Spintronic neural systems

Abstract

Neural computing, guided by brain-inspired computational frameworks, promises to realize various cognitive and perception-related tasks. Complementary metal–oxide–semiconductor-based computing machines use orders-of-magnitude more computational resources than the brain on cognitive tasks that humans efficiently perform every day. As a result, we are witnessing a seismic shift in the field of computation. Research efforts are being directed to develop artificial intelligence (AI) hardware that mimics the human brain from a bottom-up perspective — through devices that are more naturally suited to neural computation — and thereby improves the efficiency of performing cognitive tasks. In the attempt to bridge the gap between neuroscience and electronics, here we report on developments in the field of spintronic devices for AI hardware. The dynamics of spintronic devices that can be used for the realization of neural and synaptic functionalities are discussed. A cross-layer perspective extending from the device to the circuit and system levels as a pathway towards efficient neural computing systems is also presented. Spintronic devices for artificial intelligence hardware can bridge the gap between neuroscience and electronics. Here we discuss the dynamics of such devices, enabling neural and synaptic functionalities, alongside a cross-layer approach — from devices to circuits and systems — for efficient neural computing systems.