Sartori wins award for research on error tolerance in processors
ECE graduate student John M. Sartori has been trying to sell the processor community on an unconventional idea – being OK with processors that make mistakes. Sartori’s paper on the topic won the Best Paper Award at the 2011 International Conference on Compilers, Architectures and Synthesis of Embedded Systems (CASES).
Normally, processors are built to do their jobs perfectly, whether it be in cars, computers, or cell phones. In order to get the chips to work properly, they all have to be operated at a slow frequency or a high voltage or both. “We’re losing performance or power or both in order to make sure that a lot of these chips work,” Sartori explained.
Also, many applications that run on processors do not require the processor to be correct all the time. “So an application itself may be able to tolerate errors,” he said, “but we design every processor so that it has to always produce correct results.”
The results of these findings led to Sartori’s paper, titled “Architecting Processors to Allow Voltage/Reliability Tradeoffs.” The processors would have relaxed correctness – in other words, would allow errors on the chip – in order to save energy. “It was based on how you should architect the processor, in order to allow it to make reliability trade-offs.”
The paper was chosen as the best out of 23 papers presented at CASES. Sartori said on winning: “I was excited. They announced the candidates for the best paper award the day before, and I was surprised to hear my paper there. On the day that they announced the best paper award, they mentioned all the other candidates in order, and when they didn’t announce mine, I sort of figured that I won.”
Assistant Professor Rakesh Kumar, Sartori’s PhD adviser and co-author of the winning paper, explained that CASES had a committee pick the top four presented papers to be candidates, and Sartori won out of those four papers.
One real world impact of Sartori’s stochastic processor design could be on cell phones. “If I can make a processor that consumes a lot less energy, then the battery can also last a lot longer,” he explained.
Because Sartori’s approach is a new, different take on processors, winning the Best Paper award was an important recognition of his work, explained Kumar, a researcher in the Coordinated Science Lab. “It validates an approach which was fairly unconventional and was difficult to get across, at least in the initial stages of research,” Kumar said. “He [Sartori] wasn’t having an easy time convincing people.”
This was not the first paper that Sartori and Kumar had written on the subject. “We’re doing research on this topic, so we’ve written a few other papers relating to the same idea of building really low-power chips by allowing errors on them,” Sartori explained. However, this was Sartori’s first paper to gain major recognition. Another student in Professor Kumar’s group, Joseph Sloan, also recently had his paper recognized as the Best Paper in Session at the SRC TECHCON.
Overall, Kumar was just pleased to see his student doing well. “I was really happy for him,” Kumar said. “John has worked very hard over the past few years on his thesis and this was a good recognition of all the hard work he has put in.”
After Sartori receives his PhD from ECE, he wants to pursue an academic career and continue his research as a professor.