Special Topic on Emerging Hardware for Cognitive Computing

Abstract

Emerging materials and physics can be leveraged for new device-inherent behavior that can have system-level benefits. Motivation for device, circuit, and system behavior can be drawn from how the human brain processes certain data-intensive tasks adaptively and quickly, such as canonical image recognition. The field of neuromorphic computing has made great strides in implementing multi-weight synaptic behavior, as well as neuronal behavior such as integrate-and-fire and stochastic switching, and implementation of such behaviors in deep neural network (DNN) processing. Using CMOS, emerging resistive memories, and other device types as the basis, neuromorphic computing is innovating vertically from devices, to circuits, to systems, to redefine how computation can be done.