecast NAND revenue will increase 19% in 2013, driven by a 40% increase in unit shipments, offset partially by 15% decline in ASPs.
Fig. 27: NAND Market Revenue, 2000–2014E
35,000
30,000
25,000
20,000
15,000
10,000
5,000
0
Source: Gartner, Nomura estimates
Unlike the DRAM industry in the past two decades, the NAND industry has fewer competitors. Over the past five years, the NAND industry has the same six suppliers that accounted for 99% of the market. In 2012, Samsung was the largest NAND supplier with 33% share, followed by Toshiba (24%), SanDisk (16%), Micron (12%), SK Hynix (10%), and Intel (5%).
Fig. 28: Market Share in NAND, 2012
Others 5% SK Hynix 10%
Samsung 33% Micron 12%
SanDisk 16%
Toshiba 24% Samsung Toshiba SanDisk Micron SK Hynix Others
Source: Gartner, Nomura research
245
Nomura | U.S. Semiconductors Primer December 11, 2013
Multiple Applications Drive Demand Growth NAND flash is a relatively new industry, with revenue exceeding $1bn only in 2000. New applications emerged as flash memory prices came down, driving bit shipments up at a 130% CAGR over the past 10 years. NAND flash was initially targeted at digital cameras. It was then used in USB drives that ultimately displaced floppy disks. The Apple iPod revolutionized digital media players and replaced compact discs (CD). More recently, NAND flash became the de facto standard for storage in smartphones and tablets. We believe the next big opportunity is in solid state drives used mostly in notebook PCs and also in servers and storage systems. Gartner forecasts that NAND bit shipments will grow at a 40% CAGR between 2012 and 2017. Smartphone was the biggest end market for NAND flash in 2012, accounting for 20% of bit shipments. By 2017, Gartner expects solid state drives to surpass mobile phones and account for 40% of total bit shipments.
Fig. 29: A History of Disruption
Source: SanDisk
Fig. 30: SSDs Drive NAND Shipment Growth in the Next 5 Years
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2009 2010 2011 2012 2013E 2014E 2015E 2016E 2017E Solid State Drives Mobile Phones Tablets Memory Cards USB Drives Portable Media Players PCs and servers Other
Source: Gartner, Nomura research
246
Nomura | U.S. Semiconductors Primer December 11, 2013
The strong bit shipment growth was made possible by NAND suppliers aggressively lowering costs. NAND prices declined by an average of 45% per year over the past 10 years owing to rapid process technology migration, as well as the use of multiple bits per cell (MLC) and three bits-per-cell (TLC) technologies. NAND flash is currently manufactured using leading-edge process technologies at 19–21nm, with Micron starting to ramp 16nm by the end of 2013. By comparison, DRAM is produced using 25–35nm and microprocessors are made using 22–32nm. The rapid price decline of NAND has historically been the driving force of bit shipment growth. We expect the growth rate of NAND shipments to slow to 40–50% per year due to the law of large numbers and technology challenges. We also expect slower price decline of 20–25% per year, which should support revenue growth of 10–15% per year over the next few years.
Fig. 31: ASP Decline for NAND Flash, 2000–2014E
0%
-10%
-20%
-30%
-40%
ASP Decline YoY -50%
-60%
-70% 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Source: Gartner, Nomura estimates
Fig. 32: Strong Bit Growth Reflects Price Elasticity
300%
250%
200%
150%
Bit growth YoY Bit 100%
50%
0% 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Source: Gartner, Nomura estimates
247
Nomura | U.S. Semiconductors Primer December 11, 2013
Moving Faster than Moore’s Law In the past decade, process technology migrations for NAND flash averaged 12–18 months, which is faster than what Moore’s Law suggests, i.e. doubling transistor count every 18–24 months. Each generation of process technology increases the density by 40–50%, which approximately doubles the NAND capacity for the same die size. As a result, the average capacity of a monolithic die increased at a CAGR of about 50%—from 6Gb in 2007 to 40Gb in 2012. More recently, technology transitions have become increasingly challenging beyond the 19–21nm node due to physical limitations of the floating gate technology. Consequently, the time between one generation of process technology and the next has lengthened, resulting in lower cost reduction per year.
Fig. 33: SanDisk’s Process Technology Migration
180 160nm 160 130nm 140 120 90nm 100 70nm 80 55nm 60 43nm 32nm 40 24nm 19nm 1y-nm 1z-nm 20 0 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Source: SanDisk, Nomura research
Fig. 34: Density per NAND Chip Increases
9,000 8,000 7,000 6,000 5,000 4,000
Units (in mn) 3,000 2,000 1,000 0 2010 2011 2012 2013E 2014E 2015E 2016E 2017E
<4Gb 8Gb 16Gb 32Gb 64Gb 128Gb 256Gb 512Gb
Source: Gartner, Nomura estimates
248
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 35: Samsung’s Technology Migration and Bit-per-Wafer Growth at each Generation
120% 110% 113%
100% 85% 88% 86% 78% 80% 65% 57% 61% 60%
40% 30%
20%
0% 512Mb 1G SLC 2G SLC 4G MLC 8G MLC 16G MLC 32G MLC 32G MLC 64G MLC 64G MLC SLC (120nm) (90nm) (90nm) (63nm) (51nm) (42nm) (27nm) (21nm) (19nm) (160nm) 2003 2004 2005 2005 2006 2007 2008 2010 2011 2012
Source: Nomura estimates
The primary reason for the deceleration of cost reduction is the dwindling number of electrons stored in the NAND flash cell’s floating gate as the gate’s volume shrinks with each process node. The number of electrons stored in the floating gate is rapidly headed towards a couple of hundred and the leakage loss of even 10 electrons is becoming problematic. Furthermore, the use of multiple bits per memory cell also reduces the endurance and reliability of the NAND component. To address this issue, NAND suppliers need to dedicate more resources to error correction code (ECC) in the controller or include extra storage (or over-provisioning), which will add additional cost to the system. Alternatively, some suppliers could change the characteristics of the flash memory cells and tune for specific usage, making NAND components less commoditized than they may otherwise be.
Fig. 36: Flash Performance Degrades with Every Process Shrink
Source: SanDisk, Nomura research
249
Nomura | U.S. Semiconductors Primer December 11, 2013
More Bits per Cell In addition to normal scaling through process technology migration, NAND suppliers are able to increase bit production by storing more bits per memory cell. The most common technology today is 2 bits-per-cell (MLC) NAND, which represents about 70% of all unit shipments in 2012. We expect 3 bits-per-cell (TLC or 8LC) NAND, which accounted for 25% of total shipments in 2012, as a percentage of total shipments to increase in the coming years, driven primarily by Samsung. SanDisk pioneered MLC technology in early 2000, and was the first supplier to massively deploy TLC NAND in 2008. In 2012, more than 50% of SanDisk’s bit shipments were TLC NAND, primarily serving the memory card and USB drive markets. In 2013, Samsung plans to increase the proportion of TLC technology to 50% from 25% in 4Q12 with the help of its improved NAND controller technology. Theoretically, conversion to TLC NAND would allow for 30–50% higher storage than MLC NAND with identical chip size. We estimate the cost of TLC NAND per bit to be 20% below MLC NAND.
Fig. 37: Nearly 70% of the NAND Market Is MLC
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2008 2009 2010 2011 2012
SLC NAND MLC NAND 8LC NAND Other
Source: Gartner, Company data, Nomura research
Unlike geometry migration, where a NAND maker shrinks the size of die to make more chips out of a wafer and a familiar way of increasing bit per wafer so far, MLC (multi-level cell) and TLC (triple-level cell) increases the number of bits per cell. For MLC, two bits of electron are stored per cell, meaning instead of two voltage stages (i.e., 0 and 1), there are four stages (ie, 00, 01, 10, and 11). TLC takes that a step further and stores three bits per cell, to come up with eight voltage stages (i.e., 000, 001, 010, 011, 100, 101, 110, and 111). As an example, a NAND array with 16bn transistors (one transistor is required per cell), i.e., 16 gigabits (Gb), can be turned into either single-layer cell (SLC), MLC, or TLC. In the case of SLC flash, only one bit of data will be stored in one cell, hence the final product has a 16Gb capacity. Upping the bits per cell to two (MLC) increases capacity to 32Gb because now there are two bits per cell and there are still 16bn transistors. Likewise, three bits per cell (TLC) yields 48Gb. Thus, the cost per bit declines significantly from SLC to MLC and further to TLC. Theoretically, the rate of cost reduction from MLC to TLC is more than 30%.
250
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 38: Program/Erase Cycle Comparison TLC takes longer to store/retrieve than SLC and MLC 5x-nm 3x-nm 2x-nm Single Layer Cell (SLC) 100,000 100,000 N/A Multi-Layer Cell (MLC) 10,000 5,000 3,000 Triple Layer Cell (TLC) 2,500 1,250 750
Source: Nomura research
However, there are critical downsides of the TLC technology; it is slower and its lifetime is shorter than lower-bit-per-cell NANDs (i.e., SLC and MLC). With eight voltage levels in TLC, in a single physical layer, random reads take more time:100μs for TLC (vs. 25μs for SLC and 50μs for MLC). The speed of TLC’s programming is also known to be slower. In addition to having slower speed, TLC also has worse endurance than both MLC and SLC. SLC only has two program states, "0" and "1". Hence, either a high or low voltage is required. When the number of bits goes up, more voltage stages are needed. There are four states with MLC, and eight states with TLC. The problem is that the silicon oxide layer is only about 10nm thick and does not last forever; it wears out every time it is used in the tunneling process. The fewer bits per cell, the more voltage room. In other words, SLC can tolerate more changes in the voltage states because it has only two states. In TLC, there are eight, so the margin for error is a lot smaller. The advance in NAND controller technologies could increase the usefulness of TLC NAND. The role of the NAND controller includes over-provisioning (setting aside spare capacity for use as some cells die), reducing the number of writes with better ways of dealing with I/O, and retrieving useful data from cells previously classed as defunct. Basically, a good controller can compensate for the shorter working life of TLC. As such, we note that NAND suppliers are relying on better controller technology to counter the disadvantages of TLC. In spite of the disadvantages, there is still significant motivation to migrate towards TLC, as it is, given the challenging migration to finer geometries. Realistically, it is the easiest way to increase bits per wafer, or to lower the unit bit costs.
251
Nomura | U.S. Semiconductors Primer December 11, 2013
NAND Capex Set to Rise NAND is a capital intensive industry. A new fab with capacity of 200,000 wafers per month would cost as much as $8bn, while the cost of technology migration costs about $40mn per year for every 10,000 wafers per month of capacity. Over the past 10 years, the industry re-invested 40% of sales as capital investment. The capex-to-sales reached a high of 75% in 2007 and a low of 19% in 2009. After two years of disciplined investments in 2012–2013, we expect industry capex to increase by 35% in 2014E to support higher capital intensity for process technology migrations and the build-out of new technologies. The implied capex-to-sales ratio of 35% in 2014E is an increase from 30% in 2013E. Furthermore, we estimate that industry wafer capacity will increase 10% y/y in 2014 after being flat since 2011. The primary sources of wafer capacity come from Samsung ramping its China fab for 3D NAND and Micron converting some DRAM capacity to NAND.
Fig. 39: NAND Industry Capex and Capex as Percentage of Sales
14000 80%
12000 70%
60% 10000 50% 8000 40% 6000 30% 4000 20% NAND Industry Capex ($mn) Industry Capex NAND 2000 10%
0 0% 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Capex ($mn) Capex as % of Sales
Source: Company data, Nomura estimates
Fig. 40: NAND Wafer Capacity, 2005–2014E
1,400
1,200
1,000
800
600
400
NAND w afe r capacity200 (000s wpm)
0 2005 2006 2007 2008 2009 2010 2011 2012 2013E 2014E
Source: Company data, Nomura estimates
252
Nomura | U.S. Semiconductors Primer December 11, 2013
Bargaining Power of the NAND Industry Sources of demand for NAND are mainly derived from three groups: Apple, Samsung, and others. Each customer requests its own specifications, thus, NAND chips fabricated for a particular customer can hardly be diverted to another customer—i.e., NAND is becoming a specialty product. Samsung’s handset operation purchases NAND mostly from Samsung, while Apple buys from non-Samsung NAND suppliers. Given non-Samsung NAND suppliers cannot supply to Samsung’s handset operation, Apple has become the single-most important customer for non-Samsung suppliers. We estimate that Apple consumes 25–30% of worldwide NAND flash production. We further estimate that NAND shipments to Apple contribute 60% of Hynix’s NAND output, 40% of Toshiba’s output, and 25% of SanDisk’s output. We note that SanDisk has been gaining more share at Apple since 2H 2012, filling the place previously taken by Samsung. Apple has thus far managed its NAND supply chain to maximize smartphone/tablet margins. On the flip side, multiple non-Samsung NAND players are competing over a single customer – Apple, resulting in substantially lower bargaining power and lean NAND profitability. If this supply/demand structure in mobile NAND storage remains unchanged, it could limit the upside potential for the profitability of NAND players with high exposure to Apple.
Fig. 41: Apple’s Consumption of NAND Flash in 2012
Unit (mn) Avg GB Total GB (mn) iPhone 136 32 4,352 iPad 66 24 1,584 MacBook 11 150 1,650 iPod 32 15 480 Total Demand for Apple 8,066 Worldwide Shipment 32,712 Apple as % of total market 25%
Source: Nomura estimates
Fig. 42: NAND shipment proportion to Apple (4Q12) Fig. 43: NAND shipment market share at Apple (4Q12)
70% Micron Samsung 6% 10%
60% SanDisk 20% 50%
40%
30% Hynix 33% Mostly to 20% Apple's SSD
10%
Toshiba 0% 31% Samsung Hynix Toshiba SanDisk Micron
Source: Nomura estimates Source: Nomura estimates
253
Nomura | U.S. Semiconductors Primer December 11, 2013
SSD is the Next Big Growth Driver While smartphones and tablets have provided strong growth for NAND flash in the past few years, growth in the next few years will be dominated by solid state drives (SSD). Between 2012 and 2014E, we expect NAND flash used in solid state drives to grow at a 75% CAGR, driven by 46% increase in unit shipment and 21% increase in average content per system. As a result, we expect SSD contribution to increase from 20% in 2012 to 30% in 2014E.
Fig. 44: Solid State Drives Account for 30% of Industry Bit Shipments in 2014E
35%
30%
25%
20%
15% 30% 26% 10% 20% % of Industry% Shipment
12% 5% 6% 4% 0% 2007 2008 2009 2010 2011 2012 2013E 2014E
Source: Nomura estimates
Fig. 45: SSD Unit and Average Capacity Growth, 2007–2014E
120 250 220 GB
100 184 GB 200
80 151 GB 150
60 108 GB
100 73 GB 40
Unit Shipment (in mn) 46 GB 50 GB) (in Content Average 20 20 GB 10 GB
0 0 2007 2008 2009 2010 2011 2012 2013E 2014E
Source: Nomura estimates
254
Nomura | U.S. Semiconductors Primer December 11, 2013
Despite slower-than-expected adoption of NAND flash in PCs, 2012 was a breakout year for SSDs, as mainstream PC SSDs reached a key milestone of below $1 per gigabyte in the second half of the year. Clearly, at $1 per gigabyte, NAND prices are still 8–10 times more expensive than hard disk drives on a per gigabyte basis, but it is a key psychological price points for many consumers. We believe the next catalyst for SSD adoption is when SSD unit cost drops to $50 for a reasonable amount of capacity. We believe this could be achieved in 2014 for a 120GB SSD that is enough to store the operating system and several large programs such as Microsoft Office and Adobe Photoshop. This price will be equivalent to the lowest unit cost for a hard disk drive, although HDD will carry much higher capacity (estimated 500GB). With the same unit cost, users will need to consider how much storage capacity is needed, versus all the benefits (e.g. faster performance, lighter weight, more flexible form factor, instant on, etc) that solid state drives bring. Consumers that need bulk storage for photos, videos and music could choose to store them in external hard drives. We also think Intel’s push for Ultrabooks, which use flash memory for primary storage or caching device, will see more success as the end device cost will likely come down from $1,000 in 2012 to the $600–700 price range. The enterprise SSD market (server and storage) is equally important as the client SSD market. Gartner estimates that the total SSD market will grow from $7bn in 2012 to $17bn in 2016E, with the market almost equally split between client SSD and enterprise SSD. The client SSD market is significantly larger than the enterprise SSD market in terms of unit shipments by a factor of 5–10 times. However, the much stringent requirements for enterprise SSDs in terms of performance, endurance, and reliability are driving a significant price premium and lucrative profits for SSD suppliers. In both segments, we expect suppliers that are vertically integrated with competitive software/firmware offering to have a competitive advantage.
Fig. 46: SSD Market Split about Equally between Client SSD and Enterprise SSD
10000
9000
8000
7000
6000
5000 PC Enterprise 4000
3000
Market Opportunity (in $mn) (in Opportunity Market 2000
1000
0 2010 2011 2012 2013E 2014E 2015E 2016E
Source: Gartner, Nomura research
We also believe controller technology is a key differentiator in the SSD market. Samsung, SanDisk, Micron, and Toshiba have developed their own controller technology, while Hynix recently acquired Link-A-Media (LAMD), a fabless NAND controller designer. As well, there are fabless players including LSI/SandForce, Marvell, Fusion-io, STEC, Phison, and Silicon Motion. Samsung seems the most aggressive vs. peers in internalizing the controller technologies, while SanDisk and Micron employ third-party controllers (SanDisk uses LSI/SandForce and Micron uses Marvell) for most of their client SSD products. The controller ASPs of fabless players are at ~$10, while NAND players with controller technology can make a controller at the unit cost of $5. As a result, the fabless
255
Nomura | U.S. Semiconductors Primer December 11, 2013 players seem to be at competitive disadvantage. In 2012, Samsung introduced the industry first TLC-based SSD (840 series), which we believe advanced controller technology was a key enabler. Previously, it was believed that TLC NAND is not suitable for SSD due to the significantly lower performance and read/write cycle than MLC NAND. As controller technologies improve, we expect more TLC-based SSD in the market in the next two years.
In 2012, Samsung had the highest share in client SSDs and a strong presence in enterprise SSDs. Intel had the second highest share, with strength skewed toward enterprise SSD. Toshiba focused primarily on the client SSD market, while Fusion-IO had strong share in the enterprise market only. Micron and SanDisk both had about 10% share in the client SSD market, but we expect both companies to gain share in both client and enterprise markets driven by their vertically integrated models.
Fig. 47: SSD Market Share, 2012
Source: Micron Technology, Nomura research
256
Nomura | U.S. Semiconductors Primer December 11, 2013
Scaling Challenges Ahead: 3D NAND and Other Options Facing scaling challenges beyond 20nm, NAND suppliers have a few options: (1) continue to scale 2D NAND to smaller geometries; (2) shift production from MLC to TLC; and (3) shift from 2D NAND to 3D NAND (or Vertical NAND), with 3D NAND likely to have larger industry ramifications. Our understanding is that continued scaling of 2D NAND at mature yields will have the lowest cost per bit, while TLC is cheaper than MLC NAND but comes with performance and reliability issues, and 3D NAND is still in development stage and won’t be ready for volume production until 2014. We note that the technology roadmap differs across the NAND players. Samsung seems cautious on geometry migration while aggressive on TLC and 3D NAND. The rest of the NAND players—Toshiba/SanDisk, Micron, and Hynix—continue to focus on geometry migration in the next 1–2 years. It’s still too early to predict who will prevail as the technology directions abovementioned—the geometry migration, TLC, and 3D NAND— are all challenging. We believe the end results will determine the future of the respective players and that the cost of failure will be substantial.
(1) 2D NAND Scaling: The roadmap for 2D scaling is different for every supplier. Micron has started transitioning from 20nm to 16nm at the end of 2013, and believes the transition is relatively easy given most of the material changes were made with the 20nm transition. SK Hynix also has plans to convert to 16nm in 2014. Toshiba and SanDisk have the longest roadmap for 2D NAND scaling, which includes the ramp of 1y-nm in 2014 and 1z-nm in 2015 before switching to 3D NAND in 2016. Samsung had previously planned to migrate from 21nm to 16nm, but faced too high a technology barrier and pulled back to 19nm. Instead, Samsung decided to focus on TLC and 3D NAND.
Fig. 48: SanDisk Technology Roadmap
Source: SanDisk analyst meeting (May 2013), Nomura research
(2) Migration from MLC to TLC: Historically, SanDisk was the only supplier that use 3 bits-per-cell technology extensively primarily due to its high exposure to retail business, such as memory cards and USB drives. Those products do not have high requirements for reliability and endurance, and thus TLC is good enough to address those markets. Other suppliers have spent less effort on TLC, as they argue that it is more costly and time consuming to develop TLC on the current process node than to move to the next process node on MLC. With MLC cost reduction slowing, suppliers are re-thinking that strategy. Among other suppliers, Samsung seems to the most aggressive in terms of increasing TLC percentage. Samsung’s wafer proportion for TLC NAND increased from 15% in 2012 to 40% in 2013, and we expect it to grow to more than 60% in 2014 as TLC-based SSDs ramps in volume. We believe strong controller technology allows Samsung to utilize TLC in SSDs. We estimate that the migration from 21nm MLC to 19nm MLC will increase bit per wafer by 30%, while the migration from 21nm MLC to 19nm TLC will allow bit-per-wafer growth of 70%.
257
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 49: Bit Growth Comparison between MLC and TLC
80%
70%
60%
50%
40%
30%
20%
10%
0% Samsung Samsung From 21nm MLC to 19nm MLC from 21nm MLC to 19nm TLC
Source: Nomura estimates
(3) Migration from 2D NAND to 3D NAND: The basic concept of 3D NAND is that, rather than laying the cells flat on the surface, higher densities can be achieved by stacking them on top of each other. 3D NAND is true vertical NAND cell stacking, unlike chip stacking in multi-chip packages. In 3D NAND, NAND layers are stacked in a single chip. In many respects, 3D NAND is evolutionary, not revolutionary. The benefits of 3D NAND are continued cost reduction, smaller die sizes and more capacity. We believe the first generation of 3D NAND could be produced in process nodes that are a few generations behind the leading edge technology for 2D NAND. In addition, installed toolsets can be mostly reused, thereby extending the useful life of fab equipment. The problem is that manufacturing 3D NAND could be challenging given that many the aspect ratio of 3D NAND could be 5–10x of conventional etching systems that deals with aspect ratio of 3:1 or 4:1. In addition, 3D NAND is still basically a floating gate NAND with all its inherent limitations of data reliability and performance. As such, we expect 3D NAND to be an intermediate step between 2D NAND and post-NAND technology, and, hence should have a relative short life span of 5 to 8 years.
Fig. 50: Variants of 3D NAND
Source: Company data, Nomura research
258
Nomura | U.S. Semiconductors Primer December 11, 2013
Today, every major NAND supplier is developing some variants of 3D NAND (see chart above). While most suppliers are focusing more on the migration to 1y-nm and 1z-nm, Samsung seems to be at the forefront of 3D NAND technology. The company released its first 3D NAND chips in mid-2013, and we expect the first batch of commercial 3D NAND chips to be produced from its China fab that is scheduled to come on line in early 2014. Should commercial production at the China fab be successful, conversion from 2D NAND fabs to 3D NAND fabs is likely to begin in 2H14. We expect Samsung’s 3D NAND capacity at its China fab to increase to 80k wafer per month by the end of 2014. SK Hynix and Micron are sampling 3D NAND in early 2014 with mass production likely in 2015, while Toshiba and SanDisk should go into mass production in 2016. The upfront cost of 3D NAND is high, yet the upgrade cost is cheaper than 2D NAND. On our estimates, converting 10kpm of existing 2D NAND to 3D NAND would require $300mn of capex while, once converted, the costs for a one-generation upgrade become lower – $30–40mn for 10kpm 3D NAND vs. $50–60mn for 2D NAND with the same wafer capacity. Meantime, installing a new 10kpm 3D NAND fab costs $500mn, vs. $300mn for a new 2D NAND fab. Beyond 3D NAND, many companies, including both current NAND suppliers and other memory startups, are working on a variety of next-generation memory technologies, such as FeRAM, MRAM, PCM and ReRAM. Many of these new technologies require new tools known as extreme ultraviolet (EUV) lithography. The chart below compares a number of characteristics of these technologies. We do not expect these emerging technologies to significantly change the memory competitive landscape in the next few years as long as 2D and 3D NAND continue to scale.
Fig. 51: Post-NAND Technologies
Source: Micron, Nomura research
259
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 52: Nomura NAND Industry Model
2011 2012 1Q13 2Q13 3Q13 4Q13E 2013E 1Q14E 2Q14E 3Q14E 4Q14E 2014E Total demand (8Gb eq mn) 18,978 31,591 9,874 10,709 12,048 14,642 47,273 13,978 15,072 17,122 19,979 66,152 % y-y 62.0% 66.5% 58.5% 43.6% 47.5% 50.4% 49.6% 41.6% 40.7% 42.1% 36.5% 39.9% % q-q 1.4% 8.4% 12.5% 21.5% -4.5% 7.8% 13.6% 16.7% Total supply (8Gb eq mn) 20,196 33,648 10,103 10,752 12,508 13,878 47,240 13,755 15,462 17,599 19,782 66,597 % y-y 69.4% 66.6% 43.7% 29.2% 46.5% 42.2% 40.4% 36.1% 43.8% 40.7% 42.5% 41.0% % q-q 3.5% 6.4% 16.3% 11.0% -0.9% 12.4% 13.8% 12.4% Demand/supply 0.94 0.94 0.98 1.00 0.96 1.06 1.00 1.02 0.97 0.97 1.01 0.99 Total shipment value ($mn) 24,418 23,554 6,415 6,666 7,380 7,633 28,094 7,015 7,422 8,095 8,506 31,038 % y-y 12.3% -3.5% 8.6% 17.8% 33.0% 18.5% 19.3% 9.3% 11.3% 9.7% 11.4% 10.5% % q-q -0.4% 3.9% 10.7% 3.4% -8.1% 5.8% 9.1% 5.1% Average unit price (8Gb eq $) 1.21 0.70 0.64 0.62 0.59 0.55 0.59 0.51 0.48 0.46 0.43 0.47 % y-y -33.7% -42.1% -24.4% -8.8% -9.2% -16.7% -15.0% -19.7% -22.6% -22.0% -21.8% -21.6% % q-q -3.8% -2.4% -4.8% -6.8% -7.3% -5.9% -4.2% -6.5%
Source: Company data, Nomura estimates
Fig. 53: NAND supply forecasts
2011 2012 13 Q1E 13 Q2E 13 Q3E 13 Q4E 13 14 Q1E 14 Q2E 14 Q3E 14 Q4E 14 Total Supply (8Gb eq. mn) 20,196 33,648 10,103 10,752 12,508 13,878 47,240 13,755 15,462 17,599 19,782 66,597 y/y 69.4% 66.6% 43.7% 29.2% 46.5% 42.2% 40.4% 36.1% 43.8% 40.7% 42.5% 41.0% q/q 3.5% 6.4% 16.3% 11.0% -0.9% 12.4% 13.8% 12.4% Samsung Electronics 6,204 9,686 3,468 3,690 4,176 4,784 16,119 5,309 5,880 6,411 7,116 24,715 y/y 74.7% 56.1% 104.3% 62.5% 62.4% 52.1% 66.4% 53.1% 59.3% 53.5% 48.7% 53.3% q/q 10.3% 6.4% 13.2% 14.6% 11.0% 10.7% 9.0% 11.0% Toshiba 4,580 7,033 2,305 2,536 2,714 2,904 10,459 2,749 3,161 3,639 4,050 13,599 y/y 62.0% 53.6% 23.2% 69.4% 77.7% 35.8% 48.7% 19.2% 24.7% 34.1% 39.5% 30.0% q/q 7.8% 10.0% 7.0% 7.0% -5.3% 15.0% 15.1% 11.3% Sandisk (Captive) 3,788 6,125 1,639 1,721 1,974 2,172 7,507 2,020 2,222 2,555 2,938 9,735 y/y 83.9% 61.7% 36.3% 36.7% 15.3% 11.3% 22.6% 23.2% 29.1% 29.4% 35.3% 29.7% q/q -16.0% 5.0% 14.7% 10.0% -7.0% 10.0% 15.0% 15.0% SK Hynix 2,461 3,823 1,122 1,432 1,616 1,366 5,536 1,347 1,703 2,207 2,453 7,710 y/y 152.2% 55.3% 35.0% 57.5% 70.0% 20.7% 44.8% 20.0% 18.9% 36.6% 79.5% 39.3% q/q -0.9% 27.6% 12.8% -15.4% -1.4% 26.5% 29.6% 11.1% IM Flash Technologies (MU / INTC) 3,036 5,672 1,467 1,472 1,728 1,952 6,619 2,030 2,196 2,487 2,926 9,639 y/y 70.6% 86.8% 23.5% -11.4% 17.0% 45.1% 16.7% 38.4% 49.2% 43.9% 49.9% 45.6% q/q 14.0% 0.3% 17.4% 12.9% 4.0% 8.1% 13.3% 17.6%
Source: Company data, Nomura estimates
2 60
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 54: NAND demand forecasts Units in GB 2011 2012 1Q13 2Q13 3Q13 4Q13E 2013E 1Q14E 2Q14E 3Q14E 4Q14E 2014E Total Demand (8Gb eq. mn) 18,978 31,591 9,874 10,709 12,048 14,642 47,273 13,978 15,072 17,122 19,979 66,152 y/y 62.0% 66.5% 58.5% 43.6% 47.5% 50.4% 49.6% 41.6% 40.7% 42.1% 36.5% 39.9% q/q 1.4% 8.4% 12.5% 21.5% -4.5% 7.8% 13.6% 16.7% USB Flash Drives 2,146 2,017 465 452 484 506 1,907 411 394 410 427 1,642 y/y 40.5% -6.0% 22.8% -10.2% -13.1% -12.4% -5.4% -11.5% -12.9% -15.3% -15.7% -13.9% q/q -19.5% -2.9% 7.2% 4.5% -18.7% -4.3% 4.2% 4.0% Shipment (units in 000) 294,955 220,000 46,110 44,200 46,740 48,140 185,190 38,732 36,686 37,859 38,993 152,271 y/y 3.6% -25.4% -13.0% -15.0% -18.0% -17.0% -15.8% -16.0% -17.0% -19.0% -19.0% -17.8% q/q -20.5% -4.1% 5.7% 3.0% -19.5% -5.3% 3.2% 3.0% Average contents per box (MB) 7,451 9,388 10,329 10,463 10,609 10,769 10,546 10,876 10,985 11,095 11,206 11,041 y/y 35.6% 26.0% 41.1% 5.6% 6.0% 5.5% 12.3% 5.3% 5.0% 4.6% 4.1% 4.7% q/q 1.2% 1.3% 1.4% 1.5% 1.0% 1.0% 1.0% 1.0% Flash Cards 5,418 6,582 1,606 1,635 1,768 1,894 6,903 1,719 1,762 1,861 1,940 7,282 y/y 45.5% 21.5% -2.7% -1.4% 14.7% 9.4% 4.9% 7.0% 7.8% 5.2% 2.4% 5.5% q/q -7.2% 1.8% 8.2% 7.1% -9.2% 2.5% 5.6% 4.2% Shipment (units in 000) 1,014,092 915,000 218,400 209,664 220,147 226,752 874,963 201,809 203,827 209,942 214,141 829,718 y/y 1.9% -9.8% -2.9% -2.5% -6.3% -5.5% -4.4% -7.6% -2.8% -4.6% -5.6% -5.2% q/q -9.0% -4.0% 5.0% 3.0% -11.0% 1.0% 3.0% 2.0% Average contents per box (MB) 5,471 7,366 7,532 7,984 8,224 8,553 8,079 8,724 8,854 9,076 9,275 8,987 y/y 42.7% 34.6% 0.2% 1.1% 22.4% 15.8% 9.7% 15.8% 10.9% 10.4% 8.5% 11.2% q/q 2.0% 6.0% 3.0% 4.0% 2.0% 1.5% 2.5% 2.2% Solid-State Drives 2,409 6,598 2,473 2,929 3,416 3,850 12,668 4,369 4,887 5,619 6,268 21,143 y/y 218.0% 173.9% 99.7% 88.5% 91.5% 90.5% 92.0% 76.7% 66.8% 64.5% 62.8% 66.9% q/q 22.3% 18.5% 16.6% 12.7% 13.5% 11.9% 15.0% 11.6% Shipment (units in 000) 22,829 44,678 14,500 16,500 18,500 20,000 69,500 21,750 23,100 25,180 26,620 96,650 y/y 113.6% 95.7% 54.6% 50.0% 56.8% 60.0% 55.6% 50.0% 40.0% 36.1% 33.1% 39.1% q/q 16.0% 13.8% 12.1% 8.1% 8.7% 6.2% 9.0% 5.7% Average contents per box (MB) 108,045 151,214 174,633 181,793 189,065 197,145 186,653 205,701 216,634 228,491 241,132 224,010 y/y 48.8% 40.0% 29.2% 25.7% 22.1% 19.0% 23.4% 17.8% 19.2% 20.9% 22.3% 20.0% q/q 5.4% 4.1% 4.0% 4.3% 4.3% 5.3% 5.5% 5.5% Cache NAND in Laptop 1 52 22 30 36 32 120 31 29 27 37 125 y/y N/A 7423.8% 310.0% 252.0% 130.3% 43.6% 131.4% 41.2% -2.6% -23.8% 16.2% 4.1% q/q -1.8% 36.6% 19.5% -10.4% -3.4% -5.8% -6.5% 36.6% Shipment (units in 000) 393 16,591 5,500 7,000 8,000 6,872 27,372 6,500 6,000 5,500 7,369 25,369 y/y N/A 4123.9% 120.0% 100.0% 70.2% 16.7% 65.0% 18.2% -14.3% -31.3% 7.2% -7.3% q/q -6.6% 27.3% 14.3% -14.1% -5.4% -7.7% -8.3% 34.0% Average contents per box (MB) 1,800 3,207 4,100 4,400 4,600 4,800 4,499 4,900 5,000 5,100 5,200 5,054 y/y N/A 78.1% 86.4% 76.0% 35.3% 23.1% 40.3% 19.5% 13.6% 10.9% 8.3% 12.3% q/q 5.1% 7.3% 4.5% 4.3% 2.1% 2.0% 2.0% 2.0% Cell Phone (ex. Smart Phone) 533 539 124 123 124 129 499 113 110 111 115 449 y/y 26.6% 1.0% -3.4% -5.1% -8.2% -11.7% -7.3% -8.9% -10.6% -10.0% -10.4% -10.0% q/q -14.8% -1.4% 0.9% 4.2% -12.1% -3.3% 1.6% 3.7% Shipment (units in 000) 1,302,822 1,063,092 215,776 205,739 197,837 196,929 816,281 164,445 157,836 151,474 151,396 625,151 y/y 0.4% -18.4% -21.5% -22.5% -23.5% -25.5% -23.2% -23.8% -23.3% -23.4% -23.1% -23.4% q/q -18.4% -4.7% -3.8% -0.5% -16.5% -4.0% -4.0% -0.1% Average contents per box (MB) 419 519 590 610 640 670 626 705 711 752 781 736 y/y 26.1% 23.8% 23.1% 22.4% 20.0% 18.5% 20.8% 19.5% 16.5% 17.5% 16.5% 17.5% q/q 4.3% 3.4% 4.9% 4.7% 5.2% 0.8% 5.8% 3.8% Smart Phone 4,553 8,266 2,820 3,202 3,618 4,364 14,004 4,367 4,813 5,305 6,240 20,725 y/y 77.1% 81.5% 65.9% 77.5% 72.9% 63.5% 69.4% 54.8% 50.3% 46.6% 43.0% 48.0% q/q 5.7% 13.5% 13.0% 20.6% 0.1% 10.2% 10.2% 17.6% Shipment (units in 000) 471,743 677,547 210,046 224,341 240,845 278,601 953,833 273,243 288,602 308,351 344,999 1,215,195 y/y 57.9% 43.6% 42.9% 46.0% 42.4% 34.2% 40.8% 30.1% 28.6% 28.0% 23.8% 27.4% q/q 1.1% 6.8% 7.4% 15.7% -1.9% 5.6% 6.8% 11.9% Average contents per box (MB) 9,884 12,493 13,750 14,615 15,382 16,040 15,034 16,365 17,076 17,619 18,521 17,464 y/y 12.2% 26.4% 16.1% 21.6% 21.4% 21.9% 20.3% 19.0% 16.8% 14.5% 15.5% 16.2% q/q 4.5% 6.3% 5.2% 4.3% 2.0% 4.3% 3.2% 5.1% White box Smart Phone 4 189 71 78 95 107 351 105 115 129 140 489 y/y N/A 4968.3% 199.6% 114.1% 68.0% 49.1% 86.2% 48.2% 46.3% 36.8% 30.2% 39.2% q/q -2.0% 11.0% 20.5% 13.6% -2.6% 9.7% 12.7% 8.2% Shipment (units in 000) 8,500 73,500 21,000 22,000 25,500 27,600 96,100 25,000 26,000 27,500 28,000 106,500 y/y N/A 764.7% 75.0% 37.5% 21.4% 12.7% 30.7% 19.0% 18.2% 7.8% 1.4% 10.8% q/q -14.3% 4.8% 15.9% 8.2% -9.4% 4.0% 5.8% 1.8% Average contents per box (MB) 448 2,627 3,445 3,652 3,798 3,988 3,742 4,288 4,521 4,817 5,118 4,700 y/y N/A 486.1% 71.2% 55.7% 38.4% 32.4% 42.4% 24.5% 23.8% 26.8% 28.3% 25.6% q/q 14.4% 6.0% 4.0% 5.0% 7.5% 5.4% 6.6% 6.2%
Source: Nomura estimates
261
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 55: NAND demand forecasts (cont’d) Units in GB 2011 2012 1Q13 2Q13 3Q13 4Q13E 2013E 1Q14E 2Q14E 3Q14E 4Q14E 2014E Portable Media Players 1,114 1,581 394 367 433 591 1,786 476 443 523 709 2,150 y/y 18.8% 41.9% 36.6% 13.2% 15.0% -0.1% 12.9% 20.7% 20.7% 20.7% 19.8% 20.4% q/q -33.4% -7.0% 18.2% 36.5% -19.5% -7.0% 18.2% 35.5% Shipment (units in 000) 128,406 126,800 28,272 27,094 28,272 34,162 117,800 26,576 25,468 26,576 31,880 110,500 y/y -11.6% -1.3% -7.1% -7.1% -7.1% -7.1% -7.1% -6.0% -6.0% -6.0% -6.7% -6.2% q/q -23.1% -4.2% 4.3% 20.8% -22.2% -4.2% 4.3% 20.0% Average contents per box (MB) 8,887 12,770 14,287 13,861 15,696 17,727 15,525 18,344 17,797 20,153 22,762 19,928 y/y 34.5% 43.7% 47.0% 21.9% 23.8% 7.6% 21.6% 28.4% 28.4% 28.4% 28.4% 28.4% q/q -13.3% -3.0% 13.2% 12.9% 3.5% -3.0% 13.2% 12.9% Digital Still Camera 106 135 24 28 34 38 123 27 33 38 39 138 y/y 29.2% 27.6% -15.2% -12.2% -8.9% -2.6% -9.1% 15.9% 20.3% 13.5% 3.5% 12.4% q/q -38.7% 16.7% 22.4% 11.4% -27.1% 21.1% 15.5% 1.6% Shipment (units in 000) 115,534 98,150 13,863 14,615 17,248 17,816 63,542 11,704 12,805 14,260 13,431 52,200 y/y -4.9% -15.0% -36.2% -38.4% -37.5% -29.0% -35.3% -15.6% -12.4% -17.3% -24.6% -17.8% q/q -44.8% 5.4% 18.0% 3.3% -34.3% 9.4% 11.4% -5.8% Average contents per box (MB) 938 1,410 1,749 1,935 2,007 2,165 1,978 2,401 2,657 2,756 2,972 2,708 y/y 35.8% 50.2% 33.0% 42.5% 45.8% 37.3% 40.3% 37.3% 37.3% 37.3% 37.3% 36.9% q/q 10.9% 10.7% 3.7% 7.9% 10.9% 10.7% 3.7% 7.9% Digital Camcorder (Memory) 125 196 58 56 56 68 238 69 72 75 77 292 y/y 113.8% 56.6% 45.7% 22.0% 7.3% 18.1% 21.7% 18.8% 28.4% 33.4% 12.0% 22.5% q/q 0.7% -4.3% 0.4% 22.0% 1.3% 3.4% 4.4% 2.4% Shipment (units in 000) 16,570 17,310 4,263 4,238 4,212 4,187 16,900 4,093 4,068 4,044 4,020 16,220 y/y -2.6% 4.5% 0.1% -1.5% -3.2% -4.8% -2.4% -4.0% -4.0% -4.0% -4.0% -4.0% q/q -3.1% -0.6% -0.6% -0.6% -2.3% -0.6% -0.6% -0.6% Average contents per box (MB) 7,727 11,582 13,996 13,474 13,613 16,712 14,442 17,327 18,020 18,921 19,489 18,438 y/y 119.6% 49.9% 45.4% 23.9% 10.8% 24.0% 24.7% 23.8% 33.7% 39.0% 16.6% 27.7% q/q 3.9% -3.7% 1.0% 22.8% 3.7% 4.0% 5.0% 3.0% Portable Navigation System 231 272 70 68 71 76 284 78 75 74 73 300 y/y 25.9% 17.4% 11.8% -1.4% 1.8% 7.3% 4.7% 11.4% 11.4% 4.2% -3.7% 5.6% q/q -1.3% -3.0% 5.8% 6.0% 2.4% -3.0% -1.1% -2.0% Shipment (units in 000) 38,025 32,768 7,643 7,415 7,193 6,978 29,228 7,108 6,896 6,689 6,489 27,182 y/y -9.3% -13.8% -11.5% -10.3% -11.0% -10.3% -10.8% -7.0% -7.0% -7.0% -7.0% -7.0% q/q -1.7% -3.0% -3.0% -3.0% 1.9% -3.0% -3.0% -3.0% Average contents per box (MB) 6,233 8,489 9,333 9,333 10,176 11,119 9,967 11,181 11,181 11,404 11,518 11,316 y/y 38.8% 36.2% 26.4% 9.9% 14.4% 19.7% 17.4% 19.8% 19.8% 12.1% 3.6% 13.5% q/q 0.4% 0.0% 9.0% 9.3% 0.6% 0.0% 2.0% 1.0% Tablet PC 1,903 3,650 1,351 1,295 1,415 2,414 6,475 1,739 1,818 2,389 3,267 9,213 y/y 302.1% 91.8% 132.3% 55.7% 41.3% 95.4% 77.4% 28.8% 40.3% 68.9% 35.3% 42.3% q/q 9.3% -4.1% 9.2% 70.6% -27.9% 4.5% 31.4% 36.7% Shipment (units in 000) 66,994 129,518 49,200 45,100 47,600 78,100 220,000 55,180 56,990 72,513 95,318 280,000 y/y 264.1% 93.3% 142.4% 59.4% 36.7% 69.4% 69.9% 12.2% 26.4% 52.3% 22.0% 27.3% q/q 6.7% -8.3% 5.5% 64.1% -29.3% 3.3% 27.2% 31.5% Average contents per box (MB) 29,085 28,862 28,112 29,412 30,441 31,645 30,137 32,278 32,665 33,743 35,093 33,694 y/y 10.5% -0.8% -4.2% -2.3% 3.4% 15.3% 4.4% 14.8% 11.1% 10.8% 10.9% 11.8% q/q 2.4% 4.6% 3.5% 4.0% 2.0% 1.2% 3.3% 4.0% white box Tablet PC 45 515 176 207 238 272 893 243 268 289 336 1,136 y/y N/A 1048.3% 221.3% 85.8% 62.4% 34.3% 73.3% 38.0% 29.2% 21.4% 23.8% 27.2% q/q -13.0% 17.7% 15.0% 14.1% -10.6% 10.1% 8.0% 16.4% Shipment (units in 000) 7,000 52,000 15,000 16,000 17,000 18,000 66,000 16,000 17,000 18,000 20,000 71,000 y/y N/A 642.9% 114.3% 33.3% 13.3% 0.0% 26.9% 6.7% 6.3% 5.9% 11.1% 7.6% q/q -16.7% 6.7% 6.3% 5.9% -11.1% 6.3% 5.9% 11.1% Average contents per box (MB) 6,564 10,146 12,012 13,254 14,345 15,455 13,853 15,545 16,114 16,442 17,224 16,382 y/y N/A 54.6% 49.9% 39.3% 43.3% 34.3% 36.5% 29.4% 21.6% 14.6% 11.4% 18.3% q/q 4.3% 10.3% 8.2% 7.7% 0.6% 3.7% 2.0% 4.8%
Source: Nomura estimates
262
Nomura | U.S. Semiconductors Primer December 11, 2013
Company Models
263
Nomura | U.S. Semiconductors Primer December 11, 2013
Analog Devices (ADI, Neutral, TP $42)
Fig. 1: ADI Income Statement Summary Year End: October FY13 FY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E CY12 CY13E CY14E
INCOME STATEMENT Total Revenue $622.1 $659.3 $674.2 $678.1 $627.3 $658.6 $671.8 $705.4 $2,684 $2,637 $2,692 QoQ -10.5% 6.0% 2.3% 0.6% -7.5% 5.0% 2.0% 5.0% YoY -4.0% -2.3% -1.3% -2.4% 0.8% -0.1% -0.4% 4.0% -8.7% -1.7% 2.1% COGS 231.9 237.1 239.1 233.3 222.7 233.8 235.1 243.4 956 935 943 Gross Profit 390.3 422.2 435.1 444.9 404.6 424.8 436.7 462.0 1,728 1,702 1,749
SG&A 97.6 99.7 97.8 98.2 97.2 98.2 99.2 101.2 396 393 398 R&D 125.2 124.8 128.9 131.0 129.0 130.0 131.0 133.0 513 513 526 Extraordinary Expense 14.1 0.0 0.0 (69.7) 0.0 0.0 0.0 0.0 16 (65) 0 Operating Expenses 222.7 224.5 226.7 229.2 226.2 228.2 230.2 234.2 908 906 924
Operating Income 167.6 197.7 208.3 215.6 178.4 196.6 206.4 227.8 820 796 825
Other Income (Expense) (3.4) (3.7) (3.1) (2.8) (3.0) (3.0) (3.0) (3.0) (11) (13) (12)
Pretax Income 164.2 193.9 205.2 212.8 175.4 193.6 203.4 224.8 810 784 813
Income Tax Expense 27.9 32.0 24.1 15.4 22.8 25.2 26.4 29.2 154 96 106
Net Income 136.3 161.9 181.2 197.5 152.6 168.4 177.0 195.6 656 688 707
Stock Option Expense 13.1 18.0 12.7 13.1 9.5 9.9 9.8 10.3 53 55 39 % Dilution to EPS 8.7% 10.0% 6.6% 6.2% 5.9% 5.5% 5.2% 5.0% 7.5% 7.4% 5.2%
GAAP Net Income 131.1 164.5 176.2 201.6 152.6 168.4 177.0 195.6 646 688 707
Shares Outstanding 310.3 313.4 315.3 317.2 318.2 319.2 320.2 321.2 307 315 320
Proforma EPS (Incl ESO) $0.44 $0.52 $0.57 $0.62 $0.48 $0.53 $0.55 $0.61 $2.14 $2.18 $2.21 QoQ -24.5% 17.6% 11.2% 8.3% -23.0% 10.0% 4.8% 10.2% YoY -5.5% -2.9% -0.1% 7.0% 9.1% 2.1% -3.8% -2.2% -17.6% 2.0% 1.2%
GAAP EPS $0.42 $0.52 $0.56 $0.64 $0.48 $0.53 $0.55 $0.61 $2.10 $2.18 $2.21
Percent of Sales Gross Margin 62.7% 64.0% 64.5% 65.6% 64.5% 64.5% 65.0% 65.5% 64.4% 64.5% 65.0% SG&A 15.7% 15.1% 14.5% 14.5% 15.5% 14.9% 14.8% 14.3% 14.7% 14.9% 14.8% R&D 20.1% 18.9% 19.1% 19.3% 20.6% 19.7% 19.5% 18.9% 19.1% 19.4% 19.5% Operating Margin 26.9% 30.0% 30.9% 31.8% 28.4% 29.8% 30.7% 32.3% 30.6% 30.2% 30.6% Effective Tax Rate 17.0% 16.5% 11.7% 7.2% 13.0% 13.0% 13.0% 13.0% 19.0% 12.2% 13.0% Net Margin 21.9% 24.6% 26.9% 29.1% 24.3% 25.6% 26.3% 27.7% 24.4% 26.1% 26.3%
Source: Company data, Nomura estimates
264
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 2: ADI Balance Sheet Summary Year End: October FY13E FY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E CY12 CY13E CY14E Assets 5,717 5,841 6,139 6,382 6,470 6,661 6,824 7,036 5,717 6,470 7,125 Cash, Equivs, & ST Investments 3,987 4,172 4,450 4,683 4,790 4,943 5,090 5,264 3,987 4,790 5,381 Inventories 307 299 284 283 272 285 287 297 307 272 286 Total Current Assets 4,814 4,963 5,244 5,472 5,553 5,736 5,891 6,091 4,814 5,553 6,180
Total Non-Current Assets 903 878 894 909 916 925 934 944 903 916 945
Liabilities 1,393 1,380 1,495 1,642 1,634 1,639 1,640 1,644 1,393 1,634 1,640 Other Current Liabilities00000000000 Total Current Liabilities 509 509 492 571 562 568 568 573 509 562 568
Long-Term Debt 760 758 872 872 872 872 872 872 760 872 872 Total Non-Current Liabilities 884 871 1,004 1,072 1,072 1,072 1,072 1,072 884 1,072 1,072
Stockholders' Equity 4,324 4,461 4,643 4,740 4,836 5,021 5,184 5,391 4,324 4,836 5,486
Total Liabilities and Equity 5,717 5,841 6,139 6,382 6,470 6,661 6,824 7,036 5,717 6,470 7,125
Source: Company data, Nomura estimates
Fig. 3: ADI Cash Flow Summary Year End: October FY13E FY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E CY12 CY13E CY14E Cash Flow from Operations 158 252 220 282 175 233 226 260 777 924 904 Depreciation 28 27 27 28 30 31 31 32 109 112 126
Cash Flow from Investing 160 (412) (448) (250) (38) (40) (40) (42) (962) (1,082) (157) Capital Expenditures (18) (26) (30) (49) (38) (40) (40) (42) (127) (136) (157)
Cash Flow from Financing (53) (40) 92 (100) (31) (40) (39) (44) (308) (86) (159) Increase (Decrease) in Debt 0 0 0 0 0 0 505 0 0 505 0 Repurchase of common stock (17) (5) 0 (43) 0 0 0 0 (120) (53) 0 Payment of Dividends (91) (104) (105) (106) (108) (109) (109) (109) (356) (418) (436)
Free Cash Flow 140 226 190 234 138 193 185 218 651 788 747
Source: Company data, Nomura estimates
265
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 4: ADI Segment Analysis
Year End: October FY13E FY13E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E CY12 CY13E CY14E Revenue Communications 126 124 139 140 130 136 140 145 530 532 556 Consumer 107 101 100 95 85 89 94 101 461 388 375 Automotive 108 122 120 131 123 127 129 132 455 492 519 Industrial 282 312 314 313 289 306 308 327 1,238 1,225 1,242 Total 622 659 674 678 627 659 672 705 2,684 2,637 2,692
QoQ Growth Communications -11% -2% 12% 0% -7% 5% 3% 3% Consumer -22% -5% -1% -6% -10% 5% 5% 8% Automotive -2% 14% -2% 9% -6% 3% 2% 2% Industrial -8% 11% 1% 0% -8% 6% 1% 6% Total -10% 6% 2% 1% -8% 5% 2% 5%
YoY Growth Communications 3% -1% 1% -2% 3% 10% 1% 4% -9% 0% 5% Consumer -6% -6% -6% -31% -20% -12% -6% 7% -18% -16% -3% Automotive -11% 4% 5% 19% 14% 4% 7% 1% 5% 8% 6% Industrial -3%-4%-3%2%3%-2%-2%5%-9%-1%1% Total -4% -2% -1% -2% 1% 0% 0% 4% -9% -2% 2%
Percent of Revenue Converters 45% 46% 45% 45% 45% 45% 45% 45% 44% 45% 45% Amplifiers 25% 25% 25% 26% 26% 26% 26% 26% 26% 26% 26% Power Management 6% 7% 7% 7% 7% 7% 7% 7% 7% 7% 7% Other Analog 15% 14% 14% 14% 14% 14% 14% 14% 15% 14% 14% Total Analog 92% 91% 91% 91% 91% 91% 91% 91% 92% 91% 91% Total DSP 8%9%9%9%9%9%9%9%8%9%9% Total 100%100%100%100%100%100%100%100%100%100%100%
Source: Company data, Nomura estimates
Fig. 5: ADI Valuation Ratios
Year End: October FY13E FY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E CY12 CY13E CY14E Profitability Ratios Return on Equity 13% 15% 16% 17% 13% 14% 14% 15% 16% 15% 14% Return on Assets 10% 11% 12% 13% 9% 10% 11% 11% 12% 11% 10%
Efficiency Ratios Days Sales Outstanding 4946464545454545434444 Inventory Turns 0.7 0.8 0.8 0.8 0.8 0.8 0.8 0.8 3.2 3.2 3.4 Days of Inventory 122 117 111 111 111 111 111 111 115 113 108
Liquidity Ratios Current Ratio 9.5 9.7 10.7 9.6 9.9 10.1 10.4 10.6 9.5 9.9 10.9 Quick Ratio 8.5 8.8 9.8 8.8 9.1 9.3 9.5 9.8 8.5 9.1 10.1 Debt/Equity 18% 17% 19% 18% 18% 17% 17% 16% 18% 18% 16% LT Debt/Total Assets 13% 13% 14% 14% 13% 13% 13% 12% 13% 13% 12%
Cash & Book Value/Share Book Value/Share 13.94 14.23 14.73 14.94 15.20 15.73 16.19 16.78 14.08 15.33 17.12 Net Cash/Share 10.40 10.90 11.35 12.01 12.31 12.75 13.17 13.67 10.51 12.42 14.07 FCF/Share 0.450.720.600.740.430.610.580.682.122.502.33
Source: Company data, Nomura estimates
266
Nomura | U.S. Semiconductors Primer December 11, 2013
Altera Corp. (ALTR, Reduce, TP $27)
Fig. 6: ALTR Income Statement Summary 2013E 2014E 2012 2013E 2014E ($ in millions) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14
INCOME STATEMENT Total Revenue 410.5 421.8 445.9 441.4 445.9 454.5 474.9 475.9 1,783 1,720 1,851 QoQ -6.6% 2.7% 5.7% -1.0% 1.0% 1.9% 4.5% 0.2% YoY 7.0% -9.3% -9.9% 0.4% 8.6% 7.8% 6.5% 7.8% -13.6% -3.6% 7.7%
COGS 126.1 135.1 141.5 139.0 142.7 145.4 152.0 152.3 541.5 541.7 592.4
Gross profit 284.4 286.7 304.4 302.3 303.2 309.1 322.9 323.6 1,241.5 1,177.8 1,258.8
R&D 87.9 95.5 95.3 110.5 105.5 105.5 105.5 106.5 360.4 389.3 423.0 SG&A 78.6 77.9 78.9 82.5 77.5 75.5 75.5 76.5 289.9 317.9 305.0 Total operating expenses 166.5 174.3 176.1 193.0 183.0 181.0 181.0 183.0 650.3 709.9 728.0
Operating income 117.9 112.4 128.3 109.3 120.2 128.1 141.9 140.6 591.2 467.9 530.8
Interest and Other (0.8) (0.6) (0.3) (0.3) (0.2) (0.2) (0.2) (0.2) 0.7 (1.8) (0.8)
Pretax income 117.1 111.8 128.1 109.1 120.0 127.9 141.7 140.4 591.9 466.1 530.0
Provision for Taxes (3.1) 10.3 8.6 12.5 13.8 14.7 16.3 16.1 35.1 28.4 60.9
Net income before 1-time charges 120.2 101.5 119.4 96.5 106.2 113.2 125.4 124.3 556.8 437.7 469.0
Net income - proforma 120.2 101.5 119.4 96.5 106.2 113.2 125.4 124.3 556.8 437.7 469.0 Net income - GAAP 120.2 101.5 119.4 96.5 106.2 113.2 125.4 124.3 556.8 437.7 469.0
EPS - proforma (incl. ESO) $0.37 $0.31 $0.37 $0.30 $0.33 $0.35 $0.39 $0.38 $1.72 $1.35 $1.45 EPS - GAAP $0.37 $0.31 $0.37 $0.30 $0.33 $0.35 $0.39 $0.38 $1.72 $1.35 $1.45
Shares outstanding - fully diluted 323.0 323.5 323.5 323.0 323.5 324.0 324.5 325.0 324.5 323.3 324.3
Percent of Sales Gross Margin 69.3% 68.0% 68.3% 68.5% 68.0% 68.0% 68.0% 68.0% 69.6% 68.5% 68.0% R&D 21.4% 22.6% 21.4% 25.0% 23.7% 23.2% 22.2% 22.4% 20.2% 22.6% 22.9% SG&A 19.1% 18.5% 17.7% 18.7% 17.4% 16.6% 15.9% 16.1% 16.3% 18.5% 16.5% Operating Margin 28.7% 26.6% 28.8% 24.8% 27.0% 28.2% 29.9% 29.5% 33.2% 27.2% 28.7% Effective Tax Rate -2.6% 9.2% 6.7% 11.5% 11.5% 11.5% 11.5% 11.5% 5.9% 6.1% 11.5% Net Margin 29.3% 24.1% 26.8% 21.9% 23.8% 24.9% 26.4% 26.1% 31.2% 25.5% 25.3%
Source: Company data, Nomura estimates
267
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 7: ALTR Balance Sheet Summary Year End: Dec 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E
Assets 4,805 4,879 5,020 5,057 5,107 5,168 5,258 5,343 4,658 5,057 5,343 Cash, Equiv, and ST Investments 3,125 2,954 3,107 3,147 3,192 3,246 3,313 3,401 3,018 3,147 3,401 Inventories 139 134 158 155 159 162 170 170 153 155 170 Total Current Assets 3,825 3,763 3,884 3,915 3,969 4,034 4,128 4,217 3,680 3,915 4,217
Liabilities 1,355 1,431 1,432 1,427 1,424 1,426 1,428 1,427 1,324 1,427 1,427 Total Current Liabilities 569 631 657 651 648 651 653 652 543 651 652
Long-Term Debt 500 500 500 500 500 500 500 500 500 500 500 Total Non-Current Liabilities 785 801 775 775 775 775 775 775 781 775 775
Shareholders' Equity 3,450 3,448 3,588 3,631 3,683 3,742 3,830 3,916 3,333 3,631 3,916
Total Liabilities & Equity 4,805 4,879 5,020 5,057 5,107 5,168 5,258 5,343 4,658 5,057 5,343
Source: Company data, Nomura estimates
Fig. 8: ALTR Cash Flow Summary Year End: Dec 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E
Cash Flows from Operations 149 65 245 112 107 116 112 134 589 571 468 Depreciation & Amortization 11 11 12 12 12 12 12 12 37 46 47
Cash Flows from Investing (46) (136) (21) (8) (8) (8) (8) (8) (767) (212) (32) Capital Expenditures (15) (9) (8) (8) (8) (8) (8) (8) (61) (39) (32)
Cash Flows from Financing (26) (93) (57) (54) (54) (54) (38) (38) (315) (230) (183) Increase (Decrease) in Debt0(22)0000000(22)0 Repurchase of Common Stock 0 (55) (5) (5) (5) (5) (5) (5) (229) (66) (21) Payment of Dividends (32) (32) (48) (48) (49) (49) (32) (33) (116) (161) (162)
Free Cash Flow 135 56 238 104 99 108 104 126 528 532 437
Source: Company data, Nomura estimates
268
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 9: ALTR Segment Analysis
Year End: Dec 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Revenues Telecom & Wireless 170 177 182 182 182 186 195 195 787 711 758 Industrial Military & Auto 89 92 102 102 104 106 109 109 368 385 429 Networking, Computer & Storage 74 74 84 86 88 90 97 98 299 318 374 Other 77 79 78 71 72 72 73 73 329 306 291 Total 411 422 446 441 446 454 475 476 1,782 1,720 1,851
Percent of Revenues Telecom & Wireless 41% 42% 41% 41% 41% 41% 41% 41% 44% 41% 41% Industrial Military & Auto 22% 22% 23% 23% 23% 23% 23% 23% 21% 22% 23% Networking, Computer & Storage 18% 18% 19% 20% 20% 20% 20% 21% 17% 18% 20% Other 19% 19% 18% 16% 16% 16% 15% 15% 18% 18% 16% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Telecom & Wireless -13% 4% 3% 0% 0% 2% 5% 0% - - - Industrial Military & Auto -2% 3% 11% 0% 2% 2% 3% 0% - - - Networking, Computer & Storage 1% 0% 13% 3% 2% 3% 8% 1% - - - Other -2%2%-1%-9%1%1%1%0%- - - Total -7%3%6%-1%1%2%4%0% ---
YoY Growth Telecom & Wireless 8% -16% -19% -7% 7% 5% 7% 7% -12% -10% 7% Industrial Military & Auto 3% 2% 2% 12% 17% 15% 7% 7% -22% 5% 11% Networking, Computer & Storage 18% -7% 0% 18% 19% 22% 16% 14% -12% 6% 18% Other 2% -8% -11% -10% -7% -8% -6% 3% -8% -7% -5% Total 7% -9% -10% 0% 9% 8% 7% 8% -14% -4% 8%
Source: Company data, Nomura estimates
Fig. 10: ALTR Valuation Ratios
Year End: Dec 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Profitability Ratios Return on Equity 14.2% 11.8% 13.6% 10.7% 11.6% 12.2% 13.3% 12.8% 17.6% 12.6% 12.4% Return on Assets 10.0% 8.3% 9.5% 7.6% 8.3% 8.8% 9.5% 9.3% 12.0% 8.7% 8.8%
Efficiency Ratios Days Sales Outstanding 81.2 102.0 89.1 89.1 89.1 89.1 89.1 89.1 67.0 89.1 89.1 Inventory Turns 3.6 4.0 3.6 3.6 3.6 3.6 3.6 3.6 3.5 3.6 3.6 Days of Inventory 100.5 90.5 101.9 101.9 101.9 101.9 101.9 101.9 104.2 101.9 101.9
Liquidity Ratios Current Ratio 6.7 6.0 5.9 6.0 6.1 6.2 6.3 6.5 6.8 6.0 6.5 Quick Ratio 6.1 5.4 5.4 5.5 5.6 5.7 5.8 5.9 6.2 5.5 5.9 Debt/Capital 28.2% 29.3% 28.5% 28.2% 27.9% 27.6% 27.2% 26.7% 28.4% 28.2% 26.7% Debt/Assets 28.2% 29.3% 28.5% 28.2% 27.9% 27.6% 27.2% 26.7% 28.4% 28.2% 26.7%
Book & Cash Value Book Value/Share $10.7 $10.7 $11.1 $11.2 $11.4 $11.5 $11.8 $12.1 $10.3 $11.2 $12.1 Net Cash/Share $10.3 $9.6 $10.3 $10.4 $10.5 $10.7 $10.9 $11.1 $9.9 $10.4 $11.1 FCF/Share $0.4 $0.2 $0.7 $0.3 $0.3 $0.3 $0.3 $0.4 $1.6 $1.6 $1.3
Source: Company data, Nomura estimates
269
Nomura | U.S. Semiconductors Primer December 11, 2013
Advanced Micro Devices (AMD, Neutral, TP $4)
Fig. 11: AMD Income Statement Summary Year End: December 2013E 2014E 2012 2013E 2014E ($ in millions) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14
INCOME STATEMENT Total Revenue 1,088 1,161 1,461 1,535 1,404 1,446 1,599 1,655 5,422 5,245 6,103 QoQ -5.8% 6.7% 25.8% 5.0% -8.5% 3.0% 10.6% 3.5% YoY -31.4% -17.8% 15.1% 32.9% 29.0% 24.5% 9.4% 7.8% -17.4% -3.3% 16.4%
COGS 643 702 940 998 916 945 1,049 1,091 4,187 3,283 4,002 Gross profit 445 459 521 537 488 501 550 563 1,235 1,962 2,102
R & D 312 308 288 293 296 298 302 304 1,354 1,201 1,200 SG&A 179 171 155 157 157 161 163 164 823 662 645 Amort. of acquired intangibles54555555131920 Restructuring charges 475-2200000101300 Total operating expenses 543 488 426 455 458 464 470 473 2,291 1,912 1,865
Operating income -98 -29 95 82 30 37 80 90 -1,056 50 237 Operating income - proforma -46 -20 78 87 35 42 85 95 39 99 257
One time losses / (gains) 00000000000 Net Interest Expense / (Income) 46 42 44 44 44 44 44 44 161 176 176 Pretax income -144 -71 51 38 -14 -7 36 46 -1,217 -126 61
Provision for Taxes 23333344-341114 Net income/(loss) -146 -74 48 35 -17 -10 32 42 -1,183 -137 47
Net income - proforma -94 -65 30 39 -13 -6 36 46 -114 -90 63
GAAP EPS -$0.19 -$0.10 $0.06 $0.04 -$0.02 -$0.01 $0.04 $0.05 -$1.58 -$0.18 $0.06 Proforma EPS -$0.13 -$0.09 $0.04 $0.05 -$0.02 -$0.01 $0.05 $0.06 -$0.15 -$0.12 $0.08
Diluted shares proforma 749 752 764 767 770 773 776 779 749 758 775
Percent of Sales Gross margin 40.9% 39.5% 35.7% 35.0% 34.7% 34.6% 34.4% 34.0% 22.8% 37.4% 34.4% Gross margin (non-GAAP) 40.9% 39.5% 35.7% 35.0% 34.7% 34.6% 34.4% 34.0% 40.9% 37.4% 34.4% R&D 28.7% 26.5% 19.7% 19.1% 21.1% 20.6% 18.9% 18.4% 25.0% 22.9% 19.7% SG&A 16.5% 14.7% 10.6% 10.2% 11.2% 11.1% 10.2% 9.9% 15.2% 12.6% 10.6% Operating Margin (proforma) -4.2% -1.7% 5.3% 5.7% 2.5% 2.9% 5.3% 5.8% 0.7% 1.9% 4.2% Tax Rate -1.4% -4.2% 5.9% 8% -21.1% -40.9% 11.2% 8.6% 2.8% -8.7% 23.1% Net Income Margin -13.4% -6.4% 3.3% 2.3% -1.2% -0.7% 2.0% 2.6% -21.8% -2.6% 0.8% Net margin (non-GAAP) -8.6% -5.6% 2.1% 2.5% -0.9% -0.4% 2.2% 2.8% -2.1% -1.7% 1.0%
Source: Company data, Nomura estimates
2 70
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 12: AMD Balance Sheet Summary Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Assets 3,797 3,897 4,317 4,440 4,353 4,399 4,572 4,686 4,000 4,440 4,686 Cash, Equivalents, & ST Investments 1,003 968 1,060 1,338 1,462 1,481 1,508 1,578 1,002 1,338 1,578 Inventories 613 711 922 765 653 673 747 777 562 765 777 Total Current Assets 2,338 2,458 2,939 3,104 3,037 3,102 3,295 3,428 2,265 3,104 3,428
Total Non-Current Assets 1,459 1,439 1,378 1,336 1,317 1,297 1,278 1,258 1,735 1,336 1,258
Liabilities 3,382 3,538 3,883 3,948 3,856 3,889 4,007 4,055 3,462 3,948 4,055 Other Current Liabilities 43 26 21 21 21 21 21 21 63 21 21 Total Current Liabilities 1,321 1,451 1,762 1,827 1,735 1,768 1,886 1,934 1,397 1,827 1,934
Long Term Debt 2,039 2,042 2,044 2,044 2,044 2,044 2,044 2,044 2,037 2,044 2,044 Total Non-Current Liabilities 2,061 2,087 2,121 2,121 2,121 2,121 2,121 2,121 2,065 2,121 2,121
Shareholders' Equity 415 359 434 492 498 510 565 631 538 492 631
Total Liabilities & Equity 3,797 3,897 4,317 4,440 4,353 4,399 4,572 4,686 4,000 4,440 4,686
Source: Company data, Nomura estimates
Fig. 13: AMD Cash Flow Summary Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Cash Flows from Operations (155) (35) 21 293 161 57 64 107 (338) 124 389 Depreciation 66 59 57 57 57 57 57 57 260 239 228
Cash Flows from Investing 47 (74) 186 (15) (38) (38) (38) (38) (19) 144 (150) Capital Expenditures (20) (28) (15) (15) (38) (38) (38) (38) (133) (78) (150)
Cash Flows from Financing - 2 2 - - - - - 37 4 - Increase (Decrease) in Debt (1) - (2) - - - - - 3 (3) - Repurchase of Common Stock 1 ------14 1 - Payment of Dividends ------
Free Cash Flow (175) (63) 6 278 123 19 27 69 (471) 46 239
Source: Company data, Nomura estimates
271
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 14: AMD Segment Analysis
Year End: December 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Revenues Computing Solutions 751 841 790 778 726 731 752 752 4,005 3,160 2,961 Graphics and Visual Solutions 337 320 671 757 678 715 847 903 1,416 2,084 3,143 Total revenue 1,088 1,161 1,461 1,535 1,404 1,446 1,599 1,655 5,422 5,245 6,103
Percent of Revenue Computing Solutions 69.0% 72.4% 54.1% 50.7% 51.7% 50.6% 47.0% 45.4% 73.9% 60.3% 48.5% Graphics 31.0% 27.6% 45.9% 49.3% 48.3% 49.4% 53.0% 54.6% 26.1% 39.7% 51.5% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
QoQ Growth Computing Solutions -9.4% 11.9% -6.0% -1.6% -6.6% 0.7% 2.8% 0.0% - - - Graphics 3.3% -5.0% 109.6% 12.8% -10.4% 5.5% 18.6% 6.6% - - - Total -5.8% 6.7% 25.8% 5.0% -8.5% 3.0% 10.6% 3.5% - - -
YoY Growth Computing Solutions -37.5% -19.6% -14.8% -6.2% -3.3% -13.1% -4.9% -3.4% -19.9% -21.1% -6.3% Graphics -11.8% -12.8% 96.4% 132.1% 101.2% 123.3% 26.3% 19.4% -9.5% 47.1% 50.8% Total -31.3% -17.9% 15.2% 32.9% 29.0% 24.5% 9.4% 7.8% -17.5% -3.3% 16.4%
Source: Company data, Nomura estimates
Fig. 15: AMD Valuation Ratios
Year End: December 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Profitability Ratios Return on Equity -140.7% -82.5% 44.2% 28.4% -13.8% -8.1% 22.5% 26.8% -219.9% -27.9% 7.4% Return on Assets -6.8% -9.2% -5.2% -2.1% -0.2% 1.2% 1.2% 1.4% -2.5% -6.8% -9.2%
Efficiency Ratios Days Sales Outstanding 54.1 52.7 54.5 54.5 54.5 54.5 54.5 54.5 49.8 54.5 54.5 Inventory Turns 4.2 3.9 4.1 5.2 5.6 5.6 5.6 5.6 7.0 5.2 5.6 Days of Inventory 87.0 92.4 89.5 70.0 65.0 65.0 65.0 65.0 52.5 70.0 65.0
Liquidity Ratios Current Ratio 1.8 1.7 1.7 1.7 1.8 1.8 1.7 1.8 1.6 1.7 1.8 Quick Ratio 0.9 0.8 0.8 1.0 1.0 1.0 1.0 1.1 0.8 1.0 1.1 Total Debt/Equity 5.0 5.8 4.9 4.3 4.3 4.2 3.8 3.4 3.8 4.3 3.4 Debt/Capital 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
Cash & Book Value/Share Book Value/Share $0.5 $0.5 $0.6 $0.6 $0.6 $0.6 $0.7 $0.8 $0.7 $0.6 $0.8 Cash/Share -$1.4 -$1.4 -$1.3 -$0.9 -$0.7 -$0.7 -$0.7 -$0.6 -$1.4 -$0.9 -$0.6 FCF/Share -$0.2 -$0.1 $0.0 $0.4 $0.2 $0.0 $0.0 $0.1 -$0.6 $0.1 $0.3
Source: Company data, Nomura estimates
272
Nomura | U.S. Semiconductors Primer December 11, 2013
Atmel Corp. (ATML, Neutral, TP $8)
Fig. 16: ATML Income Statement Summary 2013E 2014E ($ in millions) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E
INCOME STATEMENT Total Revenue 329.1 347.8 356.3 357.0 342.7 363.3 399.6 399.6 1,432.1 1,390.2 1,505.1 Q/Q -5% 6% 2% 0% -4% 6% 10% 0% Y/Y -8% -6% -1% 3% 4% 4% 12% 12% -21% -3% 8%
Total COGS 197.8 199.9 212.8 202.1 187.1 192.9 203.8 201.8 830.8 812.6 785.6
Gross Profit (GAAP) 131.3 147.9 143.5 154.9 155.6 170.4 195.8 197.8 601.3 577.6 719.6 Gross Profit (Non-GAAP) 133.1 148.0 153.5 156.9 157.6 172.4 197.8 199.8 620.0 591.6 727.6
R&D 63.6 58.9 55.8 56.0 57.5 59.0 61.0 61.5 251.5 234.3 239.0 SG&A 68.3 67.4 66.8 68.0 69.5 71.0 73.0 73.5 275.3 270.5 287.0 Acquisition-related charges 2.3 1.8 1.7 2.0 2.0 2.0 2.0 2.0 7.4 7.7 8.0 One-time charges 59.5 0.5 8.1 0.0 0.0 0.0 0.0 0.0 19.8 68.2 0.0 Total Operating Expenses 193.7 128.5 132.4 126.0 129.0 132.0 136.0 137.0 554.0 580.6 534.0 Total Operating Exp (non-GAAP) 119.0 120.4 115.6 111.0 114.0 117.0 121.0 122.0 462.4 466.0 474.0
Operating income (62.4) 19.4 11.1 28.9 26.6 38.4 59.8 60.8 47.4 (3.0) 185.6 Operating income (non-GAAP) 14.2 27.6 37.9 45.9 43.6 55.4 76.8 77.8 157.6 125.6 253.6
Interest income/(expense) 0.4 (0.7) 1.4 (1.5) (1.5) (1.5) (1.5) (1.5) (5.1) (0.5) (6.0)
Pretax Income (62.0) 18.7 12.5 27.4 25.1 36.9 58.3 59.3 42.2 (3.5) 179.6 Provision for taxes (14.4) 5.7 7.0 1.5 2.5 3.2 4.5 4.6 11.8 (0.1) 14.9 Provision for taxes (non-GAAP) 1.0 1.6 1.6 1.5 2.5 3.2 4.5 4.6 7.4 5.7 14.9
GAAP net income (47.7) 13.0 5.4 25.9 22.6 33.6 53.8 54.7 30.4 (3.3) 164.7 Non-GAAP net income (incl ESO) 14.1 13.7 24.2 27.9 24.6 35.6 55.8 56.7 68.3 80.0 172.7 Non-GAAP net income (ex-ESO) 13.6 25.3 37.7 42.9 39.6 50.6 70.8 71.7 145.1 119.5 232.7
GAAP EPS ($0.11) $0.03 $0.01 $0.06 $0.05 $0.08 $0.13 $0.13 $0.07 ($0.01) $0.39 Non-GAAP EPS (incl ESO) $0.03 $0.03 $0.06 $0.06 $0.06 $0.08 $0.13 $0.13 $0.15 $0.18 $0.40 Non-GAAP EPS (ex-ESO) $0.03 $0.06 $0.09 $0.10 $0.09 $0.12 $0.16 $0.17 $0.32 $0.27 $0.54
Diluted shares (non-GAAP) 442.8 439.7 438.8 433.8 433.8 433.8 433.8 433.8 446.5 438.8 433.8
Percent of Sales Gross Margin 39.9% 42.5% 40.3% 43.4% 45.4% 46.9% 49.0% 49.5% 42.0% 41.5% 47.8% Gross Margin (non-GAAP) 40.5% 42.6% 43.1% 44.0% 46.0% 47.5% 49.5% 50.0% 43.3% 42.6% 48.3% SG&A 20.8% 19.4% 18.7% 19.0% 20.3% 19.5% 18.3% 18.4% 19.2% 19.5% 19.1% R&D 19.3% 16.9% 15.7% 15.7% 16.8% 16.2% 15.3% 15.4% 17.6% 16.9% 15.9% Operating Margin -19.0% 5.6% 3.1% 8.1% 7.8% 10.6% 15.0% 15.2% 3.3% -0.2% 12.3% Operating Margin (non-GAAP) 4.3% 7.9% 10.6% 12.9% 12.7% 15.2% 19.2% 19.5% 11.0% 9.0% 16.8% Pretax Margin -18.8% 5.4% 3.5% 7.7% 7.3% 10.1% 14.6% 14.8% 2.9% -0.3% 11.9% Tax Rate (non-GAAP) 6.7% 6.0% 4.0% 3.4% 6.0% 6.0% 6.0% 6.0% 4.8% 4.5% 6.0% Net Margin -14.5% 3.7% 1.5% 7.3% 6.6% 9.3% 13.5% 13.7% 2.1% -0.2% 10.9%
Source: Company data, Nomura estimates
273
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 17: ATML Balance Sheet Summary Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Assets 1,397 1,373 1,368 1,399 1,425 1,461 1,515 1,576 1,434 1,399 1,576 Cash, Equiv, and ST Investments 245 227 271 312 361 394 428 500 296 312 500 Inventories 336 324 288 287 271 265 272 279 348 287 279
Liabilities 454 426 429 434 437 440 440 446 437 434 446 Total Current Liabilities 342 315 309 314 317 320 320 326 337 314 326 Long Term Debt 0 0 0 0 0 0 0 0 0 0 0 Total Non-Current Liabilities 112 112 120 120 120 120 120 120 100 120 120
Shareholders' Equity 943 947 939 965 988 1,021 1,075 1,130 997 965 1,130
Total Liabilities & Equity 1,397 1,373 1,368 1,399 1,425 1,461 1,515 1,576 1,434 1,399 1,576
Source: Company data, Nomura estimates
Fig. 18: ATML Cash Flow Summary Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Cash Flows from Operations (12) 9 82 69 76 60 62 99 201 148 298 Depreciation 20 19 18 19 19 19 19 19 77 76 76
Cash Flows from Investing (26) (11) (13) (13) (13) (13) (13) (13) (52) (63) (50) Capital Expenditures (4) (9) (15) (13) (13) (13) (13) (13) (38) (41) (50)
Cash Flows from Financing (12) (17) (28) (15) (15) (15) (15) (15) (183) (71) (60) Increase (Decrease) in Debt 0 0 0 0 0 0 0 0 0 0 0 Repurchase of Common Stock (15) (14) (34) (20) (20) (20) (20) (20) (180) (84) (80) Payment of Dividends 0 0 0 0 0 0 0 0 0 0 0
Free Cash Flow (16) (1) 68 56 64 47 50 87 162 107 248
Source: Company data, Nomura estimates
274
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 19: ATML Segment Analysis
Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Revenues Microcontrollers 215 231 227 232 220 238 273 273 893 905 1,004 Nonvolatile memory 27 27 31 29 29 30 30 30 171 114 118 RF and auto 47 41 45 51 49 51 51 50 174 184 200 ASIC 40 48 53 46 45 45 46 47 194 187 183 Total 329 348 356 357 343 363 400 400 1,432 1,390 1,505
Percent of Revenues Microcontrollers 65% 66% 64% 65% 64% 65% 68% 68% 62% 65% 67% Nonvolatile memory 8% 8% 9% 8% 8% 8% 7% 7% 12% 8% 8% RF and auto 14% 12% 13% 14% 14% 14% 13% 12% 12% 13% 13% ASIC 12% 14% 15% 13% 13% 12% 12% 12% 14% 13% 12% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Microcontrollers -6% 7% -2% 2% -5% 8% 15% 0% Nonvolatile memory -15% 1% 14% -7% 0% 3% 0% 0% RF and auto 17% -11% 10% 12% -3% 3% 0% -2% ASIC -10% 21% 9% -13% -3% 1% 2% 2% Total -5% 6% 2% 0% -4% 6% 10% 0%
YoY Growth Microcontrollers -1% 5% 0% 1% 2% 3% 20% 18% -20% 1% 11% Nonvolatile memory -43% -42% -29% -10% 7% 9% -4% 3% -33% -33% 3% RF and auto 7% -13% 5% 27% 6% 23% 11% -2% -14% 6% 9% ASIC -18% -9% 11% 3% 12% -7% -13% 3% -16% -4% -2% Total -8% -6% -1% 3% 4% 4% 12% 12% -21% -3% 8%
Source: Company data, Nomura estimates
Fig. 20: ATML Valuation Ratios
Year End: December 2013E 2014E 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Profitability Ratios Return on Equity -20% 5% 2% 11% 9% 13% 20% 19% 3% 0% 15% Return on Assets -14% 4% 2% 7% 6% 9% 14% 14% 2% 0% 10%
Efficiency Ratios Days Sales Outstanding 54.4 54.1 50.7 49.7 51.5 52.7 52.3 49.6 48.0 51.1 52.7 Inventory Turns 2.4 2.5 3.0 2.8 2.8 2.9 3.0 2.9 2.4 2.8 2.8 Days of Inventory 154.9 147.8 123.3 129.6 132.3 125.6 121.9 126.2 153.0 128.9 129.7
Liquidity Ratios Current Ratio 2.6 2.8 2.8 2.9 3.0 3.1 3.3 3.4 2.8 2.9 3.4 Total Debt/Equity 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Total Debt/Assets 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Cash & Book Value/Share Book Value/Share $2.2 $2.2 $2.2 $2.3 $2.3 $2.4 $2.5 $2.7 $2.3 $2.3 $2.7 Cash/Share $0.6 $0.5 $0.6 $0.7 $0.9 $0.9 $1.0 $1.2 $0.7 $0.7 $1.2 FCF/Share $0.0 $0.0 $0.2 $0.1 $0.1 $0.1 $0.1 $0.2 $0.4 $0.2 $0.6
Source: Company data, Nomura estimates
275
Nomura | U.S. Semiconductors Primer December 11, 2013
Avago Technologies (AVGO, Buy, TP $46)
Fig. 21: AVGO Income Statement Summary Year End: October FY13 FY2014E CY12 CY13E CY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E
INCOME STATEMENT Total Revenue $576 $562 $644 $738 $705 $688 $727 $774 $2,377 $2,649 $2,941 QoQ -6.8% -2.4% 14.6% 14.6% -4.4% -2.5% 5.7% 6.5% YoY 2.3% -2.6% 6.3% 19.4% 22.5% 22.4% 12.9% 4.9% 1.2% 11.5% 11.0%
COGS (GAAP) 284 274 322 371 349 340 358 379 1,165 1,316 1,445
Gross Profit (Non-GAAP) 292 288 328 373 356 347 369 395 1,213 1,345 1,497
R&D (GAAP) 93 95 101 108 109 110 111 112 346 413 447 SG&A (GAAP) 53 52 57 60 61 61 62 63 202 230 250 Restructuring and Other one-time 6 7 6 7 11 7 7 7 26 31 28 Operating Expenses (non-GAAP) 130 132 141 149 145 150 152 154 495 567 612
Operating Income (non-GAAP) 162 156 187 224 211 197 217 241 718 778 885
Net Interest and Other Income 2 0 4 11 2 2 2 2 7 17 8 Other One Time Gains/Losses 0 0 0 (1) 0 0 0 0 0 (1) 0
Pretax Income (non-GAAP) 164 156 191 235 213 199 219 243 725 795 893
Income Tax Expense 1 5 2 8 10 7 8 8 18 25 31
GAAP Net Income 125 113 142 172 144 143 163 185 563 571 664 Proforma Net Income (incl ESO) 145 134 169 205 179 168 188 210 647 687 764 Proforma Net Income (ex ESO) 163 151 189 227 203 192 212 234 707 770 861
GAAP EPS $0.50 $0.45 $0.56 $0.68 $0.57 $0.56 $0.63 $0.72 $2.23 $2.25 $2.56 Proforma EPS (incl ESO) $0.58 $0.53 $0.67 $0.80 $0.70 $0.65 $0.72 $0.81 $2.56 $2.71 $2.95 Proforma EPS (ex ESO) $0.65 $0.60 $0.75 $0.89 $0.79 $0.75 $0.82 $0.90 $2.80 $3.04 $3.32
Shares Outstanding (non-GAAP) 252 251 252 255 257 258 259 260 253 254 260
Percent of Sales Gross Margin (non-GAAP) 50.7% 51.2% 50.9% 50.5% 50.5% 50.5% 50.8% 51.0% 51.0% 50.8% 50.9% R&D 16.1% 16.9% 15.7% 14.6% 15.5% 16.0% 15.3% 14.5% 14.6% 15.6% 15.2% SG&A 9.2% 9.3% 8.9% 8.1% 8.6% 8.9% 8.5% 8.1% 8.5% 8.7% 8.5% Operating Expenes (non-GAAP) 22.6% 23.5% 21.9% 20.2% 20.6% 21.8% 20.9% 19.9% 20.8% 21.4% 20.8% Operating Margin (non-GAAP) 28.1% 27.8% 29.0% 30.4% 29.9% 28.7% 29.9% 31.1% 30.2% 29.4% 30.1% Tax Rate 0.6% 3.2% 1.0% 3.4% 4.7% 3.5% 3.5% 3.5% 2.5% 3.1% 3.5% Net Margin 25.2% 23.8% 26.2% 27.8% 25.4% 24.5% 25.8% 27.2% 27.2% 25.9% 26.0%
Source: Company data, Nomura estimates
276
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 22: AVGO Balance Sheet Summary Year End: October FY2013E FY2014E CY12 CY13E CY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Assets 2,892 3,039 3,204 3,415 3,543 3,686 3,867 4,073 2,892 3,543 4,244 Cash, Equivs, and ST Investments 1,151 1,219 863 985 1,169 1,322 1,464 1,624 1,151 1,169 1,810 Inventories 208 229 284 285 268 262 275 291 208 268 282 Total Current Assets 1,713 1,818 1,641 1,818 1,941 2,079 2,255 2,456 1,713 1,941 2,622
Total Non-Current Assets 1,179 1,221 1,563 1,597 1,602 1,607 1,612 1,617 1,179 1,602 1,622
Liabilities 374 428 478 529 507 501 513 529 374 507 520 Other Current Liabilities 32 30 32 46 46 46 46 46 32 46 46 Total Current Liabilities 277 330 380 423 401 395 407 423 277 401 414
Long-Term Debt 0 0 0 00000000 Total Non-Current Liabilities 97 98 98 106 106 106 106 106 97 106 106
Shareholders' Equity 2,518 2,611 2,726 2,886 3,036 3,185 3,353 3,544 2,518 3,036 3,723
Total Liabilities & Equity 2,892 3,039 3,204 3,415 3,543 3,686 3,867 4,073 2,892 3,543 4,244
Source: Company data, Nomura estimates
Fig. 23: AVGO Cash Flow Summary Year End: October FY2013E FY2014E CY12 CY13E CY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Cash Flow from Operations 185 191 137 212 264 234 222 241 739 804 964 Depreciation & Amortization 41 42 46 58 58 58 58 58 159 204 232
Cash Flow from Investing (76) (84) (438) (54) (60) (60) (60) (60) (273) (636) (240) Capital Expenditures (67) (47) (65) (57) (63) (63) (63) (63) (261) (232) (252)
Cash Flow from Financing (42) (39) (55) (36) (20) (20) (20) (21) (134) (150) (82) Increase (Decrease) in Debt00000000100 Repurchase of Common Stock (13) 0 (49) (33) (33) (33) (33) (33) (44) (115) (132) Payment of Dividends - Convertibles (42) (47) (52) (57) (41) (41) (41) (42) (150) (197) (166)
Free Cash Flow 118 144 72 155 201 171 159 178 478 572 712
Source: Company data, Nomura estimates
277
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 24: AVGO Segment Analysis
Year End: October FY2013E FY2014E CY12 CY13E CY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Revenues Handsets 258 234 244 300 294 265 278 333 917 1072 1202 Wireless Infrastructure 52 47 46 47 45 47 50 49 206 185 194 Wired Infrastructure 144 154 204 248 233 240 249 244 642 838 971 Industrial and auto 118 123 145 140 132 135 148 145 524 541 567 Consumer and Computing 5443222289138 Total 576 562 644 738 705 688 727 774 2377 2650 2941
Percent of Revenue Handsets 45% 42% 38% 41% 42% 38% 38% 43% 39% 40% 41% Wireless Infrastructure 9% 8% 7% 6% 6% 7% 7% 6% 9% 7% 7% Wired Infrastructure 25% 27% 32% 34% 33% 35% 34% 32% 27% 32% 33% Industrial and auto 20% 22% 23% 19% 19% 20% 20% 19% 22% 20% 19% Consumer and Computing 1% 1% 1% 0% 0% 0% 0% 0% 4% 1% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Handsets -2% -9% 5% 23% -2% -10% 5% 20% Wireless Infrastructure -2% -9% -2% 2% -5% 5% 6% -1% Wired Infrastructure -10% 7% 32% 22% -6% 3% 4% -2% Industrial and auto -11% 4% 18% -4% -6% 2% 10% -2% Consumer and Computing -60% -20% 0% -20% -40% 0% 0% 0% Total -7% -2% 15% 15% -4% -2% 6% 6%
YoY Growth Handsets 27% 14% 27% 15% 14% 13% 14% 11% 26% 17% 12% Wireless Infrastructure 2% -8% -8% -10% -14% 0% 8% 5% -4% -10% 4% Wired Infrastructure -11% -8% 18% 56% 62% 56% 22% -1% -3% 31% 16% Industrial and auto 3% -2% -2% 6% 12% 9% 2% 3% -17% 3% 5% Consumer and Computing -84% -86% -90% -74% -62% -52% -52% -40% -23% -85% -41% Total 2% -3% 6% 19% 22% 22% 13% 5% 1% 11% 11%
Source: Company data, Nomura estimates
Fig. 25: AVGO Valuation Ratios
Year End: October FY2013E FY2014E CY12 CY13E CY14E ($ in millions) Jan-13 Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Profitability Ratios Return on Equity 23% 21% 25% 29% 24% 22% 23% 24% 26% 23% 21% Return on Assets 20% 18% 22% 25% 21% 19% 20% 21% 22% 19% 18%
Efficiency Ratios Days Sales Outstanding 48 44 45 48 48 48 48 48 44 44 48 Inventory Turns 1.4 1.2 1.1 1.3 1.3 1.3 1.3 1.3 5.6 4.9 5.1 Days of Inventory 67 76 80 70 70 70 70 70 65 74 71
Liquidity Ratios Current Ratio 6.2 5.5 4.3 4.3 4.8 5.3 5.5 5.8 6.2 4.8 6.3 Quick Ratio 5.1 4.5 3.2 3.3 3.8 4.3 4.5 4.8 5.1 3.8 5.3 Total Debt/Equity 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Total Debt/Assets 3% 3% 3% 3% 3% 3% 3% 3% 3% 3% 2%
Cash & Book Value/Share Book Value/Share 10.0 10.4 10.8 11.3 11.8 12.3 12.9 13.6 10.0 12.0 14.3 Net Cash/Share 4.6 4.8 3.4 3.9 4.5 5.1 5.6 6.2 4.6 4.6 7.0 FCF/Share 0.5 0.6 0.3 0.6 0.8 0.7 0.6 0.7 1.9 2.3 2.7
Source: Company data, Nomura estimates
278
Nomura | U.S. Semiconductors Primer December 11, 2013
Broadcom Corp. (BRCM, Buy, TP $33)
Fig. 26: BRCM Income Statement Summary Year End: December 2013E 2014E 2012 2013E 2014E ($ in millions) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14
INCOME STATEMENT Total Revenue 2,005 2,090 2,146 2,025 1,995 2,100 2,265 2,148 8,007 8,266 8,509 QoQ -3.6% 4.2% 2.7% -5.6% -1.5% 5.3% 7.8% -5.2% YoY 9.7% 6.0% 0.9% -2.7% -0.5% 0.5% 5.6% 6.1% 8.4% 3.2% 2.9%
COGS 988 1,030 1,044 997 998 1,050 1,137 1,078 4,027 4,059 4,263
Total gross profit 1,017 1,060 1,102 1,028 998 1,050 1,128 1,070 3,980 4,207 4,246
R&D 615 619 609 647 662 652 637 630 2,318 2,490 2,581 SG&A 179 174 181 185 185 178 174 172 696 719 709 Amortization of purch intangibles 15 14 14 14 14 14 14 14 113 57 56 Impairment of intangible assets 10 501 ------90 511 - Restructuring - - 12 12 - - - - 7 24 - Settlement costs (75) 79 (75) - Other operating expense - - 25 ------25 - Total operating expenses (GAAP) 819 1,308 766 858 861 844 825 816 3,303 3,751 3,346
Operating income 198 (248) 336 170 137 206 303 254 677 456 900
Interest income/(expense) (8) (9) (7) (9) (9) (9) (9) (9) (30) (33) (36) Other income/(expense) 3 3 (4) 3 3 3 3 3 10 5 12
Pretax income 193 (254) 325 164 131 200 297 248 657 428 876
Provision for Taxes 2 (3) 9 4 4 4 4 4 (63) 12 16
Net income - proforma, incl-ESO 260 300 335 229 184 253 350 301 1,218 1,124 1,088 Net income - proforma, ex-ESO 400 436 460 355 322 384 475 425 1,761 1,651 1,606 Net income - GAAP 191 (251) 316 160 127 196 293 244 720 416 860
EPS - proforma, incl-ESO $0.42 $0.48 $0.55 $0.38 $0.31 $0.42 $0.58 $0.49 $2.02 $1.84 $1.79 EPS - proforma, ex ESO $0.65 $0.70 $0.76 $0.59 $0.54 $0.64 $0.78 $0.70 $2.92 $2.71 $2.65 EPS - GAAP $0.33 ($0.43) $0.55 $0.27 $0.22 $0.33 $0.49 $0.41 $1.25 $0.71 $1.45
Shares outstanding - proforma 614 622 605 600 600 605 608 611 602 610 606
Percent of Sales Total gross margin 50.7% 50.7% 51.4% 50.8% 50.0% 50.0% 49.8% 49.8% 49.7% 50.9% 49.9% Product gross margin (GAAP) 49.6% 49.7% 51.4% 50.8% 50.0% 50.0% 49.8% 49.8% 48.3% 50.4% 49.9% Product GM (proforma) 52.2% 52.1% 53.6% 53.1% 52.5% 52.3% 51.9% 52.0% 52.1% 52.8% 52.2% R&D 30.7% 29.6% 28.4% 31.9% 33.2% 31.0% 28.1% 29.3% 29.0% 30.1% 30.3% SG&A 8.9% 8.3% 8.4% 9.1% 9.3% 8.5% 7.7% 8.0% 8.7% 8.7% 8.3% Product operating margin (proforma) 18.7% 19.8% 22.4% 18.0% 16.6% 18.8% 21.4% 20.3% 20.2% 19.8% 19.3% EffectiveTax Rate 1.0% 1.2% 2.8% 2.4% 3.1% 2.0% 1.3% 1.6% -9.6% 2.8% 1.8% Net margin 9.5% -12.0% 14.7% 7.9% 6.3% 9.3% 12.9% 11.3% 9.0% 5.0% 10.1%
Source: Company data, Nomura estimates
279
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 27: BRCM Balance Sheet Summary Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Assets 11,400 11,208 11,345 11,160 10,980 10,893 10,916 10,807 11,208 11,160 10,807 Cash, Equiv, & ST Investments 2,469 2,481 2,404 2,355 2,251 2,159 2,135 2,168 2,374 2,355 2,168 Inventories 523 610 541 517 517 544 589 559 527 517 559 Total Current Assets 3,897 3,995 3,927 3,806 3,690 3,667 3,754 3,709 3,781 3,806 3,709
Total Non-Current Assets 7,503 7,213 7,418 7,354 7,290 7,226 7,162 7,098 7,427 7,354 7,098
Liabilities 3,368 3,312 3,428 3,401 3,400 3,430 3,479 3,445 3,369 3,401 3,445 Total Current Liabilities 1,675 1,662 1,801 1,774 1,773 1,803 1,852 1,818 1,682 1,774 1,818
LT Debt 1,394 1,394 1,394 1,394 1,394 1,394 1,394 1,394 1,393 1,394 1,394 Total Non-Current Liabilities 1,693 1,650 1,627 1,627 1,627 1,627 1,627 1,627 1,687 1,627 1,627
Shareholders' Equity 8,032 7,896 7,917 7,759 7,580 7,463 7,437 7,362 7,839 7,759 7,362
Total Liabilities & Equity 11,400 11,208 11,345 11,160 10,980 10,893 10,916 10,807 11,208 11,160 10,807
Source: Company data, Nomura estimates
Fig. 28: BRCM Cash Flow Summary Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Cash Flows from Operations 388 334 672 431 375 388 457 512 1,932 1,825 1,732 Depreciation and Amortization 39 39 44 44 44 44 44 44 134 166 176 41% -6% -5% Cash Flows from Investing (121) (418) (351) (64) (64) (64) (64) (64) (4,796) (954) (256) Capital Expenditures (41) (67) (64) (64) (64) (64) (64) (64) (244) (236) (256)
Cash Flows from Financing (142) (13) (442) (416) (416) (416) (416) (416) 336 (1,013) (1,664) Repurchase of Common Stock (107) (110) (378) (378) (378) (378) (378) (378) (32) (973) (1,512) Increase in Debt (Decrease) ------492 - - Payment of Dividends (63) (64) (63) (63) (63) (63) (63) (63) (224) (253) (252)
Free Cash Flow 347 267 608 367 311 324 393 448 1,688 1,589 1,476
Source: Company data, Nomura estimates
280
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 29: BRCM Segment Analysis
Year End: December 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Revenue Broadband Communications 537 568 567 533 533 565 582 564 2,157 2,205 2,244 Mobile & Wireless 996 969 1,015 934 915 961 1,086 1,010 3,812 3,914 3,972 Infrastructure and Networking 429 510 564 558 547 575 598 574 1,825 2,061 2,293 All Other 43 43 ------213 86 - Total 2,005 2,090 2,146 2,025 1,995 2,100 2,265 2,148 8,007 8,266 8,509
Percent of Revenue Broadband Communications 26.8% 27.2% 26.4% 26.3% 26.7% 26.9% 25.7% 26.3% 26.9% 26.7% 26.4% Mobile & Wireless 49.7% 46.4% 47.3% 46.1% 45.9% 45.7% 47.9% 47.0% 47.6% 47.3% 46.7% Infrastructure and Networking 21.4% 24.4% 26.3% 27.6% 27.4% 27.4% 26.4% 26.7% 22.8% 24.9% 26.9% All Other 2.1% 2.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 2.7% 1.0% 0.0% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
QoQ Growth Broadband Communications -4.5% 5.8% -0.2% -6.0% 0.0% 6.0% 3.0% -3.0% - - - Mobile & Wireless -1.6% -2.7% 4.7% -8.0% -2.0% 5.0% 13.0% -7.0% - - - Infrastructure and Networking -5.6% 18.9% 10.6% -1.0% -2.0% 5.0% 4.0% -4.0% - - - All Other -17.3% 0.0% -100.0% NA NA NA NA NA -- - Total -3.6% 4.2% 2.7% -5.6% -1.5% 5.3% 7.8% -5.2% -- -
YoY Growth Broadband Communications 8.5% 4.5% 1.8% -5.2% -0.7% -0.5% 2.6% 5.9% 5.8% 2.2% 1.8% Mobile & Wireless 13.8% 7.5% -0.8% -7.7% -8.1% -0.8% 7.0% 8.1% 9.9% 2.7% 1.5% Infrastructure and Networking 7.1% 7.9% 13.3% 22.9% 27.6% 12.7% 5.9% 2.7% 10.3% 13.0% 11.2% All Other -24.1% -20.4% -100.0% -100.0% -100.0% -100.0% NM NM -6.9% -59.6% -100.0% Total 9.7% 6.0% 0.9% -2.7% -0.5% 0.5% 5.6% 6.1% 8.4% 3.2% 2.9%
Source: Company data, Nomura estimates
Fig. 30: BRCM Valuation Ratios
Year End: December 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Profitability Ratios Return on Equity 19.9% 22.1% 23.2% 18.3% 17.0% 20.6% 25.6% 23.1% 9.2% 5.4% 11.7% Return on Assets 14.0% 15.6% 16.2% 12.7% 11.7% 14.1% 17.4% 15.7% 15.7% 14.8% 14.9%
Efficiency Ratios Days Sales Outstanding 34.2 33.2 36.2 36.2 36.2 36.2 36.2 36.2 32.5 36.2 36.2 Inventory Turns 7.6 6.8 7.7 7.7 7.7 7.7 7.7 7.7 7.8 7.7 7.7 Days of Inventory 48.3 54.0 47.3 47.3 47.3 47.3 47.3 47.3 46.9 47.3 47.3
Liquidity Ratios Current Ratio 2.3 2.4 2.2 2.1 2.1 2.0 2.0 2.0 2.2 2.1 2.0 Quick Ratio 1.9 2.0 1.8 1.8 1.7 1.7 1.6 1.7 1.9 1.8 1.7 Debt/Equity 0.4 0.4 0.4 0.4 0.4 0.5 0.5 0.5 0.4 0.4 0.5 Debt/Capital 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3
Cash & Book Value/Share Book Value/Share $13.7 $13.7 $13.7 $13.3 $13.0 $12.7 $12.5 $12.3 $13.5 $13.3 $12.3 Cash/Share $4.3 $4.8 $5.1 $5.0 $4.8 $4.6 $4.5 $4.5 $4.0 $5.0 $4.5 FCF/Share $0.6 $0.5 $1.1 $0.6 $0.5 $0.6 $0.7 $0.7 $2.9 $2.7 $2.5
Source: Company data, Nomura estimates
281
Nomura | U.S. Semiconductors Primer December 11, 2013
Cavium Inc. (CAVM, Reduce, TP $28)
Fig. 31: CAVM Income Statement Summary 2013E 2014E 2012 2013E 2014E ($ in millions) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14
INCOME STATEMENT Total Revenue 69.5 74.2 79.1 81.1 80.4 85.1 91.7 92.6 235 304 350 QoQ 4.8% 6.7% 6.6% 2.5% -0.9% 5.9% 7.8% 1.0% YoY 31.8% 34.2% 29.5% 22.2% 15.6% 14.7% 15.9% 14.2% -9.2% 29.1% 15.1%
COGS 26.2 30.9 28.5 29.0 28.9 31.1 33.9 35.2 102.6 114.6 129.1 - - - Gross profit 43.4 43.3 50.6 52.2 51.4 54.0 57.8 57.4 132.9 189.4 220.7
R&D (GAAP) 32.4 32.4 33.6 34.0 34.4 34.6 34.8 35.0 108.7 132.5 138.9 SG&A (GAAP) 15.2 16.1 14.8 14.9 15.1 15.3 15.4 15.6 71.8 61.2 61.5 Other ------27.7 - - Total operating expenses (GAAP) 47.7 48.6 48.5 49.0 49.6 50.0 50.3 50.7 208.2 193.6 200.5 Total operating expenses (non-GAAP) 34.3 35.0 35.2 36.3 39.3 39.7 39.7 40.1 131.2 140.7 158.9 15% 7% 13% Operating income (GAAP) (4.3) (5.3) 2.1 3.2 1.9 4.1 7.5 6.8 (75.3) (4.3) 20.2
Interest and Other (0.6) (0.8) (0.5) (0.5) (0.6) (0.6) (0.6) (0.6) (0.8) (2.4) (2.4)
Pretax income (GAAP) (4.9) (6.1) 1.6 2.7 1.3 3.5 6.9 6.2 (76.1) (6.7) 17.8 Pretax income (non-GAAP) 10.5 13.1 16.5 17.2 13.4 15.6 19.4 18.7 16.8 57.3 67.1
Provision for Taxes 0.4 0.7 0.7 0.7 0.9 1.1 1.4 1.3 36.3 2.5 4.7
Net income (GAAP) (3.2) (4.3) 4.3 2.0 0.3 2.4 5.6 4.9 (111.8) (1.2) 13.1 Net income - (non-GAAP) 10.2 12.6 16.0 17.5 13.7 15.7 19.2 18.6 22.8 56.3 67.2
EPS - proforma (ex ESO) $0.19 $0.23 $0.29 $0.31 $0.24 $0.28 $0.34 $0.32 $0.43 $1.02 $1.18 EPS - GAAP ($0.06) ($0.08) $0.08 $0.04 $0.01 $0.04 $0.10 $0.09 ($2.23) ($0.03) $0.24
Shares outstanding - GAAP 51.0 51.5 53.5 53.8 54.2 54.6 55.0 55.4 49.9 52.4 54.8 Shares outstanding - non-GAAP 54.7 54.8 55.7 56.0 56.4 56.8 57.2 57.6 53.2 55.3 57.0
Percent of Sales Gross Margin (non-GAAP) 65.4% 65.8% 66.0% 66.5% 66.3% 65.7% 65.1% 64.1% 62.6% 66.0% 65.3% R&D 46.6% 43.7% 42.5% 42.0% 42.8% 40.7% 38.0% 37.8% 46.2% 43.6% 39.7% SG&A 21.9% 21.8% 18.7% 18.4% 18.8% 18.0% 16.8% 16.9% 30.5% 20.1% 17.6% Operating Margin (GAAP) -6.2% -7.2% 2.7% 3.9% 2.3% 4.8% 8.2% 7.3% -32.0% -1.4% 5.8% Effective Tax Rate 4.2% 5.4% 4.0% 4.0% 7.0% 7.0% 7.0% 7.0% -47.7% NM NM Net Margin (non-GAAP) 14.7% 17.0% 20.3% 21.6% 17.0% 18.5% 21.0% 20.1% 9.7% 18.5% 19.2%
Source: Company data, Nomura estimates
282
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 32: CAVM Balance Sheet Summary Year End: Dec 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E
Assets 329 343 358 357 373 390 410 428 332 357 428 Cash, Equiv, and ST Investments 87 98 113 114 132 146 160 178 77 114 178 Inventories 4140424343465052474352 Total Current Assets 168 186 206 209 226 246 269 289 165 209 289
Liabilities 7688907374767776887376 Total Current Liabilities 4355674951535453554953
Long-Term Debt 59000000400 Total Non-Current Liabilities 33 33 23 23 23 23 23 23 33 23 23
Shareholders' Equity 253 255 267 285 298 314 333 352 243 285 352
Total Liabilities & Equity 329 343 358 357 373 390 410 428 332 357 428
Source: Company data, Nomura estimates
Fig. 33: CAVM Cash Flow Summary Year End: Dec 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E
Cash Flows from Operations 8 6 14 14 15 11 12 17 29 43 55 Depreciation & Amortization98555555322720
Cash Flows from Investing 0 (3) (3) (3) (3) (3) (3) (3) (18) (7) (10) Capital Expenditures (1) (3) (3) (3) (3) (3) (3) (3) (13) (8) (10)
Cash Flows from Financing10000000310 Increase (Decrease) in Debt(6)0000000(10)(6)0 Repurchase of Common Stock00000000000 Payment of Dividends 00000000000
Free Cash Flow 8 4 11 12 13 8 10 15 16 34 45
Source: Company data, Nomura estimates
283
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 34: CAVM Segment Analysis
Year End: Dec 2013E 2014E 2012 2013E 2014E ($ in millions) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Revenues Enterprise and Service Provider 58 63 68 71 70 75 81 82 163 261 308 Broadband and Consumer 11 11 11 10 10 10 11 10 45 43 42 Software and Services ------27 - - Total 70 74 79 81 80 85 92 93 235 304 350
Percent of Revenues Enterprise and Service Provider 84% 85% 86% 87% 88% 88% 88% 89% 70% 86% 88% Broadband and Consumer 16% 15% 14% 13% 12% 12% 12% 11% 19% 14% 12% Software and Services 0% 0% 0% 0% 0% 0% 0% 0% 12% 0% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Enterprise and Service Provider 13% 8% 8% 4% -1% 6% 8% 2% - - - Broadband and Consumer -27% -1% -2% -7% -2% 5% 6% -6% - - - Software and Services NM 0% 0% 0% 0% 0% 0% 0% - - - Total 5% 7% 7% 3% -1% 6% 8% 1% ---
YoY Growth Enterprise and Service Provider 75% 63% 53% 51% 21% 18% 18% 16% NM NM 18% Broadband and Consumer -7% 6% 11% -16% -11% -6% 2% 3% NM NM -3% Software and Services -100%NMNMNMNMNMNMNMNMNMNM Total 33% 34% 29% 22% 16% 15% 16% 14% -9% 29% 15%
Source: Company data, Nomura estimates
Fig. 35: CAVM Valuation Ratios
Year End: Dec 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Profitability Ratios Return on Equity 16.5% 19.8% 24.6% 25.4% 18.7% 20.5% 23.7% 21.7% 8.2% 21.3% 21.1% Return on Assets 12.4% 14.7% 17.9% 19.6% 14.7% 16.1% 18.7% 17.4% 6.9% 15.8% 15.7%
Efficiency Ratios Days Sales Outstanding 47.3 54.8 53.5 53.5 53.5 53.5 53.5 53.5 46.0 53.5 53.5 Inventory Turns 2.5 3.1 2.7 2.7 2.7 2.7 2.7 2.7 2.2 2.7 2.7 Days of Inventory 144.0 116.3 135.5 135.5 135.5 135.5 135.5 135.5 169.0 135.5 135.5
Liquidity Ratios Current Ratio 3.9 3.4 3.1 4.2 4.4 4.7 5.0 5.5 3.0 4.2 5.5 Quick Ratio 2.8 2.6 2.4 3.3 3.5 3.7 4.0 4.4 2.0 3.3 4.4 Debt/Capital 23.2% 25.6% 25.3% 20.3% 19.9% 19.5% 18.8% 17.8% 26.7% 20.3% 17.8% Debt/Assets 23.2% 25.6% 25.3% 20.3% 19.9% 19.5% 18.8% 17.8% 26.7% 20.3% 17.8%
Book & Cash Value Book Value/Share $4.6 $4.7 $4.8 $5.1 $5.3 $5.5 $5.8 $6.1 $4.6 $5.1 $6.2 Net Cash/Share $1.3 $1.4 $1.8 $1.8 $2.1 $2.4 $2.6 $2.9 $1.1 $1.8 $2.9 FCF/Share $0.1 $0.1 $0.2 $0.2 $0.2 $0.1 $0.2 $0.3 $0.3 $0.6 $0.8
Source: Company data, Nomura estimates
284
Nomura | U.S. Semiconductors Primer December 11, 2013
Cypress Semiconductor Corp. (CY, Buy, TP $12)
Fig. 36: CY Income Statement Summary 2013E 2014E ($ in millions) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E
INCOME STATEMENT Total Revenue 172.7 193.5 188.7 166.1 166.1 182.7 204.6 201.5 769.7 721.0 754.9 Q/Q -4% 12% -4% -12% 0% 10% 12% -2% Y/Y -7% -4% -7% -8% -4% -6% 8% 21% -23% -6% 5%
Total COGS 93.7 102.0 97.1 81.7 80.9 87.8 95.9 95.5 376.9 374.5 360.0
Gross Profit (GAAP) 79.0 91.4 91.7 84.4 85.2 94.9 108.7 106.0 392.8 346.5 394.9 Gross Profit (Proforma) 87.5 102.7 101.5 87.2 88.0 97.7 111.5 108.8 426.7 378.9 406.1
R&D 49.3 48.8 50.4 46.5 47.0 47.5 48.0 47.5 189.9 195.1 190.0 SG&A 45.4 48.1 45.5 42.0 42.5 43.0 43.5 43.0 212.0 181.0 172.0 Amort of acquired intangibles 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 4.0 8.0 7.9 One-time charges 11.4 0.7 3.7 0.0 0.0 0.0 0.0 0.0 5.9 15.8 0.0 Total Operating Exp 108.2 99.6 101.6 90.5 91.5 92.5 93.5 92.5 411.7 399.9 369.9 Total Operating Exp (proforma) 80.9 79.2 76.4 72.0 73.0 74.0 75.0 74.0 329.7 308.6 296.0
Operating income (29.2) (8.1) (10.0) (6.1) (6.3) 2.4 15.2 13.5 (18.9) (53.4) 24.9 Operating income (proforma) 6.6 23.5 25.1 15.2 15.0 23.7 36.5 34.8 97.0 70.4 110.1
Interest and Other Inc/(exp) 1.1 2.6 0.9 (1.5) (1.5) (1.5) (1.5) (1.5) (1.1) 3.0 (6.0)
Pretax Income (28.0) (5.6) (9.1) (7.6) (7.8) 0.9 13.7 12.0 (20.0) (50.4) 18.9 Provision for taxes (proforma) 0.3 0.6 1.9 1.2 0.4 0.7 1.1 1.0 0.9 4.0 3.1 Minority interest (0.6) (0.4) (0.4) (0.4) (0.4) (0.4) (0.4) (0.4) (1.6) (1.9) (1.7) GAAP Net income (28.8) 3.3 (8.8) (9.3) (8.6) (0.2) 12.2 10.6 (24.0) (43.6) 14.1
GAAP Net income to CY (28.2) 3.8 (8.4) (8.8) (8.2) 0.3 12.7 11.0 (22.4) (41.6) 15.8 Proforma Net income to CY (ex-ESO) 4.579 21.6 22.0 13.9 14.5 23.0 35.4 33.7 93.1 62.1 106.7
GAAP EPS ($0.19) $0.02 ($0.06) ($0.06) ($0.05) $0.00 $0.08 $0.07 ($0.14) ($0.28) $0.11 Proforma EPS (ex-ESO) $0.03 $0.14 $0.14 $0.08 $0.09 $0.14 $0.21 $0.20 $0.56 $0.39 $0.65
Proforma shares 158.3 159.8 159.8 165.0 165.0 165.0 165.0 165.0 165.3 160.7 165.0
Percent of Sales Gross Margin (proforma) 50.7% 53.1% 53.8% 52.5% 53.0% 53.5% 54.5% 54.0% 55.4% 52.6% 53.8% Operating Exp (proforma) 46.9% 41.0% 40.5% 43.3% 44.0% 40.5% 36.7% 36.7% 42.8% 42.8% 39.2% Operating Margin (proforma) 3.8% 12.1% 13.3% 9.2% 9.0% 13.0% 17.8% 17.3% 12.6% 9.8% 14.6% Pretax Margin -16.2% -2.9% -4.8% -4.6% -4.7% 0.5% 6.7% 6.0% -2.6% -7.0% 2.5% Tax Rate (proforma) 4.4% 2.1% 7.2% 9.0% 3.0% 3.0% 3.0% 3.0% 0.9% 5.4% 3.0% Net Margin (proforma) 2.7% 11.2% 11.7% 8.4% 8.8% 12.6% 17.3% 16.7% 12.1% 8.6% 14.1%
Source: Company data, Nomura estimates
285
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 37: CY Balance Sheet Summary Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Assets 836 810 796 782 776 783 803 816 832 782 816 Cash, Equiv, and ST Investments 102 102 101 120 115 105 104 120 117 120 120 Inventories 109 99 101 85 84 91 100 99 128 85 99
Liabilities 684 645 618 610 610 613 618 618 655 610 618 Total Current Liabilities 381 360 337 328 328 332 336 336 349 328 336 Long Term Debt 232 227 227 227 227 227 227 227 232 227 227 Total Non-Current Liabilities 303 285 282 282 282 282 282 282 306 282 282
Shareholders' Equity 152 165 178 172 166 169 185 199 177 172 199
Total Liabilities & Equity 836 810 796 782 776 783 803 816 832 782 816
Source: Company data, Nomura estimates
Fig. 38: CY Cash Flow Summary Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Cash Flows from Operations 8 30 8 41 22 17 25 42 135 88 106 Depreciation 13 12 12 10 10 10 10 10 51 47 40
Cash Flows from Investing (1) 4 (2) (6) (10) (10) (10) (10) (113) (5) (40) Capital Expenditures (9) (8) (11) (6) (10) (10) (10) (10) (33) (34) (40)
Cash Flows from Financing (13) (21) 0 (17) (17) (17) (17) (17) (58) (50) (66) Increase (Decrease) in Debt 0 (14) 9 0 0 0 0 0 232 (5) 0 Repurchase of Common Stock 0 0 0 0 0 0 0 0 (209) 0 0 Payment of Dividends (16) (16) (16) (17) (17) (17) (17) (17) (63) (65) (66)
Free Cash Flow (1) 23 (3) 35 12 7 15 32 102 54 66
Source: Company data, Nomura estimates
286
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 39: CY Segment Analysis
Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Revenues Programmable System Division (PSD) 66 81 78 63 66 75 91 86 345 287 318 Data Communication (DCD) 23 21 19 18 17 18 19 20 87 81 74 Memory Products (MPD) 82 88 89 83 78 83 88 86 331 342 336 Core Semi 170 191 186 163 161 177 198 192 763 710 727 Emerging Tech 2 3 3 3 5 6 7 10 7 11 27 Total 173 193 189 166 166 183 205 202 770 721 755
Percent of Revenues Programmable System Division (PSD) 38% 42% 41% 38% 40% 41% 44% 43% 45% 40% 42% Data Communication (DCD) 13% 11% 10% 11% 10% 10% 9% 10% 11% 11% 10% Memory Products (MPD) 48% 46% 47% 50% 47% 45% 43% 43% 43% 47% 45% Core Semi 99% 99% 98% 98% 97% 97% 97% 95% 99% 98% 96% Emerging Tech 1% 1% 2% 2% 3% 3% 3% 5% 1% 2% 4% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Programmable System Division (PSD) -20% 24% -4% -20% 5% 15% 20% -5% Data Communication (DCD) 13% -6% -11% -5% -5% 6% 6% 2% Memory Products (MPD) 6% 7% 1% -7% -5% 6% 6% -2% Core Semi -5% 12% -3% -12% -1% 10% 12% -3% Emerging Tech 75% 21% 9% 5% 61% 21% 13% 42% Total -4% 12% -2% -12% 0% 10% 12% -2%
YoY Growth Programmable System Division (PSD) -18% -13% -14% -23% 0% -7% 16% 38% -28% -17% 11% Data Communication (DCD) 4% -8% -11% -11% -25% -15% 1% 9% -23% -7% -9% Memory Products (MPD) 0% 6% 1% 7% -5% -6% -1% 5% -16% 3% -2% Core Semi -7% -4% -7% -9% -6% -7% 6% 18% -23% -7% 2% Emerging Tech 30% 49% 28% 141% 122% 121% 129% 211% 50% 54% 149% Total -7% -4% -7% -8% -4% -6% 8% 21% -23% -6% 5%
Source: Company data, Nomura estimates
Fig. 40: CY Valuation Ratios
Year End: December 2013E 2014E 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E 2012A 2013E 2014E Profitability Ratios Return on Equity -76% 8% -20% -22% -21% 0% 27% 21% -14% -25% 7% Return on Assets -14% 2% -4% -5% -4% 0% 6% 5% -3% -6% 2%
Efficiency Ratios Days Sales Outstanding 70.1 54.1 52.3 52.3 52.3 52.3 52.3 52.3 39.3 48.2 55.9 Inventory Turns 3.4 4.1 3.8 3.8 3.8 3.8 3.8 3.8 3.0 4.4 3.6 Days of Inventory 106.4 88.2 95.0 95.0 95.0 95.0 95.0 95.0 123.6 82.9 100.8
Liquidity Ratios Current Ratio 1.0 1.0 1.0 1.0 1.0 1.0 1.1 1.1 1.1 1.0 1.1 Total Debt/Equity 1.5 1.4 1.3 1.3 1.4 1.3 1.2 1.1 1.3 1.3 1.1 Total Debt/Assets 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3
Cash & Book Value/Share Book Value/Share $1.0 $1.1 $1.2 $1.1 $1.1 $1.1 $1.2 $1.3 $1.1 $1.1 $1.3 Cash/Share $0.7 $0.7 $0.7 $0.8 $0.8 $0.7 $0.7 $0.8 $0.8 $0.8 $0.8 FCF/Share $0.0 $0.1 $0.0 $0.2 $0.1 $0.0 $0.1 $0.2 $0.6 $0.3 $0.4
Source: Company data, Nomura estimates
287
Nomura | U.S. Semiconductors Primer December 11, 2013
Intel Corp. (INTC, Neutral, TP $24)
Fig. 41: INTC Income Statement Summary 2013E 2014E 2012 2013E 2014E ($ in millions) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14
INCOME STATEMENT Total Revenue 12,580 12,811 13,483 13,700 12,703 12,902 13,434 13,544 53,341 52,574 52,583 QoQ -6.7% 1.8% 5.2% 1.6% -7.3% 1.6% 4.1% 0.8% YoY -2.5% -5.1% 0.2% 1.7% 1.0% 0.7% -0.4% -1.1% -1.2% -1.4% 0.0%
COGS 5,514 5,341 5,069 5,343 5,272 5,161 5,239 5,417 20,190 21,267 21,089
Gross Income 7,066 7,470 8,414 8,357 7,431 7,741 8,195 8,126 33,151 31,307 31,493
R&D 2,527 2,516 2,742 2,750 2,760 2,775 2,780 2,790 10,148 10,535 11,105 SG&A 1,947 2,165 1,970 1,950 1,955 1,955 1,958 1,963 8,057 8,032 7,831 Amort. of Intangibles, IPR&D 73 70 74 70 70 70 70 70 308 287 280 Total operating expenses 4,547 4,751 4,910 4,870 4,785 4,800 4,808 4,823 18,513 19,078 19,216
Operating Income 2,519 2,719 3,504 3,487 2,646 2,941 3,387 3,303 14,638 12,229 12,277
Net interest expense/(income) 50 37 32 25 35 35 35 35 (94) 144 140 Loss/(Gain) on equity investments 26 (11) (452) (25) (23) (23) (23) (23) (141) (462) (92)
Pretax Income 2,443 2,693 3,924 3,487 2,634 2,929 3,375 3,291 14,873 12,547 12,229
Provision for income tax 398 693 974 872 659 732 844 823 3,868 2,937 3,057
Net income 2,045 2,000 2,950 2,615 1,976 2,197 2,531 2,468 11,005 9,610 9,172 Net income (non-GAAP) 2,202 2,139 3,043 2,708 2,069 2,290 2,624 2,561 11,580 10,092 9,544
EPS (GAAP) $0.40 $0.39 $0.58 $0.51 $0.39 $0.43 $0.50 $0.49 $2.13 $1.89 $1.80 EPS (non-GAAP) $0.43 $0.42 $0.60 $0.53 $0.41 $0.45 $0.52 $0.50 $2.24 $1.98 $1.88
Shares outstanding - basic 4,948 4,978 4,981 4,981 4,981 4,981 4,981 4,981 4,996 4,972 4,981 Shares outstanding - fully diluted 5,080 5,106 5,100 5,095 5,090 5,085 5,080 5,075 5,161 5,096 5,082
Percent of Sales Gross Margin 56.2% 58.3% 62.4% 61.0% 58.5% 60.0% 61.0% 60.0% 62.1% 59.5% 59.9% R&D 20.1% 19.6% 20.3% 20.1% 21.7% 21.5% 20.7% 20.6% 19.0% 20.0% 21.1% SG&A 15.5% 16.9% 14.6% 14.2% 15.4% 15.2% 14.6% 14.5% 15.1% 15.3% 14.9% Operating Margin 20.0% 21.2% 26.0% 25.5% 20.8% 22.8% 25.2% 24.4% 27.4% 23.3% 23.3% Pretax Margin 19.4% 21.0% 29.1% 25.5% 20.7% 22.7% 25.1% 24.3% 27.9% 23.9% 23.3% Tax Rate 16.3% 25.7% 24.8% 25.0% 25.0% 25.0% 25.0% 25.0% 26.0% 23.4% 25.0% Net Margin 16.3% 15.6% 21.9% 19.1% 15.6% 17.0% 18.8% 18.2% 20.6% 18.3% 17.4%
Source: Company data, Nomura estimates
288
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 42: INTC Balance Sheet Summary
Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Assets 83,083 85,661 90,551 91,895 92,384 93,075 94,214 95,354 84,351 91,895 95,354 Cash, Equiv, & ST Investments 17,073 17,350 19,146 19,341 19,411 19,388 19,502 19,644 18,162 19,341 19,644 Inventories 4,358 4,542 4,533 4,778 4,714 4,615 4,685 4,845 4,734 4,778 4,845 Total Current Assets 28,677 29,048 31,350 31,850 31,581 31,514 31,845 32,177 31,358 31,850 32,177
Total Non-Current Assets 54,406 56,613 59,201 60,045 60,803 61,561 62,369 63,177 52,993 60,045 63,177
Liabilities 31,889 31,821 35,099 35,261 35,219 35,153 35,200 35,305 33,148 35,261 35,305 Total Current Liabilities 11,798 11,389 13,875 14,037 13,995 13,929 13,976 14,081 12,898 14,037 14,081
Long Term Debt 13,143 13,150 13,157 13,157 13,157 13,157 13,157 13,157 13,136 13,157 13,157 Long Term Liabilities 20,091 20,432 21,224 21,224 21,224 21,224 21,224 21,224 20,250 21,224 21,224
Shareholders' Equity 51,194 53,840 55,452 56,634 57,166 57,922 59,014 60,049 51,203 56,634 60,049
Total Liabilities & Equity 83,083 85,661 90,551 91,895 92,384 93,075 94,214 95,354 84,351 91,895 95,354
Source: Company data, Nomura estimates
Fig. 43: INTC Cash Flow Summary Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Cash Flows from Operations 4,285 4,722 8,193 4,909 4,669 4,576 4,713 4,741 18,884 22,109 18,699 Depreciation 1,682 1,712 1,729 1,730 1,700 1,700 1,650 1,650 6,357 6,853 6,700
Cash Flows from Investing (5,317) (5,512) (3,497) (2,866) (2,750) (2,750) (2,750) (2,750) (14,060) (17,192) (11,000) Capital Expenditures (2,174) (2,723) (2,866) (2,866) (2,750) (2,750) (2,750) (2,750) (11,027) (10,629) (11,000)
Cash Flows from Financing (1,738) (1,132) (1,378) (1,849) (1,849) (1,849) (1,849) (1,849) (1,408) (6,097) (7,395) Increase (Decrease) in Debt (224) 175 7 - - - - - 6,064 (42) - Repurchase of Common Stock (559) (796) (536) (1,000) (1,000) (1,000) (1,000) (1,000) (5,110) (2,891) (4,000) Payments of Dividends (1,114) (1,123) (1,121) (1,121) (1,121) (1,121) (1,121) (1,121) (4,350) (4,479) (4,483)
Free Cash Flow 2,111 1,999 5,327 2,043 1,919 1,826 1,963 1,991 7,857 11,480 7,699
Source: Company data, Nomura estimates
289
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 44: INTC Segment Analysis
Year End: December 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Revenue PC Client Group 7,992 8,100 8,387 8,340 7,624 7,700 8,016 7,917 34,503 32,819 31,257 Data Center Group 2,585 2,743 2,912 3,330 3,127 3,219 3,291 3,359 10,511 11,571 12,997 Other IA groups 978 942 1,067 1,030 932 919 1,025 1,132 4,378 4,017 4,008 Software and Services Group 588 610 621 541 579 625 675 722 2,381 2,360 2,601 All Other 437 416 495 460 441 439 427 413 1,567 1,808 1,720 Total 12,580 12,811 13,483 13,700 12,703 12,902 13,434 13,544 53,341 52,574 52,583
Percent of Revenue PC Client Group 63.5% 63.2% 62.2% 60.9% 60.0% 59.7% 59.7% 58.5% 64.7% 62.4% 59.4% Data Center Group 20.6% 21.4% 21.6% 24.3% 24.6% 25.0% 24.5% 24.8% 19.7% 22.0% 24.7% Other IA groups 7.8% 7.4% 7.9% 7.5% 7.3% 7.1% 7.6% 8.4% 8.2% 7.6% 7.6% Software and Services Group 4.7% 4.8% 4.6% 3.9% 4.6% 4.8% 5.0% 5.3% 4.5% 4.5% 4.9% All Other 3.5% 3.2% 3.7% 3.4% 3.5% 3.4% 3.2% 3.0% 2.9% 3.4% 3.3% Total 100.0% 100.0% 100.0% 100.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0%
QoQ Growth PC Client Group -6.6% 1.3% 3.5% -0.6% -8.6% 1.0% 4.1% -1.2% - - - Data Center Group -6.9% 6.1% 6.2% 14.3% -6.1% 3.0% 2.2% 2.1% - - - Other IA groups -4.0% -3.7% 13.3% -3.5% -9.5% -1.4% 11.5% 10.5% --- Software and Serivces Group -8% 4% 2% -13% 7% 8% 8% 7% --- All Other -10.2% -4.8% 19.0% -7.2% -4.0% -0.5% -2.7% -3.3% --- Total -6.7% 1.8% 5.2% 1.6% -7.3% 1.6% 4.1% 0.8% - - -
YoY Growth PC Client Group -6.0% -7.5% -3.5% -2.6% -4.6% -4.9% -4.4% -5.1% -2.5% -4.9% -4.8% Data Center Group 7.5% 0.3% 12.2% 20.0% 20.9% 17.3% 13.0% 0.9% 3.8% 10.1% 12.3% Other IA groups -9.0% -15.0% -9.3% 1.1% -4.7% -2.4% -4.0% 9.9% -12.5% -8.3% -0.2% Software and Serivces Group 3.0% 4.1% 5.6% -15.0% -1.6% 2.5% 8.7% 33.5% 27.3% -0.9% 10.2% All Other 22.7% 30.6% 22.0% -5.5% 1.0% 5.5% -13.8% -10.2% -1.4% 15.4% -4.9% Total -2.5% -5.1% 0.2% 1.7% 1.0% 0.7% -0.4% -1.1% -1.2% -1.4% 0.0%
Source: Company data, Nomura estimates
Fig. 45: INTC Valuation Ratios
Year End: December 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14 Profitability Ratios Return on Equity 16.0% 14.9% 21.3% 18.5% 13.8% 15.2% 17.2% 16.4% 21.5% 17.0% 15.3% Return on Assets 9.8% 9.3% 13.0% 11.4% 8.6% 9.4% 10.7% 10.4% 13.0% 10.5% 9.6%
Efficiency Ratios Days Sales Outstanding 25.6 24.7 25.2 25.2 25.2 25.2 25.2 25.2 26.0 25.2 25.2 Inventory Turns 5.1 4.7 4.5 4.5 4.5 4.5 4.5 4.5 4.8 4.5 4.5 Days of Inventory 72.1 77.6 81.6 81.6 81.6 81.6 81.6 81.6 76.3 81.6 81.6
Liquidity Ratios Current Ratio 2.4 2.6 2.3 2.3 2.3 2.3 2.3 2.3 2.4 2.3 2.3 Quick Ratio 1.7 1.8 1.6 1.6 1.6 1.6 1.7 1.7 1.7 1.6 1.7 Debt/Equity 0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6 Debt/Capital 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4
Cash & Book Value/Share Book Value/Share $10.1 $10.5 $10.9 $11.1 $11.2 $11.4 $11.6 $11.8 $9.9 $11.1 $11.8 Cash/Share $1.7 $1.9 $2.5 $2.5 $2.5 $2.5 $2.5 $2.6 $1.8 $2.5 $2.6 FCF/Share $0.4 $0.4 $1.0 $0.4 $0.4 $0.4 $0.4 $0.4 $1.5 $2.3 $1.5
Source: Company data, Nomura estimates
2 90
Nomura | U.S. Semiconductors Primer December 11, 2013
Linear Technology Corp. (LLTC, Neutral, TP $32)
Fig. 46: LLTC Income Statement Summary Year End: June FY2014E FY2015E ($ in millions) Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E CY12 CY13E CY14E
INCOME STATEMENT Total Revenue $340.4 $333.5 $346.9 $360.8 $368.0 $368.0 $382.7 $398.0 $1,283 $1,316 $1,444 QoQ 4.0% -2.0% 4.0% 4.0% 2.0% 0.0% 4.0% 4.0% YoY 1.6% 9.3% 10.3% 10.2% 8.1% 10.3% 10.3% 10.3% -4.0% 2.6% 9.7%
COGS 84.0 82.7 84.9 87.2 88.2 88.2 91.8 95.4 321 327 349 Gross Profit 256.4 250.9 262.0 273.5 279.7 279.7 290.9 302.6 962 988 1,095
SG&A 38.7 39.2 40.7 42.2 43.2 43.2 44.7 46.2 150 155 169 R&D 61.5 62.5 64.5 66.5 68.0 68.0 70.0 72.0 233 243 267 Restructuring/ One-time charges 5.4 5.4 5.4 5.4 5.4 5.4 5.4 5.4 20 22 22 Total Operating Expenses 100.2 101.7 105.2 108.7 111.2 111.2 114.7 118.2 383 398 436 1.3% Operating Income 156.2 149.2 156.8 164.9 168.6 168.6 176.3 184.4 579 591 659
Net Interest and Other Income (5.9) (5.9) (5.9) (4.0) 1.0 1.0 1.0 1.0 (23) (23) (8) One-time Non-Ope benefits/(exp) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0
Pretax Income 150.2 143.3 150.9 160.9 169.6 169.6 177.3 185.4 556 567 651 Income Tax Expense 42.4 39.4 41.5 44.2 47.5 47.5 49.6 51.9 161 140 181
Proforma Net Income (ex ESO) 123.7 119.7 125.2 132.5 137.9 137.9 143.5 149.3 457 491 534 Profroma Net Income (incl ESO) 107.9 103.9 109.4 116.6 122.1 122.1 127.6 133.5 395 427 470 GAAP Net Income 107.9 102.7 108.2 115.9 122.9 122.9 128.4 134.3 396 423 470
Proforma EPS (ex ESO) $0.52 $0.50 $0.52 $0.55 $0.57 $0.57 $0.59 $0.61 $1.94 $2.05 $2.21 Proforma EPS (incl ESO) $0.45 $0.43 $0.46 $0.48 $0.51 $0.50 $0.53 $0.55 $1.67 $1.79 $1.95 GAAP EPS $0.45 $0.43 $0.45 $0.48 $0.51 $0.51 $0.53 $0.55 $1.68 $1.77 $1.95
Shares Outstanding 239.3 239.8 240.3 240.8 241.3 241.8 242.3 242.8 236 239 241
Stock Based Comp Expense 15.9 15.9 15.9 15.9 15.9 15.9 15.9 15.9 63 64 63
Percent of Sales Gross Margin 75.3% 75.2% 75.5% 75.8% 76.0% 76.0% 76.0% 76.0% 75.0% 75.1% 75.8% R&D 18.1% 18.7% 18.6% 18.4% 18.5% 18.5% 18.3% 18.1% 18.2% 18.5% 18.5% SG&A 11.4% 11.7% 11.7% 11.7% 11.7% 11.7% 11.7% 11.6% 11.7% 11.8% 11.7% Operating Margin 45.9% 44.7% 45.2% 45.7% 45.8% 45.8% 46.1% 46.3% 45.2% 44.9% 45.6% Tax Rate (GAAP) 25.5% 25.5% 25.6% 25.4% 25.1% 25.1% 25.3% 25.4% 26.1% 22.4% 25.3% Net Margin 31.7% 31.1% 31.5% 32.3% 33.2% 33.2% 33.3% 33.5% 30.8% 32.5% 32.6%
Source: Company data, Nomura estimates
291
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 47: LLTC Balance Sheet Summary
Year End: June FY2014E FY2015E (in $mn) Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E CY12 CY13E CY14E Asssets 2,190 2,224 2,266 1,473 1,523 1,572 1,627 1,688 1,923 2,224 1,572 Cash, Equivs, & ST Investments 1,772 1,820 1,872 1,090 1,151 1,213 1,278 1,348 1,299 1,659 1,036 Inventories 8786889092929599858692 Total Current Assets 1,894 1,941 1,996 1,216 1,278 1,341 1,409 1,482 1,599 1,941 1,341
Liabilities 1,137 1,141 1,146 304 305 305 305 306 1,131 1,141 305 Total Current Liabilities 1,045 1,043 1,043 211 212 212 212 213 163 1,043 212
Long Term Debt 0 5 10 0 0 0 0 0 816 5 0 Total Noncurrent Liabilities 93 98 103 93 93 93 93 93 968 98 93
Shareholders' Equity 1,052 1,083 1,119 1,169 1,218 1,267 1,322 1,383 793 1,083 1,267
Total Liability & Equity 2,190 2,224 2,266 1,473 1,523 1,572 1,627 1,688 1,923 2,224 1,572
Source: Company data, Nomura estimates
Fig. 48: LLTC Cash Flow Summary Year End: June FY2014E FY2015E (in $mn) Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E CY12 CY13E CY14E Cash Flows from Operations 124 160 135 141 149 153 149 154 546 575 578 Depreciation & Amortization 13 13 13 13 13 13 13 13 57 52 51
Cash from Investing (5) (9) (9) 791 (10) (10) (10) (10) (337) (230) 763 Capital Expenditures (4) (9) (9) (9) (10) (10) (10) (10) (17) (22) (37)
Cash from Financing (55) (81) (81) (921) (81) (81) (81) (81) (267) (191) (1,164) Increase (Decrease) in Debt 000 Repurchase of Common Stock (15) (40) (40) (40) (40) (40) (40) (40) (57) (125) (160) Payment of Dividends (62) (62) (62) (63) (63) (63) (63) (63) (297) (186) (251)
Free Cash Flow 120 151 127 132 139 143 139 144 529 552 541
Source: Company data, Nomura estimates
292
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 49: LLTC Segment Analysis
Year End: June FY2014E FY2015E Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E CY12 CY13E CY14E Revenue Communications 69 68 70 73 75 75 78 81 275 266 293 Computer 31 30 30 31 31 30 31 32 151 124 122 Consumer 99999999383636 Industrial 147 145 151 158 162 163 170 178 523 568 634 Automotive 65 63 66 69 70 70 73 76 211 242 274 Military 20 19 20 21 21 21 22 23 80 75 84 Total 341 334 347 361 369 369 383 399 1,283 1,316 1,444
Percent of Revenue Communications 20% 20% 20% 20% 20% 20% 20% 20% 21% 20% 20% Computer 9% 9% 9% 9% 8% 8% 8% 8% 12% 9% 8% Consumer 3% 3% 3% 3% 2% 2% 2% 2% 3% 3% 2% Industrial 43% 43% 44% 44% 44% 44% 44% 45% 41% 43% 44% Automotive 19% 19% 19% 19% 19% 19% 19% 19% 16% 18% 19% Military 6% 6% 6% 6% 6% 6% 6% 6% 6% 6% 6% Total 100%100%100%100%100%100%100%100%100%100%100%
QoQ Growth Communications 4%-2%4%4%2%0%4%4% Computer -6% -4% 2% 2% 0% -2% 2% 2% Consumer 2% -4% 2% 2% 0% -2% 2% 2% Industrial 4%-2%4%4%2%0%4%4% Automotive 10% -2% 4% 4% 2% 0% 4% 4% Military 4%-2%4%4%2%0%4%4% Total 4% -2% 4% 4% 2% 0% 4% 4%
YoY Growth Communications` -3% 9% 10% 10% 8% 10% 10% 10% -9% -3% 10% Computer -24% -12% -4% -6% 1% 3% 3% 3% 3% -18% -2% Consumer -10% -5% -6% 2% 0% 2% 2% 1% -23% -7% -1% Industrial 10% 13% 12% 12% 10% 12% 12% 12% -6% 9% 12% Automotive 14% 30% 20% 16% 8% 10% 10% 10% 10% 15% 13% Military -2% -9% 16% 10% 8% 10% 10% 10% -12% -6% 11% Total 2% 9% 10% 10% 8% 10% 10% 10% -4% 3% 10%
Source: Company data, Nomura estimates
Fig. 50: LLTC Valuation Ratios
Year End: June FY2014E FY2015E Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E CY12 CY13E CY14E Profitability Ratios Return on Equity 42.4% 38.9% 39.7% 40.8% 40.9% 39.3% 39.4% 39.5% 49.8% 39.5% 37.1% Return on Assets 20.1% 18.8% 19.5% 25.0% 32.6% 31.6% 31.9% 32.2% 20.5% 19.2% 29.9%
Efficiency Ratios Days Sales Outstanding 43.943.943.943.943.943.943.943.940.542.442.7 Inventory Turnover 3.9 3.8 3.9 3.9 3.9 3.9 3.9 3.9 3.9 3.8 3.9 Days of Inventory 94.6 94.6 94.6 94.6 94.6 94.6 94.6 94.6 93.2 95.3 92.8
Liquidity Ratios Current Ratio 1.8 1.9 1.9 5.8 6.0 6.3 6.6 7.0 9.8 1.9 6.3 Quick Ratio 1.7 1.7 1.8 5.2 5.4 5.7 6.0 6.3 8.9 1.7 5.7 Debt/Equity 0.4 0.4 0.4 0.0 0.0 0.0 0.0 0.0 0.5 0.4 0.0 Debt/Assets 0.4 0.4 0.4 0.0 0.0 0.0 0.0 0.0 0.4 0.4 0.0
Book Value and Cash Book Value/Share $4.4$4.5$4.7$4.9$5.0$5.2$5.5$5.7$3.4$4.5$5.3 Net Cash/Share $6.6 $6.9 $7.1 $3.8 $4.0 $4.3 $4.5 $4.8 $2.1 $6.9 $4.3 FCF/Share $0.5$0.6$0.5$0.5$0.6$0.6$0.6$0.6$2.2$2.3$2.2
Source: Company data, Nomura estimates
293
Nomura | U.S. Semiconductors Primer December 11, 2013
Marvell Technology (MRVL, Neutral, TP $12)
Fig. 51: MRVL Income Statement Summary Year End: January FY2014E FY2015E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E
INCOME STATEMENT Total Revenues $734 $807 $931 $900 $855 $905 $960 $931 $3,169 $3,373 $3,651 QoQ -5% 10% 15% -3% -5% 6% 6% -3% YoY -8% -1% 19% 16% 16% 12% 3% 3% -7% 6% 8%
Total COGS 335 386 465 453 430 446 468 454 1,493 1,639 1,798
Gross profit (non-GAAP) 401 428 469 450 428 462 494 479 1,692 1,747 1,863
R&D (incl ESO) 279 293 296 292 295 298 300 301 1,057 1,160 1,194 SG&A (incl ESO) 6666646062646766270256259 Amortization of Intangibles 11111188888534032 Extraordinary Items 00000000000 Total oper exp (non-GAAP, ex-ESO) 313 320 320 315 320 325 330 330 1,206 1,267 1,305
Operating inc (non-GAAP, ex-ESO) 88 108 155 135 108 137 164 149 485 486 558
Other Income (Expense) 3821111116144
Pretax income (non-GAAP) 91 116 157 136 109 138 165 150 501 500 562
Income Taxes Expense (7)(2)(6)555554(10)20
GAAP Net Income 53 62 103 84 56 85 113 98 307 302 352 Proforma Net Income (incl ESO) 65 77 120 92 64 93 121 106 371 353 384 Proforma Net Income (ex ESO) 98 118 163 131 104 133 160 145 498 511 542
GAAP EPS $0.11 $0.12 $0.21 $0.17 $0.11 $0.17 $0.23 $0.20 $0.54 $0.60 $0.70 Proforma EPS (Incl ESO) $0.12 $0.15 $0.23 $0.18 $0.12 $0.18 $0.24 $0.21 $0.64 $0.68 $0.75 Proforma EPS (ex ESO) $0.19 $0.23 $0.32 $0.25 $0.20 $0.26 $0.31 $0.28 $0.86 $0.99 $1.06
Shares outstanding 505 501 501 502 501 500 499 498 563 502 499 Shares outstanding (non-GAAP) 522 516 514 515 514 513 512 511 579 517 513
Percent of Sales Gross Margin (non-GAAP) 54.6% 53.0% 50.3% 50.0% 50.0% 51.0% 51.5% 51.5% 53.4% 51.8% 51.0% R&D, excl ESO 34.8% 32.7% 28.6% 29.3% 31.2% 29.8% 28.3% 29.3% 30.6% 31.1% 29.6% SG&A, excl ESO 7.9% 6.9% 5.7% 5.7% 6.2% 6.1% 6.0% 6.1% 8.5% 7.6% 7.1% Oper margin (non-GAAP, ex-ESO) 12.0% 13.4% 16.7% 15.0% 12.6% 15.1% 17.1% 16.0% 15.3% 14.4% 15.3% Effective Tax Rate -15.6% -2.6% -6.6% 5.6% 8.2% 5.6% 4.2% 4.9% 1.2% -3.5% 5.4% Net Margin 13.4% 14.6% 17.5% 14.6% 12.1% 14.6% 16.7% 15.6% 15.7% 15.1% 14.8%
Source: Company data, Nomura estimates
294
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 52: MRVL Balance Sheet Summary
Year End: January FY2014E FY2015E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Assets 5,121 5,204 5,348 5,411 5,450 5,549 5,682 5,770 5,262 5,411 5,770 Cash, Equivalents & ST Investments 1,733 1,726 1,804 1,889 1,981 2,054 2,152 2,278 1,919 1,889 2,278 Inventories 271 335 380 370 352 365 383 371 250 370 371 Total Current Assets 2,453 2,559 2,718 2,792 2,842 2,954 3,098 3,198 2,585 2,792 3,198
Total Non-Current Assets 2,668 2,646 2,631 2,619 2,607 2,596 2,584 2,572 2,676 2,619 2,572
Liabilities 769 812 885 862 843 857 875 863 777 862 863 Total Current Liabilities 613 674 757 735 716 729 747 736 608 735 736
Total Non-Current Liabilities 156 138 128 128 128 128 128 128 169 128 128
Stockholders' Equity 4,352 4,393 4,463 4,549 4,606 4,693 4,807 4,906 4,485 4,549 4,906
Total Liabilities and Equity 5,121 5,204 5,348 5,411 5,450 5,549 5,682 5,770 5,262 5,411 5,770
Source: Company data, Nomura estimates
Fig. 53: MRVL Cash Flow Summary Year End: January FY2014E FY2015E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Cash Flow from Operations 84 86 177 145 153 133 158 186 729 493 630 Depreciation and Amortization 25 26 26 26 26 26 26 26 91 102 104
Cash Flow from Investing 0 (23) 105 (14) (14) (14) (14) (14) 179 68 (57) Capital Expenditures (20) (19) (14) (14) (14) (14) (14) (14) (68) (68) (57)
Cash Flow from Financing (242) (66) (77) (46) (46) (46) (46) (46) (941) (431) (184) Increase (Decrease) in Debt 00000000000 Stock Buybacks (217) (88) (71) (50) (50) (50) (50) (50) (937) (426) (200) Payment of Dividends (30) (30) (30) (30) (30) (30) (30) (30) (99) (120) (120)
Free Cash Flow 58 65 157 131 138 119 144 172 626 411 573
Source: Company data, Nomura estimates
295
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 54: MRVL Segment Analysis
Year End: January FY2014E FY2015E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Revenue Storage 390 423 435 418 397 417 433 425 1,503 1,666 1,672 Communications 173 167 162 157 157 162 167 165 704 659 650 Mobile and Wireless 134 174 284 276 253 279 312 294 826 867 1,138 Other 38 44 50 50 48 48 48 48 135 182 190 Total 734 807 931 900 855 905 960 931 3,169 3,373 3,651
Percent of Revenue Storage 53% 52% 47% 46% 46% 46% 45% 46% 47% 49% 46% Communications 24% 21% 17% 17% 18% 18% 17% 18% 22% 20% 18% Mobile and Wireless 18% 22% 31% 31% 30% 31% 33% 32% 26% 26% 31% Other 5% 5% 5% 6% 6% 5% 5% 5% 4% 5% 5% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Storage 0%8%3%-4%-5%5%4%-2% Communications -2% -4% -3% -3% 0% 3% 3% -1% Mobile and Wireless -24% 30% 63% -3% -8% 10% 12% -6% Other 2% 15% 15% 0% -5% 0% 0% 0% Total -5% 10% 15% -3% -5% 6% 6% -3%
YoY Growth Storage 8% 10% 17% 7% 2% -1% 0% 2% -4% 11% 0% Communications 1% -7% -9% -11% -9% -3% 3% 5% 1% -6% -1% Mobile and Wireless -42% -21% 44% 57% 89% 60% 10% 7% -16% 5% 31% Other 15% 30% 44% 47% 26% 9% -5% -5% -3% 34% 5% Total -8% -1% 19% 16% 16% 12% 3% 3% -7% 6% 8%
Source: Company data, Nomura estimates
Fig. 55: MRVL Valuation Ratios
Year End: January FY2014E FY2015E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Profitability Ratios Return On Equity 6% 7% 11% 8% 6% 8% 10% 9% 8% 8% 8% Return On Assets 5% 6% 9% 7% 5% 7% 9% 7% 7% 7% 7%
Efficiency Ratios Days Sales Outstanding 45.9 48.6 45.6 47.1 47.1 47.1 47.1 47.1 38.0 50.4 48.2 Inventory Turns 5.0 4.6 4.9 4.9 4.9 4.9 4.9 4.9 6.0 4.4 4.8 Days of Inventory 73.4 79.0 74.4 74.4 74.4 74.4 74.4 74.4 61.2 82.5 75.4
Liquidity Ratios Current Ratio 4.0 3.8 3.6 3.8 4.0 4.1 4.1 4.3 4.3 3.8 4.3 Quick Ratio 3.63.33.13.33.53.63.63.83.83.33.8 Total Debt/Equity 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Total Debt/Capital 0%0%0%0%0%0%0%0%0%0%0%
Book & Cash Value Book Value/Share 8.33 8.52 8.68 8.83 8.96 9.15 9.39 9.60 7.75 8.80 9.57 Net Cash/Share 3.35 3.38 3.54 3.70 3.89 4.04 4.23 4.49 3.35 3.69 4.48 FCF/Share 0.11 0.13 0.30 0.25 0.27 0.23 0.28 0.34 1.08 0.79 1.12
Source: Company data, Nomura estimates
296
Nomura | U.S. Semiconductors Primer December 11, 2013
Maxim Integrated Products (MXIM, Neutral, TP $28)
Fig. 56: MXIM Income Statement Summary Year End: June FY2014E FY2015E CY12 CY13E CY14E ($ in Millions) Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E
INCOME STATEMENT Total Revenue $585 $620 $620 $648 $684 $684 $684 $715 $2,405 $2,419 $2,636 QoQ -3.8% 6.0% 0.0% 4.5% 5.5% 0.0% 0.0% 4.5% YoY -6.1% 2.5% 2.6% 6.6% 16.9% 10.2% 10.2% 10.2% -2.3% 0.6% 9.0%
COGS (proforma) 230 245 245 256 274 274 274 286 910 925 1,048 Gross Profit (proforma) 355 375 375 392 410 410 410 429 1,494 1,494 1,588
SG&A 77 87 88 89 92 90 88 90 318 329 360 R&D 130 146 144 142 144 146 144 144 539 542 576 Oper expenses (proforma) 207 233 232 231 236 236 232 234 857 871 937
Operating Income (proforma) 148 142 143 161 174 174 178 194 637 623 652
Net Interest and Other Income (3) (3) (3) (3) (3) (3) (3) (3) 23 (14) (14)
Pretax Income 144 138 139 157 170 170 174 191 660 609 638
Income Tax Expense (proforma) 25 25 25 28 31 31 31 34 131 110 115
Proforma Net Income 138 132 133 148 159 159 162 175 555 570 598 Proforma Net Income (incl ESO) 119 114 114 129 140 140 143 157 498 497 523 GAAP Net Income 103 102 103 118 129 129 132 145 370 456 479
GAAP Shares Outstanding 290 288 286 284 282 280 278 276 299 294 283
Proforma EPS (Incl ESO) $0.41 $0.39 $0.40 $0.45 $0.50 $0.50 $0.51 $0.57 $1.66 $1.69 $1.85 GAAP EPS $0.36 $0.36 $0.36 $0.41 $0.46 $0.46 $0.47 $0.53 $1.23 $1.55 $1.69
Percent of Sales Gross Margin (proforma) 60.7% 60.5% 60.5% 60.5% 60.0% 60.0% 60.0% 60.0% 62.1% 61.8% 60.2% SG&A 13.2% 14.1% 14.2% 13.8% 13.5% 13.2% 12.9% 12.6% 13.2% 13.6% 13.7% R&D 22.2% 23.6% 23.2% 21.9% 21.1% 21.4% 21.1% 20.2% 22.4% 22.4% 21.9% Operating Margin (proforma) 25.3% 22.9% 23.0% 24.8% 25.4% 25.4% 26.0% 27.2% 26.5% 25.7% 24.7% Effective Tax Rate 17.3% 18.0% 18.0% 18.0% 18.0% 18.0% 18.0% 18.0% 19.8% 18.1% 18.0% Net Margin 20.4% 18.3% 18.4% 19.9% 20.4% 20.4% 20.9% 21.9% 20.7% 20.6% 19.8%
Source: Company data, Nomura estimates
297
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 57: MXIM Balance Sheet Summary
Year End: June FY2014E FY2015E CY12 CY13E CY14E (in $mn) Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E Assets 3,764 3,749 3,752 3,775 3,811 3,841 3,875 3,928 3,739 3,749 3,841 Cash, Equiv, and ST Investmen 1,035 1,005 1,021 1,029 1,038 1,078 1,122 1,153 1,030 1,005 1,078 Inventories 278 296 296 309 329 329 329 344 258 296 329 Total Current Assets 1,782 1,781 1,797 1,832 1,879 1,919 1,963 2,024 1,724 1,781 1,919
Liabilities 1,366 1,374 1,374 1,379 1,387 1,387 1,387 1,392 1,157 1,374 1,387 Total Current Liabilities 331 339 339 344 352 352 352 357 369 339 352
Long-Term Debt 506 506 506 506 506 506 506 506 309 506 506 Total Non-Current Liabilities 529 529 529 529 529 529 529 529 480 529 529
Shareholders' Equity 2,398 2,375 2,378 2,396 2,425 2,455 2,488 2,536 2,581 2,375 2,455
Total Liabilities & Equity 3,764 3,749 3,752 3,775 3,811 3,841 3,875 3,928 3,739 3,749 3,841
Source: Company data, Nomura estimates
Fig. 58: MXIM Cash Flow Summary Year End: June FY2014E FY2015E CY12 CY13E CY14E (in $mn) Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E Cash Flows from Operations 96 152 173 166 168 199 202 191 778 674 705 Depreciation & Amortization 51 51 51 51 51 51 51 51 213 204 205
Cash Flows from Investing (33) (37) (37) (39) (41) (41) (41) (43) (189) (102) (158) Capital Expenditures (36) (37) (37) (39) (41) (41) (41) (43) (259) (177) (158)
Cash Flows from Financing (228) (120) (119) (119) (118) (118) (117) (117) (376) (523) (474) Increase (Decrease) in Debt0000000004940 Repurchase of Common Stock (154) (50) (50) (50) (50) (50) (50) (50) (201) (464) (200) Payment of Dividends (74) (75) (74) (74) (73) (73) (72) (72) (269) (289) (295)
Free Cash Flow 63 115 136 127 127 158 161 148 539 525 547
Source: Company data, Nomura estimates
298
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 59: MXIM Segment Analysis
Year End: June FY2014E FY2015E CY12 CY13E CY14E ($ in millions) Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E Revenue Consumer 250 263 273 295 304 298 304 325 1,062 1,081 1,170 Computing 72 104 99 99 111 111 94 94 350 321 419 Industrial 171 163 163 166 173 173 180 187 612 661 674 Communications 92 91 86 88 97 103 107 109 380 356 374 Total 585 620 620 648 684 684 684 715 2,405 2,419 2,636
Percent of Revenues Consumer 43% 42% 44% 46% 44% 44% 44% 45% 44% 45% 44% Computing 12% 17% 16% 15% 16% 16% 14% 13% 15% 13% 16% Industrial 29% 26% 26% 26% 25% 25% 26% 26% 25% 27% 26% Communications 16% 15% 14% 14% 14% 15% 16% 15% 16% 15% 14% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Consumer -7% 5% 4% 8% 3% -2% 2% 7% Computing -2% 45% -5% 0% 12% 0% -15% 0% Industrial -3% -5% 0% 2% 4% 0% 4% 4% Communications 3%-1%-6%3%9%6%4%2% Total -4% 6% 0% 4% 5% 0% 0% 4%
YoY Growth Consumer -14% -5% -9% 10% 21% 13% 11% 10% 15% 2% 8% Computing -19% 22% 38% 35% 54% 6% -5% -5% -17% -8% 31% Industrial 15% 8% 8% -6% 1% 6% 10% 12% -9% 8% 2% Communications -2% -1% 3% -1% 5% 13% 24% 23% -15% -6% 5% Total -6% 2% 3% 7% 17% 10% 10% 10% -2% 1% 9%
Source: Company data, Nomura estimates
Fig. 60: MXIM Valuation Ratios
Year End: June FY2014E FY2015E CY12 CY13E CY14E Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Jun-15E Profitability Ratios Return on Equity 19.5% 19.0% 19.3% 21.6% 23.2% 22.9% 23.2% 24.9% 19.6% 20.1% 21.7% Return on Assets 12.4% 12.1% 12.2% 13.7% 14.7% 14.6% 14.8% 16.1% 13.7% 13.3% 13.8%
Efficiency Ratios Days Sales Outstanding 45.5 45.5 45.5 45.5 45.5 45.5 45.5 45.5 40.2 46.7 47.2 Inventory Turns 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 3.7 3.2 3.3 Days of Inventory 106.6 106.6 106.6 106.6 106.6 106.6 106.6 106.6 99.2 112.9 111.2
Liquidity Ratios Current Ratio 5.4 5.2 5.3 5.3 5.3 5.5 5.6 5.7 4.7 5.2 5.5 Quick Ratio 4.5 4.4 4.4 4.4 4.4 4.5 4.6 4.7 4.0 4.4 4.5 Debt/Capital 17.4% 17.6% 17.5% 17.4% 17.3% 17.1% 16.9% 16.6% 10.7% 17.6% 17.1% Debt/Assets 13.4% 13.5% 13.5% 13.4% 13.3% 13.2% 13.1% 12.9% 0.0% 0.0% 0.0%
Book & Cash Value Book Value/Share $8.3 $8.2 $8.3 $8.4 $8.6 $8.8 $8.9 $9.2 $8.6 $8.1 $8.7 Net Cash/Share $1.8 $1.7 $1.8 $1.8 $1.9 $2.0 $2.2 $2.3 $2.4 $1.7 $2.0 FCF/Share $0.2 $0.4 $0.5 $0.4 $0.4 $0.6 $0.6 $0.5 $1.8 $1.8 $1.9
Source: Company data, Nomura estimates
299
Nomura | U.S. Semiconductors Primer December 11, 2013
Micron Technology (MU, Buy, TP $30)
Fig. 61: MU Income Statement Summary Year End: August Nov-13 Feb-14 May-14 Aug-14 Nov-14 Feb-15 May-15 Aug-15 ($ in millions) 1Q14E 2Q14E 3Q14E 4Q14E 1Q15E 2Q15E 3Q15E 4Q15E CY13E CY14E CY15E
INCOME STATEMENT Total Revenue 3,828 3,778 3,702 3,852 4,202 3,985 3,970 4,011 11,067 15,534 16,122 QoQ 35% -1% -2% 4% 9% -5% 0% 1% YoY 109% 82% 60% 36% 10% 5% 7% 4% 39% 40% 4%
Cost of Goods Sold 2,602 2,392 2,430 2,622 2,855 2,716 2,689 2,710 8,211 10,300 10,923
Gross Profit 1,226 1,385 1,272 1,230 1,347 1,269 1,281 1,301 2,856 5,234 5,199 Gross Profit (Pro Forma) 1,341 1,385 1,272 1,230 1,347 1,269 1,281 1,301 3,012 5,234 5,199
R&D 340 330 330 330 330 320 320 320 1,047 1,320 1,285 SG&A 190 180 180 185 185 175 175 180 633 730 715 Other 5060000000197600 Operating expenses 580 570 510 515 515 495 495 500 1,877 2,110 2,000
Oprating income 646 815 762 715 832 774 786 801 979 3,124 3,199
Other Non-Op Exp/(Inc) 00000000(13)00 One-time charges 00000000(1,312)00 Net Interest Exp/(Inc) 8080808080808080243320320 Pretax income 566 735 682 635 752 694 706 721 2,061 2,804 2,879
Tax Exp/(Inc) 757575757575757570300300 Minority int in TECH Singapore00000000000 Eq inc/(loss) from eq method inv40404040404040409160160 Other 00000000(4)00
GAAP Net income 531 700 647 600 717 659 671 686 1,996 2,664 2,739 GAAP Net Inc attrib to Micron 531 700 647 600 717 659 671 686 1,996 2,664 2,739
GAAP EPS $0.47 $0.63 $0.58 $0.53 $0.64 $0.58 $0.59 $0.61 $1.85 $2.38 $2.42 Adjusted EPS $0.71 $0.77 $0.67 $0.63 $0.73 $0.68 $0.69 $0.70 $1.15 $2.80 $2.80
Diluted Shares 1,126 1,118 1,120 1,122 1,124 1,126 1,128 1,130 1,080 1,121 1,129
Percent of Sales Gross Margin 32.0% 36.7% 34.4% 31.9% 32.1% 31.8% 32.3% 32.4% 25.8% 33.7% 32.2% Gross Margin (Pro Forma) 35.0% 36.7% 34.4% 31.9% 32.1% 31.8% 32.3% 32.4% 27.2% 33.7% 32.2% R&D 8.9%8.7%8.9%8.6%7.9%8.0%8.1%8.0%9.5%8.5%8.0% SG&A 5.0% 4.8% 4.9% 4.8% 4.4% 4.4% 4.4% 4.5% 5.7% 4.7% 4.4% Operating expenses 13.8% 13.5% 13.8% 13.4% 12.3% 12.4% 12.5% 12.5% 15.2% 13.2% 12.4% Operating Margin 16.9% 21.6% 20.6% 18.6% 19.8% 19.4% 19.8% 20.0% 8.8% 20.1% 19.8% Pretax Margin 14.8% 19.5% 18.4% 16.5% 17.9% 17.4% 17.8% 18.0% 18.6% 18.1% 17.9% Effective Tax Rate 13.3% 10.2% 11.0% 11.8% 10.0% 10.8% 10.6% 10.4% 3.4% 10.7% 10.4% Net Margin 20.9% 22.9% 20.3% 18.3% 19.6% 19.2% 19.5% 19.7% 11.2% 20.2% 19.6%
Source: Company data, Nomura estimates
3 00
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 62: MU Balance Sheet Summary
Year End: August FY14E FY15E ($ in millions) Nov-13E Feb-14E May-14E Aug-14E Nov-14E Feb-15E May-15E Aug-15E CY13E CY14E CY15E Assets 19,610 20,445 21,141 21,957 22,946 23,611 24,265 25,096 19,610 22,946 25,980 Cash, Equiv, and ST Investments 3,783 4,736 5,304 5,608 6,167 7,059 7,703 8,152 3,783 6,167 8,850 Inventories 2,566 2,360 2,397 2,586 2,816 2,679 2,652 2,673 2,566 2,816 2,770 Fixed Assets 7,766 7,896 8,016 8,126 8,226 8,366 8,496 8,616 7,766 8,226 8,726
Liabilities 9,043 9,147 9,167 9,353 9,594 9,571 9,524 9,639 9,043 9,594 9,770 Total Current Liabilities 4,006 4,060 4,030 4,166 4,357 4,284 4,187 4,252 4,006 4,357 4,333 Total Non-Current Liabilities 5,037 5,087 5,137 5,187 5,237 5,287 5,337 5,387 5,037 5,237 5,437 Long Term Debt 6,087 6,137 6,187 6,237 6,287 6,337 6,387 6,437 6,087 6,287 6,487
Shareholders' Equity 9,703 10,433 11,110 11,740 12,487 13,176 13,877 14,593 9,703 12,487 15,346
Total Liabilities & Equity 19,610 20,445 21,141 21,957 22,946 23,611 24,265 25,096 19,610 22,946 25,980
Source: Company data, Nomura estimates
Fig. 63: MU Cash Flow Summary Year End: August FY14E FY15E ($ in millions) Nov-13E Feb-14E May-14E Aug-14E Nov-14E Feb-15E May-15E Aug-15E CY13E CY14E CY15E Cash Flows from Operations 1,382 1,654 1,268 1,004 1,258 1,643 1,393 1,199 2,957 5,184 5,684 Depreciation 560 570 580 590 600 610 620 630 2,001 2,340 2,500
Cash Flows from Investing (700) (700) (700) (700) (700) (750) (750) (750) (1,773) (2,800) (3,000) Capital Expenditures (700) (700) (700) (700) (700) (750) (750) (750) (1,510) (2,800) (3,000)
Cash Flows from Financing 0 0 0 0 0 0 0 0 276 0 0 Increase (Decrease) in Debt0000000025700 Repurchase of Common Stock 0 0 0 0 0 0 0 0 0 0 0 Payment of Dividends 00000000000
Free Cash Flow 682 954 568 304 558 893 643 449 1,447 2,384 2,684
Source: Company data, Nomura estimates
301
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 64: MU Segment Analysis
Year End: August FY14E FY15E ($ in millions) Nov-13E Feb-14E May-14E Aug-14E Nov-14E Feb-15E May-15E Aug-15E CY13E CY14E CY15E Revenue DRAM 2,650 2,659 2,581 2,672 2,919 2,752 2,767 2,788 6,291 10,830 11,170 NAND 1,028 989 991 1,039 1,134 1,103 1,073 1,083 3,810 4,153 4,401 NOR 130 110 110 121 127 108 108 119 707 469 461 Other 20 20 20 20 22 22 22 22 258 82 90 Total Revenue 3,828 3,778 3,702 3,852 4,202 3,985 3,970 4,011 11,066 15,534 16,122
DRAM Revenue growth 61% 0% -3% 4% 9% -6% 1% 1% 102% 72% 3% Bit growth 50%0%7%15%15%0%4%5%100%84%29% ASP growth 7% 0% -9% -10% -5% -6% -4% -4% 1% -6% -20% Gross margin 33% 40% 37% 33% 33% 32% 33% 34% 26.2% 35.4% 33.3%
NAND Revenue growth 5%-4%0%5%9%-3%-3%1%8%9%6% Bit growth 13% 4% 8% 13% 18% 4% 4% 9% 17% 46% 42% ASP growth -7% -8% -7% -7% -7% -7% -7% -8% -7% -25% -25% Gross margin 33% 33% 32% 33% 33% 33% 33% 33% 32% 33% 33%
Source: Company data, Nomura estimates
Fig. 65: MU Valuation Ratios
Year End: August FY14E FY15E Nov-13E Feb-14E May-14E Aug-14E Nov-14E Feb-15E May-15E Aug-15E CY13E CY14E CY15E Profitability Ratios Return on Equity 33% 33% 27% 24% 26% 23% 22% 22% 13% 25% 21% Return on Assets 16% 17% 14% 13% 14% 13% 13% 13% 6% 14% 12%
Efficiency Ratios Days Sales Outstanding 50 50 51 54 52 50 48 54 50 52 52 Inventory Turns 4.1 4.1 4.1 4.1 4.1 4.1 4.1 4.1 4.1 4.1 4.1 Days of Inventory 90 90 90 90 90 90 90 90 90 90 90
Liquidity Ratios Current Ratio 2.3 2.5 2.6 2.7 2.8 3.0 3.2 3.3 2.3 2.8 3.4 Total Debt/Equity 0.5 0.4 0.4 0.4 0.4 0.4 0.3 0.3 0.5 0.4 0.3 Total Debt/Assets 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2
Cash & Book Value/Share Book Value/Share $8.6 $9.3 $9.9 $10.5 $11.1 $11.7 $12.3 $12.9 $8.6 $11.1 $13.6 Net Cash/Share -$1.6 -$0.8 -$0.3 -$0.1 $0.3 $1.1 $1.6 $2.0 -$1.7 $0.3 $2.5 FCF/Share $0.6$0.9$0.5$0.3$0.5$0.8$0.6$0.4$1.3$2.1$2.4
Source: Company data, Nomura estimates
302
Nomura | U.S. Semiconductors Primer December 11, 2013
Nvidia Corp. (NVDA, Buy, TP $17)
Fig. 66: NVDA Income Statement Summary Year End: January FY2014E FY2014E CY12 CY13E CY14E ($ in million) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E
INCOME STATEMENT Total Revenues 955 977 1,054 1,050 968 1,099 1,205 1,146 4,280 4,036 4,417 QoQ -13.7% 2.4% 7.9% -0.4% -7.8% 13.5% 9.6% -4.9% YoY 3.2% -6.4% -12.5% -5.2% 1.4% 12.5% 14.3% 9.1% 7.1% -5.7% 9.5%
COGS 436 432 470 481 440 495 548 521 2,054 1,818 2,004
Gross Profit (GAAP) 519 546 584 569 528 604 656 624 2,226 2,218 2,413
R&D 327 332 340 345 355 351 353 355 1,147 1,344 1,415 SG&A 109 108 103 105 109 105 103 104 411 425 422 Restructuring 00000000000 Other Total opex (GAAP) 436 440 443 450 464 456 456 459 1,578 1,770 1,837 Total opex (non-GAAP) 396 401 403 410 422 417 417 420 Operating income (GAAP) 83 106 141 119 63 148 200 165 648 448 576 Operating income (non-GAAP) 125 149 184 162 109 191 243 208
Interest & other expense / (income) -5 -6 0 0 0 0 0 0 -14 -13 -2
Pretax income 88 112 141 119 64 149 201 166 662 460 578 Provision for Taxes/(Benefit) 10 15 23 19 10 24 32 26 100 67 93
GAAP Net Income 78 96 119 100 53 125 168 139 563 393 486 Proforma Net Income (incl ESO) 85 106 125 106 60 131 175 145 613 421 510 Proforma Net Income (excl ESO) 114 133 154 136 91 159 203 174 728 536 627
GAAP EPS $0.13 $0.16 $0.20 $0.17 $0.09 $0.21 $0.29 $0.24 $0.90 $0.66 $0.82 Proforma EPS (incl. ESO) $0.14 $0.18 $0.21 $0.18 $0.10 $0.22 $0.30 $0.25 $0.98 $0.71 $0.87 Proforma EPS (excl ESO) $0.18 $0.23 $0.26 $0.23 $0.15 $0.27 $0.35 $0.30 $1.17 $0.90 $1.07
Basic Shares 617 585 581 581 581 581 581 581 619 591 581 Shares O/S Fully Diluted 619 592 589 589 589 589 589 589 624 597 589
Percent of Sales Gross Margin 54.3% 55.8% 55.4% 54.2% 54.5% 55.0% 54.5% 54.5% 52.0% 54.9% 54.6% Gross Margin (ex licensing) 50.9% 52.6% 52.5% 51.1% 51.2% 52.1% 51.9% 51.7% 48.9% 51.8% 51.7% Gross Margin (Proforma) 54.6% 56.3% 55.7% 54.5% 54.8% 55.3% 54.8% 54.8% 52.3% 55.3% 54.9% R&D 34.3% 33.9% 32.3% 32.9% 36.7% 32.0% 29.3% 31.0% 26.8% 33.3% 32.0% SG&A 11.4% 11.1% 9.8% 10.0% 11.3% 9.6% 8.6% 9.1% 9.6% 10.5% 9.5% Operating expenses 45.6% 45.0% 42.1% 42.9% 48.0% 41.5% 37.9% 40.1% 36.4% 43.8% 41.6% Operating margin 8.7% 10.8% 13.4% 11.3% 6.5% 13.5% 16.6% 14.4% 15.1% 11.1% 13.0% Pretax Margin 9.2% 11.4% 13.4% 11.3% 6.6% 13.5% 16.6% 14.4% 15.5% 11.4% 13.1% Effective Tax Rate 11.5% 13.7% 16.1% 16.1% 16.0% 16.0% 16.0% 16.0% 15.0% 14.6% 16.0% Net Income Margin (GAAP) 8.2% 9.9% 11.3% 9.5% 5.5% 11.4% 14.0% 12.1% 13.1% 9.7% 11.0%
Source: Company data, Nomura estimates
303
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 67: NVDA Balance Sheet Summary
Year End: January FY14E FY15E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Assets 6,280 5,570 5,711 5,776 5,781 5,957 6,258 6,309 6,412 5,776 6,309 Cash, Equiv, & ST Investments 3,713 2,936 3,033 3,168 3,144 3,256 3,538 3,686 3,728 3,168 3,686 Inventories 371 378 380 343 367 404 394 372 420 343 372 Total Current Assets 4,607 3,923 4,050 4,113 4,130 4,319 4,633 4,698 4,775 4,113 4,698
Total Non-Current Assets 1,673 1,647 1,662 1,664 1,651 1,638 1,625 1,612 2,916 3,285 3,301
Liabilities 1,455 1,397 1,391 1,365 1,324 1,385 1,528 1,450 1,585 1,365 1,450 Other Current Liabilities 0 0 0 0 0 0 0 0 0 0 0 Total Current Liabilities 926 936 986 960 919 980 1,123 1,046 976 960 1,046
Long Term Debt 0 0 0 0 0 0 0 0 0 0 0 Total Non-Current Liabilities 529 462 405 405 405 405 405 405 608 405 405
Shareholders' Equity 4,825 4,172 4,321 4,411 4,457 4,572 4,730 4,859 4,828 4,411 4,859
Total Liabilities and Equity 6,280 5,570 5,711 5,776 5,781 5,957 6,258 6,309 6,412 5,776 6,309
Source: Company data, Nomura estimates
Fig. 68: NVDA Cash Flow Summary Year End: January FY14E FY15E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Cash Flow From Operations 176 97 130 244 71 207 376 242 631 646 895 Depreciation and Amortization 60 62 63 63 63 63 63 63 227 248 252
Cash Flow from Investing (231) 712 (65) (65) (50) (50) (50) (50) (668) 351 (200) Capital Expenditures (66) (85) (65) (65) (50) (50) (50) (50) (201) (281) (200)
Cash Flow from Financing (116) (778) (44) (44) (44) (44) (44) (44) 87 (983) (177) Increase (Decrease) in Debt (1)(1)000000(2)(1)0 Repurchase of Common Stock (100) (750) 0 0 0 0 0 0 0 (850) 0 Payment of Dividends (46) (43) (44) (44) (44) (44) (44) (44) 0 (178) (177)
Free Cash Flow 110 12 65 179 21 157 326 192 430 365 695
Source: Company data, Nomura estimates
304
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 69: NVDA Segment Analysis
Year End: January FY14E FY15E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Revneue GPU 786 859 877 853 800 881 932 923 3,251 3,375 3,536 Tegra 103 53 111 130 102 152 207 156 764 397 617 All Others 66 66 66 66 66 66 66 66 264 264 264 Total 955 977 1,054 1,050 968 1,099 1,205 1,146 4,280 4,036 4,417
Percent of Revenue GPU 82% 88% 83% 81% 83% 80% 77% 81% 76% 84% 80% Tegra 11% 5% 11% 12% 11% 14% 17% 14% 18% 10% 14% All Others 7% 7% 6% 6% 7% 6% 5% 6% 6% 7% 6% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth GPU -6% 9% 2% -3% -6% 10% 6% -1% - - - Tegra -51% -49% 111% 17% -22% 50% 36% -24% - - - All Others 0% 0% 0% 0% 0% 0% 0% 0% - - - Total -14% 2% 8% 0% -8% 14% 10% -5% ---
YoY Growth GPU 8% 8% -2% 3% 2% 3% 6% 8% 2% 4% 5% Tegra -23% -71% -54% -37% -1% 190% 86% 20% 29% -48% 56% All Others 0% 0% 0% 0% 0% 0% 0% 0% 21% 0% 0% Total 3% -6% -12% -5% 1% 12% 14% 9% 7% -6% 9%
Source: Company data, Nomura estimates
Fig. 70: NVDA Valuation Ratios
Year End: January FY14E FY15E CY12 CY13E CY14E ($ in millions) Apr-13 Jul-13 Oct-13 Jan-14E Apr-14E Jul-14E Oct-14E Jan-15E FY13 FY14E FY15E Profitability Ratios Return On Equity 9% 13% 14% 12% 8% 14% 17% 14% 12% 9% 10% Return On Assets 7% 10% 11% 9% 6% 11% 13% 11% 11% 9% 10%
Efficiency Ratios Days Sales Outstanding 3339393641393936373339 Inventory Turns 4.7 4.6 4.9 5.3 5.4 5.4 5.3 5.3 5.0 4.7 4.6 Days of Inventory 7880746968676969737880
Liquidity Ratios Current Ratio 4.98 4.19 4.11 4.28 4.50 4.41 4.13 4.49 4.89 4.28 4.49 Quick Ratio 4.39 3.58 3.53 3.73 3.89 3.80 3.61 3.96 4.28 3.73 3.96 Total Debt/Equity 30% 33% 32% 31% 30% 30% 32% 30% 33% 31% 30% Debt/Capital 23% 25% 24% 24% 23% 23% 24% 23% 25% 24% 23%
Book & Cash Value Book Value/Share 7.79 7.05 7.34 7.49 7.57 7.77 8.03 8.25 7.73 7.39 8.25 Cash/Share 6.00 4.96 5.15 5.38 5.34 5.53 6.01 6.26 5.97 5.30 6.26 FCF/Share 0.18 0.02 0.11 0.30 0.03 0.27 0.55 0.33 0.69 0.61 1.18
Source: Company data, Nomura estimates
305
Nomura | U.S. Semiconductors Primer December 11, 2013
Qualcomm Inc. (QCOM, Buy, TP $85)
Fig. 71: QCOM Income Statement Summary Year End: September FY2013 FY2014E CY2012 CY2013 CY2014 ($ in millions) Dec-12 Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-12 Dec-13E Dec-14E
INCOME STATEMENT QCT Revenue 4,120 3,916 4,222 4,457 4,553 4,301 4,359 4,684 13,177 17,148 18,584 Q/Q 32% -5% 8% 6% 2% -6% 1% 7% Y/Y 34% 28% 47% 42% 11% 10% 3% 5% 34% 30% 8% QCT as % of sales 68% 64% 68% 69% 69% 64% 67% 67% 64% 67% 67% QTL Revenue 1,757 2,057 1,867 1,873 1,921 2,296 2,034 2,151 6,645 7,718 8,652 Q/Q 12% 17% -9% 0% 3% 19% -11% 6% Y/Y 22% 19% 17% 19% 9% 12% 9% 15% 14% 16% 12% QTL as % of sales 29% 34% 30% 29% 29% 34% 31% 31% 32% 30% 31% Other Segments 141 151 154 149 150 150 160 160 636 604 630 Total Revenue (pro-forma) 6,018 6,124 6,243 6,480 6,624 6,747 6,552 6,994 20,458 25,471 27,866 Q/Q 23.5% 1.8% 1.9% 3.8% 2.2% 1.9% -2.9% 6.7% Y/Y 28.6% 23.9% 35.0% 33.0% 10.1% 10.2% 5.0% 7.9% 25.6% 24.5% 9.4%
COGS (pro-forma) 2,154 2,279 2,415 2,638 2,653 2,488 2,503 2,674 7,441 9,985 10,646
Gross profit (pro-forma) 3,864 3,845 3,828 3,844 3,972 4,259 4,049 4,321 13,010 15,489 17,219
R&D (pro-forma) 949 1,057 1,130 1,182 1,186 1,200 1,110 1,156 3,566 4,555 4,752 SG&A (pro-forma) 468 556 505 547 530 530 530 550 1,937 2,138 2,140 Other (pro-forma) - - 158 173 - - - - (83) 331 0 Total operating expenses (pro-forma) 1,417 1,613 1,793 1,902 1,716 1,730 1,640 1,706 5,420 7,024 6,892
Operating income (EBIT) (pro-forma) 2,447 2,232 2,035 1,942 2,255 2,528 2,409 2,615 7,590 8,464 10,327
Other 0 000 000011500 000 Investment income (pro-forma) 248 221 175 233 180 180 180 180 1,095 809 720
Pretax income (pro-forma) 2,695 2,453 2,210 2,175 2,435 2,708 2,589 2,795 8,685 9,273 11,047
Income tax expense (pro-forma) 491 465 387 355 438 488 466 503 1,671 1,645 1,988 1,988 1,823 1,820 Net income - GAAP 1,906 1,866 1,580 1,500 1,753 1,972 1,871 2,032 6,614 6,699 8,047 Net income - Pro-forma 2,204 2,066 1,823 1,817 2,063 2,287 2,189 2,358 6,996 7,769 9,323 Net income - Pro-forma (incl ESO) 1,985 1,846 1,601 1,591 1,843 2,062 1,961 2,122 6,161 6,881 8,407
EPS - GAAP $1.09 $1.06 $0.90 $0.86 $1.02 $1.15 $1.09 $1.19 $3.78 $3.84 $4.71 EPS - Pro-forma $1.26 $1.17 $1.03 $1.05 $1.20 $1.33 $1.28 $1.38 $4.00 $4.45 $5.46 EPS - Pro-forma (incl ESO) $1.13 $1.05 $0.91 $0.92 $1.07 $1.20 $1.15 $1.24 $3.52 $3.94 $4.92
Shares outstanding - basic 1,709 1,722 1,727 1,703 1,685 1,680 1,675 1,670 1,709 1,709 1,673 Shares outstanding - fully diluted 1,751 1,763 1,765 1,738 1,720 1,715 1,710 1,705 1,749 1,747 1,708
Percent of Sales Gross margin (pro-forma) 64.2% 62.8% 61.3% 59.3% 60.0% 63.1% 61.8% 61.8% 63.6% 60.8% 61.8% Gross margin (pro-forma) - excl QTL 49.4% 44.0% 44.8% 42.7% 43.6% 44.1% 44.6% 44.8% 46.1% 43.8% 44.6% R&D (pro-forma) 15.8% 17.3% 18.1% 18.2% 17.9% 17.8% 16.9% 16.5% 17.4% 17.9% 17.1% SG&A (pro-forma) 7.8% 9.1% 8.1% 8.4% 8.0% 7.9% 8.1% 7.9% 9.6% 8.5% 7.7% Operating Margin (pro-forma) 40.7% 36.4% 32.6% 30.0% 34.0% 37.5% 36.8% 37.4% 36.6% 34.5% 37.1% Operating Margin (QCT) 25.9% 17.4% 17.5% 15.8% 18.0% 17.5% 19.7% 20.5% 19.9% 17.1% 20.0% Pretax income (pro-forma) 44.8% 40.1% 35.4% 33.6% 36.8% 40.1% 39.5% 40.0% 35.0% 31.5% 35.3% Tax rate (pro-forma) 18.2% 19.0% 17.5% 16.3% 18.0% 18.0% 18.0% 18.0% 19.2% 17.7% 18.0% Net margin (pro-forma) 36.6% 33.7% 29.2% 28.0% 31.1% 33.9% 33.4% 33.7% 34.2% 30.5% 33.5%
Source: Company data, Nomura estimates
306
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 72: QCOM Balance Sheet Summary
Year End: September FY2013E FY2014E CY2012 CY2013 CY2014 (in $mn) Dec-12 Mar-13 Jun13 Sep-13 Dec-13E Mar-14E Jun14E Sep-14E Dec-12 Dec-13E Dec-14E Assets 44,841 47,599 46,809 45,516 46,426 47,495 48,572 49,910 44,841 46,426 51,456 Cash, Equiv, & ST Investments 13,275 13,493 11,461 14,966 15,871 16,987 18,136 19,275 13,275 15,871 20,503 Inventories 1,277 1,484 1,727 1,302 1,280 1,202 1,210 1,292 1,277 1,280 1,439 Total Current Assets 17,103 17,766 15,972 19,555 20,486 21,565 22,657 24,025 17,103 20,486 25,590
Total Non-Current Assets 27,738 29,833 30,837 25,961 25,940 25,931 25,915 25,885 27,738 25,940 25,867
Liabilities 9,487 9,971 9,226 9,429 9,331 9,166 9,104 9,131 9,487 9,331 9,234 Other Current Liabilities 1,997 2,442 1,983 2,319 2,319 2,319 2,319 2,319 1,997 2,319 2,319 Total Current Liabilities 5,005 5,263 5,028 5,213 5,187 5,093 5,103 5,201 5,005 5,187 5,376
Long-Term Debt ------Total Non-Current Liabilities 4,482 4,708 4,198 4,216 4,144 4,073 4,001 3,929 4,482 4,144 3,858
Shareholders' Equity 35,354 37,628 37,578 36,087 37,095 38,329 39,468 40,780 35,354 37,095 42,222
Total Liabilities & Equity 44,841 47,599 46,809 45,516 46,426 47,495 48,572 49,910 44,841 46,426 51,456
Source: Company data, Nomura estimates
Fig. 73: QCOM Cash Flow Summary Year End: September FY2013E FY2014E CY2012 CY2013 CY2014 (in $mn) Dec-12 Mar-13 Jun13 Sep-13 Dec-13E Mar-14E Jun14E Sep-14E Dec-12 Dec-13E Dec-14E Cash Flows from Operations 1,975 2,216 2,077 2,522 2,180 2,402 2,426 2,401 6,194 8,996 9,729 Depreciation & Amortization 241 248 255 273 273 273 273 273 930 1,049 1,092
Cash Flows from Investing (1,227) (2,444) (2,316) 4,449 (252) (264) (257) (243) (6,036) (563) (1,019) Capital Expenditures (205) (289) (314) (200) (252) (264) (257) (243) (1,130) (1,055) (1,019)
Cash Flows from Financing (275) 14 (1,272) (3,365) (1,024) (1,022) (1,020) (1,019) (841) (5,647) (4,078) Increase (Decrease) in Debt - - (492) - - - - - (591) (492) - Repurchase of Common Stock (250) - (1,039) (3,321) (1,000) (1,000) (1,000) (1,000) (1,464) (5,360) (4,000) Payment of Dividends (428) (431) (604) (592) (585) (583) (581) (580) (1,649) (2,212) (2,322)
Free Cash Flow 1,770 1,927 1,763 2,322 1,928 2,139 2,169 2,158 5,064 7,941 8,710
Source: Company data, Nomura estimates
307
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 74: QCOM Segment Analysis
Year End: September FY2013E FY2014E CY2012 CY2013 CY2014 Dec-12 Mar-13 Jun13 Sep-13 Dec-13E Mar-14E Jun14E Sep-14E Dec-12 Dec-13E Dec-14E Revenues Qualcomm CDMA Technologies (QCT) 4,120 3,916 4,222 4,457 4,553 4,301 4,359 4,684 13,177 17,148 18,584 Qualcomm Technology Licensing (QTL) 1,757 2,057 1,867 1,873 1,921 2,296 2,034 2,151 6,645 7,718 8,652 Qualcomm Wireless & Internet (QWI) 141 151 154 149 150 150 160 160 636 604 630 Total 6,018 6,124 6,243 6,479 6,624 6,747 6,552 6,994 20,458 25,471 27,866
Percent of Revenues Qualcomm CDMA Technologies (QCT) 68.5% 63.9% 67.6% 68.8% 68.7% 63.8% 66.5% 67.0% 64.4% 67.3% 66.7% Qualcomm Technology Licensing (QTL) 29.2% 33.6% 29.9% 28.9% 29.0% 34.0% 31.0% 30.8% 32.5% 30.3% 31.0% Qualcomm Wireless & Internet (QWI) 2.4% 2.5% 2.5% 2.4% 2.3% 2.2% 2.4% 2.3% 3.1% 2.4% 2.3% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
QoQ Growth Qualcomm CDMA Technologies (QCT) 31.7% -4.9% 7.8% 5.6% 2.2% -5.5% 1.3% 7.5% - - - Qualcomm Technology Licensing (QTL) 11.8% 17.0% -9.2% 0.3% 2.6% 19.5% -11.4% 5.8% - - - Qualcomm Wireless & Internet (QWI) -9.3% 6.2% 1.9% -2.5% -2.6% 0.0% 6.7% 0.0% - - - Total 23.6% 1.8% 1.9% 3.8% 2.2% 1.9% -2.9% 6.7% - - -
YoY Growth Qualcomm CDMA Technologies (QCT) 33.5% 28.0% 47.1% 42.4% 10.5% 9.8% 3.2% 5.1% 34.1% 30.1% 8.4% Qualcomm Technology Licensing (QTL) 22.0% 19.4% 17.2% 19.2% 9.3% 11.6% 8.9% 14.8% 14.5% 16.2% 12.1% Qualcomm Wireless & Internet (QWI) -3.9% -2.5% -1.3% -4.3% 2.7% -3.2% 1.3% 3.9% -1.6% -1.4% 2.1% Total 28.6% 23.9% 34.9% 33.0% 10.1% 10.2% 5.0% 8.0% 25.6% 24.5% 9.4%
Source: Company data, Nomura estimates
Fig. 75: QCOM Valuation Ratios
Year End: September FY2013E FY2014E CY2012 CY2013 CY2014 Dec-12 Mar-13 Jun13 Sep-13 Dec-13E Mar-14E Jun14E Sep-14E Dec-12 Dec-13E Dec-14E Profitability Ratios Return on Equity 21.6% 19.8% 16.8% 16.6% 18.9% 20.6% 19.0% 19.9% 18.7% 18.1% 19.1% Return on Assets 17.0% 15.7% 13.5% 13.2% 15.1% 16.6% 15.4% 16.3% 14.7% 14.4% 15.6%
Efficiency Ratios Days Sales Outstanding 25.0 28.1 28.5 30.2 30.2 30.2 30.2 30.2 29.4 31.4 32.8 Inventory Turns 7.0 6.4 5.8 8.3 8.3 8.3 8.3 8.3 5.9 6.4 6.2 Days of Inventory 52.1 57.1 63.1 43.8 43.8 43.8 43.8 43.8 61.5 57.2 58.7
Efficiency Ratios Current Ratio 3.4 3.4 3.2 3.8 3.9 4.2 4.4 4.6 3.4 3.9 4.8 Quick Ratio 3.0 2.9 2.7 3.3 3.5 3.8 4.0 4.2 3.0 3.6 4.4 Debt/Equity 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Debt/Capital 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Cash & Book Value/Share Book Value/Share $20.2 $21.3 $21.3 $20.8 $21.6 $22.3 $23.1 $23.9 $20.2 $21.2 $24.7 Cash/Share $16.2 $17.3 $17.2 $16.9 $17.6 $18.3 $19.1 $19.8 $16.2 $17.4 $20.5 FCF/Share $1.0 $1.1 $1.0 $1.3 $1.1 $1.2 $1.3 $1.3 $2.9 $4.5 $5.1
Source: Company data, Nomura estimates
308
Nomura | U.S. Semiconductors Primer December 11, 2013
SanDisk Corp. (SNDK, Reduce, TP $56)
Fig. 76: SNDK Income Statement Summary 2013E 2014E ($ in millions) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E CY12A CY13E CY14E
INCOME STATEMENT Total Revenue 1,341 1,476 1,625 1,694 1,508 1,566 1,691 1,788 5,053 6,136 6,553 Q/Q -13% 10% 10% 4% -11% 4% 8% 6% Y/Y 11% 43% 28% 10% 12% 6% 4% 6% -11% 21% 7%
Total COGS 809 799 823 877 793 836 913 996 3,369 3,309 3,538
Gross Profit 532 677 802 817 715 730 778 793 1,683 2,827 3,016 Gross Profit (Non-GAAP) 543 689 815 830 728 743 791 806 1,733 2,877 3,067
R&D 171 172 184 195 190 195 205 215 603 722 805 SG&A 104 110 121 143 123 128 133 143 374 479 527 Amort of acquired intangibles22555555101420 Impairment charges 0083000000830 Total Operating Exp 278 284 394 343 318 328 343 363 987 1,299 1,352 Total Operating Exp (non-GAAP) 255 260 282 315 290 300 315 335 906 1,112 1,239
Operating Income 254 393 408 473 397 402 435 429 696 1,528 1,663 Operating Income (non-GAAP) 288 429 533 515 438 443 476 471 827 1,764 1,828
Interest and Other Inc/(exp) (20) (9) (5) (5) (5) (5) (5) (5) (69) (39) (20)
Pretax Income 234 383 404 469 392 397 430 424 627 1,490 1,643 Pretax Income (non-GAAP) 291 436 538 520 443 448 481 476 848 1,786 1,848
Provision for taxes 68 122 127 150 125 127 138 136 210 466 526 Provision for taxes (non-GAAP) 85 137 167 165 140 142 153 151 265 554 586
GAAP Net Income 166 262 277 319 266 270 292 289 417 1,024 1,117 Non-GAAP Net Income 207 299 371 355 303 306 329 325 582 1,231 1,262
GAAP EPS $0.68 $1.06 $1.18 $1.37 $1.16 $1.18 $1.29 $1.29 $1.70 $4.23 $4.90 Non-GAAP EPS (ex-ESO) $0.84 $1.21 $1.59 $1.54 $1.33 $1.35 $1.47 $1.46 $2.38 $5.16 $5.60
Diluted Shares (Non-GAAP) 246 246 233 230 228 226 224 222 245 239 225
Percent of Sales Gross Margin (Non-GAAP) 40.5% 46.7% 50.1% 49.0% 48.3% 47.4% 46.8% 45.0% 34.3% 46.9% 46.8% SG&A 7.8% 7.5% 7.5% 8.4% 8.2% 8.2% 7.9% 8.0% 7.4% 7.8% 8.0% R&D 12.8% 11.7% 11.3% 11.5% 12.6% 12.4% 12.1% 12.0% 11.9% 11.8% 12.3% Expense Ratio (non-GAAP) 19.0% 17.6% 17.4% 18.6% 19.2% 19.1% 18.6% 18.7% 17.9% 18.1% 18.9% Operating Margin (non-GAAP) 21.5% 29.0% 32.8% 30.4% 29.1% 28.3% 28.2% 26.3% 16.4% 28.8% 27.9% Pretax Margin (GAAP) 17.4% 26.0% 24.8% 27.7% 26.0% 25.3% 25.4% 23.7% 12.4% 24.3% 25.1% Tax Rate (non-GAAP) 29.0% 31.5% 31.1% 31.7% 31.7% 31.7% 31.7% 31.7% 31.3% 31.0% 31.7% Net Margin (non-GAAP) 15.4% 20.3% 22.8% 20.9% 20.1% 19.5% 19.4% 18.2% 11.5% 20.1% 19.3%
Source: Company data, Nomura estimates
309
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 77: SNDK Balance Sheet Summary
Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E CY12A CY13E CY14E Assets 10,376 9,619 8,945 9,008 8,980 8,994 9,042 9,082 10,339 9,008 9,082 Cash, Equiv, and ST Investments 6,193 5,355 4,289 4,371 4,415 4,218 4,091 4,088 5,711 4,371 4,088 Inventories 734 723 776 746 742 836 904 916 750 746 916
Liabilities 3,057 2,274 2,342 2,363 2,344 2,363 2,393 2,418 3,080 2,363 2,418 Total Current Liabilities 1,873 1,108 1,193 1,203 1,175 1,184 1,204 1,219 1,882 1,203 1,219 Long Term Debt 800 810 820 830 839 849 859 869 790 830 869 Total Non-Current Liabilities 1,184 1,166 1,149 1,159 1,169 1,179 1,189 1,199 1,198 1,159 1,199
Shareholders' Equity 7,320 7,345 6,602 6,645 6,636 6,631 6,649 6,664 7,260 6,645 6,664
Total Liabilities & Equity 10,376 9,619 8,945 9,008 8,980 8,994 9,042 9,082 10,339 9,008 9,082
Source: Company data, Nomura estimates
Fig. 78: SNDK Cash Flow Summary Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E CY12A CY13E CY14E Cash Flows from Operations 474 391 382 547 495 254 324 447 530 1,794 1,520 Depreciation 118 111 108 108 108 108 108 108 335 446 433
Cash Flows from Investing (339) 646 538 (163) (150) (150) (150) (150) (574) 682 (600) Capital Expenditures (48) (71) (51) (125) (150) (150) (150) (150) (488) (296) (600)
Cash Flows from Financing 12 (1,131) (1,073) (302) (301) (301) (300) (300) (125) (2,494) (1,203) Increase (Decrease) in Debt 0 (928) 0 0 0 0 0 0 0 (928) 0 Repurchase of Common Stock (90) (280) (1,070) (250) (250) (250) (250) (250) (250) (1,690) (1,000) Payment of Dividends 0 0 0 0 0 0 0 0 0 0 0
Free Cash Flow 425 319 332 422 345 104 174 297 42 1,498 920
Source: Company data, Nomura estimates
3 10
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 79: SNDK Segment Analysis
Year End: December 2013E 2014E (in $mn) 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E CY12A CY13E CY14E Revenue Product revenue 1,239 1,381 1,530 1,599 1,413 1,476 1,596 1,688 4,679 5,749 6,173 License & Royalties 102 95 95 95 95 90 95 100 374 387 380 Total Revenue 1,341 1,476 1,625 1,694 1,508 1,566 1,691 1,788 5,053 6,136 6,553
Sequential Growth Product revenue -14.3% 11.5% 10.8% 4.5% -11.7% 4.5% 8.1% 5.8% -11.5% 22.9% 7.4% GB sold -16.0% 5.0% 14.7% 10.0% -7.0% 10.0% 15.0% 15.0% 61.7% 22.6% 29.7% ASP per GB 2.1% 6.2% -3.4% -5.0% -5.0% -5.0% -6.0% -8.0% -45.3% 0.3% -17.2%
Product GM, Proforma 35.6% 43.0% 47.1% 45.9% 44.8% 44.2% 43.6% 41.8% 29.1% 43.3% 43.5%
Source: Company data, Nomura estimates
Fig. 80: SNDK Valuation Ratios
Year End: December 2013E 2014E 1Q13A 2Q13A 3Q13A 4Q13E 1Q14E 2Q14E 3Q14E 4Q14E CY12A CY13E CY14E Profitability Ratios Return on Equity 9% 14% 17% 19% 16% 16% 18% 17% 6% 15% 17% Return on Assets 6% 11% 12% 14% 12% 12% 13% 13% 4% 11% 12%
Efficiency Ratios Days Sales Outstanding 29.9 39.3 38.5 34.1 31.8 35.0 35.8 33.3 42.5 38.2 36.9 Inventory Turns 4.4 4.4 4.2 4.8 4.3 4.1 4.1 4.4 4.5 4.4 3.9 Days of Inventory 82.8 82.6 86.0 76.5 84.3 90.0 89.1 82.8 81.3 82.3 94.5
Liquidity Ratios Current Ratio 2.6 3.9 3.3 3.3 3.3 3.3 3.2 3.2 2.4 3.3 3.2 Total Debt/Equity 0.4 0.3 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 Total Debt/Assets 0.3 0.2 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3
Cash & Book Value/Share Book Value/Share $29.8 $29.9 $28.1 $28.6 $28.8 $29.1 $29.4 $29.7 $29.6 $27.4 $29.2 Cash/Share $25.2 $21.8 $18.2 $18.8 $19.2 $18.5 $18.1 $18.2 $23.3 $18.1 $17.9 FCF/Share $1.7 $1.3 $1.4 $1.8 $1.5 $0.5 $0.8 $1.3 $0.2 $6.2 $4.0
Source: Company data, Nomura estimates
311
Nomura | U.S. Semiconductors Primer December 11, 2013
Texas Instruments Inc. (TXN, Reduce, TP $33)
Fig. 81: TXN Income Statement Summary Year End: December 2013E 2014E 2012 2013E 2014E ($ in millions) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Dec-12 Dec-13 Dec-14
INCOME STATEMENT Total Revenue 2,885 3,047 3,244 2,980 2,955 3,207 3,400 3,197 12,700 12,156 12,759 QoQ -3.2% 5.6% 6.5% -8.1% -0.8% 8.5% 6.0% -6.0% YoY -5.6% -8.6% -2.6% 0.0% 2.4% 5.2% 4.8% 7.3% -7.5% -4.3% 5.0%
COGS 1,511 1,477 1,465 1,386 1,374 1,459 1,530 1,455 6,458 5,839 5,818 Gross profit 1,374 1,570 1,779 1,594 1,581 1,748 1,870 1,742 6,242 6,317 6,941
R&D 419 389 368 357 367 387 402 390 1,877 1,533 1,546 SG&A 459 471 465 442 452 467 477 462 1,801 1,837 1,858 Other 101 (196) 102 86 86 86 86 86 715 93 344 Total operating expenses 979 664 935 885 905 940 965 938 4,393 3,463 3,747
Operating income 395 906 844 710 676 808 906 804 1,849 2,855 3,194 Operating income (proforma) 496 710 946 796 762 894 992 890 2,564 2,948 3,538
One-time Charges / (gains) (2) 0 0 0 0 0 0 0 (164) (2) 0 Other expenses / (income) 0 0 4 0 0 0 0 0 (8) 4 0 Interest on loans exp / (income) 23 24 24 24 24 24 24 24 85 95 96
Pretax income 374 882 816 686 652 784 882 780 1,936 2,758 3,098
Provision for Taxes (benefit) 12 222 187 165 170 204 229 203 177 586 805
Net inc, before discontinued ops 362 660 629 521 483 580 652 577 1,759 2,172 2,292 G/L from discountinued ops 0 0 0 0 0 0 0 0 0 0 0
GAAP net income 362 660 629 521 483 580 652 577 1,759 2,172 2,292 Proforma net income 363 533 686 577 539 636 708 633 1,852 2,159 2,516
GAAP EPS $0.32 $0.58 $0.56 $0.46 $0.43 $0.53 $0.60 $0.53 $1.51 $1.92 $2.09 Proforma EPS $0.32 $0.47 $0.61 $0.51 $0.48 $0.58 $0.65 $0.59 $1.59 $1.90 $2.30
Shares outstanding - basic 1,107 1,103 1,096 1,086 1,076 1,066 1,056 1,046 1,132 1,098 1,061 Shares outstanding - fully diluted 1,123 1,117 1,111 1,101 1,091 1,081 1,071 1,061 1,146 1,113 1,076
Percent of Sales Gross Margin 47.6% 51.5% 54.8% 53.5% 53.5% 54.5% 55.0% 54.5% 49.1% 52.0% 54.4% R&D 14.5% 12.8% 11.3% 12.0% 12.4% 12.1% 11.8% 12.2% 14.8% 12.6% 12.1% SG&A 15.9% 15.5% 14.3% 14.8% 15.3% 14.6% 14.0% 14.5% 14.2% 15.1% 14.6% Operating Margin 13.7% 29.7% 26.0% 23.8% 22.9% 25.2% 26.6% 25.2% 14.6% 23.5% 25.0% Operating Margin (proforma) 17.2% 23.3% 29.2% 26.7% 25.8% 27.9% 29.2% 27.8% 20.2% 24.2% 27.7% Pretax income margin 13.0% 28.9% 25.2% 23.0% 22.1% 24.4% 25.9% 24.4% 15.2% 22.7% 24.3% Tax Rate 3.2% 25.2% 22.9% 24.0% 26.0% 26.0% 26.0% 26.0% 9.1% 21.2% 26.0% Net Margin 12.5% 21.7% 19.4% 17.5% 16.3% 18.1% 19.2% 18.1% 13.9% 17.9% 18.0% Net Margin (proforma) 12.6% 17.5% 21.2% 19.4% 18.2% 19.8% 20.8% 19.8% 14.6% 17.8% 19.7%
Source: Company data, Nomura estimates
312
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 82: TXN Balance Sheet Summary
Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Assets 19,701 19,398 19,244 19,022 18,811 18,765 18,785 18,633 20,064 19,022 18,633 Cash, Equiv, and ST Investments 3,862 3,244 3,593 3,741 3,706 3,590 3,584 3,768 3,965 3,741 3,768 Inventories 1,700 1,720 1,726 1,656 1,642 1,743 1,828 1,738 1,757 1,656 1,738 Total Current Assets 8,205 8,038 8,101 8,055 7,994 8,098 8,268 8,266 8,273 8,055 8,266
Total Non-Current Assets 11,496 11,360 11,143 10,967 10,817 10,667 10,517 10,367 11,791 10,967 10,367
Liabilities 8,749 8,330 8,193 8,139 8,131 8,189 8,237 8,186 9,103 8,139 8,186 Other Current Liabilities 805 902 730 730 730 730 730 730 1,005 730 730 Total Current Liabilities 3,110 2,802 2,723 2,669 2,661 2,719 2,767 2,716 3,473 2,669 2,716
Long Term Debt 4,183 4,165 4,161 4,161 4,161 4,161 4,161 4,161 4,186 4,161 4,161 Total Non-Current Liabilities 5,639 5,528 5,470 5,470 5,470 5,470 5,470 5,470 5,630 5,470 5,470
Shareholders' Equity 10,952 11,068 11,051 10,883 10,680 10,576 10,548 10,447 10,961 10,883 10,447
Total Liabilities & Equity 19,701 19,398 19,244 19,022 18,811 18,765 18,785 18,633 20,064 19,022 18,633
Source: Company data, Nomura estimates
Fig. 83: TXN Cash Flow Summary Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Cash Flows from Operations 360 675 1,151 961 801 718 824 1,012 3,410 3,147 3,355 Depreciation 228 221 217 217 217 217 217 217 957 883 868
Cash Flows from Investing 22 309 (212) (124) (150) (150) (150) (150) (1,038) (5) (600) Capital Expenditures (84) (98) (124) (124) (150) (150) (150) (150) (494) (430) (600)
Cash Flows from Financing (405) (1,197) (684) (689) (686) (683) (681) (678) (1,948) (2,975) (2,728) Increase (Decrease) in Debt 0 (514) 0 0 0 0 0 0 117 (514) 0 Repurchase of Common Stock (679) (721) (734) (734) (734) (734) (734) (734) (1,800) (2,868) (2,936) Payment of Dividends (232) (309) (308) (304) (301) (298) (296) (293) (819) (1,153) (1,188)
Free Cash Flow 276 577 1,027 837 651 568 674 862 2,916 2,717 2,755
Source: Company data, Nomura estimates
313
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 84: TXN Segment Analysis
Year End: December 2013E 2014E 2012 2013E 2014E (in $mn) Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Revenue Analog 1,648 1,745 1,931 1,839 1,839 1,968 2,086 2,024 6,998 7,163 7,917 Embedded Processing 561 618 668 638 638 689 737 700 2,257 2,485 2,764 Other 676 684 645 503 478 550 577 473 3,570 2,508 2,078 Total 2,885 3,047 3,244 2,980 2,955 3,207 3,400 3,197 12,825 12,156 12,759
Percent of Revenues Analog 57.1% 57.3% 59.5% 61.7% 62.2% 61.4% 61.3% 63.3% 54.6% 58.9% 62.0% Embedded Processing 19.4% 20.3% 20.6% 21.4% 21.6% 21.5% 21.7% 21.9% 17.6% 20.4% 21.7% Other 23.4% 22.4% 19.9% 16.9% 16.2% 17.1% 17.0% 14.8% 27.8% 20.6% 16.3% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
QoQ Growth Analog -1.3% 5.9% 10.7% -4.8% 0.0% 7.0% 6.0% -3.0% Embedded Processing 2.7% 10.2% 8.1% -4.5% 0.0% 8.0% 7.0% -5.0% Other -11.5% 1.2% -5.7% -22.0% -5.0% 15.0% 5.0% -18.0% Total -3.2% 5.6% 6.5% -8.1% -0.8% 8.5% 6.0% -6.0%
YoY Growth Analog -2.3% -3.1% 4.8% 10.2% 11.6% 12.8% 8.0% 10.0% 9.8% 2.4% 10.5% Embedded Processing 3.9% 6.6% 13.0% 16.8% 13.7% 11.5% 10.4% 9.8% -5.2% 10.1% 11.2% Other -24.5% -28.4% -32.5% -34.1% -29.3% -19.6% -10.5% -5.9% -28.3% -29.7% -17.2% Total -7.6% -8.6% -4.3% 0.0% 2.4% 5.2% 4.8% 7.3% -6.6% -5.2% 5.0%
Source: Company data, Nomura estimates
Fig. 85: TXN Valuation Ratios
Year End: December 2013E 2014E 2012 2013E 2014E Mar-13 Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Profitability Ratios Return on Equity 13% 24% 23% 19% 18% 22% 25% 22% 16% 20% 22% Return on Assets 7% 14% 13% 11% 10% 12% 14% 12% 9% 11% 12%
Efficiency Ratios Days Sales Outstanding 42.2 44.7 42.9 42.9 42.9 42.9 42.9 42.9 37.7 42.9 42.9 Inventory Turns 3.6 3.5 3.4 3.3 3.3 3.3 3.3 3.3 3.5 3.3 3.3 Days of Inventory 101.3 104.8 106.0 109.0 109.0 109.0 109.0 109.0 103.1 109.0 109.0
Liquidity Ratios Current Ratio 2.6 2.9 3.0 3.0 3.0 3.0 3.0 3.0 2.4 3.0 3.0 Total Debt/Equity 0.8 0.8 0.7 0.7 0.8 0.8 0.8 0.8 0.8 0.7 0.8 Total Debt/Assets 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.5 0.4 0.4
Cash & Book Value/Share Book Value/Share $9.9 $10.0 $10.1 $10.0 $9.9 $9.9 $10.0 $10.0 $9.8 $10.0 $10.0 Cash/Share $3.6 $3.1 $3.4 $3.6 $3.6 $3.5 $3.5 $3.7 $3.6 $3.6 $3.7 FCF/Share $0.2 $0.5 $0.9 $0.8 $0.6 $0.5 $0.6 $0.8 $2.5 $2.4 $2.6
Source: Company data, Nomura estimates
314
Nomura | U.S. Semiconductors Primer December 11, 2013
Xilinx Inc. (XLNX, Buy, TP $50)
Fig. 86: XLNX Income Statement Summary Year end: June FY14E FY15E CY12 CY13E CY14E ($ in million) Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E
INCOME STATEMENT Total revenue 579.0 598.9 599.0 624.5 673.1 693.6 678.1 704.1 2,195 2,309 2,669 QoQ 8.8% 3.5% 0.0% 4.2% 7.8% 3.0% -2.2% 3.8% YoY -0.7% 10.1% 17.5% 17.3% 16.3% 15.8% 13.2% 12.7% -3.3% 5.2% 15.6%
COGS 179.7 182.8 185.7 192.0 205.3 209.8 205.1 213.0 744.2 728.8 812.3
Gross profit 399.3 416.1 413.3 432.5 467.8 483.8 473.0 491.1 1,451.3 1,580.3 1,857.0
R&D 111.5 125.0 123.0 125.0 125.0 125.0 125.0 125.0 479.6 470.7 500.0 SG&A 92.4 96.3 99.6 98.1 100.1 102.1 102.1 102.1 366.2 379.1 402.6 Goodwill amort. 2.4 2.4 2.4 2.4 2.5 2.5 2.5 2.5 9.0 9.7 9.9 Restructuring Charges / Other 0.0 28.6 0.0 0.0 0.0 0.0 0.0 0.0 15.4 28.6 0.0 Total operating expenses 206.3 252.4 225.1 225.6 227.6 229.6 229.6 229.6 870.2 888.1 912.5 2% 3% Operating income 192.9 163.8 188.3 206.9 240.2 254.1 243.3 261.5 581.0 692.2 944.5
Interest and other expense, net (9.9) (11.0) (9.0) (9.0) (9.0) (9.0) (9.0) (9.0) (32.0) (38.8) (36.0) Pretax income 183.0 152.8 179.3 197.9 231.2 245.1 234.3 252.5 549.1 653.3 908.5
Provision for Taxes 26.0 11.3 23.3 25.7 30.1 31.9 30.5 32.8 69.7 68.3 118.1 0.0 0.0 Net income (proforma) 162.1 171.9 161.1 177.3 206.6 218.8 209.4 225.2 504.2 630.9 812.1 Net income (GAAP) 157.0 141.5 156.0 172.2 201.1 213.3 203.9 219.6 479.3 585.1 790.4
EPS (Proforma) $0.58 $0.59 $0.55 $0.61 $0.71 $0.75 $0.71 $0.76 $1.85 $2.21 $2.78 EPS (GAAP) $0.56 $0.49 $0.54 $0.59 $0.69 $0.73 $0.69 $0.75 $1.76 $2.05 $2.70
Shares outstanding (fully diluted) 280.3 290.7 290.7 290.7 291.7 292.7 293.7 294.7 272.8 284.7 292.2
Percent of Sales Gross Margin 69.0% 69.5% 69.0% 69.3% 69.5% 69.8% 69.8% 69.8% 66.1% 68.4% 69.6% R&D 19.3% 20.9% 20.5% 20.0% 18.6% 18.0% 18.4% 17.8% 21.8% 20.4% 18.7% SG&A 16.0% 16.1% 16.6% 15.7% 14.9% 14.7% 15.1% 14.5% 16.7% 16.4% 15.1% Operating margin 33.3% 27.3% 31.4% 33.1% 35.7% 36.6% 35.9% 37.1% 26.5% 30.0% 35.4% Effective Tax Rate 14.2% 7.4% 13.0% 13.0% 13.0% 13.0% 13.0% 13.0% 12.7% 10.4% 13.0% Net Margin 27.1% 23.6% 26.0% 27.6% 29.9% 30.7% 30.1% 31.2% 21.8% 25.3% 29.6%
Source: Company data, Nomura estimates
315
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 87: XLNX Balance Sheet Summary
Year End: June FY14E FY15E CY12 CY13E CY14E (in $mn) Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E
Assets 4,761 5,000 5,081 5,190 5,338 5,481 5,597 5,754 4,613 5,081 5,597 Cash, Equiv, and ST Investments 1,842 2,264 2,362 2,460 2,580 2,717 2,853 2,998 1,687 2,362 2,853 Inventories 187 184 176 182 194 199 194 202 226 176 194 Total Current Assets 2,355 2,801 2,890 3,007 3,162 3,312 3,436 3,601 2,264 2,890 3,436
Liabilities 2,034 2,128 2,132 2,145 2,171 2,179 2,170 2,186 1,797 2,132 2,170 Total Current Liabilities 1,402 1,482 1,487 1,499 1,525 1,534 1,525 1,540 336 1,487 1,525
Long-Term Debt 0000000091900 Total Non-Current Liabilities 631 646 646 646 646 646 646 646 1,461 646 646
Shareholders' Equity 2,727 2,872 2,949 3,046 3,167 3,302 3,427 3,569 2,816 2,949 3,427
Total Liabilities & Equity 4,761 5,000 5,081 5,190 5,338 5,481 5,597 5,754 4,613 5,081 5,597
Source: Company data, Nomura estimates
Fig. 88: XLNX Cash Flow Summary Year End: June FY14E FY15E CY12 CY13E CY14E (in $mn) Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E
Cash Flows from Operations 144 239 211 209 234 250 248 257 691 768 941 Depreciation & Amortization 19 19 19 19 19 19 19 19 74 75 75
Cash Flows from Investing (226) 186 (11) (11) (11) (11) (11) (11) (416) (227) (45) Capital Expenditures (11) (11) (11) (11) (11) (11) (11) (11) (44) (40) (45)
Cash Flows from Financing (32) (104) (102) (99) (103) (102) (101) (101) (313) (237) (405) Increase (Decrease) in Debt00000000000 Repurchase of Common Stock 0 (70) (70) (70) (70) (70) (70) (70) (198) (140) (280) Payment of Dividends (66) (67) (67) (67) (67) (67) (67) (67) (223) (258) (269)
Free Cash Flow 133 227 200 197 223 239 237 246 648 727 896
Source: Company data, Nomura estimates
316
Nomura | U.S. Semiconductors Primer December 11, 2013
Fig. 89: XLNX Segment Analysis
Year End: June FY14E FY15E CY12 CY13E CY14E Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Revenue by End Market Communications & Data Center 256 261 272 286 315 327 311 329 1,007 1,026 1,239 Industrial & A&D 217 229 219 227 241 246 246 251 761 862 960 Broadcast, Consumer & Auto 90 94 92 95 100 104 104 106 333 362 402 Other 16 15 16 16 18 17 18 18 95 60 69 Total 579 599 599 624 673 694 678 704 2,195 2,309 2,669
Percent of Revenues Communications 44% 44% 45% 46% 47% 47% 46% 47% 46% 44% 46% Data Processing 3% 3% 3% 3% 3% 2% 3% 3% 4% 3% 3% Consumer & Automotive 15% 16% 15% 15% 15% 15% 15% 15% 15% 16% 15% Industrial & Other 37% 38% 36% 36% 36% 35% 36% 36% 35% 37% 36% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%
QoQ Growth Communications 8% 2% 5% 5% 10% 4% -5% 6% ‐‐‐ Data Processing 27%-5%4%2%10%-5%5%2% ‐‐‐ Consumer & Automotive 4% 5% -2% 3% 5% 4% 0% 2% ‐‐‐ Industrial & Other 10%6%-5%4%6%2%0%2% ‐‐‐ Total 9%3%0%4%8%3%-2%4% ‐‐‐
YoY Growth Communications 5% -3% 14% 21% 0% 26% 14% 15% -4% 2% 21% Data Processing -48% -36% 33% 28% 11% 11% 12% 12% -27% -37% 15% Consumer & Automotive -5% 18% 23% 11% 11% 10% 12% 11% -1% 9% 11% Industrial & Other 10% 33% 18% 15% 11% 7% 12% 10% 1% 13% 11% Total -1% 10% 18% 17% 16% 16% 13% 13% -3% 5% 16%
Source: Company data, Nomura estimates
Fig. 90: XLNX Valuation Ratios
Year End: June FY14E FY15E CY12 CY13E CY14E Jun-13 Sep-13 Dec-13E Mar-14E Jun-14E Sep-14E Dec-14E Mar-15E Profitability Ratios Return on Equity 22.3% 19.9% 21.7% 23.8% 26.7% 27.4% 25.2% 26.1% 20.0% 22.3% 26.9% Return on Assets 11.4% 12.1% 12.9% 13.4% 13.9% 14.5% 15.0% 15.5% 11.1% 11.0% 11.4%
Efficiency Ratios Days Sales Outstanding 42.1 42.2 42.2 42.2 42.2 42.2 42.2 42.2 41.5 42.2 42.2 Inventory Turns 3.5 3.6 3.9 4.1 4.2 4.2 4.2 4.2 3.6 3.9 4.2 Days of Inventory 94.6 91.4 86.4 86.4 86.4 86.4 86.4 86.4 120.6 86.4 86.4
Liquidity Ratios Current Ratio 1.7 1.9 1.9 2.0 2.1 2.2 2.3 2.3 6.7 1.9 2.3 Quick Ratio 1.5 1.7 1.8 1.8 1.9 2.0 2.1 2.2 5.7 1.8 2.1 Debt/Capital 42.7% 42.6% 42.0% 41.3% 40.7% 39.8% 38.8% 38.0% 39.0% 42.0% 38.8% Debt/Assets 19.5% 18.6% 18.3% 17.9% 17.4% 17.0% 16.6% 16.2% 19.9% 18.3% 16.6%
Book & Cash Value Book Value/Share $9.7 $9.9 $10.1 $10.5 $10.9 $11.3 $11.7 $12.1 $10.4 $10.1 $11.7 Net Cash/Share $12.4 $12.7 $13.1 $13.4 $13.8 $14.2 $14.6 $15.0 $8.5 $13.3 $14.7 FCF/Share $0.5 $0.8 $0.7 $0.7 $0.8 $0.8 $0.8 $0.8 $2.4 $2.6 $3.1
Source: Company data, Nomura estimates
317
Nomura | U.S. Semiconductors Primer December 11, 2013
Appendix A-1
Analyst Certification
We, Romit Shah, Sidney Ho and Sanjay Chaurasia, hereby certify (1) that the views expressed in this Research report accurately reflect our personal views about any or all of the subject securities or issuers referred to in this Research report, (2) no part of our compensation was, is or will be directly or indirectly related to the specific recommendations or views expressed in this Research report and (3) no part of our compensation is tied to any specific investment banking transactions performed by Nomura Securities International, Inc., Nomura International plc or any other Nomura Group company.
Issuer Specific Regulatory Disclosures
The term "Nomura Group" used herein refers to Nomura Holdings, Inc. or any of its affiliates or subsidiaries, and may refer to one or more Nomura Group companies.
Materially mentioned issuers
Issuer Ticker Price Price date Stock rating Previous rating Date of change Sector rating Analog Devices ADI US USD 48.91 10-Dec-2013 Neutral Reduce 03-Oct-2011 Neutral Altera Corporation ALTR US USD 31.41 10-Dec-2013 Reduce Not Rated 06-Mar-2012 Neutral Advanced Micro Devices, Inc. AMD US USD 3.72 10-Dec-2013 Neutral Buy 19-Sep-2011 Neutral Atmel Corporation ATML US USD 7.49 10-Dec-2013 Neutral Buy 24-Oct-2012 Neutral Avago Technologies Limited AVGO US USD 46.60 10-Dec-2013 Buy Neutral 09-Jan-2012 Neutral Broadcom Corporation BRCM US USD 28.51 10-Dec-2013 Buy Neutral 01-Jun-2011 Neutral Cavium Inc. CAVM US USD 35.62 10-Dec-2013 Reduce Not Rated 15-Jul-2013 Neutral Cypress Semiconductors Corporation CY US USD 9.70 10-Dec-2013 Buy Not Rated 22-Mar-2012 Neutral Intel Corporation INTC US USD 24.82 10-Dec-2013 Neutral Reduce 11-Dec-2013 Neutral Linear Technology Corporation LLTC US USD 43.92 10-Dec-2013 Neutral Buy 18-Oct-2012 Neutral Marvell Technology Group Ltd. MRVL US USD 13.46 10-Dec-2013 Neutral Reduce 04-Sep-2012 Neutral Micron Technology MU US USD 23.14 10-Dec-2013 Buy Neutral 11-Dec-2013 Neutral Maxim Integrated Products Inc. MXIM US USD 28.26 10-Dec-2013 Neutral Buy 10-Oct-2013 Neutral NVIDIA Corporation NVDA US USD 15.56 10-Dec-2013 Buy Neutral 07-May-2012 Neutral Qualcomm, Inc. QCOM US USD 73.38 10-Dec-2013 Buy Neutral 05-Nov-2012 Neutral SanDisk Corporation SNDK US USD 68.96 10-Dec-2013 Reduce Neutral 11-Dec-2013 Neutral Texas Instruments Inc. TXN US USD 43.41 10-Dec-2013 Reduce Neutral 11-Dec-2013 Neutral Xilinx Inc. XLNX US USD 43.67 10-Dec-2013 Buy Neutral 11-Dec-2013 Neutral
Important Disclosures Online availability of research and conflict-of-interest disclosures Nomura research is available on www.nomuranow.com/research, Bloomberg, Capital IQ, Factset, MarkitHub, Reuters and ThomsonOne. Important disclosures may be read at http://go.nomuranow.com/research/globalresearchportal/pages/disclosures/disclosures.aspx or requested from Nomura Securities International, Inc., on 1-877-865-5752. If you have any difficulties with the website, please email [email protected] for help.
The analysts responsible for preparing this report have received compensation based upon various factors including the firm's total revenues, a portion of which is generated by Investment Banking activities. Unless otherwise noted, the non-US analysts listed at the front of this report are not registered/qualified as research analysts under FINRA/NYSE rules, may not be associated persons of NSI, and may not be subject to FINRA Rule 2711 and NYSE Rule 472 restrictions on communications with covered companies, public appearances, and trading securities held by a research analyst account.
Nomura Global Financial Products Inc. (“NGFP”) Nomura Derivative Products Inc. (“NDPI”) and Nomura International plc. (“NIplc”) are registered with the Commodities Futures Trading Commission and the National Futures Association (NFA) as swap dealers. NGFP, NDPI, and NIplc are generally engaged in the trading of swaps and other derivative products, any of which may be the subject of this report.
Any authors named in this report are research analysts unless otherwise indicated. Industry Specialists identified in some Nomura International plc research reports are employees within the Firm who are responsible for the sales and trading effort in the sector for which they have coverage. Industry Specialists do not contribute in any manner to the content of research reports in which their names appear. Marketing Analysts identified in some Nomura research reports are research analysts employed by Nomura International plc who are primarily responsible
318
Nomura | U.S. Semiconductors Primer December 11, 2013 for marketing Nomura’s Equity Research product in the sector for which they have coverage. Marketing Analysts may also contribute to research reports in which their names appear and publish research on their sector.
Distribution of ratings (Global) The distribution of all ratings published by Nomura Global Equity Research is as follows: 44% have been assigned a Buy rating which, for purposes of mandatory disclosures, are classified as a Buy rating; 41% of companies with this rating are investment banking clients of the Nomura Group*. 46% have been assigned a Neutral rating which, for purposes of mandatory disclosures, is classified as a Hold rating; 54% of companies with this rating are investment banking clients of the Nomura Group*. 10% have been assigned a Reduce rating which, for purposes of mandatory disclosures, are classified as a Sell rating; 21% of companies with this rating are investment banking clients of the Nomura Group*. As at 30 September 2013. *The Nomura Group as defined in the Disclaimer section at the end of this report.
Explanation of Nomura's equity research rating system in Europe, Middle East and Africa, US and Latin America, and Japan and Asia ex-Japan from 21 October 2013 The rating system is a relative system, indicating expected performance against a specific benchmark identified for each individual stock, subject to limited management discretion. An analyst’s target price is an assessment of the current intrinsic fair value of the stock based on an appropriate valuation methodology determined by the analyst. Valuation methodologies include, but are not limited to, discounted cash flow analysis, expected return on equity and multiple analysis. Analysts may also indicate expected absolute upside/downside relative to the stated target price, defined as (target price - current price)/current price.
STOCKS A rating of 'Buy', indicates that the analyst expects the stock to outperform the Benchmark over the next 12 months. A rating of 'Neutral', indicates that the analyst expects the stock to perform in line with the Benchmark over the next 12 months. A rating of 'Reduce', indicates that the analyst expects the stock to underperform the Benchmark over the next 12 months. A rating of 'Suspended', indicates that the rating, target price and estimates have been suspended temporarily to comply with applicable regulations and/or firm policies. Securities and/or companies that are labelled as 'Not rated' or shown as 'No rating' are not in regular research coverage. Investors should not expect continuing or additional information from Nomura relating to such securities and/or companies. Benchmarks are as follows: United States/Europe/Asia ex- Japan: please see valuation methodologies for explanations of relevant benchmarks for stocks, which can be accessed at: http://go.nomuranow.com/research/globalresearchportal/pages/disclosures/disclosures.aspx; Global Emerging Markets (ex-Asia): MSCI Emerging Markets ex-Asia, unless otherwise stated in the valuation methodology; Japan: Russell/Nomura Large Cap.
SECTORS A 'Bullish' stance, indicates that the analyst expects the sector to outperform the Benchmark during the next 12 months. A 'Neutral' stance, indicates that the analyst expects the sector to perform in line with the Benchmark during the next 12 months. A 'Bearish' stance, indicates that the analyst expects the sector to underperform the Benchmark during the next 12 months. Sectors that are labelled as 'Not rated' or shown as 'N/A' are not assigned ratings. Benchmarks are as follows: United States: S&P 500; Europe: Dow Jones STOXX 600; Global Emerging Markets (ex-Asia): MSCI Emerging Markets ex-Asia. Japan/Asia ex-Japan: Sector ratings are not assigned.
Explanation of Nomura's equity research rating system in Japan and Asia ex-Japan prior to 21 October 2013 STOCKS Stock recommendations are based on absolute valuation upside (downside), which is defined as (Target Price - Current Price) / Current Price, subject to limited management discretion. In most cases, the Target Price will equal the analyst's 12-month intrinsic valuation of the stock, based on an appropriate valuation methodology such as discounted cash flow, multiple analysis, etc. A 'Buy' recommendation indicates that potential upside is 15% or more. A 'Neutral' recommendation indicates that potential upside is less than 15% or downside is less than 5%. A 'Reduce' recommendation indicates that potential downside is 5% or more. A rating of 'Suspended' indicates that the rating and target price have been suspended temporarily to comply with applicable regulations and/or firm policies in certain circumstances including when Nomura is acting in an advisory capacity in a merger or strategic transaction involving the subject company. Securities and/or companies that are labelled as 'Not rated' or shown as 'No rating' are not in regular research coverage of the Nomura entity identified in the top banner. Investors should not expect continuing or additional information from Nomura relating to such securities and/or companies.
SECTORS A 'Bullish' rating means most stocks in the sector have (or the weighted average recommendation of the stocks under coverage is) a positive absolute recommendation. A 'Neutral' rating means most stocks in the sector have (or the weighted average recommendation of the stocks under coverage is) a neutral absolute recommendation. A 'Bearish' rating means most stocks in the sector have (or the weighted average recommendation of the stocks under coverage is) a negative absolute recommendation.
Target Price A Target Price, if discussed, reflects in part the analyst's estimates for the company's earnings. The achievement of any target price may be impeded by general market and macroeconomic trends, and by other risks related to the company or the market, and may not occur if the company's earnings differ from estimates.
Disclaimers This document contains material that has been prepared by the Nomura entity identified at the top or bottom of page 1 herein, if any, and/or, with the sole or joint contributions of one or more Nomura entities whose employees and their respective affiliations are specified on page 1 herein or identified elsewhere in the document. The term "Nomura Group" used herein refers to Nomura Holdings, Inc. or any of its affiliates or subsidiaries and may refer to one or more Nomura Group companies including: Nomura Securities Co., Ltd. ('NSC') Tokyo, Japan; Nomura International plc ('NIplc'), UK; Nomura Securities International, Inc. ('NSI'), New York, US; Nomura International (Hong Kong) Ltd. (‘NIHK’), Hong Kong; Nomura Financial Investment (Korea) Co., Ltd. (‘NFIK’), Korea (Information on Nomura analysts registered with the Korea Financial Investment Association ('KOFIA') can be found on the KOFIA Intranet at http://dis.kofia.or.kr); Nomura Singapore Ltd. (‘NSL’), Singapore (Registration number 197201440E, regulated by the Monetary Authority of Singapore); Nomura Australia Ltd. (‘NAL’), Australia (ABN 48 003 032 513), regulated by the Australian Securities and Investment Commission ('ASIC') and holder of an Australian financial services licence number 246412; P.T. Nomura Indonesia (‘PTNI’), Indonesia; Nomura Securities Malaysia Sdn. Bhd. (‘NSM’), Malaysia; NIHK, Taipei Branch
319
Nomura | U.S. Semiconductors Primer December 11, 2013
(‘NITB’), Taiwan; Nomura Financial Advisory and Securities (India) Private Limited (‘NFASL’), Mumbai, India (Registered Address: Ceejay House, Level 11, Plot F, Shivsagar Estate, Dr. Annie Besant Road, Worli, Mumbai- 400 018, India; Tel: +91 22 4037 4037, Fax: +91 22 4037 4111; SEBI Registration No: BSE INB011299030, NSE INB231299034, INF231299034, INE 231299034, MCX: INE261299034) and NIplc, Madrid Branch (‘NIplc, Madrid’). ‘CNS Thailand’ next to an analyst’s name on the front page of a research report indicates that the analyst is employed by Capital Nomura Securities Public Company Limited (‘CNS’) to provide research assistance services to NSL under a Research Assistance Agreement. CNS is not a Nomura entity. THIS MATERIAL IS: (I) FOR YOUR PRIVATE INFORMATION, AND WE ARE NOT SOLICITING ANY ACTION BASED UPON IT; (II) NOT TO BE CONSTRUED AS AN OFFER TO SELL OR A SOLICITATION OF AN OFFER TO BUY ANY SECURITY IN ANY JURISDICTION WHERE SUCH OFFER OR SOLICITATION WOULD BE ILLEGAL; AND (III) BASED UPON INFORMATION FROM SOURCES THAT WE CONSIDER RELIABLE, BUT HAS NOT BEEN INDEPENDENTLY VERIFIED BY NOMURA GROUP. Nomura Group does not warrant or represent that the document is accurate, complete, reliable, fit for any particular purpose or merchantable and does not accept liability for any act (or decision not to act) resulting from use of this document and related data. To the maximum extent permissible all warranties and other assurances by Nomura group are hereby excluded and Nomura Group shall have no liability for the use, misuse, or distribution of this information. Opinions or estimates expressed are current opinions as of the original publication date appearing on this material and the information, including the opinions and estimates contained herein, are subject to change without notice. Nomura Group is under no duty to update this document. Any comments or statements made herein are those of the author(s) and may differ from views held by other parties within Nomura Group. Clients should consider whether any advice or recommendation in this report is suitable for their particular circumstances and, if appropriate, seek professional advice, including tax advice. Nomura Group does not provide tax advice. Nomura Group, and/or its officers, directors and employees, may, to the extent permitted by applicable law and/or regulation, deal as principal, agent, or otherwise, or have long or short positions in, or buy or sell, the securities, commodities or instruments, or options or other derivative instruments based thereon, of issuers or securities mentioned herein. Nomura Group companies may also act as market maker or liquidity provider (within the meaning of applicable regulations in the UK) in the financial instruments of the issuer. Where the activity of market maker is carried out in accordance with the definition given to it by specific laws and regulations of the US or other jurisdictions, this will be separately disclosed within the specific issuer disclosures. This document may contain information obtained from third parties, including ratings from credit ratings agencies such as Standard & Poor’s. Reproduction and distribution of third party content in any form is prohibited except with the prior written permission of the related third party. Third party content providers do not guarantee the accuracy, completeness, timeliness or availability of any information, including ratings, and are not responsible for any errors or omissions (negligent or otherwise), regardless of the cause, or for the results obtained from the use of such content. Third party content providers give no express or implied warranties, including, but not limited to, any warranties of merchantability or fitness for a particular purpose or use. Third party content providers shall not be liable for any direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs, expenses, legal fees, or losses (including lost income or profits and opportunity costs) in connection with any use of their content, including ratings. Credit ratings are statements of opinions and are not statements of fact or recommendations to purchase hold or sell securities. They do not address the suitability of securities or the suitability of securities for investment purposes, and should not be relied on as investment advice. Any MSCI sourced information in this document is the exclusive property of MSCI Inc. (‘MSCI’). Without prior written permission of MSCI, this information and any other MSCI intellectual property may not be reproduced, re-disseminated or used to create any financial products, including any indices. This information is provided on an "as is" basis. The user assumes the entire risk of any use made of this information. MSCI, its affiliates and any third party involved in, or related to, computing or compiling the information hereby expressly disclaim all warranties of originality, accuracy, completeness, merchantability or fitness for a particular purpose with respect to any of this information. Without limiting any of the foregoing, in no event shall MSCI, any of its affiliates or any third party involved in, or related to, computing or compiling the information have any liability for any damages of any kind. MSCI and the MSCI indexes are services marks of MSCI and its affiliates. Investors should consider this document as only a single factor in making their investment decision and, as such, the report should not be viewed as identifying or suggesting all risks, direct or indirect, that may be associated with any investment decision. Nomura Group produces a number of different types of research product including, among others, fundamental analysis, quantitative analysis and short term trading ideas; recommendations contained in one type of research product may differ from recommendations contained in other types of research product, whether as a result of differing time horizons, methodologies or otherwise. Nomura Group publishes research product in a number of different ways including the posting of product on Nomura Group portals and/or distribution directly to clients. Different groups of clients may receive different products and services from the research department depending on their individual requirements. Clients outside of the US may access the Nomura Research Trading Ideas platform (Retina) at http://go.nomuranow.com/equities/tradingideas/retina/ Figures presented herein may refer to past performance or simulations based on past performance which are not reliable indicators of future performance. Where the information contains an indication of future performance, such forecasts may not be a reliable indicator of future performance. Moreover, simulations are based on models and simplifying assumptions which may oversimplify and not reflect the future distribution of returns. Certain securities are subject to fluctuations in exchange rates that could have an adverse effect on the value or price of, or income derived from, the investment. The securities described herein may not have been registered under the US Securities Act of 1933 (the ‘1933 Act’), and, in such case, may not be offered or sold in the US or to US persons unless they have been registered under the 1933 Act, or except in compliance with an exemption from the registration requirements of the 1933 Act. Unless governing law permits otherwise, any transaction should be executed via a Nomura entity in your home jurisdiction. This document has been approved for distribution in the UK and European Economic Area as investment research by NIplc. NIplc is authorised by the Prudential Regulation Authority and regulated by the Financial Conduct Authority and the Prudential Regulation Authority. NIplc is a member of the London Stock Exchange. This document does not constitute a personal recommendation within the meaning of applicable regulations in the UK, or take into account the particular investment objectives, financial situations, or needs of individual investors. This document is intended only for investors who are 'eligible counterparties' or 'professional clients' for the purposes of applicable regulations in the UK, and may not, therefore, be redistributed to persons who are 'retail clients' for such purposes. This document has been approved by NIHK, which is regulated by the Hong Kong Securities and Futures Commission, for distribution in Hong Kong by NIHK. This document has been approved for distribution in Australia by NAL, which is authorized and regulated in Australia by the ASIC. This document has also been approved for distribution in Malaysia by NSM. In Singapore, this document has been distributed by NSL. NSL accepts legal responsibility for the content of this document, where it concerns securities, futures and foreign exchange, issued by their foreign affiliates in respect of recipients who are not accredited, expert or institutional investors as defined by the Securities and Futures Act (Chapter 289). Recipients of this document in Singapore should contact NSL in respect of matters arising from, or in connection with, this document. Unless prohibited by the provisions of Regulation S of the 1933 Act, this material is distributed in the US, by NSI, a US-registered broker-dealer, which accepts responsibility for its contents in accordance with the provisions of Rule 15a-6, under the US Securities Exchange Act of 1934. This document has not been approved for distribution to persons other than ‘Authorised Persons’, ‘Exempt Persons’ or ‘Institutions’ (as defined by the Capital Markets Authority) in the Kingdom of Saudi Arabia (‘Saudi Arabia’) or 'professional clients' (as defined by the Dubai Financial Services Authority) in the United Arab Emirates (‘UAE’) or a ‘Market Counterparty’ or ‘Business Customers’ (as defined by the Qatar Financial
320
Nomura | U.S. Semiconductors Primer December 11, 2013
Centre Regulatory Authority) in the State of Qatar (‘Qatar’) by Nomura Saudi Arabia, NIplc or any other member of Nomura Group, as the case may be. Neither this document nor any copy thereof may be taken or transmitted or distributed, directly or indirectly, by any person other than those authorised to do so into Saudi Arabia or in the UAE or in Qatar or to any person other than ‘Authorised Persons’, ‘Exempt Persons’ or ‘Institutions’ located in Saudi Arabia or 'professional clients' in the UAE or a ‘Market Counterparty’ or ‘Business Customers’ in Qatar . By accepting to receive this document, you represent that you are not located in Saudi Arabia or that you are an ‘Authorised Person’, an ‘Exempt Person’ or an ‘Institution’ in Saudi Arabia or that you are a 'professional client' in the UAE or a ‘Market Counterparty’ or ‘Business Customers’ in Qatar and agree to comply with these restrictions. Any failure to comply with these restrictions may constitute a violation of the laws of the UAE or Saudi Arabia or Qatar. NO PART OF THIS MATERIAL MAY BE (I) COPIED, PHOTOCOPIED, OR DUPLICATED IN ANY FORM, BY ANY MEANS; OR (II) REDISTRIBUTED WITHOUT THE PRIOR WRITTEN CONSENT OF A MEMBER OF NOMURA GROUP. If this document has been distributed by electronic transmission, such as e-mail, then such transmission cannot be guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or contain viruses. The sender therefore does not accept liability for any errors or omissions in the contents of this document, which may arise as a result of electronic transmission. If verification is required, please request a hard-copy version.
Nomura Group manages conflicts with respect to the production of research through its compliance policies and procedures (including, but not limited to, Conflicts of Interest, Chinese Wall and Confidentiality policies) as well as through the maintenance of Chinese walls and employee training. Additional information is available upon request and disclosure information is available at the Nomura Disclosure web page: http://go.nomuranow.com/research/globalresearchportal/pages/disclosures/disclosures.aspx Copyright © 2013 Nomura Securities International, Inc.. All rights reserved.
321