Publication
IEDM 2022
Conference paper

Gradient descent-based programming of analog in-memory computing cores

Abstract

Precise programming of crossbar arrays is crucial for high matrix-vector-multiplication (MVM) accuracy in analog in-memory computing cores. By minimizing the MVM error using gradient descent, our method significantly reduces the MVM error compared with conventional iterative programming, eliminates the need for high-resolution ADCs, and improves the experimental accuracy of ResNet-9.