Gonugondla wins Best Paper Award at IEEE Conference
Circuits are the glue that keep all kinds of systems connected, working behind the scenes to power the technology we use every day. Researchers at CSL are working to develop a technique that can make integrated circuits operate faster, consume less energy, and maintain a high level of correctness at the system level, even when there are errors at the circuit level.
These findings are detailed in the paper, “Perfect Error Compensation via Algorithmic Error Cancellation,” which won the Best Student Paper Award at the 41st IEEE International Conference on Acoustics, Speech, and Signal Processing, in Shanghai, China.
First author Sujan Kumar Gonugondla is a CSL and ECE graduate student, advised by CSL Professor Naresh R Shanbhag, the Jack S. Kilby Professor of Electrical and Computer Engineering and the director of the Illinois-led SONIC Center. Gonugondla collaborated with Professor Byonghyo Shim from Seoul National University, a former CSL student who was also advised by Shanbhag.
“Computing systems that process massive amounts of data today consume a lot of energy, and such systems are not usually suitable for portable platforms. So what we proposed is an algorithmic technique to reduce the energy consumption without compromising performance,” said Gonugondla.
Specifically, Gonugondla developed a technique to pack more information processing into increasingly smaller chips without sacrificing processing speed or error detection and correction rates. The result is an integrated circuit that is 25 percent faster, using 31 percent less energy, and correctly performing at the system level even while making errors at a rate of 73 percent at the circuit level.
“As circuits become smaller, this threatens their reliability. In order to make them reliable, we developed an error compensation technique," said Gonugondla. "The difference between our technique and others is that ours is very good at finding and handling errors, while maximizing the performance and reliability with limited energy overhead.”
To make the integrated circuits of the future, Gonugondla combined statistical techniques with implementation constraints.
“We're using expertise from statistical signal processing theory to solve the reliability problems in integrated circuits today,” said Gonugondla. “This work will help make future circuits more energy efficient and reliable.”
Gonugondla works in the SONIC (Systems on Nanoscale Information fabriCs) Center, a multi-university research center led by the University of Illinois that focuses on the design of robust, energy-efficient, and intelligent computing platforms using emerging nanoscale devices.