5G Fixed Wireless Access Array and RF Front-End Trade-Offs Bror Peterson and David Schnaufer Qorvo, Greensboro, N.C

Cover Feature Invited Paper

Editor’s Note: At the end of December, the 3GPP approved the 5G non-standalone new radio (NSA NR) specification, which defines how enhanced broadband services can be deployed using a 5G NR leveraging the existing LTE network. This NSA architecture will first be fielded—later this year—for fixed wireless access (FWA) services using mmWave spectrum, i.e., 28 and 39 GHz. Qorvo and Anokiwave are two companies leading the development of the mmWave front-end technology for the active phased arrays that will power these FWA services. Each company has analyzed the system requirements and defined a unique approach to meeting them. Qorvo has chosen GaN, Anokiwave silicon. We are fortunate that this issue of Microwave Journal features articles from both, each stating the case for its technology choice. Regardless of which argument you favor, no doubt you will agree that both companies are doing excellent technology and product development, a key step to making 5G viable. 5G Fixed Wireless Access Array and RF Front-End Trade-Offs Bror Peterson and David Schnaufer Qorvo, Greensboro, N.C.

he vision of next-gener- gigabit fixed wireless access (FWA) testbed toward a truly mobile ation 5G networks is to services to houses, apartments broadband experience. Not surpris- deliver an order-of-magni- and businesses, in a fraction of the ingly, Verizon, AT&T and other car- tude improvement in ca- time and cost of traditional cable riers are aggressively trialing FWA, Tpacity, coverage and connectivity and fiber to the home installations. with the goal of full commercializa- compared to existing 4G networks, Carriers are also using FWA as the tion in 2019. all at substantially lower cost per bit to carriers and consumers. The Device-to-Device Communications many use cases and services en- Automobile-to-Automobile Densi cation Communications abled by 5G technology and networks are shown in Figure 1. In this Smart Grid first phase of 5G new radio (NR) standardization, the primary focus Smart Home has been on defining a radio access Enhanced Mission Critical technology (RAT) that takes advan- Mobile Broadband Services tage of new wideband frequency Fixed Wireless Access allocations, both sub-6 GHz and above 24 GHz, to achieve the huge Massive Internet Critical/Emergency Broadcast on of Things Services peak throughputs and low latencies Mobile Device proposed by the International Mo- bile Telecommunications vision for Augmented 2020 and beyond.1 Reality & Virtual Reality Mobile network operators are capitalizing on the improvements IoT Smart Cities introduced by NR RAT, particularly Machine-to-Machine in the mmWave bands, to deliver  Fig. 1 5G use cases.

Fig. 2 Global 5G bands above 24 GHz. • Random Dallas Suburb  - 800 Houses/km2 - 500 m ISD - 9 Cell Sites - 23 Sectors - ~35 Houses/Sector Active Antenna System • Capacity Per Sector - 35 Houses/Sector - 5x Oversubscription - 1 Gbps Service - Capacity ~5 Gbps • BTS Parameters - Capacity ~5 Gbps

Customer Premise - 400 MHz BW Equipment - 16-QAM w/LDPC: 3 bps/Hz Mobile - 4 Spatial Streams/Layers Equipment Customer Premise Equipment • Business Case - 35% Take Rate - $100/Month for 1 Gbps SLA Central Data - $14k/Sector/Year Edge Center - $177k/km2/Year Data Center  Fig. 4 FWA in a suburban  Fig. 3 End–to–end FWA network. environment. In this article, we analyze the ar- and come in several form factors: all tions that maximize range, ensure chitecture, semiconductor technol- outdoor, split-mount and all indoor initial customer satisfaction and al- ogy and RF front-end (RFFE) de- desktop and dongle-type units. Mo- low time for BTS and CPE equip- sign needed to deliver these new bile-handset form factors will follow. ment to reach the needed cost and mmWave FWA services. We discuss Global mmWave spectrum avail- performance targets. the link budget requirements and ability is shown in Figure 2. In the Large coverage is essential to the walk through an example of subur- U.S., most trials are in the old block success of the FWA business case. ban deployment. We address the A LMDS band between 27.5 and To illustrate this, consider a subur- traits and trade-offs of hybrid beam- 28.35 GHz, but the plan-of-record ban deployment with 800 homes/ forming versus all-digital beamform- of carriers is to deploy nationwide km2, as shown in Figure 4. For BTS ing for the base transceiver station in the wider 39 GHz band, which is inter-site distance (ISD) of 500 m, we (BTS) and analyze the semiconduc- licensed on a larger economic area need at least 20 sectors, each cov- tor technology and RFFE compo- basis. These candidate bands have ering 35 houses from nine cell sites. nents that enable each. Finally, we been assigned by 3GPP and, except Assuming 33 percent of the custom- discuss the design of a GaN-on-SiC for 28 GHz, are being harmonized ers sign up for 1 Gbps service and a front-end module (FEM) designed globally by the International Tele- 5x network oversubscription ratio, an specifically for the 5G FWA market. communications Union.2 average aggregate BTS capacity of 3 FWA describes a wireless con- Gbps/sector is needed. This capacity FWA DEPLOYMENT nection between a centralized sec- is achieved with a 400 MHz band- A clear advantage of using torized BTS and numerous fixed or width, assuming an average spectral mmWave is the availability of un- nomadic users (see Figure 3). Sys- efficiency of 2 bps/Hz and four layers derutilized contiguous spectrum at tems are being designed to lever- of spatial multiplexing. If customers low cost. These bands allow wide age existing tower sites and support pay $100 per month, the annual rev- component carrier bandwidths up a low-cost, self-install CPE build- enue will be $280,000/km2/year. Of to 400 MHz and commercial BTSs out. Both are critical to keeping course, without accounting for re- are being designed with carrier ag- initial deployment investment low curring costs, it is not clear FWA is a gregation supporting up to 1.2 GHz while the business case for FWA is good business, but we can conclude of instantaneous bandwidth. Cus- validated. Early deployments will be that as ISD increases, the business tomer premise equipment (CPE) mostly outdoor-to-outdoor and case improves. To that end, carriers will support peak rates over 2 Gbps use professional roof-level installa- are driving equipment vendors to CoverFeature

TABLE 1

FCC POWER LIMITS FOR 28 AND 100 39 GHz BANDS 500 m ISD ~333 m Cell Range 90 Equipment Class Power (EIRP) 80 Pro Install Self Install CPE NEC 70 CATT Qualcomm Base Station 75 dBm/100 MHz 60 ZTE 50 Huawei Mobile Station 43 dBm 40 Samsung Ericsson 30 Transportable 55 dBm Intel 20 China Telecom P athloss is L ess T han A bscissa (%) Station 10 0 build BTS and CPE equipment that 180 160 140 120 100

robability robability Path Loss (dB) operate up to regulatory limits to P maximize coverage and profitability. In the U.S., the Federal Com-  Fig. 5 Statistical path loss simulation for urban-macro environment with 500 m ISD. munications Commission has defined very high effective isotropic sive professional roof-level installa- radiated power (EIRP) limits for the tions. The distribution curve shows 3 75 28 and 39 GHz bands, shown in the maximum system path loss to 160 dB Table 1. The challenge becomes be 165 dB. 70 165 dB

EIRP ( d B m ) 170 dB building systems that meet these Closing the link depends on 65 targets within the cost, size, weight many variables, including transmit 60

ransmit 5 10 15 20 25 30 35 and power budgets expected by EIRP, receive antenna gain, receiv- T carriers. Selecting the proper front- er noise figure (NF) and minimum Receive G/NF (dB) end architecture and RF semicon- edge-of-coverage throughput. To ductor technology are key to get- avoid overdesign of the cost-sen-  Fig. 6 Transmit EIRP and receive G/ ting there. sitive CPE equipment and shift the NF vs. path-loss for 1 Gbps edge-of- coverage throughput. FWA Link Budget burden toward the BTS, the link design begins at the CPE receiver The standards community has and works backward to arrive at the sustain a 1 Gbps link at 165 dB of been busy defining the perfor- BTS transmitter requirements. In path loss when the CPE receiver G/ mance requirements and evaluat- lieu of the conventional G/T (the ra- NF is ≥ 21 dBi. ing use cases over a broad range of tio of antenna gain to system noise Next, we consider the impact of mmWave frequencies. The urban- temperature) figure-of-merit (FOM), receiver NF by plotting the mini- macro scenario is the best represen- we define a more convenient G/NF mum number of array elements tation of a typical FWA deployment: FOM: the peak antenna gain (includ- needed to achieve G/NF of 21 dB having large ISD of 300 to 500 m and ing beamforming gain) normalized (see Figure 7). We also plot the to- providing large path-loss budgets by the NF of the receiver. Figure 6 tal low noise amplifier (LNA) power that overcome many of the propa- illustrates the required EIRP for the consumption. By adjusting the axis gation challenges at mmWave fre- range of receive G/NF to overcome range, we can overlap the two and quencies. To understand the need- a targeted path loss delivering an see the impact NF has on array size, ed link budget, consider a statistical edge-of-coverage throughput of complexity and power. For this ex- path-loss simulation using detailed 1 Gbps, assuming the modulation ample, each LNA consumes 40 mW, large-scale channel models that ac- spectral efficiency is effectively which is typical for phased arrays. count for non-line-of-site conditions 2 bps/Hz and demodulation SNR is The NFs of RFFEs, including the T/R and outdoor-to-indoor penetra- 8 dB. From the graph, the BTS EIRP switch losses, are shown for 130 nm tion, like those defined by 3GPP.4 for a range of CPE receiver’s G/NF SiGe BiCMOS, 90 nm GaAs PHEMT Figure 5 shows the result for a 500 can be determined. For example, and 150 nm GaN HEMT at 30 GHz. m ISD urban-macro environment 65 dBm BTS EIRP will be needed to The compound semiconductor performed by equipment vendors and operators. For this simulation, 256 28 GHz channel models were used 30 T otal LNA P dc ( W ) with 80 percent of the randomly 192 dropped users falling indoors and 20 20 percent outdoors. Of the indoor 128 SiGe users, 50 percent were subject to 1.5 dB 10 64 high penetration-loss models and GaAs/GaN # of A rray E lements 50 percent lower loss. Long-term, 0 0 carriers desire at least 80 percent 2 3 4 5 6 7 8 of their potential users to be self- Noise Figure (dB) installable to minimize more expen-  Fig. 7 Array size vs. front-end NF and power consumption for G/NF = 21 dB. CoverFeature

TABLE 2 15-25 m APPROXIMATE PERFORMANCE FOR CORPORATELY FED ELEMENTS (a) Column Array Size Beamwidth (°) Gain (dB) Single Element 102 5

2-Element 51 8

4-Element 26 11

(b) 8-Element 13 14  Fig. 8 Array complexity depends on the scanning range needed for the Let’s start by first understanding per-element array supports wider deployment: suburban (a) or urban (b). the azimuth and elevation scanning scan angles but needs 4x as many requirements and whether two-di- PAs, phase shifters and variable mensional beamforming is required gain components for an antenna for a typical FWA deployment or if with four elements. To achieve the a lower complexity, one-dimensional same EIRP, the PA driving a column- (azimuth only) beamforming array is fed array with four antennas will sufficient. This decision impacts the need to provide at least 4x the out- power amplifier (PA).Figure 8 shows put power, which can easily change two FWA deployment scenarios. In the semiconductor selection. It is the suburban deployment, the tower reasonable to assume a suburban • Nx Fewer Components heights range from 15 to 25 m and BTS will use antennas with 6 to 9 dB • Nx Larger PA the cell radius is 500 to 1000 m, with • Higher Feed Losses higher passive antenna gain com- • Fixed Elevation Pattern an average house height of 10 m. pared to an urban deployment. As (a) Just as with traditional macro cellu- a result, the phased array needs far lar systems, there is no need for fully fewer active channels to achieve the adaptive elevation scanning. The el- same EIRP, significantly reducing ac- evation beam can be focused down tive component count and integra- by corporately feeding several pas- tion complexity. sive antenna elements, as shown in Figure 9a. This vertically stacked Array Front-End Density

1:4 S plitter column of radiating elements is de- Early mmWave FWA BTS designs signed to minimize radiation above used separate, single-polarization the houses and fill in any nulls along transmit and receive antenna arrays, the ground. Further, the gain pat- which allowed significantly more • Nx More Components board area for components. These de- • Nx Smaller PAs tern is designed to increase at rela- • Lower Feed Losses tively the same rate as the path loss. signs avoided the additional insertion • Elevation Beam Steering loss and linearity challenges of a T/R (b) This provides more uniform coverage for both near and far users. The switch. However, a major architecture  Fig. 9 Column-fed (a) and per- nominal half-power beamwidth can trend is integrated T/R, dual-polariza- element (b) active arrays. be approximated as 102°/NANT and tion arrays (see Figure 10), which is the array gain by 10log (N ) + driving RFFE density. The key rea- technology provides ≥ 1.5 dB ad- 10 ANT 5 dBi. With passively combined an- son is spatial correlation. Adaptive vantage, translating to a 30 percent tennas, the elevation beam pattern beamforming performance depends savings in array size, power and, ul- is focused and the fixed antenna on the ability to calibrate the receive timately, CPE cost. gain increases, as shown in Table 2. and transmit arrays relative to one To explore architecture trades For the suburban FWA deployment, another. As such, it is important to that are key to technology selec- a 13 to 26 degree beamwidth is suf- integrate the transmit and receive tion and design of the RFFE com- ficient, with the passively combined channels for both polarizations, so ponents, we start by understanding column array from four to eight the array shares a common set of the antenna scanning requirements. elements. In the urban scenario, antenna elements and RF paths. The We highlight the circuit density and however, the elevation scanning re- net result is a requirement for the packaging impact for integrated, quirements are greater, and systems RFFE to have 4x the circuit density of dual-polarization receive/transmit will be limited to one or two passive earlier systems. arrays. Finally, we investigate all- elements. At mmWave frequencies, the digital beamforming and hybrid RF Figure 9b illustrates the per- lattice spacing between phased- beamforming architectures and the element active array. Both the per- array elements becomes small, e.g., requirements for each. element and column-fed array ar- 3.75 mm at 39 GHz. To minimize 1D or 2D Scanning chitectures have the same antenna feed loss, it is important to locate The number of active channels in gain, but the column-fed array has the front-end components close to the array depends on many things. a fixed elevation beam pattern. The the radiating elements. Therefore, it CoverFeature is necessary to shrink the RFFE foot- ALL-DIGITAL VS. HYBRID between the number of baseband print and integrate multiple func- ARRAYS channels and the number of active tions, either monolithically on the It was natural for BTS vendors to RF channels. This approach better die or within the package, using a first explore extending the current, balances analog beamforming gain multi-chip module. Tiling all these sub-6 GHz, all-digital beamform- and baseband processing. The fol- functions in a small area requires ing, massive MIMO platforms to lowing sections analyze the two either very small PAs, requiring a mmWave. This preserves the basic architectures and discuss the RFFE many-fold increase in array size, or architecture and the advanced sig- approaches needed for each. using high-power density technolo- nal processing algorithms for beam- Digital Beamforming gies like GaN. Further, it is critical formed spatial multiplexing. How- to use a semiconductor technology ever, due to the dramatic increase Assuming large elevation scan- that can withstand high junction in channel bandwidths offered by ning is not required for suburban temperatures. The reliability of SiGe mmWave and the need for many FWA and a well-designed, column degrades rapidly above 150°C, but active channels, there is a valid antenna provides gain of up to GaN on SiC is rated to 225°C. This concern that the power dissipation 14 dBi, we start with a mmWave 75°C advantage in junction temper- and cost of such a system would be BTS transceiver design targeting an ature has a large impact on the ther- prohibitive. Therefore, vendors are EIRP of 65 dBm and compute the mal design, especially for outdoor, exploring hybrid beamformed ar- power consumption using off-the- passively-cooled phased arrays. chitectures,5 which allows flexibility shelf point-to-point microwave radio

T Array R Array T/R Array Dual-Polarization T/R Array

Isolation 10 cm > 40 dB

1:N Splitter 1:N Combiner 1:N Combiner/Splitter 2x the Circuit Density 1:N Combiner/Splitter 4x the Circuit Density

Transitioning From Separate Arrays Integrated T/R Integrated T/R and Dual Polarization

 Fig. 10 FWA antenna arrays are evolving from separate T and R arrays to integrated T/R arrays with dual polarization.

RF-DAC Hybrid IQ Mixer 9 W GaN PA 14-bit 4.5 Gbps DVG A VGA Circulator IRF BPF Driver

RF-DAC IQ Mixer 14-bit 4.5 Gbps DVG A Hybrid 9 W GaN PA Circulator IRF BPF VGA Driver C orporate F eed DUC DAC 90° C olumn- A ntenna 0°

RF-ADC C olumn- A ntenna 14-bit 3 Gbps Gain Block DVG A LNA + IQ Mixer AAF BPF C olumn- A ntenna JES D204B C orporate F eed C olumn- A ntenna DDC ADC 90°

 Fig. 11 Array design using digital beamforming and commercial, off-the-shelf components. CoverFeature components that have been avail- Hybrid Beamforming There is an important trade un- able for years, including a high-pow- The basic block diagram for folding, whether SiGe front-ends er, 28 GHz GaN balanced amplifier. a hybrid beamforming active ar- can provide sufficient output power The multi-slat array and transceiver ray is shown in Figure 14. Here, N and efficiency to avoid the need for are shown in Figure 11. Assuming baseband channels are driving RF higher performance III-V technol- circulator and feed-losses of 1.5 dB, analog beamformers, which divide ogy like GaAs or GaN. With good the power at the antenna port is the signal M-ways and provide dis- packaging and integration, both 27 dBm. From the following equa- crete phase and amplitude control. approaches can meet the tight an- tions, achieving 65 dBm EIRP re- FEMs drive each M-element subar- tenna lattice-spacing requirements. quires 16 transceivers that, com- ray panel. The number of baseband bined, provide 12 dB of digital paths and subarray panels is de- Tx Total/Channel = 13 W beamforming gain: termined by the minimum number of spatial streams or beams that Other: 0.5 RF-DAC: 1 EIRP=+ GBF () dB GANT ( dBi ) +are needed. The number of beam- DVGA: 0.5 PAVE _ TOTAL () dBm former branches and elements in each subarray panel is a function Final PA: VGA: 1.2 EIRP=+ 10log10() NCOLUMNS of the targeted EIRP and G/NF. 8.8 10log10() NPAs++ G ANT While a popular design ratio is to Driver: 1 PAVE/CHANNEL ()dBm have one baseband path for every 16 to 64 active elements, it really The power consumption for each depends on the deployment sce- (a) transceiver is shown in Figure 12. nario. For example, with a hot-spot Rx Total/Channel = 4 W The total power dissipation (PDISS) small cell (or on the CPE terminal at 80 percent transmit duty cycle for side), a 1:16 ratio single panel is ap- Down-Converter/LNA: 0.8 all 16 slats will be 220 W per polar- propriate. A macro BTS would have ization, and a dual-polarized system two to four subarray panels with 64 will require 440 W. For all outdoor active elements, where each panel RF-ADC: tower-top electronics, where pas- is dual-polarized, totaling four to 2.2 sive cooling is required, it is chal- eight baseband paths and 256 to DVGA: 0.9 lenging to thermally manage more 512 active elements. The digital than 300 W from the RF subsystem, and analog beamforming work to- Gain Block: 0.2 suggesting an all-digital beamform- gether, to maximize coverage or (b) ing architecture using today’s off- independently, to provide spatially the-shelf components is impractical. separated beams to multiple users.  Fig. 12 Power dissipation of the However, new GaN FEMs are on transmit (a) and receive (b) chains. the horizon to help address this. As shown in Figure 13, the GaN PAs integrated in the FEM apply the tried- and-true Doherty efficiency-boost- ing technique to mmWave. With Doherty PAs, digital pre-distortion (DPD) is needed; however, the ad- jacent channel power ratio (ACPR) Transceiver requirements defined for mmWave bands are significantly more re- laxed, enabling a much “lighter” C orporate F eed DPD solution. The estimated power dissipation of a 40 dBm PSAT, symmetric, multi-stage Doherty PA can (a) be reduced more than 50 percent. In the above system, this improve- 40 ment alone drops the total PDISS 36 below 300 W. Combined with 32 power savings from next-genera- 28 24 tion RF-sampling digital-to-analog 20 and analog-to-digital converters, 16 advancement in mmWave CMOS 12 24 26 28 30 32 34 36 38 40 42 transceivers and increased levels of (%) A dded E f ciency P ower- Output Power (dBm) small-signal integration, it will not (b) be long before we see more all-digital beamforming solutions being  Fig. 13 Integrated FEM with symmetric GaN Doherty PA and switch-LNA (a) and PA deployed. performance from 27.5 to 29.5 GHz (b). CoverFeature

RF Beamformer Front-Ends Digital Processing Mixed Signal IF-RF Conversion

D/A Subarray DUC Panel 1 DUC A/D

N:Number of 1:M/N Baseband Channels M:N DUC D/A D igital B eamformer DUC A/D Subarray Panel N

CMOS SiGe-BiCMOS GaAs-/GaN

 Fig. 14 Active array using hybrid beamforming.

As array size gets large (more than 50 35 EIRP = 65 dBm 512 active elements), the power per

45 A ntenna rray G ain (d B i) element becomes small enough to 40 f = 28 GHz 30 35 y/2 = 5.4 mm allow SiGe, which can be integrated GaN emax = 90% 30 25 into the core beamformer RFIC. In 4πemaxDarray2 25 Array Gain ≈ contrast, by using GaN for the front- y2 20 20 end, the same EIRP can be achieved GaAs 15 with 8 to 16x fewer channels. 10 15 System Power Dissipation 5 SiGe

verage T x P ower per E lement (d B m) 0 10 For an array delivering 64 dBm A 32 64 96

128 256 512 EIRP, Figure 16 shows an analysis Number of Active Elements 1024 of the total PDISS of the beamformer  Fig. 15 Optimum RFFE technology vs. array size. plus the front-end as a function of the number of active elements in each subarray panel. The P is 110 40.0 DISS Element Gain = 8 dBi EVM = 8% EVM = 6% shown for several error vector mag- 100 35.0

EVM = 4% P nitude (EVM) levels, since the EVM

GaN av e Pave/Channel 30.0

90 / C hannel ( d B m ) determines the power back-off and 25.0 efficiency achieved by the front- 80 20.0 end. We assume each beamformer 70 15.0 branch consumes 190 mW, which is 2-Stage the typical power consumption of 60 GaAs 10.0 P ower D issipated (W) core beamformers in the market.6 3-Stage 50 SiGe 5.0 The system on the far right of the 40 0 figure represents an all-SiGe solu- 16 40 64 88

112 136 160 184 208 232 256 280 304 328 352 376 400 424 448 472 496 512 tion with 512 elements, with an out- Number of Active Channels put power per element of 2 dBm  Fig. 16 System power dissipation vs. array size and EVM for 64 dBm EIRP. and consuming approximately 100 W. Moving left, the number of elements decreases, the PAVE per chan- FRONT-END SEMICONDUCTOR of array size and antenna gain for a nel increases and PDISS is optimized CHOICES uniform rectangular array delivering to a point where beamforming gain The technology choice for the 65 dBm EIRP. The graph is overlaid starts to roll off sharply, and the with an indication of the power rang- RFFE depends on the EIRP and PDISS to maintain the EIRP rapidly G/NF requirements of the system. es best suited for each semiconduc- increases. The small steps in the dis- Both are a function of beamforming tor technology. The limits were set sipation curves represent where the gain, which is a function of the ar- based on benchmarks of each tech- front-end transitions from a single ray size. To illustrate this, Figure 15 nology, avoiding exotic power-com- stage to two-stage and three-stage bining or methods that degrade shows the average PA power (PAVE) designs to provide sufficient gain. per channel needed as a function component reliability or efficiency. As stages are added, the efficiency CoverFeature

TABLE 3 RELATIVE COST OF ALL SiGe AND SiGe BEAMFORMER WITH GaN FEM Parameter Units All SiGe GaN +SiGe

Average Output Power 1875 µm dBm 2 20 per Channel Power Dissipation per mW 190 1329 Channel 2700 µm Antenna Element Gain dBi 8 8 (a) Number of Active 512 64 Channels LNA EIRP dBmi 64 64 Rx1 SW PA ANT1 Total Power Dissipation W 97 97 Tx1 6 mm Beamformer Die Area per mm2 2.3 2.3 LNA Channel Rx2 SW ANT2 Front-End Die Area per PA mm2 1.2 5.2 Channel Tx2

2 Total SiGe Die Area mm 1752 144 4.5 mm (b) Total GaN Die Area mm2 0 334

Die Cost Units Notes

All SiGe System Die Cost 1752 $/x GaN + SiGe System Die 1647 $/x 4-inch GaN = 4.5x Cost (4-inch GaN) GaN + SiGe System Die 1146 $/x 6-inch GaN = 3x Cost (6-inch GaN) drops with the increase in power themselves. Using compound semi- dissipation. conductor front-ends allows an im- mediate 8x reduction in array size Designing to optimize system (c) with no increase in P . Even with PDISS without regarding com- DISS lower-cost printed antenna tech-  Fig. 17 Integrated 39 GHz GaN front- plexity or cost, an array of about end MMIC – intentionally blurred (a), 128 elements with a two-stage, nology, this is a large saving in ex- dual-channel FEM (b) and package (c). pensive antenna-quality substrate 14 dBm output PA (24 dBm P1dB) is the best choice. However, if we material. Considering component 2 lowest NF FEM for the 37 to 40 GHz strive to optimize cost, complex- cost, the current die cost per mm of band. To support the trend to inte- ity and yield for a P budget of 150 nm GaN on SiC fabricated on DISS grated transmit/receive arrays, the under 100 W, the optimum selec- 4-inch wafers is only 4.5x the cost front-end includes a PA, integrated tion is the range of 48 to 64 active of 8-inch 130 nm SiGe. As 6-inch T/R switch and a low NF LNA. The channels using a three-stage GaN GaN production lines shift into high module was designed with sufficient PA with an average output power volume, the cost of GaN relative to gain to be driven by core beamform- of 20 to 23 dBm, depending on the SiGe drops to 3x. A summary of the er RFICs, which have a typical drive EVM target. The trends shown in assumptions and a cost comparison level of 2 dBm. The FEM’s P of 23 Figure 16 are less a function of PA of the relative raw die cost of the two AVE dBm was selected from an analysis efficiency and more a function of technologies is shown in Table 3. Us- similar to that shown in Figure 16, and beamformer inefficiency. In other ing a high-power density compound the P was determined by analyz- words, the choice to increase array semiconductor like GaN on 6-inch SAT ing the needed headroom to support size 8x to allow an all-SiGe solution wafers can save up to 35 percent a back-off linearity of ≥ 33 dBc ACPR, comes with a penalty, given that in the raw die cost relative to an all- EVM ≤ 4 percent and a 400 MHz or- the input signal is divided many SiGe architecture. Even though the thogonal frequency-division multiple more ways and requires linearly bi- cost of silicon technologies is lower access (OFDMA) waveform. ased, power consuming devices to per device, the cost of the complete A key design decision was deter- amplify the signal back up. system is significantly higher. mining if GaAs or GaN or a combi- Cost Analysis GaN FRONT-END MODULES nation of both were needed. The The cost of phased arrays include To validate the concept of a GaN die size for a GaAs PA would not al- the RF components, printed circuit FEM for mmWave FWA arrays, Qorvo low the FEM to meet the tight 3.75 board material and the antennas set out to design the highest power, mm lattice spacing at 39 GHz. The CoverFeature

equivalent output power GaN PA is gain. In receive mode, the NF is 4.1 Vision–Framework and Overall Objectives of the Future Development of IMT for 2020 and 4x smaller with no sacrifice in gain dB, and receive gain is 16 dB. The Beyond,” August 2015, www.itu.int/dms_pu- and a slight benefit in efficiency. package size is 4.5 mm × 6.0 mm × brec/itu-r/rec/m/R-REC-M.2083-0-201509- Considering the LNA, the 90 nm 1.8 mm.7-8 I!!PDF-E.pdf. 2. International Telecommunications Union, Res- GaAs PHEMT process was favored olution 238 (WRC-15), “Studies on Frequen- due to its slightly superior NF. How- SUMMARY cy-Related Matters for International Mobile Telecommunications Identification Including ever, the net improvement was only FWA is rapidly approaching com- Possible Additional Allocations to the Mobile a few tenths of a dB once the addi- mercialization. This is due to the Services on a Primary Basis in Portion(s) of the tional bond wires and 50 Ω match- abundance of low-cost spectrum, Frequency Range 24.25 and 86 GHz for Fu- ture Development of IMT-2020 and Beyond,” ing networks were considered. The early regulatory and standards work 2015, www.itu.int/dms_pub/itu-r/oth/0c/0a/ trade-off analysis concluded it was and the opportunity for operators R0C0A00000C0014PDFE.pdf. better to stay with a monolithic GaN to quickly tap a new market. The re- 3. Federal Communicationws Commission, “Use of Spectrum Bands Above 24 GHz for Mobile design that allowed co-matching of maining challenge is the availability Radio Services, In the Matter of GN Docket the PA, LNA and T/R switch. Such a of equipment capable of closing the No. 14-177, IB Docket No. 15-256, RM-11664, WT Docket No. 10-112, IB Docket No. 97-95,” design was lower risk, easier to as- link at a reasonable cost. Both hybrid July 2016, apps.fcc.gov/edocs_public/attach- semble and test, and the MMIC was beamforming and all-digital beam- match/FCC-16-89A1.pdf. as compact as possible. The system forming architectures are being ex- 4. 3GPP TR 38.901, “Study on Channel Model for Frequencies from 0.5 to 100 GHz,” Sep- thermal analysis indicated that the plored. These architectures capitalize tember 2017, www.3gpp.org/ftp//Specs/ar- higher junction temperature offered on the respective strengths of com- chive/38_series/38.901/38901-e20.zip. by GaN-on-SiC was critical for pas- mercial semiconductor processes. 5. A. F. Molisch et al., “Hybrid Beamforming for Massive MIMO: A Survey,” IEEE Communi- sively-cooled arrays. The use of GaN front-ends in either cations Magazine, Vol. 55, No. 9, 2017, pp. As shown in Figure 17, the approach provides operators and 134–141. 6. B. Sadhu et al., “7.2 A 28GHz 32-Element 39 GHz FEM integrates two of the manufacturers a pathway to achiev- Phased-Array Transceiver IC with Concurrent multi-function GaN MMICs into an ing high EIRP targets while minimiz- Dual Polarized Beams and 1.4 Degree Beam- air-cavity, embedded heat-slug, ing cost, complexity, size and power steering Resolution for 5G Communication,” 2017 IEEE International Solid-State Circuits surface-mount package, sized to dissipation. To prove the feasibility, Conference (ISSCC), San Francisco, Calif, meet the array element spacing at Qorvo has developed a 39 GHz FEM 2017, pp. 128–129. 39 GHz. Each of the GaN MMICs based on a highly integrated GaN- 7. B. Kim and V. Z. Q. Li, “39 GHz GaN Front-End MMIC for 5G Applications,” 2017 IEEE Com- contains a three-stage linear PA, on-SiC T/R MMIC and is develop- pound Semiconductor Integrated Circuit Sym- three-stage LNA and a low-loss, ing similar FEMs for other millimeter posium (CSICS), Miami, Fla., 2017, pp. 1–4. 8. “QPF4005 37-40.5 GaN Dual Channel FEM high-linearity SPDT switch. The wave frequency bands proposed for Datasheet,” www.qorvo.com/products/d/ FEM covers 37.1 to 40.5 GHz and 5G systems.n da006271. provides 23 dBm average output power, which supports 256-QAM References 1. International Telecommunications Union, ITU- EVM levels, with 24 dB transmit R Radiocommunications Sector of ITU, “IMT