中文 English

Inspecting And Testing GaN Power Semis

With increasing expectations for EVs, GaN power semis may offer manufacturers easier solutions for entry.

popularity

As demand for new automotive battery electric vehicles (BEVs) heats up, automakers are looking for solutions to meet strict zero-defect goals in power semiconductors. Gallium nitride (GaN) and silicon carbide (SiC) wide-bandgap power semiconductors offer automakers a range of new EV solutions, but questions remain on how to meet the stringent quality goals of the automotive industry.

Among the biggest questions are how power IC makers can guarantee close to 100% reliability, with the lowest possible defectivity in automotive semiconductors. Complicating the new developments is the need to significantly lower costs of these chips to make EVs more attractive to the average consumer, and manufacturers are beginning to address this by moving to the larger 200mm wafer sizes from current 150mm wafers.

The payoff promises to be substantial for those successful in hitting the sweet spot for widening GaN applications for EVs, as well as many other consumer and smartphone rapid-charging applications. A recent report from Yole Développement estimated that the GaN consumer handset power supply market alone to hit $597 million by 2026, with a 72% CAGR from 2020-2026. And as Apple recently unveiled its 140W MagSafe charger, research firm TrendForce said it expects GaN solutions to reach a 52% penetration rate in the fast-charge market by 2025.

Because all GaN power devices today are lateral, rather than vertical, there are instances where SiC makes more sense if higher voltages are required by the application. Increasing breakdown voltage also requires proportionally larger die area plus thicker epi layers, notes George Liang, director of product and system engineering for switching power and battery-powered applications at Infineon Technologies.

Penetration of GaN in different markets will vary depending on the requirements of each application, according to Taha Ayari, technology and market analyst for compound semiconductor and emerging substrates at Yole. “In general, the remaining challenges for GaN devices are reliability and performance acceptance, price competitiveness, along with the development of high-voltage devices for high power applications. One of the main obstacles remains the epitaxy of GaN layers on silicon substrate, the lattice mismatch, and the thermal coefficient of expansion mismatch between the two materials, which generates killer defects in the GaN layer. So it requires a sophisticated buffer layer and epilayers.”

Ayari noted the epitaxy is generally related to in-house processes developed by the manufacturer, rendering epitaxy standardization quite tricky. In addition, price pressure and higher volume demand are pushing the industry to transition from the traditional six-inch platform to the eight-inch platform, which would require more epitaxy development for uniformity and higher yield purposes, he said.

GaN has practical limitations on top-end voltage, with applications limited to 900V. In April, researchers at Imec and AIXTRON, a deposition equipment supplier, announced a successful demonstration of epitaxial growth of GaN buffer layers qualified for 1,200V applications on 200mm QST substrates, with a hard breakdown exceeding 1,800V. If this development proves feasible, it would open higher voltage GaN applications in EVs that previously were possible only with SiC-based technology.

Manufacturing, test, inspection Issues
Today the main lateral GaN HEMT grown on Si or Sapphire substrate is still susceptible to surface breakdown and gate leakage current, thus several players are focusing on low current and a voltage around 650V, notes Ahmed Ben Slimane, technology and market analyst at Yole. “For higher voltages (>1,200V), other emerging substrates are attractive, for example, SOI, QST (this substrate is used in IMEC’s demonstration), or bulk GaN that allows vertical GaN devices,” he said. “However, the supply chain of these emerging substrates is still under development with low volumes and high prices, and it will likely take time for end-users to adopt the new technologies.”

Critical manufacturing hurdles remain for production of GaN-on-Si and GaN-on-sapphire. “With the GaN adoption in consumer, mass production and higher volumes with lower prices are required. This implies a transition to larger wafer sizes,” Ben Slimane added. “As of 2021, some players have 8” GaN-on-Si fabs (Innoscience and X-fab) or plan to move to 8” (Infineon, Nexperia, and TSMC) in the coming years. On the other hand, there are technical challenges in GaN epitaxy and processes in terms of thickness and Al composition uniformity, bow and warp, as well as yield losses. For GaN-on-sapphire, according to industry feedback, it’s unlikely to move to 8”; 6” will most likely remain the mainstream platform for GaN-on-sapphire with only Power Integrations as a leading player.”

As with other semiconductors, GaN gate degradation remains one of the main hurdles to overcome at the process level. Manufacturers are relying on inspection techniques to develop and qualify their products, Ben Slimane said.

Testing and inspection issues also confront manufacturers of GaN devices.

“Power semiconductors is an area where, in automotive, all of a sudden you need zero defectivity, said Ingo Schmitz, technical marketer at Bruker Nano Surfaces. “They have all these needs, and what used to be very simple devices, like MOSFETs, used to be so cheap that you couldn’t afford the metrology. Today, that’s different.”

Safety-critical reliability requirements add challenges for all chips used in those markets, and especially those involving new materials, as well as opportunities for companies that can help resolve those challenges. “In automotive, chips are tested multiple times at the wafer level, the chip level, and the package level,” said Oreste Donzella, executive vice president for KLA’s Electronics, Packaging and Components Group. “And then you do burn-in, you do reliability testing, you test again, and then you have records. But you still may have an airbag fail when you crash a car because a chip in your airbag control doesn’t work. This is because of latent defects that either escaped the testing, because the test is not 100% efficient, or because these failures are activated during the operation of the car in a very harsh environment.”

And this is where things get challenging, because the amount of testing and inspection needed to prevent these kinds of defects is increasing. Those processes take longer, costs more, and produce so much data that advanced analytics are required to identify problems. Even then, there is less history and data against which to compare materials like GaN and silicon carbide.

“We help the automotive industry to search for potential latent defects by doing a more intelligent sampling in the wafer fab inspection and test,” Donzella said. “This is where I-PAT (inline defect part average testing) fits in, because there is a lack of maturity and lack of data on some of these EV materials. We’re measuring them against silicon results, but they are not at the same level as the silicon.”

 

Fig. 1: Identifying potential latent defects. Source: KLA

Fig. 1: Identifying potential latent defects. Source: KLA

Classifying the magnitude of defects is essential. “At the epitaxy level, early detection of killer defects and creating a link between different epi-failure mechanisms and dynamic RdsON could help isolate faulty dies or wafers and improve process control, resulting in a cost reduction and time-saving,” Ben Slimane stated. “Several techniques, such as optical, can be used for defect inspection. The most common are photoluminescence and X-Rays, used for metrology to detect epilayer uniformity, Al composition, and defects. At the device level, burn-in and time-dependent dielectric breakdown (TDDB) are used to test the device’s reliability.”

Despite GaN’s proven reliability in many applications, test methodology still needs to be standardized. “It is difficult to offer a test condition that is common to all devices since different structures are prone to failures due to different mechanisms,” Ben Slimane said. “In this context, JEDEC standards and AEC-Q101 automotive qualification are adapting to new methods of testing, which are essential to make guidelines of what and how to measure. Furthermore, companies are developing internal databases or physics-based models to offer to customers with higher reliability demands, like automotive OEMs and Tier-1s.”

Still, the market for GaN continues to grow. Efficient Power Conversion (EPC), for example, offers GaN devices that it says are being used for more than 100 emerging applications. Alex Lidow, EPC’s CEO, contends that GaN on silicon is past the tipping point for most applications.

“There are few manufacturing hurdles remaining,” Lidow said. “GaN devices are produced side-by-side with silicon devices in standard fabrication facilities using standard equipment. The one area that will come down significantly in cost as new generation equipment becomes available is MOCVD epitaxial growth of GaN heterostructures. GaN devices are still far from their theoretical performance limits, so new manufacturing challenges might emerge as the boundaries are pushed farther and farther away from the aging silicon MOSFET.”

Transitioning to 200mm
Bigger wafers remain a challenge at this time, but the industry has begun a transition to 200mm from 150mm wafers for GaN production. “It is true that GaN and silicon-based devices are well established in power and RF applications,” said David Haynes, managing director of strategic marketing for the Customer Support Business Group at Lam Research. “But this is largely on wafers that are six inches or smaller, and in the case of many GaN-based devices, on substrates such as sapphire and SiC.”

A strong shift is underway to 200mm wafer processing, which will increase compatibility of these technologies with mainstream semiconductor processing, as well as improve the economics for its use in more advanced or higher-volume applications, Haynes said. “SiC is migrating to 200mm, with production set to ramp in the next two to three years as 200mm wafer cost and availability improves.”

This is more of an economic optimization than a fundamental change, however. “GaN is already in mass production,” noted Infineon’s Liang. “So there are not really hurdles, but continuous improvements. Moving to 200 mm production will be a key milestone. Sapphire is a viable substrate Infineon has considered, but is not presently pursuing due to its poor thermal conductivity among other issues.”

Reliability and cost are barriers for widespread adoption of GaN power devices, Liang added. “Our OEMs have been relatively cautious in their approach to adopting GaN until recently,” he said. “There is always concern about reliability or other unknowns when considering any new technology. But as industry standards emerge (e.g. from JEDEC), and pilot programs prove the value and reliability for the OEMs, adoption has accelerated significantly in the past year. Especially in the portable charger/adapter product space, it seems there are weekly announcements of new high-performance GaN-based adapters launched. We believe this is the tipping point that will lead the more conservative industrial sector into adoption as well.”

Navitas Semiconductor produces GaN on silicon, using commodity silicon wafers to save on cost. “What we do, along with pretty much everybody else in the GaN world, is GaN on silicon,” said Stephen Oliver, Navitas corporate vice president for marketing and investor relations. “We start with a standard commodity silicon wafer — it’s tens of dollars, compared to 20 times that for a silicon carbide start wafer — and then we put the very thin sliver of gallium nitride deposited on the top and then all the action is within that tiny, tiny layer.”

The GaN layer is about 5um, and the silicon base wafer is about 1,000um, Oliver said. “The really cool thing about GaN is that it’s a really advanced material, but you make it on really old equipment.”

Navitas currently produces GaN on silicon on six-inch equipment. “Our foundry partner, TSMC, said they’re going to eight-inch, but right now we’ve got we get five times as many chips on a wafer as you do with a silicon chip because of that performance per square area, that conduction, when you also include the fact that we do a power IC. So we’ve got gate drive, ESD protection diodes, level shifting, current sensing, and protection autonomously. It’s a real power IC.”

Transitioning to eight-inch equipment is coming. “I’m guessing it’s maybe one or two years away,” said Oliver. “But it’s not long.”

GaN’s future
Much of the product, quality and manufacturing risks have largely been removed GaN devices, contends Primit Parikh, president of Transphorm, “which is why you are seeing strong adoption of GaN across these segments—from several manufacturers in the lower power fast chargers, and in particular from Transphorm across the higher power segment. That said, GaN as a semiconductor material has a lot more to offer. At Transphorm, we have a continuous technology and product roadmap to take GaN closer to its ultimate material limit’s potential, which is several times more figure of merit than done today.

“Also, although there is lot of talk, there is limited availability of viable higher power GaN other than from Transphorm, barring some products here and there from others,” Parikh added. “The market will benefit from more suppliers with solid products, just like the SiC market. Our goal will be to continue to dominate or be among the top few in every market segment we participate in.”

For future devices, the basic figure of merit for GaN (voltage sustained for given semiconductor dimension) is quite high and similar to SiC, Parikh said. “For practical lateral GaN devices, applications up to 1,200V (that require breakdowns around 1,500 to 1,800V) are well within roadmap. One word of caution when judging hard breakdown voltage publications is the switching functionality. You can get as high breakdown as you want with lousy switching, which is of no use in a real device. The Transphorm team understands this very well and designs accordingly. For example, years ago, we had shown 1,800 V+ breakdown bi-directional devices under our ARPA-E program, so it is certainly achievable.”

Transphorm has several GaN platforms in either production or development, Parikh said. “Transphorm has qualified devices in the market up to 900 volts with GaN on silicon and we are also working on 1,200 volts in R&D, and do not see a need for GaN-on-GaN in practical power devices for the foreseeable future,” Parikh said. “Overall, as an entrepreneur and fan of GaN, I don’t want to write off anything, so all the power to folks pursuing GaN-on-GaN substrates.”

Regarding testing/inspection issues, a lot depends on how each company has set up their wafer and/or packaged testing specs and flows, to assure high-quality products, Parikh added. “One important item is assessing dynamic switching, or ‘on’ resistance performance at high voltage because, for a long time, many GaN providers did not have full awareness of this issue and hence its measurement. We have worked over years to streamline a slew of proprietary on-wafer and packaged product DC and AC measurements. Now, reasonably good testers from various test equipment providers are also available. Inspection is again a function of the wafer quality for your epi material and fab material. What partnering with our high-quality Japan foundry team has shown is that we have achieved defect densities (one of the drivers of yield) for our GaN-on-silicon wafers similar to what silicon CMOS running in the same foundry achieves.”

Conclusion
With many of the current moves underway to transition to 200mm to lower costs, widespread adoption of GaN for power semis potentially brings a lot to the table for makers of current and future hybrid and electric vehicle, consumer electronics, smartphones, and other products using GaN wide-bandgap power semiconductor technologies. Questions still remain, however, on the ability of builders to adequately lower costs and stabilize the manufacturability of the more futuristic versions of GaN technology.

Related
GaN Application Base Widens, Adoption Grows
Mature, low-cost manufacturing and proven reliability spur use in EVs, smart phones, and consumer electronics.
Revving Up SiC And GaN
These technologies are gaining traction in high-voltage applications, but economies of scale and predictability still haven’t fully kicked in.
Gearing Up For Next-Gen Power Semis
Aluminum nitride, diamond semiconductors, gallium oxide and vertical GaN are all being readied, each with its own pros and cons
Improving Reliability For GaN And SiC
Why these chips are gaining ground, and what still needs to be addressed.
China Accelerates Foundry, Power Semi Efforts
Massive expansion campaign targets wide variety of chips, but export controls limit growth at leading edge.



1 comments

Tom Somyak says:

Nice article Patrick. Keep ’em coming!

Leave a Reply


(Note: This name will be displayed publicly)