Alex Lidow, former CEO of International Rectifier and CEO of Efficient Power Conversion, is one of the few individuals to have pioneered two revolutionary power semiconductor technologies – the silicon power MOSFET and the gallium nitride HEMT. How did he come to have these two unique opportunities?
Elektronik: Although you created the foundational innovations for the commercial power MOSFET, the roots for this start when you entered graduate school. How did start all this?
Alex Lidow: Before entering graduate school, I worked in the summer of 1975 in the R&D lab at International Rectifier developing bipolar transistors. With this experience in my mind, I went to Stanford University being sure that silicon transistors for power electronics had come to a dead end. We needed something new. Therefore, I spent my graduate work looking at gallium arsenide as the power semiconductor material of the future. By that time, I recognized its positive attributes, but also its downsides, especially the manufacturability and the high costs.
At the same time, I saw a friend’s stereo equipment, I think it was from Sony, which had a power MOSFET inside. It was a very primitive kind of a MOSFET, a lateral device. It was used in the amplifier section because of its linearity. It was very interesting, very intriguing to tear down this device. And this triggered the idea that we do not need better bipolar transistors in the area of power semiconductors, but to use the field effect to modulate the current.
Together with the knowledge, I gained at the integrated circuit lab at Stanford to scale circuits in gallium arsenide. I realized that this could be a way to create very high-density power transistors much beyond what the lateral device in this stereo equipment was capable of.
Then you presented this idea to your former colleagues at the R&D lab at International Rectifier. Since you were the son of the founder and CEO of IR, they must have been excited about it, weren't they?
No, they were not. They liked their thyristors and all that other stuff. But my father had great faith in me, and he encouraged me to go ahead.
At Stanford, I was in the same group as Tom Herman, and by the time we became close friends. He graduated a few months before me and started to work for International Rectifier. He set up a lab there; we went in there and decided to use integrated circuit MOS technology to make power transistors.
Then there was set a target to achieve one ohm for a 400-volts transistor. How did it come about?
In early 1978, my father assigned Bill Collins, a very experienced gentleman, and he said one day: “What the world needs is 400 volt/1 ohm! If you can do that, you own the world, because then you can do AC to DC conversion.” Tom Herman and I looked at each other and we sketched some ideas.
That same day in the afternoon, I booked a flight to Stanford, where I graduated a just a few weeks before, and I showed our ideas to my buddies there. And they liked it. Eventually, John Shott said: “Let’s cut masks!”
Still the same day in the evening, we stood around a big light table cutting masks out of Mylar with razorblades. We worked all night, and the next morning we shot these giant translucent masks with a photo camera being some 20 to 40 feet away to shrink the mask. We repeated this manual shrinking process until we had the right format for semiconductor manufacturing.
But that was not the end. That very day you had the first prototype MOSFETs in your hands. What happened?
Across the hall some students were growing some epi-wafers, and they gave one to me. So we had everything we needed for making prototypes. I went back to Tom’s lab with a box of process-ready silicon wafers and a set of masks. Only 72 hours after Bill Collins said that there is a need for a 400-volt/1-ohm device we had first working prototypes.
Did you achieve the target of 1 ohm?
Well these first prototypes were just a proof of concept. The following weeks we worked intensely to scale it up to meet the 1-ohm target. But by the time we had first samples, they didn’t work well. They actually had 2 ohms which was a disaster. But Tom Herman upon looking at my enormously detailed draft drawings on Mylar found an error. The very first two lines of a node were too close together. What a disaster! I really felt bad. We had to start over again. We will miss the scheduled launch day.
On July 22nd of 1978, while driving to my brother’s birthday party, I realized I could turn my mistake into an advantage. By adding a certain step, we could make a much higher density product. It was to add charge in that space that I had designed too small. That became the basis of a silicon power MOSFET that was far superior to anything developed before.
Did this first product have this hexagonal structure for which IR became very famous for with the brand name HEXFET?
Actually no. On launch day, I was in New York City visiting very proudly all the editors there with our aluminum gate MOS transistor which was 10x better as anything else. At 2am at night, Tom Herman called me and he said: “I have an idea. Take out the pencil.” So we sketched out this hexagonal structure which eventually became the HEXFET. The minute we did this, we had a very mature set of inventions which was the dominant force for 20 to 30 years to come. Basically, all the planar silicon MOSFETs, the IGBTs, and the silicon carbide MOSFETs still use this structure.
With every novel component structure, robustness and reliability are questioned. What have you done to respond to this?
This was a big deal, because there were two oxides we had to worry about – the edges of the device and the gate oxide. Back in those days, bipolar transistors were glass-passivated at the edges. The chip manufacturers etched a deep trench at the edge of their devices, filled it with glass powder and heated it up so that the glass melted. This was a very dirty process step and it was not scalable. Therefore, we decided not to use that for our MOSFETs. Instead, we developed a high-voltage structure using floating wells and field plates in order to use a grown oxide to passivate the edges of the device. That was very innovative as well.
On top of that, a MOS device has a gate oxide between the gate metal and the underlying channel. Bipolar junction transistors don’t have this. This was something completely new in the field of power transistors.
When we first launched the product, most of the world ignored it, just like it happened to gallium nitride HEMTs just until recently. After a while the big manufacturers of bipolar transistors like Motorola said: “Dear customer, don’t play with that stuff. These delicate little glass structures will break, if you use real power with it.” They used that theme for quite a while, so that we had to do an enormous amount of work proving reliability. That gave me a lesson how to overcome the natural and rational fear that new technologies might fail in ways nobody predicted.
That was the original thinking behind our test to fail methodology. We stressed our devices to fail and then analyzed why and how they failed. By that, we were able to improve our devices to make them more robust and reliable.
But eventually the silicon power MOSFET gained traction and became the dominant power semiconductor technology. I remember that you said you received a phone call from Steve Wozniak, the Apple co-founder. Was that the turning point?
This was a big deal, but not the commercial breakthrough. Wozniak’s idea was to incorporate the power supply into a computer which you can put onto your desk. By that time, AC to DC power supplies were so big and bulky that they would never ever fit inside a desktop computer. But that was Wozniak’s vision. We had a quite prominent applications engineer at that time, Brian Pelly, and a very talented Indian engineer, Rutton Ruttonsha. They both figured out how to use this 400-volt/1-ohm device to build a single-ended flyback converter for a universal-input AC to DC conversion. You were able to use this converter for the 110 volt mains voltage in the US as well as the 220 volts in Europe, and this 300- or 400-watt power supply with its some 100 kilohertz switching frequency was small enough to fit inside a desktop computer. Eventually this was incorporated in the Apple IIe.
But the real turning point for the commercial success of silicon MOSFETs was that IBM adopted this technology for their famous personal computer. This meant huge volumes for our product.
After making a fortune with silicon power MOSFETs, you turned to gallium nitride. Why?
By the start of the millennium, some 22 years after inventing the silicon power MOSFET, we realized that the silicon MOSFET technology hit the Baliga limit. So what to do? Could we find innovative ways to manufacture the devices? Should we invest multiple billion dollars in the most advanced wafer fabs to scale down costs? That created certain restlessness.
By that time, Nobuaki Sekine, a good friend of mine and head of R&D at Sanken Electric, told me about a paper from Japanese researchers. They have grown device-grade layers of gallium nitride on top of a standard piece of silicon.
This immediately triggered a thinking process on my part. From my graduate work with gallium arsenide, I knew that it is extremely expensive to grow a compound semiconductor crystal. With this hetero-epitaxy process of growing a thin layer of gallium nitride on a standard silicon wafer, I saw an approach to overcome the cost barrier associated with the extremely energy-intensive growth process of wide bandgap semiconductor crystals. You will never ever make a silicon carbide or a gallium nitride wafer as cheap as a silicon wafer of the same diameter. But growing a few microns of material on a standard silicon wafer won’t you cost that much.
On top of that, gallium nitride can be processed at the same temperature levels as silicon. That means that GaN-on-silicon wafers could be processed with standard equipment. So, these pieces started to form a picture to me. The idea was to make GaN on silicon just like silicon germanium on silicon. But gallium nitride shows magnitudes better results as silicon germanium because of its much higher critical electric field, its much higher electron mobility, and its much higher saturation velocity. So, I started some research about it within International Rectifier.
But you also got help from an old Caltech buddy, didn’t you?
Yes. I came across a little start-up company. It was started by my advisor as an undergraduate student at Caltech, Tom McGill, together with a couple of graduate students at Caltech. We bought that company, and the two founders, Bob Beach and Paul Bridger, started to work for me at International Rectifier. We set up a fab line, but the progress we made was really slow. The reason for this was that IR was a big company by that time, and everybody loved their MOSFETs. Just like Motorola with bipolar transistors, the entire ecosystem at IR wanted to explain that gallium nitride was never going to work. History repeats itself. I was really struggling with it.
But then you were fired as CEO of International Rectifier.
Actually, although it certainly didn’t seem like it at the time, this was great news for me. Being the CEO of the company for 12 years, the board of directors became unhappy with me and my father, the founder of IR, and we were kicked out.
But what to do next? I was preparing myself for my very last day at IR, a little bit down, when I made a decision. One option was to introduce some form of carbon trading to account for the impact of CO2 emissions. In the case of fossil fuels, such as coal-fired power generation or cars with internal combustion engines, the environmental impact, the cost of cleanup and the effects on climate change, is not costed in. But in the very near future, we will have to do so. The other option was to pursue gallium nitride. My heart wavered back and forth between these two. But at the time I was ready to start my last day at IR, I made up my mind to go after gallium nitride. That was the start of EPC. We were incorporated just 30 days later.
I remember you saying that today the market for bipolar junction transistors is still as big as in the late 1970s. Looking into the crystal ball, will the market for silicon power MOSFETs be in ten years as big as today or will gallium nitride and silicon carbide win market share from applications which traditionally use silicon MOSFET?
There is a general misunderstanding about how technologies evolve over each other. In the early days of the MOSFET, we placed ads saying “Bipolar is Dead” with a red circle and a cross over the bipolar transistor. Today, we know that the revenue in terms of Dollars for bipolar transistors is as much as in 1978. What really happened was that the growth rate for bipolar transistors came down and eventually stopped, when all new designs were using MOSFETs. But the legacy designs clung to the bipolar transistors.
This process took some 10 years. First there were the early adopters like Steve Wozniak, and then the technology gained traction in applications for which bipolar transistors were not a viable solution. But the trigger point for the end game was when the price for MOSFET fell under the price of an equivalent bipolar transistor. That was in 1988. After that, you could not be competitive with any design and stay in business, if you did not use a MOSFET.
Gallium nitride is in the same position right now, as well as silicon carbide by the way. As being so long in this business and being through this learning curve before with MOSFETs, I knew that our strategy must be to build gallium nitride devices which have better performance at lower cost as equivalent MOSFETs in order to make this transition. Therefore, we worked on the cost angle from day 1; and by about 2015, some 8 years after founding EPC, I recognized that we have crossed the MOSFET cost barrier.
Remember, I had been the largest producer of MOSFETs in the world when I left International Rectifier. I knew about cost, I knew how brutally competitive the cost structure was with MOSFETs. Therefore, I definitely knew when we crossed that barrier with our GaN devices at EPC. And since then, we never looked back, but shrunk our devices further and improved the reliability and robustness of our products as well as our processes.
Still a lot of people have in mind that such a new technology like gallium nitride is higher prize as standard silicon MOSFETs. But more and more people are approaching me and realizing, "Your GaN devices are actually cheaper than MOSFETs! How is that possible?"
And of top of that, you can deliver your GaN devices off the shelf, whereas MOSFETs are on allocation.
Exactly! At the moment we can price our GaN devices around mid-range, because MOSFETs have some 63 weeks lead time. Of course, that will change again. MOSFETs will come again and will be deliverable off the shelf. But at the moment, frankly spoken, we don’t have to be cheaper as MOSFETs, but we can be, if we need to. But it took some 10 years, similar as with MOSFETs over bipolar transistors.
And once again, as I experienced it with the big bipolar manufacturers, all the big MOSFET people have said in the first years of GaN: “Dear customer, don’t play around with these tiny, delicate devices. They are not reliable. They will break, if you use real power with those.” It's like déjà vu for me, history repeats itself.
But all major power semiconductor ODMs are also manufacturing GaN devices. How do you make sure to stay ahead of them?
The maturity of GaN devices is now somewhere like MOSFETs were around 1988. There is a lot of room for improvement until we hit the technological limits. I have to make sure that our learning cycles are faster than those of the established majors like Infineon or onsemi. There will come the point in time when these cycles slow down and the big corporations with their massive financial capabilities will catch up. But I hope that EPC will then be one of these big corporations with massive financial capabilities.
Let’s take a quick look into the crystal ball. Do you think there is a market and a need for vertical GaN? Because today’s GaN devices are all lateral GaN-on-silicon devices.
I think there may be some small niche markets where vertical GaN would make sense, but in general no. The fundamental reason for that is that semiconductor-grade gallium nitride crystal is at least as expensive as silicon carbide. But with silicon carbide, we have over 20 years of industrial experience and investments, established processes, and an established supply chain. With vertical gallium nitride, all this investment has yet to be made. This is one point.
The other is that with vertical GaN you gain almost no electrical improvement over silicon carbide in terms of electron mobility, critical electric field and so on. But there is one parameter in which silicon carbide is three times better as gallium nitride – thermal conductivity. In essence, a vertical GaN device is nearly the same size as an equivalent SiC device but is much worse in thermal conductivity. And now think one moment about how much you have to invest just to get a device that is nearly as good as an established device technology like silicon carbide MOSFETs! This doesn’t make any sense to me.
But with a lateral device like GaN-on-silicon you have the inherent possibility to integrate more functionality and/or more than one power switch onto one die, haven’t you?
This is a fundamental advantage of the GaN-on-silicon approach. This 2-dimensional electron gas doesn’t occur in silicon. By this quantum physics phenomenon, you get a boost in charge carrier mobility and saturation velocity. This means that such a lateral HEMT can be better than any vertical device like a MOSFET, in silicon or silicon carbide.
On the other hand, a lateral GaN-on-silicon HEMT is much like any MOS integrated circuit in silicon. You can integrate complex circuitry on it together with a power switch. In silicon, you cannot do that. If you want to switch high currents and high voltages in silicon, you have to go vertical. And you cannot put two vertical power switches onto one die, because they pollute each other with electrons and holes. Therefore, in gallium nitride you can really make a real power system on a chip, which is not doable with vertical devices. This leads to a cost advantage over a solution with multiple silicon and/or silicon carbide chips that is even more obvious.
Aside from all your achievements, what has been your biggest mistake, error or fallacy in your professional career?
The biggest error was complacency. In the 1990’s, Siliconix introduced trench MOSFETs for lower voltage applications. I spent too much time trying to prove that the planar solution Tom and I had developed many years earlier was superior. We were wrong, and it took us about 5 years before we developed a trench technology that could effectively compete with trench. The result was a temporary loss of technology leadership. Our customers had come to count on International Rectifier for the best solutions, so when we failed to deliver the best, they were disappointed and did not listen to us as closely going forward.
Many thanks for your time, Alex.
This interview was conducted by Ralf Higgelke.