Choosing mmWave semiconductor technologies for a wireless transmission system is a system-level decision, not simply a comparison of transistor performance. Backhaul, fronthaul and fixed broadband links impose different demands on output power, noise figure, phase noise, modulation, integration and cost. The appropriate process depends on the operating band, link distance, data requirements and expected production volume.
Available technologies span compound semiconductors such as GaAs and silicon platforms including SiGe, BiCMOS and CMOS. Each offers a different balance of RF performance, integration capability and economics. Packaging and assembly also matter: at millimetre-wave frequencies, they can affect both module performance and manufacturability.
How mmWave semiconductor technologies differ
GaAs has historically been used for millimetre-wave front ends, particularly power amplifiers (PAs) and low-noise amplifiers (LNAs). Its RF performance makes it well suited to demanding links that need strong transmit power, low receiver noise or high-order modulation. The trade-off is that building complex logic and highly integrated functions in GaAs is difficult, so a complete radio may require multiple chips and more involved assembly.
SiGe and SiGe BiCMOS provide a different approach. SiGe heterojunction bipolar transistors support high-frequency analog functions, while BiCMOS combines bipolar devices with CMOS logic on one chip. This enables greater integration of RF, analog and control functions, potentially reducing external components and simplifying module assembly. SiGe solutions can be attractive where performance requirements are moderate or where integration and bill of materials are important.
CMOS can support even higher levels of integration, including RF, converters and digital functions. Its cost advantages depend on achieving sufficient volume to justify development and manufacturing investment. The source comparison identifies lower transmit power and efficiency, as well as device-isolation limitations, as considerations for CMOS at mmWave frequencies. Fully depleted SOI (FD-SOI) is another developing silicon option, with low-voltage operation and body-bias control offering potential benefits for power and circuit performance.
In general, silicon does not match III-V materials such as GaAs for low-noise performance, linearity and output power at higher frequencies. But silicon-based RF ICs can integrate application-specific functions and calibration more readily. A hybrid architecture can therefore make sense: use compound-semiconductor devices for the most demanding transmit and receive functions, and silicon for mixed-signal, control and frequency-conversion functions.
Match the RF platform to the link
Macro-cell backhaul and fronthaul
Longer macro-cell links place demanding requirements on transmit power, receiver sensitivity and modulation quality. E-band systems using high-order QAM need sufficient linearity and phase-noise performance to preserve signal quality. The whitepaper describes GaAs as a common choice for demanding RF front ends, while combinations of SiGe or BiCMOS with GaAs PAs, LNAs or voltage-controlled oscillators (VCOs) can divide functions according to their performance needs.
Fronthaul has a related RF profile but a distinct capacity challenge. The cited use cases include transport of CPRI digital signals, which are not optimized in the same way as some other traffic and require higher throughput. Consequently, the system specification and channel capacity must inform semiconductor selection, rather than assuming that a backhaul design will transfer unchanged.
Small cells and enterprise links
Shorter small-cell links can operate at lower output power than macro links, making integrated silicon solutions more practical. The paper identifies SiGe, BiCMOS and CMOS as potential options for small-cell backhaul, with GaAs available where additional power or noise performance is needed. For enterprise, surveillance and similar links, the stated design goal is often a lower-cost system that can meet the required range and throughput without using costly high-performance amplifiers when they are unnecessary.
V-band and fixed broadband
V-band operation around 60 GHz is associated with shorter ranges because of atmospheric oxygen absorption. That attenuation can also support frequency reuse, helping make the band relevant to small cells and fixed broadband. Beamforming may be needed for point-to-point or point-to-multipoint links, including line-of-sight and non-line-of-sight conditions. The paper notes that BiCMOS and RF CMOS can be attractive for these applications where integration, power consumption and overall bill of materials matter.
Integration, packaging and production economics
Higher integration can reduce the number of external components and simplify assembly, but it also makes a chip more application-specific and may increase development effort. The right balance depends on required output power, phase noise, noise figure, power consumption and available board area. For high-performance applications, splitting functions across technologies may be more suitable than pursuing a single-chip solution.
Packaging choices include approaches such as QFN, chip-scale packaging, flip-chip and embedded wafer-level ball grid array (eWLB). The whitepaper describes eWLB as enabling integration of multiple dies and passive components, with the possibility of package-integrated antennas. Other approaches, including through-silicon vias and die embedding, are being investigated, although cost remains a barrier to broader use.
Production volume is another key constraint. Silicon processes generally rely on larger markets to support development investment and manufacturing economics. The paper’s analysis of the market at the time of publication highlights that mmWave backhaul volumes were small relative to the capacity of a typical silicon fab. Adjacent, higher-volume applications such as WiGig were therefore relevant to the business case for silicon development. For engineers and decision-makers, the implication is to assess both technical fit and the likely supply-chain and development economics of a chipset.
Planning for frequencies above 90 GHz
Future backhaul and fronthaul capacity requirements point toward wider contiguous spectrum, including frequencies above 90 GHz. The whitepaper describes development toward operation as high as 300 GHz, while noting that semiconductor capability is only part of the challenge. Commercial use also depends on system requirements and the regulatory framework for these bands. Design teams evaluating future platforms should therefore track both device development and spectrum availability.
Practical implications for system designers
Begin with the link requirements: distance, band, modulation, bandwidth, availability and target cost. Then translate them into RF specifications for transmit power, linearity, phase noise and noise figure. Use these requirements to determine where integration is valuable and where a higher-performance discrete device is justified.
- Consider GaAs for RF front ends when output power or noise performance is a primary constraint.
- Evaluate SiGe or BiCMOS where integration and moderate-to-high RF performance must be balanced.
- Consider CMOS or FD-SOI for integration-oriented, volume-sensitive designs when their RF performance fits the link budget.
- Include package and assembly effects in the design decision, rather than treating packaging as a late-stage detail.
- Check whether projected volumes support the development and sourcing strategy for the selected process.
The practical lesson is that no single mmWave semiconductor technology is optimal for every transmission use case. Successful designs align device performance, integration, packaging and production economics with the actual demands of the link. As systems move toward higher capacity and frequencies above 90 GHz, that same discipline will remain essential—alongside attention to spectrum and regulatory constraints.