THz semiconductor technologies sit at the intersection of ambitious wireless requirements and practical manufacturing constraints. As communications and sensing systems move above 100 GHz, circuit designers must deliver sufficient RF output power and signal quality while integrating digital processing, calibration, and memory. No single semiconductor process makes every part of that task easy: high-speed compound semiconductors offer strong RF capabilities, while CMOS is attractive for complex digital circuitry and cost-conscious, higher-volume products.
The resulting design question is not simply which transistor is fastest. It is how to combine device performance, integration, packaging, and manufacturing economics in a workable system.
Why systems above 100 GHz need more than RF speed
Millimeter-wave and higher-frequency systems support demanding communication and sensing applications. The 2018 IEEE presentation describes 5G as using established sub-6 GHz bands alongside newer millimeter-wave bands up to 90 GHz, with increasing backhaul and fronthaul capacity needs. It also identifies future wireless links above 100 GHz as a path toward data rates beyond 100 Gb/s.
Sensing is another driver. Wider bandwidth can make greater radar range resolution easier to achieve, while shorter wavelengths can improve velocity and angular resolution. Automotive radar operating at 76–81 GHz was already moving toward mature silicon implementations, including BiCMOS and single-chip CMOS, while systems above 100 GHz remained a future development area.
These applications place several demands on the electronics at once:
- Power amplifiers (PAs): adequate output power supports the link budget and range.
- Voltage-controlled oscillators and phase-locked loops (VCOs and PLLs): low phase noise helps maintain error vector magnitude (EVM) and spectral efficiency.
- System integration: RF circuitry must work alongside digital calibration, physical-layer processing, and memory.
- Compact packaging: higher integration can reduce overall size and cost, but integration and packaging create their own engineering challenges.
THz semiconductor technologies: performance versus integration
CMOS has a central advantage when a design needs large digital circuits and targets affordable production. However, its RF performance becomes a constraint well above 100 GHz. The presentation notes that the transit frequency, or fT, of silicon-based field-effect transistors is not expected to rise substantially with continued scaling. It also identifies FinFET implementations as intrinsically slower than planar devices in the context discussed.
Alternative device technologies can offer a stronger fit for demanding RF functions. III-V compound semiconductors—including materials such as GaAs, GaN, and InP—are used in high-electron-mobility transistors (HEMTs) and heterojunction bipolar transistors (HBTs). The presentation cites III-V HEMTs with fT above 500 GHz at relaxed gate lengths, while noting their specialty-process cost as a consideration. GaN is highlighted for stronger driving capability, and GaN and GaAs devices are associated with high output power and efficiency at high frequencies.
SiGe HBTs are another option. The presentation describes them as outperforming CMOS in speed and power handling and identifies SiGe BiCMOS as a way to combine bipolar RF devices with silicon logic. It also points to RF-SOI as offering higher fT than FinFET in the comparison presented, with body bias and the possibility of stacking devices in power amplifiers as relevant features.
Choosing an implementation strategy
Technology selection depends on the balance of RF capability, digital complexity, cost, and integration. All-silicon approaches can simplify the path to incorporating substantial logic, while III-V devices may better address demanding high-frequency RF functions. Co-integration offers a potential middle ground: pair devices suited to RF with silicon-based digital circuitry. The presentation notes that integrating III-V materials on silicon is complex, though research groups were exploring the approach.
Manufacturing scale is part of the decision, not an afterthought. The source identifies large wafer capability, including 300 mm wafers, as an important consideration for making an approach commercially viable. It also raises the open question of whether the market will adopt devices co-integrated with CMOS or whether CMOS circuit techniques and digital compensation will be sufficient to avoid non-CMOS devices.
What bulk CMOS demonstrations show
Although CMOS faces limitations at very high frequencies, the presentation includes circuit examples that show what bulk CMOS can achieve. One reported 140 GHz power amplifier used a 28 nm high-performance mobile process, operated at a 0.9 V supply, and occupied 0.11 mm2. Its reported DC power consumption was 81 mW.
A separate example was a 140 GHz frequency-modulated continuous-wave (FMCW) radar implemented in bulk CMOS with on-chip antennas. The reported receiver included 47 dB of programmable baseband gain. The transmitter, with two power amplifiers active, had a reported EIRP of 9 dBm. These demonstrations establish that CMOS can support specific circuits at these frequencies; they do not remove the broader trade-offs around output power, system integration, and application requirements.
Practical implications for system designers
For engineers and technical decision-makers, the main takeaway is to evaluate the full system rather than select a process on transistor speed alone. Start by defining the required frequency, link budget, phase-noise and EVM targets, digital processing needs, and anticipated production scale. Then determine whether those needs can be met in a silicon process or whether a hybrid or co-integrated approach is warranted.
Above 100 GHz, the choice among MOSFET, SiGe HBT, III-V HBT, and HEMT technologies remains an engineering and commercial trade-off. CMOS supports dense digital integration, while other device families can provide advantages in speed or power handling. The practical path will depend on circuit design, integration and packaging, and whether the manufacturing approach can meet cost and volume requirements.
As scaling delivers less improvement in silicon transistor speed, system architects should treat semiconductor choice as a strategic part of RF architecture. Comparing technologies at the level of complete functions—RF, digital, and packaging—offers a more useful basis for decisions than considering any device characteristic in isolation.