What Happened To GaN And SiC?

Early predictions were overly optimistic, but these technologies are starting to make inroads.


About five years ago, some chipmakers claimed that traditional silicon-based power MOSFETs had hit the wall, prompting the need for a new power transistor technology.

At the time, some thought that two wide-bandgap technologies—gallium nitride (GaN) on silicon and silicon carbide (SiC) MOSFETs—would displace the ubiquitous power MOSFET. In addition, GaN and SiC were supposed to pose a threat to higher-end, silicon-based insulated-gate bipolar transistors (IGBTs). Power MOSFETs and IGBTs are the workhorse chips in power electronic systems.

Compared to silicon-based devices, GaN and SiC power chips operate at higher voltages, frequencies and temperatures, helping to eliminate up to 90% of the power losses in electricity conversion. Wide bandgap refers to higher voltage electronic band gaps in devices, which are larger than 1 electronvolt (eV).

As it turns out, power MOSFETs and IGBTs are moving towards their limits. But today, the two technologies continue to dominate the landscape in applications from 5 volts to 6.5 kilovolts. In contrast, GaN-on-silicon power chip shipments are lower than expected amid a multitude of challenges. And SiC MOSFETs are shipping, but SiC also suffers from high wafer costs.

“At one time, in International Rectifier’s promotion for GaN, the company said that within 10 years the topology would be that you use silicon for anything below 5 volts,” said Stephan Ohr, an analyst with Gartner. “You would use GaN for anything from 5 volts to 600 volts to 1,000 volts. And you would use SiC for anything above 1,000 volts.”

That prediction didn’t pan out. “I am not seeing that happening now,” Ohr said. “I don’t think you can buy a GaN part today. They are all on allocation. But if you go to a distributor, you can find SiC. SiC got to the market faster than GaN.”

All told, GaN and SiC will grow faster than silicon-based power semis over the next decade. But in total, GaN and SiC are projected to have a combined share of only 13% in the overall power semiconductor market by 2024, according to Lux Research. Silicon-based power semis will continue to dominate with an 87% share by 2024, according to the firm.

Still, OEMs face some tough decisions today. Silicon-based power semis continue to work, but OEMs still want to have smaller, faster and more efficient devices and for good reason. The power losses in today’s systems range from 8% to 15%, according to experts.

So, the questions are clear. Are silicon-based power MOSFETs and IGBTs on their last legs? Will GaN and SiC power devices eventually fulfill their promises and displace silicon? And, of course, which technology will provide the ultimate performance?

The contenders
Power semis are used in the field of power electronics. Basically, power electronics make use of solid-state electronics to control and convert electric power. The conversion is performed with various semiconductor-switching devices.

The perfect switch would have infinite speeds, zero on-state resistances and infinite off-state resistances. Unfortunately, the perfect switch doesn’t exist. So, engineers must look at several factors when evaluating chips, such as voltage, current, switching speed, load and temperature.

Today, there are several devices to choose from. On the transistor front, the entry-level market is served by traditional power MOSFETs, which are used in 10- to 500-volt applications. Developed in 1976, power MOSFETs are based on a double-diffused (DMOS) architecture. They are vertical structures, meaning the current flows from the source at the top to the drain at the bottom.

Power MOSFETs are cheap and here to stay. At best, GaN and SiC could make a tiny dent in applications below 500 volts. In any case, the big and hotly contested market is now taking place in two voltage segments—600 volts and 1,200 volts. In these areas, four basic technologies are competing for some large markets, such as adapters, automotive, switching power supplies and solar inverters.

In this segment, there are two silicon-based solutions—super-junction power MOSFETs and IGBTs. Super-junction power MOSFETs, which are souped-up versions of power MOSFETs, are used in 500- to 900-volt applications. Super-junction power MOSFETs are vertical devices. They also consist of pillar structures in the body, confining the electric field in the epi region.

The IGBT, meanwhile, is a three-terminal device that combines the characteristics of MOSFETs and bipolar transistors. IGBTs are used for 400-volt to 10-kilovolt applications.

Then, there are the two wide-bandgap technologies—SiC and GaN. Based on silicon and carbon, SiC has a bandgap of 3.3 eV. Silicon has a bandgap of 1.1 eV. SiC FETs are targeted for 600-volt to 10-kilovolt applications.

Another technology, GaN, is a binary III-V material. In the power arena, GaN-on-silicon chips are used in 30- to 600-volt applications. GaN has a bandgap of 3.4 eV.

The best technology?
IGBTs, SiC and other technologies are geared for the niche-oriented markets at 1,700 volts and higher. But what is the best technology for the larger 600- and 1,200-volt markets? It’s not a simple answer. “You will likely have a co-existence of all technologies. But it also depends on the applications of the voltage range and how much a customer is willing to pay for a device, whether they will go for a silicon-based solution, GaN or SiC,” said Roland Rupp, project manager for SiC devices at Infineon, the world’s largest power semi vendor. Infineon sell chips based on all of the technologies—MOSFETs, IGBTs, GaN and SiC.

Indeed, there are tradeoffs between the technologies. For example, both super-junction MOSFETs and IGBTs are ramping up on 300mm wafers, making them less expensive than GaN and SiC. In comparison, SiC MOSFETs are ramping up on 100mm wafers, while GaN-on-silicon is running on 150mm substrates.

In addition, super-junction power MOSFETs and IGBTs continue to improve in terms of performance. For instance, in some hard-switching applications, super-junction devices are closing in on GaN or SiC. “With respect to manufacturing cost, the IGBT is clearly superior to all other power switch technologies and has the lowest T-dependence of conduction losses,” Rupp said.

But super-junction power MOSFETs hit the ceiling at around 900 volts. IGBTs are plagued by slow switching speeds. “Both super-junction and IGBT technologies are getting closer to their technological limits,” he said. “There are still new ideas to further improve the trade-off between static and switching losses and keeping short-circuit ruggedness, but they are fighting with the fact that performance improvements are counterbalanced by increased processing costs. The newly available 300mm wafer process environment for such silicon-based power switches helps with respect to this cost aspect, but will probably be the last significant productivity gain for silicon-based power electronics for the next decade.”

So, there is a keen interest in GaN and SiC. Today, SiC diodes are used in high-end power supplies for servers and telecom systems, but SiC MOSFETs are still in the early stages of market penetration. Compared to power MOSFETs, SiC has 10 times the breakdown field and three times the thermal conductivity. “Neglecting the cost differences between the various technologies would lead to a clear champion—SiC FET,” Rupp said.

But SiC also suffers from high wafer costs and low effective channel mobility. In a move to address some of the issues, suppliers hope to reduce the costs by moving to larger wafers. “We are doing production on 4-inch. We want to go to 6-inch,” said John Palmour, chief technology officer for power and RF at Cree, a supplier of SiC-based LEDs and power devices.

SiC MOSFETs are vertical devices. The channel structures also come in various configurations, including trench and planar. Trench-based SiC MOSFETs have lower conductivity loses than planar. But trench tends to suffer from gate-oxide breakdowns, prompting some to devise double-trench SiC MOSFETs.

Cree, for one, advocates the planar channel structure. In fact, Cree has rolled out its third-generation SiC technology, which could address the channel mobility issues. “It’s a die shrink,” Palmour said. “We’ve also reduced the cost-per-amp.”

All told, SiC MOSFETs have some advantages over MOSFETs and IGBTs, but SiC won’t displace silicon anytime soon. “IGBTs are not going away,” he said. “They will be around for a while.”

Like SiC, GaN is also generating steam. A GaN high electron mobility transistor (HEMT) is a lateral device. The current flows from the source to the drain on the surface. Below the surface, AlGaN and GaN layers are grown on a silicon substrate.

GaN-on-silicon is fast, but it also suffers from a lattice mismatch, making it prone to defects in the fab. It also suffers from reliability issues and low thermal conductivity. And there are also questions whether GaN-on-silicon can scale beyond 600 volts.

“This assumes you can buy GaN parts,” Gartner’s Ohr said. “I have been saying that GaN parts are still in development and experimental. But assuming you can get one, you can reduce the size of your capacitors and inductors that would go with your power supply or motor drive. But what is a GaN part going to cost you? And does that pay for the smaller size and incremental efficiency you can get from that?”

In fact, GaN has made slow progress. But one supplier, Transphorm, is making some headway. In 2013, Transphorm acquired Fujitsu’s GaN IP. At the time, Transphorm also announced a 600-volt GaN part, based on a cascode-type, normally-off technology. The device reduces energy losses by 50%, compared to silicon.

Then, earlier this year, the company moved into mass production. The chips are being made on a foundry basis within Fujitsu’s 150mm fab in Japan. “Transphorm is the only GaN provider who has announced and is actively providing 600-volt qualified products to the industry,” said Primit Parikh, president of Transphorm.

Parikh also dismissed the notion that GaN will run out of steam at 600 volts. In the lab, Transphorm has demonstrated 1,800-volt devices. “It has already scaled in demonstrations,” he said. “Our current focus is on mass production and widespread commercialization of the first set of products at 600 volts. That will be followed by higher voltage devices at 900 volts and 1,200 volts.”

Transphorm isn’t the only GaN vendor, however. In fact, more than a dozen companies have entered the GaN power chip fray in recent times. “One of the reasons for that is clear. GaN is not just a technology. GaN is becoming desirable as the power conversion platform of choice,” he added.

It’s unlikely that there is room for a dozen GaN suppliers. There also are too many SiC vendors. Time will tell if the market will see a shakeout. But clearly, GaN and SiC are shaking up the landscape.


Israel Beinglass says:

Thanks for a great review, I haven’t seen any comment on MACOM as a SiC and/or GaN on Si supplier any thoughts?

david pacholok says:

I have been a power electronics engineer for 40 something years. I remember the days of germanium power bipolar transistors! Wow look at how far we have come thanks to the semi gurus out there!
I can make a few personal observations for what they are worth:

I see real value in SiC mosfets above 650 v where Si super junction devices run out of gas. Igbts also are good up to maybe 100 kHz depending on topology at the 600 volt level. At 1200 volts Ton and Toff really suffer, so I see SiC mosfets making inroads there. The BS diode in Si FET s has always been a problem for any topology that uses it. The BS diodes in SiC FET s are a lot faster but the 3-5 volt Vf can increase dissipation in some topologies.
So since I do some work at ISM frequenc ies here is my question:
Why can Si RF mosfets like the ARF 446 switch in 2-4 nSec with a BVds of 900v while the fastest SiC FET I can buy is 5.X slower??? Is it akin to the high gate resistance in polysilicon Si gate structures? Can it be lowered in SiC like in Si by metallized gate runners? Or is SiC subject to some inherent speed limitations which seem counterintuitive given SiC material properties?

Paul E. says:

The mobility (ease or speed with which electrons can move in a semiconductor) of SiC is lower than for Si – this is a fundamental material property. The mobility can be changed to a degree, e.g. by applying a stress to the semiconductor crystal, but one can only take it so far.

Reedman Bassoon says:

The article needs to distinguish between GaN-on-Silicon (Transphorm, et al) and “GaN-on-GaN” (Avogy).

Mark LaPedus says:

Avogy and its technology were covered in this article:
Searching For The Next Power Transistor http://semiengineering.com/searching-for-the-next-power-transistor/

Harry Taylor says:

I think we have a third (or fourth if you count GaN-on-GaN as a separate technology from GaN-on-Silicon) contender in power electronics, namely AKAHN Semiconductor with it’s Diamond-on-Silicon Miraj Diamond Platform. Can a small but well funded startup compete against the other well established technologies in power electronics?

Leave a Reply

(Note: This name will be displayed publicly)