Alex Lidow has spent 50 years working in power electronics. He was one of the co-inventors of the HEXFET power MOSFET and later founded Efficient Power Conversion (EPC), where his attention shifted from silicon to gallium nitride.
At PCIM 2026, we sat down with Lidow to talk about that journey and where he thinks power electronics is heading next. The conversation covers the early days of the power MOSFET, why he believed silicon was running out of room, EPC’s difficult first steps with GaN, and the growing use of the technology in AI infrastructure, robots, drones and satellites.
A Mistake, a Two-Month Deadline and the HEXFET
Lidow’s story starts well before GaN.
Early power MOSFET layouts were drawn by hand. As he recalls in the interview, that meant sitting at a drafting table and drawing semiconductor features at roughly 200 times their actual size. A 6 mm device could take up a huge section of the table, with individual lines drawn using a ruler before the design went to a mask maker.
One of those early devices did not work as planned. It was supposed to measure around 1 ohm but came back at roughly 2 ohms.
The problem was eventually traced to the layout. Two features had been drawn too close together, pinching off the region between them and restricting the movement of electrons. Starting again would have meant redrawing the device, producing new masks and running another wafer cycle. The team was around two months from launch.
Lidow tells the rest of that story in the interview, including the idea that came to him while driving to his brother’s birthday party on the San Diego Freeway. The eventual result performed better than the device they had originally intended to make.
That work preceded the HEXFET, which took the power MOSFET into a cellular structure. Lidow describes it as a honeycomb, with the source connection at the centre, the gate around the cell and the drain underneath. The hexagonal layout increased the amount of channel that could be fitted into a given area while avoiding the sharper 90-degree corners of a square cell.
The devices are still being made today.
Why Lidow Moved Beyond Silicon
Around the turn of the century, Lidow reached the view that silicon was approaching its theoretical limits for power conversion.
The next step came from research in Japan, where device-grade GaN had been grown on silicon. That caught his attention because it offered a route to manufacture GaN devices in a silicon wafer fab without replacing the entire manufacturing base.
In the interview, Lidow goes into the device physics behind this in some detail. He explains why GaN cannot simply be grown directly onto silicon, how aluminium nitride and aluminium gallium nitride are used through the structure, and how strain in the crystal helps create the high-mobility electron channel used by a HEMT.
His view is that the technology still has a long way to go. “We’re 100x away from the limit of performance,” he says.
That is also his answer to questions about what might come after GaN. Gallium oxide, aluminium nitride and diamond are all discussed as possible future semiconductor materials, but Lidow sees little reason to take on that level of technical risk while GaN itself remains far from its limit.
Finding the Application That Actually Needs It
EPC’s first GaN devices were not an immediate success. Lidow recalls wafer runs where nothing appeared to work. One of the company’s early employees then tested tens of thousands of devices and found a single working part.
“She actually sat down and she tested tens of thousands of devices,” he says. “Every single one of them.” Then she found one.
For a company running short of both time and money, that device mattered because it proved the process could work. Within a few months, EPC was able to reproduce the result.
Commercial adoption was another problem. Lidow says the first generation attracted interest, but customers were not actually using it. That changed when Dave Hall approached EPC with an application based on firing a laser quickly and powerfully enough to create accurate three-dimensional sensing. The application was LiDAR.
Silicon had already been used, but Lidow says its switching speed limited the resulting system. GaN changed what could be done with the laser pulse and gave EPC its first major application.
His conclusion from that period is simple: “If you don’t have a killer app, you’ve got a technical curiosity.”
Why EPC Did Not Follow the 650 V GaN Market
Much of the GaN industry concentrated on 650 V devices. EPC went in another direction.
The company focuses heavily on 200 V and below, and Lidow says that choice came from his earlier experience with power MOSFETs. In his view, the 650 V market tends to become highly price-driven, with limited value placed on incremental device performance.
For a smaller company, he saw little reason to enter a market where price would quickly dominate. Instead, EPC looked for applications where higher performance changed the system enough for customers to pay for it. More recently, that has included humanoid robots.
Lidow says EPC noticed that companies building humanoid robots were buying its ICs and started asking why. The answer was the need to fit dense motor drives into joints and other constrained spaces. That work then carried into drones, e-bikes and power tools.
He describes the strategy as starting at the top of a pyramid. Find the application with the hardest requirements first, then use the same technology in adjacent markets as manufacturing improves and costs come down.
AI Power Is Moving Again
EPC took a similar approach with DC-DC conversion. Around eight or nine years ago, the company identified high-density computing and 48 V distribution as a market where GaN performance could matter. It began developing devices around that requirement and is now on its seventh generation.
Lidow makes a strong claim for Gen 7. He says it is the first generation that “completely writes over all MOSFETs” because of the gains made at lower voltages, including 40 V, 25 V and 15 V. The timing matters because data centre power architecture is moving again.
For years, 48 V distribution was one of the main talking points around high-density computing. Lidow now sees a wider spread of voltage nodes, including 12 V and 6 V, while parts of the AI infrastructure market examine 800 V distribution. That creates some very large conversion ratios.
“Let’s go straight 800 down to six,” Lidow says during the interview. “What’s the best way to do that? We’ll use GaN.” He expects GaN to proliferate through the data centre rather than sit at one fixed conversion stage.
GaN in Space
One of the less obvious parts of the conversation concerns space electronics.
Lidow argues that GaN is fundamentally better suited to radiation environments than silicon MOSFETs. His explanation starts with the oxide used in MOS devices, where high-energy electrons can become trapped and gradually shift device behaviour.
GaN devices do not have the same oxide structure. Lidow also points to the stronger chemical bond between gallium and nitrogen when discussing exposure to protons, neutrons and heavy ions.
This is an area he knows from both sides. In the interview, Lidow notes that he worked on radiation-tolerant MOSFETs in the early 1980s and says those devices required a performance compromise to improve their resistance to radiation.
His argument is that GaN does not require the same trade-off. According to Lidow, EPC has gone from zero to 35% market share in satellite electronics in four years. He sees communications constellations as the larger commercial opportunity, alongside longer-duration space exploration systems exposed to severe radiation environments.
What Comes Next for GaN?
Lidow does not describe the next phase as a simple replacement of silicon. For the next three or four years, he expects the focus to remain on maturing the technology. After that comes cost.
The strategy he describes is to stay ahead in device performance, win the applications that value that performance most, build volume and then improve the manufacturing economics. Only then does the technology move further down the pyramid into markets where cost matters more.
It is a view shaped by five decades in the same industry. Lidow saw switching power conversion held back by bipolar transistors and thyristors, helped develop the power MOSFET technology that followed, and later decided that silicon itself was running out of room.
Now he thinks GaN still has plenty left.
Watch the full interview above for Alex Lidow’s account of the HEXFET story, EPC’s first working GaN device, the applications that pushed the technology into volume, and where he expects power electronics to go next.
You may also like
