Hardware-Software Interface Seminar


Understanding the Future of Computing Performance and Energy Efficiency: A SW/HW perspective"

Wednesday, February 22, 2017 at 12:00 Noon PLEASE NOTE UNUSUAL DAY

Room 368 (CIT 3rd Floor)

Computing is entering a new era where improvements to performance and energy efficiency will be no longer dominated by better device technology. The next level of improvement will come from working across the computing stack. In this talk we will explore two intersecting paradigms at the software-hardware (SW-HW) interface that hold great promise in delivering large improvements in performance and energy efficiency, especially for emerging applications such as machine learning and artificial intelligence. The first paradigm is application or domain-specific acceleration, where custom hardware circuits are created to match application characteristics, leading to large improvements in performance and power consumption. The second paradigm is approximate computing, where circuits are deliberately under-designed to reduce power consumption at the expense of introducing application-tolerable arithmetic approximations. We will describe our work on creating approximate accelerator systems for deep neural networks with orders of magnitude improvement in performance and energy efficiency compared to existing general purpose CPUs and GPUs. Beyond deep neural networks, we will overview our work on automated methods to synthesize arbitrary approximate accelerators. Our techniques, which are inspired by software debugging techniques, automatically discover large number of approximate accelerators that trade-off accuracy and power consumption. Results will be demonstrated for various approximate accelerators in image processing, computer vision and machine learning.

Sherief Reda is an Associate Professor of Engineering and Computer Science at Brown. He joined the Computer Engineering group at Brown in 2006 after receiving his Ph.D. in computer science and engineering from University of California, San Diego. His research interests are in the areas of energy-efficient computing, thermal-power sensing and management, and low-power design techniques. Professor Reda received a number of research awards and acknowledgments, including five best paper nominations (DATE 2002, ICCAD 2005, ASPDAC 2008, ISLPED 2010 and ICCAD 2015), two best paper awards (DATE 2002 and ISLPED 2010), and a NSF CAREER award. He is a senior member of IEEE.

R. Iris Bahar is a Professor of Engineering and Computer Science at Brown. She has been on Brown’s faculty since 1996 following completion of her PhD from University of Colorado, Boulder. Her research interests include computer architecture; computer-aided design for logic synthesis, verification, and low-power applications; and design, test, and reliability analysis for nanoscale systems. Her research has been continuously funded since 1997 through various industrial and government sources, including the National Science Foundation, DARPA, DoD, the Semiconductor Research Corporation (SRC), Intel, and IBM.

Food will be served.