Scientists have made a curious breakthrough in computer chip technology. They’ve discovered that if you “prune” a chip’s design–chopping off little-used functions and actually allowing it to make errors–it can result in far more power efficient and smaller designs. This breakthrough could be crucial for mobile electronics.
The team performing this research comes from the U.S., Singapore and Switzerland, but it’s headed up by Krishna Palem, professor of computing at Rice University in Houston. He’s noted that this might be the first time a research team has taken an integrated circuit and said “let’s get rid of the part we don’t need.” This approach has been used successfully in coding. Specifically in “reduced instruction set computing,” which disposes of complex processor-level instructions, trading them for multiple simpler codes that are more efficient. That was the initial secret behind the success of ARM, famous for its low-power, efficient chips. The team is revealing the technique at the European DATE11 chip design conference this week.
The idea is deceptively simple, and will appeal to design and architecture purists: Take a complex chip designed for a specific purpose (like a chip in a hearing aid or camera) and then remove all the unnecessary circuits that are merely there to assure its function, or are left over from earlier iterations of its design. As a result the chip will still work for its intended purpose, but may make “mistakes” in processing–although by careful management of error correction code and circuitry the impact of these errors these can be minimized. The result is a smaller chip layout that’s more power efficient and, possibly, faster too.
The research team tested their ideas by taking a chip design and prototyping it on an experimental silicon slice, alongside an identical but “pruned” version to compare their performance. The results are impressive: Some test designs were twice as fast as the initial chips, using around half the electrical energy, and taking up about half as much space on the silicon substrate.
The increased error rates mean the idea isn’t applicable to all sorts of chip design–you’re unlikely to see the processor in your iPhone benefiting, but other task-specific applications like audio processors or video processors could easily benefit, which could result in important power savings on the complete circuits of portable devices. When you’re talking about smaller hearing aids that can run four times as long on the same sort of battery technology, this is a serious boon.
To read more news like this, follow Fast Company on Twitter: Click here.