Transphorm, a startup out of UC Santa Barbara, has come up with a set of chips that could substantially curb power consumption in computers, electric cars, motors and other appliances.
The big question now is whether they can pull it off.
The company essentially has created a semiconductor platform for making power converters -- AC to DC power converters, DC to AC converters, AC to AC chips and DC to DC chips -- out of gallium nitride, the same semiconductor material behind white-light LEDs. Right now, converters are made out of silicon and can be 85 percent to 90 percent efficient, i.e., 85 percent of the AC power that gets injected into them comes out as DC power. The rest gets converted into waste heat.
Transphorm can boost that figure to the upper-90-percent range.
"Silicon has reached its limits in power conversion. It has reached its physical limit. We are using gallium nitride to move away from the path set by silicon," said Umesh Mishra, Transphorm's CEO, who said that the company's chips can utilize "up to 90 percent of the remaining 10 percent" that now ends up as waste heat.
That bit of waste heat adds up. Transphorm estimates that if its solution were applied universally, it could save hundreds of terawatt-hours of energy a year. New York City, he noted, only consumes 50 terawatt-hours a year. Las Vegas consumes 33 terawatt-hours and the whole Western U.S. consumes 240-terawatt hours. And conversions are impossible to avoid. The grid delivers AC power. However, nearly every appliance you can think of -- notebooks, servers, electric motors, solar panels, batteries -- runs on DC power. (Disclosure: yesterday, we predicted that the announcement would have something to do with improving AC-DC conversions, but guessed wrong on the company. Transphorm has been in deep stealth mode and I hadn't heard of it until today.)
More efficient converters would also reduce the demand for air conditioning in buildings and data centers: a substantial portion of the world's air conditioners are on to get rid of waste heat generated by lights and servers. Transphorm's converters will also reduce the need for additional components. Thus, although the chips will sell at a premium at first, the total bill of materials will be relatively comparable and in a few years conceivably cost less than traditional components, according to Mishra.
"Inverters are everywhere. We can make power conversion cheaper and easier," said Randy Komisar, a partner from Kleiner, Perkins, who predicted that Transphorm's technology could have three times the impact that digital lighting will have.
The first products to contain Transphorm chips will come out toward the end of this year.
Besides Kleiner, the firms Foundation Capital, Lux Capital and Google Ventures have invested a total of $38 million in the company.
It was probably the most glamorous announcement ever made in the history of AC-to-DC converters. Five exits down the freeway, the CEO of AC/DC king National Semiconductor was probably cursing his fate.
So what's the catch?
Getting a new type of semiconductor to market is never easy. In fact, it is one of the most difficult tasks in the technology business. Just ask Rambus, which tried to introduce a new type of computer memory to the world in the '90s. Or IBM, which touted super-speedy germanium semiconductors for years. Or Intel, which has been talking about ovonics memory since 1970. Ovonics chips only started to come out commercially last year and only in very tiny volumes.
Transphorm's chips depend on materials, processes, circuits and a module that are unique to the company. Transphorm plans to build a factory to make its own chips. It will not be able to rely on third-party foundries for some time.
"There will be a time for foundries, but the time is not now. It is still slightly mysterious," he said.
The company's prototype factory in Santa Barbara will allow it to meet its targets through 2012 but after that it will need to expand. And to expand, it will need a lot more than $38 million. Fabrication facilities are notoriously expensive.
Then there is customer inertia and reluctance. Transphorm will have to show reliability data and volume manufacturing commitments. No server maker is going to want to have to delay a line of computers because their supplier of power conversion chips came up short one quarter. Imagine the chatter in Detroit if an EV maker had to recall a car because of a faulty AC converter. (Delays to Rambus products embarrassed many PC makers and killed interest in its line of memory offerings.)
Gallium nitride is also not easy material to work with. "You cannot mine GaN. It has to be grown on foreign substrates," said Mishra. What's more, the substrate in some situations can degrade it.
To the company's credit, it understands these issues. Despite the celebrity name recognition that Kleiner and Google have, Mishra says the company will crawl before it will walk. It is targeting very specific vertical customers at the moment. First, it will go after servers. In a few weeks, it may be able to commit to providing chips to notebook makers. Then Transphorm will go after PV inverters and electric motors after that. Only later after it is established will it go after auto makers.
"There will be substantial penetration within the decade," he said.
Another plus: Transphorm will be able to leverage the manufacturing tools developed for the LED industry. (Side note: Shuji Nakamura, who invented the white LED and is behind the startup Soraa, works at UCSB as well.)
"The full infrastructure today is completely applicable to us," Mishra said.
Like in LEDs, others will come into market. The ultimate difference between Transphorm and a new LED company, said Mishra in a hallway interview, is that the LED market is on its way.
"You want to be at the beginning," he said.