M. Hassan Najafi Wins ICCD Best Paper Award

The best paper award at the 35th IEEE International Conference on Computer Design (ICCD) conference has been awarded to “High Quality Down-Sampling for Deterministic Approaches to Stochastic Computing” authored by M. Hassan Najafi and Prof. David J. Lilja.

The paper was one of 75 accepted full-length papers out of 258 submissions at the November conference and also among the 12 top ranked ICCD papers selected for publication in the IEEE Transactions on Emerging Topics in Computing, Special Issue on Emerging Technologies in Computer Design.

Hassan is a doctoral student working under the guidance of Prof. David Lilja and his research interests include stochastic and approximate computing, fault-tolerant system design, and computer architecture. He is also the recipient of the University’s Doctoral Dissertation Fellowship that recognizes outstanding research work, for the 2017-2018 academic year.

A brief description of the paper:

Recent work on stochastic computing (SC) has shown that computation using stochastic logic can be performed deterministically and accurately by properly structuring unary-style bit-streams. The hardware cost and the latency of operations are much lower than those of the conventional random SC when completely accurate results are expected. For applications where slight inaccuracy is acceptable, however, these unary stream-based deterministic approaches must run for a relatively long time to produce acceptable results. This long processing time makes the deterministic approach energy-inefficient. While randomness was a source of inaccuracy in the conventional random stream-based SC, the authors exploited pseudo-randomness in improving the progressive precision property of the deterministic approach to SC. Completely accurate results are still produced if running the operation for the required number of cycles. When slight inaccuracy is acceptable, however, significant improvement in the processing time and energy consumption is observed compared to the prior unary stream-based deterministic approach and also the conventional random-stream based approach.