On March 17, Actel Corp., based in Mountain View, Calif., released a new low-power field-programmable gate array (FPGA) that bottoms out at a power consumption of 5 microwatts. Actel is a relatively small David struggling for recognition amid the Goliaths of Xilinx and Altera. When Xilinx and Altera compete for lowest-power FPGA kudos, Actel is always conveniently left out of the discussion, says CEO John East, even though his chip outperforms both his competitors' in terms of low power—by several orders of magnitude. Actel's chips can be found in portable equipment like defibrillators, vehicle engine control modules, and rearview mirrors, elevators, and escalators. But the one market Actel dominates, jokes East, is the ”Martian market.”
East sat down with IEEE Spectrum's Sally Adee to talk about application-specific integrated circuits (ASICs) and FPGAs and how his low-power FPGAs can help save the world.
IEEE Spectrum: All Actel FPGA development is flash-based, but you're the leader in antifuse devices. How do those work?
John East: Unlike other FPGAs, antifuse FPGAs are one-time programmable. A fuse burns out when a lot of voltage goes across it; an antifuse burns in when a lot of voltage goes across it. It starts out as an open circuit at first, and then a high voltage burns it in and it's permanently connected. Our chips work that way; you send current through and that's how it sets the chip--and you can never reprogram that chip. An antifuse FPGA can get hit by any particle and be okay.
Spectrum: All the Mars rovers are outfitted with Actel antifuse chips. Why is that?
JE: Out in space, you really want an antifuse. You're getting hit with everything--neutrons, ions, heavy elements, a lot of radiation. When heavy elements hit a silicon transistor, they knock off an electron. With enough radiation hitting it, you can make that transistor be on all the time--that's called impact ionization. You could fix that by annealing [heating a chip to 500 degrees to smooth out errors], but in space you can't stick a chip in an oven. So the circuit just quits working.
If an antifuse gets hit with a neutron, it won't flip the switches. In satellites, where you really can't afford a problem, the chips are almost 100 percent antifuse. That's why all that stuff on the rovers and other equipment on Mars uses our FPGAs.
Spectrum: How come antifuse devices aren't more in use in the terrestrial market?
JE: Antifuse has the programmability (and cost) of an ASIC but doesn't have the same reprogramming vulnerabilities as an FPGA. We foolishly thought there would be a massive market for one-time programmable chips. Here's why they're beneficial: you program the chip once and then you can never mess up what's inside it. Once you tell it what you want it to be, it's that and you can never mess it up. With a static random-access memory (SRAM) FPGA, there are many ways a flip-flop switch can be changed accidentally.
We figured the world would not tolerate the uncertainty of not knowing whether their configurations have been screwed up. So we thought the one-time programmable chip would be the clear winner. Well, the fact is, it only won in superhigh-reliability applications.
Spectrum: Why do big companies insist on reprogrammable?
JE: It's hard to build a million-gate circuit--you can mess up. The engineers wanted to build them to be reprogrammed. You want to be able to say, ”Hey, I messed up, and I'll change it right now” instead of ”Oh my God, I messed up, I need to buy a new one.” And that's how reprogrammable logic took over the commercial market.
Spectrum: You talk as if you're David against two Goliaths. What's your strategy?
JE: The problem we have is, everyone knows Altera and Xilinx and their SRAM-based devices. They say they're best at power, but they only compare themselves to each other. With our flash-based devices, we are actually shockingly better at power. We're better not by 10 percent, not four times as good--we're better by three orders of magnitude. Power matters. It equates to energy, and that will be more and more important as time goes on. Actel wants to dominate the market in the applications that need low power and power efficiency. This is not a vision for the future. Today we are already the king of low power with our existing flash-based FPGAs and programmable system chips (PSCs).
Spectrum: You're touting your new ultralow-power FPGAs as being environmentally cutting edge. What led you to want to go green?
JE: Green environments matter! Energy problems are not going away--they're getting worse, actually. There's only so much more fossil fuel. We'll run out of petroleum, certainly, but not out of coal, unfortunately.
I personally am a nuclear guy. In fact, I would argue that by not building up our nuclear infrastructure over the past 20 years, we've killed a lot of people with pollution. But we shot nuclear through the heart about 20 years ago because it was so unsafe. We're engineers--we could make nuclear safe!
Spectrum: What about renewables?
JE: Renewables all have the drawback that they only provide power at certain times--you only get solar when it's sunny, wind when it's windy, and so on. You need some kind of power generation. And we are the best at energy conservation, and that's going to matter a lot.
Spectrum: Do you think your customers care about the green aspect, or are they more interested in the money-saving aspect?
JE: Many care, but not many people are buying for environmental reasons--right now.
Spectrum: What's the next great technology around the corner?
JE: Connectivity is the new thing. Most of the functions you want are available today; there are tons of different MP3 players, game formats, database formats, shows, DVD players. Right now all these things are independent. It's tying them together that's a pain. I have a great collection of 1960s music on my MP3 player. I can't play them on my cellphone. The cellphone can sort of interface to my computer at work--but it certainly doesn't do it well. And my work computer doesn't interface that well with my home computer.
The next decade is about connectivity. Ten years from now, you'll have one device--you tell it something and everything else knows what you told it. Put a note in your cellphone and it'll be on your home computer. Download a song on your home computer and you can listen to it on your iPod on the way to work, and then on your computer at work--without having to spend two hours manually synchronizing everything. Just one device will do everything.
Spectrum: I look forward to inevitably losing that device. How does that connect to what you're doing?
JE: Battery life. That device isn't going to be plugged in all the time. You don't want to always be plugging everything in today either. A portable medical application like a defibrillator--if you're in the field or in the army, that thing has to run reliably on batteries. We offer 10 to 20 times as much battery life as our competitors. And if the device is low power, it's not draining serious power from the battery.
That's not just for medical equipment; it's for anything portable and anything that needs to be alive for long periods of time without a power source. One example is the GPS systems inside shipping containers that let FedEx track shipments. Once a day the thing has to turn itself on, blast out its location, and then turn itself off.
Those are not serious dynamic power requirements. But the static power is another story. There's no one there to flip an on-and-off switch. The device itself has to be a tiny bit on at all times, just enough to turn itself on and off again once a day. That's where you need a low-power, long-battery-life chip.
Spectrum: How are FPGAs doing in the war against ASICs?
JE: The war isn't between FPGAs and ASICs; it's between FPGAs and FPGAs and between ASICs and ASICs.
FPGAs are wonderful because there's no tooling required, no $2 million price tag up front, and they're instantaneous: as soon as you know what you want, you can have exactly what you want. And FPGAs offer the flexibility to change in real time. But every tangible thing after that, we're a little worse: the cost per unit, the performance, speed--an ASIC will beat an FPGA at every tangible attribute. If you really need the attributes of an ASIC, and you can afford it, you're going to go with an ASIC.
Spectrum: What are you working on with five years from now in mind?
JE: We're examining technologies that could improve power further, such as hafnium for high-k dielectrics. Most of your readers understand hafnium, but they might not understand the surrounding environment.
Hafnium has been touted as a solution to power, but it's misunderstood. It doesn't address any power problems we're having today--it addresses a problem we will have two to three years from now if we don't develop hafnium to deal with them: quantum tunneling. As you go from 65-nanometer process technology to 45 nm and 32 nm, tunneling is about to become a staggeringly big problem. You need a wider, thicker insulator to minimize the effect of tunneling. Hafnium is a way to do that. But that's different from it solving today's power problems. All the substantial power problems we have today will still exist.
Spectrum: Speaking of static and dynamic power consumption, are you doing anything with vertical transistors?
JE: We are not. Personally, I think that idea is kind of thin. But I've gotten really skeptical and jaundiced. When I was in college, gallium arsenide was the big deal. It was going to take over because mobility was better than silicon. Forty years after I get out of EE school, we're still making everything out of silicon. Like the old joke goes, it was the technology of the future, and it always will be.
Nine out of 10 or even 99 out of 100 things people tout as being ”the future” don't end up being the future. Right now all our effort is going into silicon, horizontal, CMOS--boring, mundane stuff that just keeps on getting better because we keep getting better at it. The physics part of it is not yet at its limits.
Spectrum: You were in the news recently for having an 8-year-old doing chip testing for you. Are you violating child labor laws?
JE: Carson Page. Yeah, he's been covered around the world; he was on the cover of EETimes. He can't qualify for their student of the year award because he's not a Ph.D. student. His father, Ray Page, is one of the intellectual-property design guys. He has test benches at home, and Carson started messing around with the chips he brought home. Carson didn't do anything fancy, but he was able to turn the chip on, make things happen. He talked to the software engineer team about what he found when he was messing with it--how intuitive it is, stuff like that.
Spectrum: Are you paying him?
JE: Yes, we have an employee contract. [Laughs] No, he was just messing around. He wants to be a fighter pilot.