An All-Electric Neural Device and Network Based on Laterally Coupled Nanomagnets for Binary Image Recognitions

Spintronic neural devices are attracting growing interest for its potential capability in processing massive data with extremely low power consumption. The devices based on magnetic domain walls (DWs) are considered as promising solutions for its nonvolatility, high density, and flexible design. Among the device mechanisms, strong chiral coupling of laterally adjacent nanomagnets through interfacial Dzyaloshinsky–Moriya interaction (DMI) enables field-free correlated switching of neighboring nanomagnets and planar DW racetrack cascading. Here, we apply this lateral coupling mechanism to realize a current-driven magnetic neural device and network. With the spin-orbit-torque (SOT)-driven DW motion and chiral coupling between laterally neighboring out-of-plane (OOP) and in-plane (IP) nanomagnets, this device achieves nonvolatile binary neuromorphic computing. By racetrack cascading of this device, we further implement a three-layer binary neural network (BNN) capable of recognizing images. This device, which realizes all-electric nonvolatile in-memory computing, may inspire the development of ultralow-power spin artificial neurons.