Professor Rakesh Kumar’s paper “On Reconfiguration-Oriented Approximate Adder Design and Its Application” that appeared at International Conference on Computer-Aided Design (ICCAD), the premier conference in electronic design automation, in 2012 has been selected to receive the William J. McCalla ICCAD Ten Year Retrospective Most Influential Paper Award. The award is given to the paper judged to be the most influential on research and industry practice in computer-aided design of integrated circuits over the ten years since its original appearance at ICCAD.
The general idea of the paper, with co-authors Rong Ye, Ting Wang, Feng Yuan, and Qiang Xu, is that a growing number of applications do not need “strict” correctness, but just “approximate” correctness. So why waste energy on situations where “approximate” would deliver proper results?
Professor Rakesh Kumar gave the example of watching a football game on television. If a small part of the screen became slightly pixelated just as a quarterback was making a touchdown, you may not even notice or know what happened after the screen righted itself. This is a situation where “approximate” results are acceptable.
Now, there are situations where you would want the results to be definite. Kumar used the example of money in your bank account. You want to make sure those numbers are accurate.
As the computation correctness (or quality) requirement of an application may vary significantly at runtime, it is preferable to design quality-configurable systems (QCSs) that are able to tradeoff computation quality and computational effort on-the-fly according to application needs. In the award paper, the team proposed a design of a quality-configurable adder – they call it gracefully-degrading adder - whose quality can be tuned at runtime such that the proper amount of energy is dissipated for the desired result accuracy. They showed that image processing hardware built using this adder can have much higher throughput and image quality than existing designs. The paper spawned a lot of follow-on works on design of quality configurable hardware both in academia and industry and eventually led to this Most Influential Paper Award recognition.
This is the fourth Most Influential Paper Award for Kumar. He has previously won Most Influential Paper Awards at ISCA, MICRO, and ASPDAC. ISCA and MICRO are premiere conferences in computer architecture. ASPDAC, like ICCAD, is a premiere conference in electronic design automation. He said he was very lucky to receive these recognitions from multiple research communities.
“I have had wonderful students, collaborators, and colleagues over the years,” said Kumar. “It is always gratifying to know that your work is having an impact,” he added.