Generally if you want to read about power dissipation in SoCs and the potential impact on limiting how much computer power we might be able to cram onto a given piece of silicon then EE Times is a good place to look. But last weekend there was a full-length article in, of all places a different Times, the New York Times, entitled Progress Hits Snag: Tiny Chips Use Outsize Power.
The article riffs off apaper at ISCA(link is PDF) titled Dark Silicon and the End of Multicore Scaling. It basically looks at what percentage of transistors will need to powered down at any one time. As early as next year, these advanced chips will need 21 percent of their transistors to go dark at any one time, according to the researchers who wrote the paper.
This is a big challenge, one that I mused about rather less scientifically just a few weeks ago here on SemiWiki. The challenge with multicore is to work out which parts can be powered down. You can’t power down a whole core (at least not all the time, or why bother to have it on the chip) which means a more subtle approach is required.
The article in the New York Times ends on an optimistic note with a quote from Dave Patterson of Berkeley:
“It’s one of those ‘If we don’t innovate, we’re all going to die’ papers,” Dr. Patterson said in an e-mail. “I’m pretty sure it means we need to innovate, since we don’t want to die!”
This reminds me of a quote from Maurice Wilkes, one of the pioneers of digital computing, who was the head of the University of Cambridge Computer Laboratory when I was an undergraduate. When someone argued that computers couldn’t get much faster due to speed of light considerations (this was in the days of room-sized computers) Professor Wilkes retorted that “I think it just means computers are going to get a lot smaller.” Which, of course, they did.
Next Generation of Systems Design at Siemens