Analog Computing-In-Memory: Beyond DNN Acceleration
Join us for our upcoming Future Computing Seminar Series
Speaker: Prof. Saibal Mukhopadhyay (Georgia Institute of Technology)
Date: June 2th, 2025, 11:00 CET
Where: ETZ E6
Abstract:
Computing-in-Memory (CIM) has emerged as an approach to reduce data movement in von-Neuman computing. Many prior works have shown that CIM can improve energy-efficiency of deep neural network (DNN) engines. This talk will introduce potential applications of CIMs, specially, analog CIMs to applications beyond DNN. First, we will present analog CIMs for in-sensor processing to extract low-dimensional digital features from high-dimensional analog signals. The application of such designs to image sensors and radars will be presented. Second, we will discuss analog CIMs for solving hard optimization problems. We will discuss a Boolean Satisfiability Solvers (SAT) and a Least-Square Solvers designed using analog CIMs. The measurement results from test-chips will be presented to demonstrate the energy-efficiency potentials of these designs. The talk will conclude with a discussions on future applications and design challenges of analog CIMs.
Bio:
Prof. Saibal Mukhopadhyay is a Joseph M. Pettit Professor in School of Electrical and Computer Engineering at Georgia Institute of Technology. He is the director of “CogniSense: A Center on Cognitive Multispectral Sensor,” a center under JUMP2.0 program supported by Semiconductor Research Corporation and Defense Advanced Research Project Agency (DARPA). His research interests include design of secure, intelligent, and energy-efficient integrated circuits and design of smart sensors.