CS Mantech: Improving the capabilities of RF GaN

Options for building better HEMTs including switching the polarity, introducing InAlN barriers and optimising buffer growth.

BY RICHARD STEVENSON, EDITOR, CS MAGAZINE

ΩAs a material for making compound semiconductor devices, GaN is peerless. It is an incredibly efficient light emitter, enabling a revolution in lighting, as well as the production of short-wavelength lasers that can read data off of discs and cut material. In addition, GaN has excellent electronic characteristics that are now being put to good use for improving the efficiency of chargers and power supplies.

With so many lucrative opportunities for GaN, one might overlook those that are not grabbing the headlines, but are still set to generate growing revenues over the coming years.

In that regard stands RF GaN, which is deployed in 4G and 5G base stations, radar and electronic warfare systems. According to market analyst Yole Group, sales of RF GaN devices for these applications are climbing, with total device revenue forecast to increase to $2 billion by 2029.

This market continues to attract investment to improve the performance of the GaN HEMT through changes to its architecture, its foundation and the epitaxial processes employed to produce this device. At this year’s CS Mantech, held in Tucson, Arizona, from 20-23 May, spokesmen for makers of devices, epiwafers and MOCVD tools all reported progress on these fronts. At this conference Sumitomo Electric Industries reported record-breaking output powers for N-polar HEMTs in the Ka-band, a partnership between Mitsubishi Electric and Furuno Electric unveiled a very efficient GaN HEMT for marine radar, WIN Semiconductors explained how an optimised buffer improves GaN HEMT linearity, and a European collaboration involving Aixtron discussed the use of depleted interfaces between the substrate and epilayers for minimising RF loss in GaN-on-silicon HEMTs.

Figure 1. Atomic force microscopy uncovers pits on the surface of III-polar HEMTs that can lead to device breakdown and current collapse (a). These pits are not observed in N-polar HEMTs.

Switching polarity
The GaN HEMTs that are winning commercial deployment feature an AlGaN buffer and a Ga-polar design. To improve performance, a number of researchers have considered refining this architecture. A team from the University of California, Santa Barbara, have enjoyed much success by turning to N-polar HEMTs, realising an output power density of 8.84 W mm^-1 at 94 GHz, at a power-added efficiency of 27 percent. Meanwhile, at team from Fujitsu has delivered a power density of 3 W mm^-1 at 96 GHz by replacing the conventional barrier with lattice matched InAlN.

At this year’s CS Mantech, Shigeki Yoshida from Sumitomo Electric Industries described efforts to combine both these valuable innovations. This team’s HEMTs, with an N-polar orientation and an InAlN barrier, are claimed to deliver a record-breaking power density of 12.8 W mm^-1 at 28 GHz. Yoshida told Compound Semiconductor that in comparison, the output power of the company’s Ga-polar GaN HEMT product is about 4 W mm^-1 in the Ka-band.

To produce their transistors, Yoshida and co-workers began by loading 100 mm semi-insulating SiC substrates into an MOCVD reactor and depositing a buffer layer, followed by an InAlN back barrier, an AlN spacer, a 12 nm-thick GaN channel and a GaN cap. On this epitaxial structure they added a HfSiO_x gate insulator, selected for its high values for permittivity and breakdown field – they are 13 and 8.5 MV cm^-1, respectively. To reduce source and drain contact resistance, the engineers introduced heavily doped n-type GaN selective growth regions. HEMTs were produced with a 200 mm gate length, a source-drain length of 2.5 µm, a gate-source length of 0.9 µm, and a gate-drain length of 1.4 µm.

Figure 2. Engineers at Sumitomo Electric Industries have benchmarked their N-polar GaN/InGaN HEMTs (blue star) against N-polar GaN/AlGaN HEMTs.

On-wafer measurements determined an electron mobility of 928 cm² V^-1 s^-1 and a surface roughness, according to atomic force microscopy, of 0.304 nm – that’s the root-mean-square roughness over a 1 µm by 1 µm area. This value for roughness is lower than that for III-polar GaN HEMTs, suggesting a suppression of electron scattering and a high electron mobility. For the surface of the III-polar GaN HEMTs, pits are seen on the surface that can cause device breakdown and current collapse (see Figure 1). Note that these imperfections are not seen on the surface of the team’s N-polar HEMTs.

One of the benefits of the N-polar HEMT over its conventional cousin is that the external electric field resulting from the gate voltage and the internal electric field due to the polarisation of the barrier are in opposite directions, rather than the same direction. This allows N-polar HEMTs to have a higher breakdown voltage, with Yoshida and co-workers estimating a breakdown voltage of more than 60 V, based on electrical measurements.