Abstract:
The memory system presents many problems in computer architecture and system design. One of the most important problems is worsening hardware variability, primarily caused by manufacturing difficulties in the nanometer nodes and increasing susceptibility to the operating environment. As a consequence, memories have become a fundamental bottleneck in the resiliency and energy-efficiency of computing platforms ranging from embedded IoT devices to the cloud and supercomputers.  To address this dual challenge, in this talk I propose the design of opportunistic memory systems that exploit and cope with hardware variation to improve both energy efficiency and resiliency. In addition to a general overview of current research in the UCLA NanoCAD Lab, I will discuss several of my research projects on this topic that were part of the NSF Variability Expedition. The key theme is that memory systems can be designed for the worst-case outcome from variability, but operate in a better-than-worst-case manner whenever possible. Together, my projects comprise a multi-layer approach spanning circuits to system software. The latter part of the talk will focus on Software-Defined Error-Correcting Codes, my most recent project, which offers a novel approach to improving memory system resiliency using hardware/software collaboration.

Bio:
Mark Gottscho is a Ph.D. Candidate in the Electrical Engineering Department at the University of California, Los Angeles, US, and is advised by Professor Puneet Gupta. His research interests generally lie in system design and computer architecture, with an emphasis on memories and resiliency. He received his M.S. in 2014 with Great Distinction and his B.S. in 2011, both from UCLA. He received an Honorable Mention from the NSF's Graduate Research Fellowship Program in 2014, and won the 2016-2017 UCLA Dissertation Year Fellowship. He was recently awarded the highly competitive 2016-2017 US Qualcomm Innovation Fellowship for his proposal on Software-Defined Error-Correcting Codes. He is a student member of both the IEEE and the ACM.