CSL researchers explore using residue number system to reduce power usage
CSL Professor Rakesh Kumar recently received a three-year, $75,000-per-year grant from Intel to explore innovative ways to disperse the power that computer chips use.
For the past decade or so, the computing industry has applied itself to reducing power use. While much research has been done in this area, Kumar’s research takes a nontraditional approach to the problem. Computers have always represented numbers in a binary number system (BNS) format, but Kumar and one of his students, Biplab Deka, are considering a residue number system (RNS) instead.
The RNS format has been used in signal processing before, but “for the first time we are trying to see if it can be used for general purpose computers,” Deka said. The main reason for using an RNS is that it helps decompose complex computations into simpler ones. Instead of performing large computations, everything is broken down into subcomputations to reduce effort and therefore power. It’s like comparing the effort that one person has to put in to multiply 56 by 216, with the effort that multiple people put in multiplying the subcomputations of those numbers using only factors of one through nine.
Using an RNS may also increase reliability. By doing more work (but still using less effort) the RNS can check for errors, just like the multiple people who work on the subcomputations can check each other’s work. This may make some of the work redundant, but it’s a trade-off for error tolerance.
Using an RNS is just the first candidate for a lower-power solution.
“The first milestone would be getting and understanding the tradeoffs of this system, then we’ll know where to go from there,” said Kumar, an assistant professor of electrical and computer engineering. Due to the exploratory nature of the research, Kumar and Deka may explore other options in the future.
“Frankly we are still trying to figure what we are trying to do,” Kumar said. “But to use a cliche, if you knew what you were doing it wouldn’t be called research.” Intel is funding this high-risk, high-reward research due to the project’s potential benefit to the company’s products that planned to be launched between 2015 and 2020.
Kumar and Deka work in the Reliable and High Performance Computing research group at CSL. Deka has been a graduate student at U of I since January of 2011. Prior to that, he earned his bachelor’s degree in electrical engineering at the Indian Institute of Technology.
“My being comfortable with a student who has been here for a small amount of time working on an open-ended and challenging problem like this speaks to the kind of students that we get here at Illinois,” he said about Deka.