Microarchitectural Challenges for Heterogeneous Computing
Modeling Challenges for Future Computing Systems
William J. Song is currently an Assistant Professor with the School of Electrical and Electronic Engineering, Yonsei University in Seoul, South Korea. He earned his Ph.D. degree in Electrical and Computer Engineering from Georgia Tech, Atlanta, GA, and B.S. degree in Electrical and Electronic Engineering from Yonsei University, Seoul, South Korea. His research focus lies in the challenges of heterogeneous architectures and processing near data for neural networks and big data problems. His interests also include solutions to power, thermal, and reliability issues in many-core microarchitectures and 3D-integrated packages. Prior to joining the faculty of Yonsei University, he worked as an engineer for Intel in Santa Clara, CA. He was a graduate research intern at Qualcomm, San Diego, CA (2015 summer), IBM T.J. Watson Research Center, Yorktown Heights, NY (2014 summer and fall), AMD Research, Bellevue, WA (2013 summer), and Sandia National Labs, Albuquerque, NM (2012, 2011, and 2010 summers). He received Distinguished Faculty Award for Teaching Excellence from Yonsei University in 2018. He was a recipient of IBM/SRC graduate fellowship from 2012 to 2015. He received the Best Student Paper Award at IEEE International Reliability Physics Symposium (IRPS) in 2015 and Best in Session Award at SRC TECHCON in 2014.
This talk discusses various engineering challenges for modeling future computing systems. Technology scaling had been a primary driver for the evolution of computing systems that integrated billions of transistors onto a die. However, the diminishing momentum of miniaturization transformed computing paradigm from naively escalating performance to enhancing power efficiency since computing systems are increasingly dominated by physical constraints such as power, thermal, and reliability. Thus, the analysis of future computing systems cannot be masked in a single scope but must be conducted in holistic manner by incorporating complicated multi-physics interactions between distinct computing elements. This presentation will review modeling efforts from devices to systems and present ongoing problems that we have to tackle.