Self-Resetting Magnetic Tunnel Junction Neuron-based Spiking Neural Networks

Spintronic devices such as the magnetic tunnel junction show significant potential for energy-efficient neuromorphic computing applications. This paper presents a spintronic magnetic tunnel junction neuromorphic device capable of integration, spike, and self-reset neuron characteristics. The spin-orbit drives the neuron magnetization dynamics, which controls the neuron characteristics. The input pixels are encoded in the amplitude of the current, which controls the spiking frequency of the neuron. We model the neuron characteristics into a compact model to integrate the proposed spiking neuron into a 3-layer SNN and CSNN architecture. We train and test the spiking neuron model to classify the MNIST and FMNIST datasets. The network achieves classification accuracy above 97% on MNIST and 91% on FMNIST. Considering the classification performance, self-resetting functionality, and nanosecond operation range, the proposed device shows a substantial potential for energy-efficient neuromorphic computing.