Alex Lidow is a third generation power engineer with ambitious targets.
“I’m third generation in power semiconductors, my grandfather and father founded International Rectifier. But I wanted to be an aerospace engineer, and went to Caltech and I fell in love with semiconductors there. Graduate school at Stanford in the 1970s was really the centre of the semiconductor universe. I remember I was walking to the Xerox room to copy technical articles and Fred Turman was there – he was a professor at Stanford that brought Hewlett and Packard together and creates the industrial park, and he talked about WBG semiconductors such as GaN back then.”
“So when I came to IR I wanted to do MOS devices, and we did. All the basic patents were in our names at the time. I’d been looking at GaAs but realised GaAs was not going to do it. In 2000 there was a group in Japan growing GaN on standard silicon, and I started that effort at IR but it was slow going. Reaching down as CEO to make something like that happen was extraordinarily difficult.”
“Having left [IR] involuntarily, I might add, we set up EPC to develop devices with higher performance and lower cost than silicon of the same power ratings,” he said.
The key is using the silicon manufacturing infrastructure to integrate other control functions on the substrate with a thin layer of GaN on top.
“We build all our devices side by side with silicon wafers in a foundry and the device is 15 to 20 times smaller than the equivalent silicon device. You can put functionality on the discrete device with almost no extra cost, a buck converter or motor phase controller in a fully integrated fashion,” he said.
Next: GaN integration
“All the analogue and digital is done at the same time, and at these levels the silicon devices have to be vertical. The 2D eGaN can conduct laterally across the surface so can integrate as easily as CMOS. The value is enormous. This gives more performance than discrete devices.”
But there has been resistance in the industry.
“When you have a new technology like GaN you have a lot of convincing to do. You create a matrix of devices, small, medium and large, and throw out that net and work the hell out of your customers, then you go find what they really need and what you can do about it.
Getting started with lidar laser imaging systems for the automotive industry was a key example of this. “We were approached by Dave Hall, founder of Velodyne, who wanted to use lidar for the Google mapping project. Five years we were putting these functions on a chip to see further and faster at lower cost. Now we are in autonomous vehicles, in robots, in unmanned aircraft.”
“We also have lots of demand in space applications. We have a whole family of rad hard devices. I developed the first rad hard MOSFETs at IR and I know that market. GaN is light years beyond silicon there. That’s how it happened in silicon 60 years ago,” he said.
Last year EPC teamed up with VPT and Dan Sable to set up a joint venture called EPC Space to make rad hard devices. VPT is part of the Heico Electronic Technologies group that specialisesin DC-DC converters for satellites.
Despite the pandemic, he is looking at automotive for growth. “The first GaN devices in 2012 were in headlamps all over the world and in lidar for autonomous vehicles. Hybrids are evolving to 48V distribution busess but you still need the 12V for legacy systems so its being designed in in 2022, 2023 for 48V distribution buses – we have six major programmes there. If there’s a 48V bus it will eventually be a GaN system.
Then there’s traction drives that current use silicon IGBTs and increasingly SiC. “The answer is I don’t know. IGBT is the best choice below 650V. SiC is the best at 900V and above. Then there’s the 650 to 900V and that’s where traction drives are either IGBT, SiC or GaN. GaN will take over that if integration becomes a key element. SiC if they can get the cost down. IGBT will take over if the other two don’t do those things.”
“I like to cultivate market, understand the roadmap, and develop a differentiated position,” said Lidow. “We did that in lidar and we are doing that in the DC-DC converter– we own the 48V node and we did it in space.
Next: GaN for data centre power
One area that has taken off is data centre power supplies. As racks of processors draw more power with currents over 1000A, so the power supplies have to deliver more power in the same footprint.
“The data centre is critical,” he said. “The 300W LLC suppliers are in volume production and these are built in the millions. It’s the beginning of the wave. People don’t pull out a MOSFET and put in GaN, it’s new designs. Three years I said I was ready for a frontal assault on silicon at 48V, it was a rallying cry. Onboarding 48V on the server, its becoming extremely important, not just in servers.
“Four of the largest server manufactures will be using 1kW GaN point of load converters in a tiny form factor,” he said. “Lots of servers are built with 2kW on the board with two converters. We focussed on custom devices for the secondary side to fit into the small form factor – you have to have certain aspect ratios.
Here it is the thermal performance that is key.
“The devices are 5x smaller than silicon but also six to ten times more thermally efficient, depending on the voltage, so we can get the same amount of heat out and typically don’t generate as much heat, so the power density is five to 10x higher,” he said.
He points to the recent Phase 12 report that confirms the reliability of the devices, an issue that silicon suppliers often raise.
“The bottom line is that the GaN devices out in the field for ten years are far more reliable than any silicon device ever made,” he said. “The statistics are adding up really quickly, we have the data now. But more importantly with GaN it was a bit of a mystery how they worked. We can take the devices and make them fail, look at how they failed, characterise them and create a model to give you an idea of how it will work in other applications, but that suffers from survivor bias.”
“With Phase 12 we didn’t just build observational models that fit the data but created a physics-based model of the crystal to construct the lifetime model of the devices. Now we can derive the behaviour, both electrical and lifetime, from the physics. It shows GaN is becoming a mature technology,” he said.
There is huge potential for the technology, he says.
“As an industry we are making trillions of silicon devices and we can make trillions of these,” he said. “EPC is private funded, not venture capital backed, and my business partner is also the owner of the foundry so we have a large foundry that can accommodate as many devices we want. We can do billions of devices a week if necessary.”
The company works with Episil in Taiwan as a foundry.
“We have plenty of reactor capacity and its not hard to get reactors but that’s the most expensive part. We have our own material capability – a couple of microns of GaN on standard silicon, that’s not consequential in the energy to create the crystal.”
GaN devices are 300x larger than they could be at the theoretical limits as a power device. When MOSFETs started they were 400x larger, he says, which gives plenty of room for improvement.
“The thing that limits GaN today is material quality, trapping electrons limits the size of the device and the size of the electric field so the devices are bigger than they need to be because of that limit. There is about 64x improvement in material quality with the electric field limitation. Not just the crystal growth but the design of the layers. We use the silicon underneath for the integration of the active devices.”
The next step is the integrated devices combining the silicon logic and GaN power.
“You will see more integrated GaN ICs in existing market and in new markets,” he said. “In space the discrete MOSFET market will fall, the power devices will go to GaN and then the integrated circuits ICs will go to GaN as they also have to be rad hard and we know how to do that.”
The key is the power density and low cost, and no one is succeeding with that in consumer, he says. For example with laptops he sees GaN enabling 60V outputs from AC-DC converters rather than 19V. “Then you can charge your laptop in 5 mins. Maybe it’s a couple of years out. That’s trillions of devices.”
“I experienced this with MOSFETs vs bipolar. Bipolar is still around, and there’s plenty of legacy markets for MOSFETs. The new markets, the 3D awareness, the satellite constellations, 5G, 6G all that’s GaN. It will take decades but it will get there.”
Other GaN articles on eeNews Power
- TI SHAKES UP POWER WITH INTEGRATED 650V FET
- MULTIPHASE CONTROLLERS AND SMART POWER FOR 1000A AI LOADS
- EPC, MICROCHIP TEAM ON THIN 48V DC-DC CONVERTER DESIGNS
- 200V FET DOUBLES PERFORMANCE