College of Engineering, Design & Computing Events
CSCI seminar series, "Improving high-performance memory through information theory"
|
12:00 PM - 01:00 PM
Room Number : 844 (Mac Lab)
Lawrence Street Center
1380 Lawrence Street
Denver, CO
Room Number : 844 (Mac Lab)
Lawrence Street Center
1380 Lawrence Street
Denver, CO
Speaker: Alireza Vahid, assistant professor of electrical engineering
"Improving High-Performance Memory through Information Theory"
Abstract:
In this talk, we describe applications of Information and Coding Theory in improving high-performance memory. We start with Flash memory where scaling down the feature size requires new coding strategies to prevent significant degradation to the lifetime of memory. We develop a new interface that acts as a bridge between theory and practice, making it possible to introduce coding schemes that extend the lifetime of Flash Memory. We derive a fundamental tradeoff between host-visible capacity and lifetime and describe several operating points. We then discuss an emerging technology, namely Racetrack memory, which suffers from a unique type of shift error. We deploy tools from wireless communications to handle this error and yield this technology practical. We will briefly talk about Memory Bandwidth Bottleneck in GPUs for Machine Learning applications.
Bio:
Alireza Vahid received his B.Sc. in Electrical Engineering from Sharif University of Technology, Iran. He obtained his M.Sc. and Ph.D. from the School of Electrical and Computer Engineering, Cornell University in 2012 and 2015 respectively, where he worked with Professor Salman Avestimehr. He was a Postdoctoral Fellow at Duke University between 2015 and 2017, where he works with Professor Robert Calderbank. He joined the Electrical Engineering department at CU Denver in August 2017. His research interests include Information and Coding Theory, Wireless Communications, Computer Architecture, and Memory Systems.
He received the Outstanding PhD Thesis Research Award in 2015, and the Director's PhD Teaching Award in 2010 from Cornell University. He was the recipient of the Qualcomm Innovation Award in 2013, and the Jacobs Scholar Fellowship in 2009. He recently received CU Denver's New Faculty Award.
"Improving High-Performance Memory through Information Theory"
Abstract:
In this talk, we describe applications of Information and Coding Theory in improving high-performance memory. We start with Flash memory where scaling down the feature size requires new coding strategies to prevent significant degradation to the lifetime of memory. We develop a new interface that acts as a bridge between theory and practice, making it possible to introduce coding schemes that extend the lifetime of Flash Memory. We derive a fundamental tradeoff between host-visible capacity and lifetime and describe several operating points. We then discuss an emerging technology, namely Racetrack memory, which suffers from a unique type of shift error. We deploy tools from wireless communications to handle this error and yield this technology practical. We will briefly talk about Memory Bandwidth Bottleneck in GPUs for Machine Learning applications.
Bio:
Alireza Vahid received his B.Sc. in Electrical Engineering from Sharif University of Technology, Iran. He obtained his M.Sc. and Ph.D. from the School of Electrical and Computer Engineering, Cornell University in 2012 and 2015 respectively, where he worked with Professor Salman Avestimehr. He was a Postdoctoral Fellow at Duke University between 2015 and 2017, where he works with Professor Robert Calderbank. He joined the Electrical Engineering department at CU Denver in August 2017. His research interests include Information and Coding Theory, Wireless Communications, Computer Architecture, and Memory Systems.
He received the Outstanding PhD Thesis Research Award in 2015, and the Director's PhD Teaching Award in 2010 from Cornell University. He was the recipient of the Qualcomm Innovation Award in 2013, and the Jacobs Scholar Fellowship in 2009. He recently received CU Denver's New Faculty Award.