Spin Hall MTJ Devices for Advanced Neuromorphic Functions

Abstract

We propose a new device based on standard spin Hall magnetic tunnel junction-based spintronic neurons, which allows them to compute multiple neural network functionalities simultaneously and in parallel, saving space and time. An approximation to the rectified linear unit transfer function and the local pooling function is computed simultaneously with the convolution operation itself. A proof-of-concept simulation thoroughly explores the behavior of the device, predicting that the operations can be performed with up to 99% precision at a cost of about 6 pJ per 3 × 3 pooling template. The simulations are remarkably robust to thermal noise, performing well even with very small magnetic layers.