SHE-MTJ Circuits for Convolutional Neural Networks

TL;DR

This paper explores the performance of SHE-MTJ spintronic circuits for convolutional neural networks, demonstrating high accuracy and low energy consumption through simulation on the MNIST dataset.

Contribution

It introduces a novel SHE-MTJ-based neural circuit architecture and evaluates its performance and robustness in image classification tasks.

Findings

Achieves 90-95% accuracy on MNIST

Consumes approximately 100 nJ per image

Analyzes effects of device variation and redundancy

Abstract

We report the performance characteristics of a notional Convolutional Neural Network based on the previously-proposed Multiply-Accumulate-Activate-Pool set, an MTJ-based spintronic circuit made to compute multiple neural functionalities in parallel. A study of image classification with the MNIST handwritten digits dataset using this network is provided via simulation. The effect of changing the weight representation precision, the severity of device process variation within the MAAP sets and the computational redundancy are provided. The emulated network achieves between 90 and 95\% image classification accuracy at a cost of ~100 nJ per image.