Skyrmion-based Leaky Integrate and Fire Neurons for Neuromorphic Applications

Abstract

Spintronics is an important emerging technology for data storage and computation. In this field, magnetic skyrmion-based devices are attractive due to their small size and energy consumption. However, controlling the creation, deletion and motion of skyrmions is challenging. Here we propose a novel energy-efficient skyrmion-based device structure, and demonstrate its use as leaky integrate (LIF) and fire neuron for neuromorphic computing. Here we show that skyrmions can be confined by patterning the geometry of the free layer in a magnetic tunnel junction (MTJ), and demonstrate that the size of the skyrmion can be adjusted by applying pulsed voltage stresses. A spiking neural network (SNN) made of such skyrmion-based LIF neurons shows the capability of classifying images from the Modified National Institute of Standards and Technology (MNIST) dataset.