5G mmWave base stations are targeting EIRPs of 60 to 65 dBm. Figure 3 compares the number of elements required in the Tx phased array to achieve 65 dBm EIRP, as a function of the average modulated output power per PA element, and the associated overall Tx array power dissipation. This figure has been compiled using publicly available information on output power and efficiency of PAs implemented in different technologies, and factors in the power consumption in the “plumbing” of the array, namely PA drivers, beamforming circuitry etc. It can be seen that at the lower end of per PA output power (corresponding to CMOS solutions), the number of array elements becomes very large, exceeding 500 elements and increasing implementation complexity and cost. On the other hand, at the higher end of per PA output power (corresponding to GaN), the number of array elements required becomes quite small (as low as 32 elements). This drives up the overall array power consumption, since the N2 array gain is small, implying that the EIRP is being achieved through raw output power generation. RF SOI CMOS sits in the “sweet spot” of complexity versus DC power, as the per PA output power achievable results in manageable array sizes of 256 elements and low overall Tx array power consumption.

So, there are three general areas where each technology has its strength. CMOS is best for very large arrays, because it achieves low cost when the power per amplifier is low. RF SOI occupies the “middle ground” where cost, power efficiency and output power are balanced, such as CPEs and urban mobile infrastructure. GaN comes into its own where the link is not uplink limited like point-to-point backhaul and at higher frequency bands (≥60 GHz) where high power per GaN amplifier can be leveraged for smaller array but still maintaining large point-to-point distance. Because mmWave networks will need to cover a wide variety of terrain and capacity requirements, it is clear that all three of these solutions will have a place: CMOS phased arrays are well suited for down link with large number of access points like in a stadium with large numbers of beams and no need to penetrate building walls or windows. GaN is well suited to long-distance transport network with wider beams and less steering requirements. RF SOI appears to be best for fixed-wireless CPEs and mobile infrastructure in urban environments.

DESIGN CHALLENGES FOR RF SOI mmWAVE FRONT-END MODULES

While RF SOI allows for superior Tx and Rx performance, as well as digital integration capability, there is significant opportunity for innovation in the circuit design as well as the system architecture, both of which are being actively pursued by companies such as MixComm (in partnership with GF in this case). On the Tx side, circuit approaches that extract best output power and efficiency from stacked SOI CMOS PAs while maintaining long-term reliability are critical. Even more important than peak efficiency is the average efficiency under modulation, with 5G NR waveforms typically dictating ~8 dB of back-off from the 1 dB compression point to achieve the required 3 percent EVM on the Tx side. In addition to the PA circuit design, the overall front-end module (FEM) architecture can also have a significant impact on the average system efficiency, and careful system architecture design and planning is critical.

Digital pre-distortion (DPD) may be employed to extract even better performance from the PAs, but is complicated by the fact that there will be both systematic and random variations between the PAs of a large-scale array due to process and temperature variations, as well as amplitude tapering in the beamforming. Therefore, new array DPD algorithms and PA architectures that are friendly to DPD will enable improved Tx performance.

The use of DPD is possible for massive MIMO arrays, but it must be a “light” DPD algorithm that consumes less DC power than is saved in the PA. For very large arrays with low power per PA path, DPD may not be worthwhile, but for small arrays, at higher RF power levels, DPD may become an important element. One strong possibility here is to have a DPD algorithm and adaptation engine that is shared among multiple RF paths, essentially updating the DPD algorithm periodically instead of continuously to save on cost and power devoted to a single PA path. This approach sacrifices the level of linearization but improves efficiency which is a more important.

Large-scale phased arrays are subject to amplitude and phase mismatches arising from process, temperature and package interface variations across the channels of a single FEM chip, as well as across chips. Built-in self-test and calibration approaches that can compensate for these mismatches are important for the realization of robust and accurate large-scale arrays. Implementation of these techniques allows for a self-aligning array, which adapts to field conditions and manufacturing variation to optimize performance in the critical RF front-end.

HOW RF SOI CAN ADDRESS CARRIERS' CHALLENGES

Figure 4

Figure 4 Comparison of median CPE uplink throughput rates for a baseline bulk CMOS-based CPE array and a MixComm GF 45 nm RF-SOI-based CPE array.

Transmitter output power is perhaps the most fundamental metric of a radio, and higher output power can be used to improve virtually every dimension of a mmWave link. Higher output power increases range, which translates to large cell radius, and consequently fewer base stations can be deployed, reducing operator CAPEX. Alternatively, for the same cell radius, it enables higher rates at the cell edge, improving quality of service. Higher per PA output power can be used by beamforming algorithms to enable “broad beams,” as opposed to the conventional narrow “pencil beams,” thus improving robustness in highly mobile scenarios. Higher per PA output power can also be used to reduce BOM cost, as a smaller array is needed to achieve the same EIRP. A smaller array also comes with natural beam broadening and associated robustness. Figure 4 compares the median CPE uplink throughput rate of a baseline bulk CMOS-based CPE array with a DPD 45 nm RF SOI-based CPE array with equal number of antennas. It can be seen that the higher output power per PA enables significantly superior link budgets, allowing a 2.7x increase in throughput rate.

WHAT'S THE IMPACT OF AN IMPROVEMENT IN THE mmWAVE AMPLIFIER?

Three challenges will dictate the success of mmWave for mobility: network cost, thermal/power budget and BOM cost. Our detailed review of semiconductor fundamentals illustrates that RF SOI brings advantages in all of these areas.

  • Higher transmitter power has a huge impact on the financial case for the operators. Adding 3 dB higher EIRP to a CPE can save 20 percent of the cost of network deployment, by allowing base stations to be deployed farther apart, and also providing higher spectral efficiency. That’s billions of dollars of savings at the network level, plus a bonus of higher capacity.
  • Almost all radios in the market today are limited by their thermal profile. Improvements to the PA efficiency have a direct impact on the real-world EIRP that is achieved. RF SOI sits in a “sweet spot” for thermal performance compared with other technologies, allowing for tradeoffs of power, linearity and efficiency that far outperform the bulk CMOS used in many CPEs today.
  • RF SOI-based radios can achieve high transmitter power without using hundreds of array elements. The RF SOI process allows for integration of the PA, LNA, and up/downconverter, keeping the BOM cost low and the supply chain simple.

Overall, it’s clear that weak power in the uplink presents the biggest problem to 5G operators today. Simply upgrading CPEs to use RF SOI amplifiers can boost uplink EIRP by 3 dB or more, improving both coverage and capacity of the network. Other products such as gNodeB arrays and handsets can also benefit in similar ways. CMOS phased arrays are best suited for down link with large number of access points like in a stadium with large numbers of beams and no need to penetrate building wall or windows and GaN is well suited to long-distance transport network with wider beams and less steering requirements.

References

  1. Industry Voices - Madden: Operators will need mmWave spectrum for 5G capacity, Fierce Wireless, https://www.fiercewireless.com/wireless/industry-voices-madden-operators-will-need-mmWave-spectrum-for-5g-capacity.
  2. We Tested 5G Across America. It’s Crazy Fast - and a Hot Mess, WSJ, https://www.wsj.com/articles/all-the-reasons-not-to-buy-a-5g-phone-right-now-11563467389.