Approximate computing is an emerging computing paradigm, where computing accuracy is relaxed for improvements in hardware metrics, such as design area and power. Approximate computing is particularly attractive in applications where the arithmetic calculations show resilience to small errors, such as in image processing, computer vision and machine learning. In circuit design, a major challenge is to design approximate circuits from input exact circuits. In this work I present an overview our recent efforts in designing approximate circuits, including (1) design of approximate arithmetic circuits (e.g., multipliers, dividers), and (2) automatic synthesis techniques for generating approximate circuits from general-purpose circuits, and (3) application-specific approximate circuits for deep learning. I will overview some of our open-source tools, including ABACUS, BLASYS and BACS.
Sherief Reda is a full Professor at the School of Engineering, Brown University. His research interests include energy efficient computing, design automation of integrated circuits, reconfigurable and embedded systems, and computer architecture. Prof. Reda has authored/edited two books and has over 120 articles. Professor Reda received three best paper awards (DATE 2002, ISLPED 2010 and IGSC 2018). Prof. Reda is a recipient of NSF CAREER award. He is a senior member of IEEE.
Zoom Link: https://ucsc.zoom.us/j/9314278