Emerging Technologies and Moore’s Law: Prospects for the Future
Mike Mayberry Director of Components Research VP, Technology and Manufacturing Group Intel Corporation
March, 2010 Economics, Technology, & Innovation
“No exponential is forever … but we can delay ‘forever’.”
Technologies on Deck
Opportunities for Future Products
Visible Horizon and Beyond
2 Moore’s Law - 1965
“Reduced cost is one of the big attractions of integrated electronics, and the cost advantage continues to increase as the technology evolves toward the production of larger and larger circuit functions on a single semiconductor substrate.” Electronics, Volume 38, Number 8, April 19 , 1965
Integrated circuits = Parallel fabrication
Parallel fabrication = Lower cost per function 3 Enabling Innovation and Cost Reductions
250nm 180nm 130nm 90nm 65nm 45nm 200mm 200mm 200mm 300mm 300mm 300mm
Twice the Same circuitry circuitry in the Option to design half the space OR same space = for optimal (cost reduction) (architectural performance/cost innovation)
Source: Intel
4 Key Economics: Scaling is Better under almos t all circums tances
Revenues drive R&D y If you can scale then which enables you should further scaling Create compelling – Compelling products command products higher prices – Leading technology delivers lower costs Business – Enables necessary investment Affordable size in R&D R&D matters y If you can’t scale then hope your competitors can’t either Lower cost and Scaling gives lower better capability cost per function enables new and improved capability products Ability to make it work
5 Economics, Technology, & Innovation
“No exponential is forever … but we can delay ‘forever’.”
Technologies on Deck
Opportunities for Future Products
Visible Horizon and Beyond
6 “We’re limited by the wavelength of light”
Resolved Rayleigh criteria
Feature >> Unresolved Feature << Wavelength Wavelength
Feature = 0.6 x Wavelength / numerical aperture
Lithography Scaling Limitations
From Broers [1] IEDM Plenary Session 1980 7 Breaking through the Wall
1000 New lens materials enabled shorter wavelengths
Optical 365nm in mid 80s Wavelength 193nm immersi on (145nm eqv) tod ay
New mask techniques enabled 100 smaller features nm Feature size π 0
EUV 13.5nm 10 Conventional Alternate 1980 1990 2000 2010 2020 Mask Structure Phase Shift 8 “We’ re hitting the power wall”
R. Schmidt, IBM 14
IBM ES9000 CMOS Prescott 12 Jayhawk(dual)
y Bipolar ) tt 2 10 T-Rex Mckinley
nsi Squadrons Fujitsu VP2000 atts/cm 8
ee IBM GP ww IBM 3090S NTT IBM RY5 6 Fujitsu M-780 IBM RY7 Pentium 4
eat Flux( Pulsar er D
HH 4 IBM 3090 IBM RY6 Start of CDC Cyber 205 Water Cooling IBM 4381 IBM RY4 2 IBM 3081 Apache Fujitsu M380 IBM 370 Merced Module Pow IBM 360 IBM 3033 Vacuum Pentium II(DSIP) 0 1950 1960 1970 1980 1990 2000 2010 Year of Announcement
Source: IBM 2004 9 Energy p er Transistor Continues to Drop
1010 10
As the number 109 rr 100 oooo of transistors 10-1 goes UP 108
-2 r transistr transist 10 107 Energy per 10-3 transistor goes 106 -4 DOWN nergy penergy pe 10 EEEE 105 10-5
10-6 104
-7 10 103 ’70 ’80 ’90 ’00 Source: WSTS/DWSTS/Dtt/Itlataquest/Intel
~ 1 Million Factor Reduction In Energy/Transistor Over 30+ Years 10 Energy Consumption is about System Optimization
5 years
1015 938 Estimated Annual Energy Consumption YearYear
))))
655 Going med permed per is is betterbetter uuuu r rrr MbilMobile (lowe(lowe 229 Wh ConsWh Cons KKK K >26x Reduction 38
Unmanaged Pentium® D Unmanaged Pentium® D Unmanaged Core™2 Duo Managed Core™2 Duo Processor Managed Core™2 Duo Processor Processor 945 with CRT display Processor 945 with LCD display Processor E6550 with LCD E6550 with LCD display T9400 mobile platform display
Energy savings comes from combination of transistor scaling and architectural innovations
Performance tests/ratings are provided assuming specific computcomputerer systems and/or components and reflect the approximate perfoperformrmanceance of Intel products as measured by those tests. Any difference in system hardware or software design or configuration may affect actual performanperformancece.. This data may vary from 11 other material generated for specific marketing requests. “We’re running out of atoms”atoms
Polysilicon Gate Electrode
1.2 nm SiO2 3-4 atoms thick
Silicon Substrate
12 New Materials and New Fabrication
100 SiON/Poly 65nm
45nm node e
g 10
1 SiON/Poly 65nm
te Leaka te 0.1
0.01 liz e d G a
aa 00010.001 HiK+MG 45nm HiK+MG 45nm 0.0001 Norm PMOS NMOS 0.00001 00001 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
VGS (V)
3.0nm HighHigh--kk 45 nm HK+MG provides >25x gate leakage reduction
13 The Timing of Research Is Not Always Predictable But a Commitment to Research is Required
90’s 1999 2001 2003 2005 2007
Strained Development Silicon ‘0101-’02 1st gen ’03 2nd gen ‘05 3rd gen ‘07
Gat Materials Metal e HighHigh--kk Metal selected gates Development High-k ‘05- ’06 Gate 2000 research Silicon 2001 Demo ‘03 Prod ‘07
Intel 45nm Highh--kk Metal Gate
Ten Years, Millions of Hours Dead Ends, Failures Unwavering Focus
Source: Intel 14 Meanwhile despite these worries ItiInnovations contitidnued on proddtuct side y Personal computing replacing shared computers y Portable computing y Connected computers, sharing data, Internet y Wireless connectivity y Smaller form factors, pocket computers y Sppppecial purpose devices …
15 Economics, Technology, & Innovation
“No exponential is forever … but we can delay ‘forever’.”
Technologies on Deck
Opportunities for Future Products
Visible Horizon and Beyond
16 Steady Technology Cadence
TECHNOLOGY GENERATION
130nm 90nm 65nm 45nm 32nm 2001 2003 2005 2007 2009
MANUFACTURING
17 Product Cadence for Sustained Leadership
TICK Pentium® D, Xeon™, Core™ processor RSRS 65nm
2 YEA2 YEA TOCK Core 2 processor, Xeon processor
SS TICK PENRYN Family RRRR 45nm
2 YEA2 YEA TOCK NEHALEM
SSSS TICK WESTMERE 32nm YEAR YEAR 2 222 TOCK SANDY BRIDGE
All product information and dates are preliminary and subject to change without notice 18 So What ?
Generational benefits don’t just add, they multiply
19 Energy Efficiency Improvement (Same Perf ormance Footpri nt)
184 single core servers 21 Intel® Xeon® 5500 2005 2009 bdbased servers
Up to Up to Annual Energy Costs 8 month 9X 92% Return on Investment Server Reduction Estimated Reduction
Source: Intel estimates as of Nov 2008. Performance comparison using SPECjbb2005 bops (business operations per second). Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance. For detailed calculations, configurations and assumptions refer to the legal information slide in backup.
20 New Fabrication enables New Product Opportunities
L3 Core L3 Core L3 Core
Core and Uncore Core Power Gates OFF Both Core and Uncore Power Gates are ON And Uncoreretention Power Gates are OFF
Westmere 6 cores, Infrared imaging showing advanced power management 21 The Opportunity for Data Center Efficiency
Projected Data Center Energy Use Under Five Scenarios
2.9% Of Projected 140 Total U.S. Electricity Use Historical 120 Trends 1.5% Of Total US. Current 100 Electricity Usage Efficiency
year) year) Trends //// Improved 80 0.8% Of Total US Operation Electricity Usage 60
ns (kWh ns (kWh BtBest PPtiractice 40
BillioBillio StateState--ofof--thethe-- 20 Art
0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Forecast
Source: EPA Report to Congress on Server and Data Center Energy Efficiency; August 2, 2007 22 Technology Advances Enable Both Better Prod ucts and a Wider Range
65nm 45nm
+25%
10X Power yy 8 Core Stand-b Quad Core Log
Dual Core Single Core
Transistor Performance (switching speed)
23 Technology Advances Enable Both Better Prod ucts and a Wider Range
65nm 45nm LkLeakage Server
+25% 1x Desktop Intel®10X Atom™
0.1x LtLaptop
0.01x
Netbook 0.001x
Transistor Performance (switching speed)
24 Extending Transistor Advantage to SoC’s FtFaster Transi itstors and BttBetter Power PPferformance
65nm 45nm 32nm LkLeakage
+25% 1x
+22% 10X
10X 0.1x
0.01x IA 32nm SoC
0.001x
Transistor Performance (switching speed)
25 Implementing a Platform on a Chip
High Speed Memory Interface
High Perf /Low Power SRAM/RF Cache Logic Transistors Memory IA Core ed Media nterface rface I ee
Gfx/Security/ Analog/ HV I/O Transistors rial Inte Inductors sic I/O Video/Audio High Sp a a ee B S Display Interface CE Embedded Handhelds
SodavillePrecision Linear Capacitors Jasper Forest LincroftDecap Port 1 E E Port 2 Vbias= Vgate Vbias= 0 V Diodes B B n+ n+ n+ n+ n+ Vertical BJT Single EndedVaractor n-well C C Varactors
26 Economics, Technology, & Innovation
“No exponential is forever … but we can delay ‘forever’.”
Technologies on Deck
Opportunities for Future Products
Visible Horizon and Beyond
27 Industry R & D is Increasing – in dollars and as a percent of revenues
40 Industry R & D $ Billion
30
20
10 How to fabricate 0
Forecast Process Product Software R & D Hardware Other
Source: IBS, Intel 28 Possible Product Possibilities y Cloud computing – Solving grand challenge problems – Enable access anyyytime anywhere y LifeLife--likelike computing – Immersive experiences ––ContextContext aware y Heterogenous networks – Computing in network – Integration with bio and sensors
29 Potential Energy Savings Outside of Information and Communications Technology
Figure ES-1. Future Electricity Scenarios for the U.S. 7,000
6,000 Frozen Efficiency Case 1.896 Billion kWh 5,000 Reference Case illion kWh BB 1.242 4,000 Billion Semiconductor Enabled Efficiency Scenario kWh
3,000 2006 2010 2014 2018 2022 2026 2030
Source: ACEEE Report Number E094, May 2009 Semiconductor Technologies: The Potential to Revolutionize U.S. Energy Productivity 30 A tr uly S ma rt G rid Requ ires t hat Mic rop rocesso r intelligence be added throughout the network
31 High Performance Computing Segment Needs Decades of Performance Increases
Massive FLOPS Weather computing Prediction power to Zeta solve complex problems
Exa Genomics Research Medical Peta Imaging
Tera
Giga
1990 2010 2030
Forecast 32 Source: Intel Investor Meeting 2009 Core Computing Segment Will Need to Increase Performance to Support New Usages
High- FLOPS performance, lifelife--likelike 3D Interactions computing
Immersive User Interface fully Computational Sensor Based Presence Modeling integrated PhotoPhoto--realisticrealistic 3D Rendering with your Tera surroundings HighHigh--definitiondefinition Audio, RealReal--TimeTime Translation Video, Animation Natural Language Processing
3D Modeling Smart Cars Smart Homes
Video Editing Touch Screens Giga Basic Video
Photo Editing
2000 2010 2020
Forecast 33 Source: Intel Investor Meeting 2009 Perceiving the User to Make Interfaces Better
User Internet
tagged user Client data interaction
world interaction
World
34 Small Computing Segment Needs More and More Performance At Low Power
Anytime, FLOPS anywhere,
RealReal--time,time, 3D3D--VideoVideo personalized, Augmented Reality contextcontext--aware,aware, HighHigh--definitiondefinition Audio, environmental Video & Animation RealReal--TimeTime computing Language Natural Translation Language Giga Processing Visual Search - Still Recognition Video Presence Mobile TV Video Gaming Camera ThTouch s Web screens MMX Music Browser & Radio Mega EE--mailmail SMS
Voice
1995 2010 2025
Forecast 35 Source: Intel Investor Meeting 2009 Speech Interfaces made Better with AudioAudio--VisualVisual Fusion
Speech recognition works, but fails in noisy environments (mobile)
Source: Silsbee & Bovik, IEEE Speech & Audio
36 Economics, Technology, & Innovation
“No exponential is forever … but we can delay ‘forever’.”
Technologies on Deck
Opportunities for Future Products
Visible Horizon and Beyond
37 Steady Technology Cadence
TECHNOLOGY GENERATION
65nm 45nm 32nm 22nm 15nm 11nm 8nm Beyond 2005 2007 2009 2011 2013 2015 2017 2020
MANUFACTURING DEVELOPMENT RESEARCH
38 “We’re running out of atoms” Take 2 45nm node
TECHNOLOGY GENERATION 65nm 45nm 32nm 22nm 2005 2007 2009 2011 60 atoms across MANUFACTURING DEVELOPMENT made by top-down except high-K
Crafting Films with Atomic Layer Deposition
x y B x y A x A Step 1 Step 3
B A A
Step 2 Step 4
39 New Materials: Enabling the Future of Silicon Technology
40 IIIIII--VV Materials have Higher Mobility
Relative mobility Relative power-performance
-17 105 10 * /Vs]/Vs] 22 InGaAs Si MOSFETs 104 10-19 IDTH [Js/m] [Js/m] IDTH IDTH TY [cmTY [cm IIII WWWW
10x 30X improvement 103 Improvement 10-21 RONRON MOBIL MOBIL GYxDELAY/ GYxDELAY/ TTTT Silicon RRRR InGaAs QWFETs ENEENE ELECELEC 102 10-23 5x1011 1012 1013 2x1013 10 100 1000 10000 SHEET CARRIER DENSITY [cm-2] GATE LENGTH [nm]
R. Chau et al, CSIC (2005)
41 Fabrication of QWFET with HighHigh--KK
Fabrication challenges – Addition of two more unique materials to the stack (7 total) – Molecular Beam Epitaxy & Atomic Layer Deposition used – Atomic deposition precision needed across five critical layers
in partnership with IQE 42 Designing Materials with Smooth Grains
Polymer MG
D
Source: A. De Silva, et al. Adv. Mater. 2008
Polymer Blend Molecular glass + Mature materials platform + Higher sensitivity at same resolution - Larger individual components - Lower mechanical strength (currently)
Need to engineer materials with components below 1nm 43 Our limit to visibility goes out ~10 years TECHNOLOGY GENERATION 45nm 32nm 22nm 15nm 11nm 8nm Beyond 2007 2009 2011 2013 2015 2017 2020
MANUFACTURING DEVELOPMENT RESEARCH Carbon Nanotube ~1nm diameter
Graphene QW III-V Device 1 atom thick
5nm
Nanowire 10 atoms across y Silicon la ttice is ~ 0. 5nm, hard to imagi ne good dev ices smaller than 10 lattices across –reached in 2020 44 Beyond 2020 and possible futures y Conventional fabrication architectures continue – Individual steps continue as 2D layers – More and more layers stacked to give increasing function
High resolution Graphene layers can couple together Bilayer graphene structure TEM of graphene and create a quantum condensate Theore tica lly >10000x less power
Source: M. Gilbert et.al J Comput Electron (2009)
45 Beyond 2020 and possible futures y Conventional fabrication architectures continue – Individual steps continue as 2D layers – More and more layers stacked to give increasing function y Increasing use of heterogeneous technologies and novel ways to combine tthechnol ogi es – Electronics fabricated conventionally but combined with bio interface grown bottomsbottoms--upup – Eliminating, reducing cost of interfaces
Source: NRI, SWAN review 2009 46 Beyond 2020 and possible futures y Conventional fabrication architectures continue – Individual steps continue as 2D layers – More and more layers stacked to give increasing function y Increasing use of heterogeneous technologies and novel ways to combine tthechnol ogi es – Electronics fabricated conventionally but combined with bio interface grown bottomsbottoms--upup – Eliminating, reducing cost of interfaces y NonNon--binarybinary or alternate state computation ––SameSame fbfabrication compl exity, more val ue per function
47 Conclusions y Moore’s Law is not a law of nature, it is an expectation of continued innovation y We expect to continue through focused research, rapid development, investment in production y Scaling research is increasingly about materials research, solving problems brings opportunities y New product opportunities will arise from continued advances in integration, connectivity
48 Closing Thought
Thomas Edison: Opportunity is missed by most people because it is dressed in overalls and looks like work.
Source: National Archives and Records Administration 49 Thank You
50