Accelerating GaN-on-silicon RF power amplification
Driving power amplification beyond 100 GHz positions the GaN-on-silicon HEMT as a strong candidate for sub-terahertz 6G communication.
BY GEOK ING NG FROM NANYANG TECHNOLOGICAL UNIVERSITY AND AGENCY FOR SCIENCE, TECHNOLOGY AND RESEARCH, SINGAPORE
As humanity enters a hyperconnected era that’s driven by phenomenal leaps in AI, it is critical to roll-out a robust infrastructure that supports this revolution. This means upgrading the capabilities of data centres and telecommunication networks, with the latter needing to accommodate extremely high 6G data rates of around 1 Tbit s-1. It’s a requirement that will be fulfilled with massive arrays of compact cells operating in the sub-terahertz range – a domain that must be employed, because channels with bandwidths as wide as tens of gigahertz are only possible in relatively unexplored sub-terahertz frequencies.
The development of technologies for producing and amplifying these frequencies will provide benefits extending beyond telecommunication networks. Progress will also aid: highly automated smart factories, by helping radar sensors to offer millimetre-level precision detection; and support the development and production of test instruments operating up to the D-band (110 GHz to 170 GHz) and beyond.
With all these factors at play, it’s not surprising that there’s tremendous interest in low-cost, high-performance semiconductors operating in the D-band.
It’s a frequency range that’s within reach for a number of established compound semiconductor technologies, notably InP HBTs, and GaN-on-SiC HEMTs.
Figure 1. Benchmarking the output power (Pout) as a function
of frequency for GaN-on-silicon. To the best of the authors’ knowledge,
this work advances the frequency limits of GaN-on-silicon HEMT power
amplification into D-band for the first time.
Another candidate is the GaN-on-silicon HEMT, a relatively new contender that is yet to receive much attention. What it offers over its rivals is the compelling combination of superior material properties associated with III-N heterostructures and the widespread availability of large-diameter (up to 300 mm) silicon substrates, which enable a sizeable cost reduction that can spur the broad adoption of these ultra-high frequency systems.
Over the years, rapid progress in GaN-on-silicon HEMT technology has failed to go hand-in-hand with an expansion in the frequency range for power amplification. This is still confined to the Ka band (26.5 GHz - 40 GHz) and W band (75 GHz - 100 GHz) – a limitation that reveals that there’s a great opportunity to advance the GaN-on-silicon HEMT for D-band power amplification (see Figure 1).
Critical to the fabrication of every high-performance GaN transistor is the epitaxial structure and device design. The D-band GaN-on-silicon HEMT is no exception.
Answering the challenge is our team, led by researchers at Nanyang Technological University and involving contributions from various institutions in Singapore, including the National Semiconductor Translation and Innovation Centre for Gallium Nitride, Agency for Science, Technology and Research (A*STAR); Institute of Microelectronics, A*STAR; Singapore-MIT Alliance for Research and Technology; and National University of Singapore.
Figure 2. RF large-signal performance at 123 GHz in continuous wave (CW) mode, and Vds=10 V. A maximum Pout of 0.67 W mm-1 is achieved.
Our latest progress with RF GaN-on-silicon HEMTs, involving amplification at unprecedented frequencies, draws on our previous expertise with these devices.
We have considerable experience employing in-situ SiN/AlN/GaN/AlGaN heterostructures by MOCVD. These epi-stacks feature an AlN barrier just 5 nm-thick – it provides strong polarisation that results in a high charge density of 1.7 x 1013 cm-2. Our design also incorporates an AlGaN back barrier that provides good ‘vertical scaling’, by ensuring excellent carrier confinement.
A feature of our HEMTs is their in-situ SiN layer, designed to serve two purposes: passivation and a gate dielectric. By applying a thin passivation layer, we hope to strike a balance between surface passivation and parasitic capacitance. We view our design as an ‘educated guess’ for an optimal D-band GaN-on-silicon HEMT.
Fabrication of our transistors began with the use of chlorine-based inductively coupled plasma reactive ion etching to provide mesa isolation. We then formed alloyed Ti/Al/Ni/Au ohmic contacts, which provide a contact resistance of 0.3 Ω mm. Note, though, that it’s possible to realise an even lower resistance with re-grown contacts formed by either MBE or MOCVD. We employed the Ni/Au stack for T-shaped gates, defined prior to atomic layer deposition of 10 nm-thick Al2O3 at 300 °C.
Figure 3. Benchmarking RF large-signal performance of GaN HEMTs and MMICs at 120 GHz – 140 GHz, in terms of PAE and Pout. The gate length and Vds are specified. The name of the foundry, if different from the publishing affiliation, is written in parentheses. All epitaxial structures are metal-polar, unless otherwise specified. All reports are pre-matched GaN MMICs, except this work, which is a GaN HEMT with external load/source impedance tuning. In the case of MMICs, Pout is normalised by the gate periphery of the final stage transistor.
Thanks to the thin passivation structure, consisting of 5 nm-thick SiN grown in situ with the AlN/GaN stack, plus 10 nm-thick Al2O3, our HEMTs have decent current collapse characteristics. This is accomplished while ensuring that parasitic capacitance, which may limit high-frequency performance, remains as low as possible.
Dimensions for our device are a gate length of 140 nm and a source-to-drain distance of 1.3 µm. We use a small gate finger width – it is 2 x 16 µm – to reduce signal propagation delay along the width direction, enabling an increase in gain.
Our devices deliver excellent DC characteristics, including a maximum drain current of 2.0 A mm-1, an on-resistance of 1.1 Ω mm, and a maximum transconductance of 0.65 S mm-1. Breakdown voltage is 35 V, limited by breakdown at the drain edge of the gate. According to RF small-signal measurements at a drain voltage of 10 V, and modelling with a small-signal model, values for the cut-off frequency and the maximum oscillation frequency are 112 GHz and 205 GHz, respectively.
The highlight of this work is the RF large-signal performance at D-band, determined with an on-wafer D-band passive load-pull system. These measurements were obtained with a vector network analyser (VNA, Agilent N5247B) with VNA extenders (VDI WR6.5) as the signal source.
We employed passive tuners (Focus W1701100BV) at the fundamental frequency of 123 GHz for source and load tuning. Calibration was conducted for the VNA extender, input coupler, tuners, probes, power source, and power receiver, in that sequence. After performing power calibration, the signal source provided power ranging from -10 to 10 dBm.
Measurements on our HEMTs involved a drain-source voltage of 5 V, which is a supply voltage that’s favourable for mobile user equipment. Using continuous wave (CW) Class AB operation, and tuning for optimal power-added efficiency (PAE), we recorded a maximum output power of 0.53 W mm-1, and an associated PAE of 2.3 percent. Increasing the drain-source voltage to 10 V boosted the maximum output power to 0.67 W mm-1, realised with an associated PAE of 1.3 percent (see Figure 2). The peak drain efficiency is 27.5 percent.
Benchmarking RF large-signal performance reveals that our GaN-on-silicon HEMTs produce a promising performance when compared with GaN-on-SiC HEMTs operating at a similar drain-source voltage (see Figure 3). What’s more, we have extended the frequency limits of GaN-on-silicon HEMT power amplification into the D-band. Thanks to these encouraging results, GaN-on-silicon HEMT technology is a worthy contender for sub-terahertz power amplification. We attribute the results to the combination of an epitaxial structure that’s suitable for high-frequency HEMTs, and an in-situ SiN layer that serves as thin passivation and a gate dielectric.
Our next step is to boost the performance of our GaN-on-silicon HEMTs for high frequency power amplification. These efforts, including refinements to epitaxy, will focus on reducing the RF loss of the GaN-on-silicon substrate and increasing mobility through a reduction in intermixing in the AlN/GaN heterostructure. We also plan to explore common features in high-frequency HEMTs, such as regrown contacts and deep transistor scaling (gate length and source-to-drain spacing). In addition, we shall optimise the device layout.
To the best of our knowledge, we have provided the first demonstration of the feasibility of GaN-on-silicon HEMT power amplification at frequencies beyond 100 GHz. This work establishes GaN-on-silicon HEMTs as strong candidates for low-cost sub-terahertz 6G cellular infrastructure, test instruments in data centres, and more. We hope this work could make a modest contribution to bringing 6G connectivity to reality.
This work was supported in part by the Agency for Science, Technology and Research (A*STAR) under Research, Innovation and Enterprise 2025 (RIE2025) Manufacturing, Trade and Connectivity (MTC) under Award M21K6b0134 and in part by the National Research Foundation (NRF)/A*STAR under RIE2025 MTC under Grant M23WSNG001.
