CASS Talks with Sandip Kundu, University of Massachusetts Amherst, USA

#CASS #Talks #2021

Abstract:

In recent years, Deep Neural Networks (DNNs) have made inroads into a number of applications involving pattern recognition ranging from facial recognition to self-driving cars. Many of these applications, such as self-driving cars, have real-time requirements, where specialized DNN hardware accelerators help meet those requirements. As the models become more capable, they also become deeper and more complex, leading to increased computational and storage demands. Model pruning and data quantization are two effective methods for mitigating this demand by compressing the network and simplifying computation. As hardware accelerators push the performance envelope while adhering to strict power constraints, reliability is frequently a casualty. In particular, power-constrained DNN accelerators are more vulnerable to transient and intermittent hardware faults due to particle hits, manufacturing variations, and fluctuations in power supply voltage and temperature. In this talk, we will discuss reliability of DNNs from various design perspectives and share results from recent improvements. 

Short Bio: 

Sandip Kundu is a Professor of Electrical and Computer Engineering at the University of Massachusetts Amherst. Recently, he also served as a program director at the National Science Foundation. Kundu began his career at IBM Research as a Research Staff Member; then worked at Intel Corporation as a Principal Engineer before joining UMass Amherst as a professor in 2005. He has published nearly 300 research papers in VLSI design and test, holds several key patents including ultra-drowsy sleep mode in processors, and has given more than a dozen tutorials at various conferences. He is a Fellow of the IEEE, Fellow of the Japan Society for Promotion of Science (JSPS), Senior International Scientist of the Chinese Academy of Sciences and was a Distinguished Visitor of the IEEE Computer Society. He has served as associate editor of a number of IEEE and ACM journals. He has been the Technical Program Chair/General Chair of multiple conferences including ICCD, ATS, ISVLSI, DFTS and VLSI Design.