Spintronic Artificial Neurons Showing Integrate-and-Fire Behavior with Reliable Cycling Operation

Abstract

The rich dynamics of magnetic materials makes them promising candidates for neural networks that, like the brain, take advantage of dynamical behaviors to efficiently compute. Here, we experimentally show that integrate-and-fire neurons can be achieved using a magnetic nanodevice consisting of a domain wall racetrack and magnetic tunnel junctions in a way that has reliable, continuous operation over many cycles. We demonstrate the domain propagation in the domain wall racetrack (integration), reading using a magnetic tunnel junction (fire), and reset as the domain is ejected from the racetrack with over 100 continuous cycles. Both the pulse amplitude and pulse number encoding are shown. By simulating a spiking neural network task, we benchmark the performance of the devices against an ideal leaky, integrate-and-fire neuron, showing that the spintronic neuron can match the performance of the ideal. These results achieve demonstration of reliable integrated-fire reset in domain wall-magnetic tunnel junction-based neuron devices for neuromorphic computing.