Quashing parasitics in RF GaN-on-silicon HEMTs
Atomera’s oxygen-based epitaxial technology is addressing problematic parasitic channels in GaN-on-silicon HEMTs.
BY RICHARD STEVENSON, EDITOR, CS MAGAZINE
Facing a dilemma are producers of GaN RF HEMTs, a class of transistor that’s widely deployed for mobile communication and defence infrastructure. If performance tops their agenda, they will produce their devices on SiC substrates, which have a high thermal conductivity that aids performance, but adds to cost. And if price is their top priority, they will select a silicon foundation, as this drives down production costs – but at the expense of a parasitic channel that introduces RF losses and drags down linearity.
Fortunately, it is now possible to address parasitic limitations and enjoy lower costs by adopting an approach pioneered by technology-licensing company Atomera. This US outfit is expanding its expertise developed in the silicon industry by demonstrating that its epitaxial technology can incorporate a much-valued oxygen-rich region into the silicon substrate beneath the III-N RF heterostructure – this addition reduces the highly detrimental parasitic charge by more than an order of magnitude.
A parasitic layer plagues GaN-on-silicon HEMTs, because during the growth of the AlN nucleation layer and the AlGaN stress-relief layers, aluminium and gallium atoms, which both act as dopants, descend into the silicon substrate. This creates a layer of charge at the substrate-epilayer interface that’s responsible for the parasitic channel, even when III-N heterostructures are grown on high-resistivity silicon.
Atomera’s Mears Silicon Technology (MST), which addresses this weakness, has already been shown to enhance a broad swathe of silicon devices, from power switches to leading-edge gate-all-around transistors.
Company’s co-founder and CTO, Robert Mears, explains that at the heart of Atomera’s trailblazing tech is a pause in silicon epitaxy, when a puff of oxygen is introduced into the growth chamber. After that, the growth of the silicon stack resumes.
“If you get the oxygen right, it's a very stable layer,” argues Mears, who adds: “It’s not big clusters of silicon dioxide.” MST retains the silicon planes and creates a structure that’s extremely effective at blocking diffusion. Ultimately, this prevents a parasitic channel from forming.
The degree of effectiveness of MST is highlighted in measurements of epitaxial structures for RF GaN-based HEMTs, grown by Edwin Piner’s team at Texas State University.
Spreading resistance probe measurements, which offer a good insight into the local conductivity that results from the parasitic charge, show that MST leads to a reduction in interfacial charge by more than an order of magnitude, compared with the conventional heterostructure. The origin of this reduction is quantified by secondary ion mass spectrometry, which shows a substantial reduction in the penetration of gallium atoms into the substrate.
Mears points out that the blocking of diffusion does not need to be perfect. Once it allows resistivity to exceed around 10 kΩ cm, the substrate is effectively lossless, and other characteristics are more worthy of consideration.
Expounding on that point, he remarks: “What really matters to all RF designers is the linearity. It's how much you can suppress the second and third harmonics for the incoming RF signal. That's where this new data is exceptional.”
According to Mears, GaN-on-silicon has traditionally fallen far short of silicon-on-insulator (SOI) technologies for supressing the second harmonic. One leader of SOI technologies, Soitec, has products eSI80, eSI90, and eSI100 – named to reflect values of -80 dBm, -90 dBm and -100 dBm for the second harmonic power level at the standard input power of 15 dBm. For essentially a first iteration of a GaN RF structure with MST technology, the comparable value is -97 dBm, according to measurements made by Incize, a Belgium-based company that provides characterisation services for the RF industry.
Today, many RF applications involve input powers of more than 15 dBm (that is just over 30 milliwatts). And in this regime, RF GaN structures with MST technology are also performing well, says Mears: “The linearity up to input powers of up to 40 dBm – 10 watts – is extremely good. That's unique for GaN-on-silicon.”
With Mears Silicon Technology (MST), aluminium and gallium atoms, which act as dopants, are prevented from descending into the silicon substrate. This prevents a parasitic channel from forming.
Built on experience
While Atomera may be a new name to many in the compound semiconductor industry, it has a long history, having been founded in the early 2000s. Initially, efforts focused on ab initio simulations of the addition of oxygen into silicon lattices, proving the stability of the structure and the epitaxial growth process for realising such structures in silicon devices.
According to company CEO and President, Scott Bibaud, the global financial crash of 2008 hampered progress, due to difficulties in raising funds. But within a few years, the team were once again progressing their technology.
“We didn't have the money to really aggressively go after customers until we went public in 2016,” explains Bibaud. “Public markets have been able to give us funding to continue developing this [technology] and bring it out to customers.”
Bibaud says that Atomera is currently working with handful of customers in the silicon industry, and claims that “virtually all” major semiconductor companies have, or are, evaluating MST technology.
For silicon CMOS processes, standard substrate orientations are (100) and (110), alongside an offcut of around 4°. So, for the work related to GaN RF HEMTs, which are typically grown on silicon (111) with essentially no off-cut, Atomera has modified its MST recipe.
During these efforts, Atomera found that its MST technology provides an extra degree of freedom in the lattice, and helps accommodate lattice mismatch between the III-Ns and silicon.
For companies that adopt MST technology, disruption should be minimal and additional cost marginal. As most firms already have epitaxial tools in their fabs, these just need to be modified to allow the injection of oxygen into the reactor.
Benchmarking various RF technologies highlights the capability of GaN-on-silicon, when it includes MST technology.
“The implementation is roughly similar to the time and effort required to put in strain engineering,” argued Bibaud, who explains that Atomera’s business model is for customers to pay a royalty on their wafers, once they have been shipped. “Generally speaking, that [royalty] is a small fraction of the economic benefit they get from the improvement we bring them.”
In addition to these chipmakers, fabless firms and foundries can deploy MST technology.
“When foundries offer our technology as a standard part of their PDK, we will be out evangelising to their fabless customers, so they understand what benefits they get by using our technology,” enthuses Bibaud.
There’s no doubt that Atomera’s technology has much promise for the production of GaN-on-silicon RF HEMTs. But there’s still a long way to go, beginning with engaging with customers, and getting them to evaluate the technology. The appeal is that significant rewards result, due to the potential to combine low costs with impressive performance figures and compatibility with silicon CMOS technology.
Main image: Scott Bibaud (left), has been Atomera’s CEO and President since 2015. During his career, he has held positions at Broadcom, Conexant and Raytheon. Robert Mears co-founder Atomera in 2001, and is the company’s CTO. As well as developing MST, he has played a key role in the developed of erbium-doped fibre amplifiers.
