SOT-MRAM based Sigmoidal Neuron for Neuromorphic Architectures

TL;DR

This paper introduces a novel SOT-MRAM based sigmoidal neuron for neuromorphic systems, achieving significant power and area efficiency improvements and demonstrating high-speed, accurate pattern recognition in a simulated multilayer perceptron.

Contribution

The paper presents a new SOT-MRAM based neuron design that is more power- and area-efficient than previous circuits and validates its effectiveness in a large-scale neuromorphic pattern recognition task.

Findings

74x reduction in power-area-product compared to previous neurons

Achieves accuracy comparable to GPU-based binarized MLPs on MNIST

Orders of magnitude faster processing speed in simulations

Abstract

In this paper, the intrinsic physical characteristics of spin-orbit torque (SOT) magnetoresistive random-access memory (MRAM) devices are leveraged to realize sigmoidal neurons in neuromorphic architectures. Performance comparisons with the previous power- and area-efficient sigmoidal neuron circuits exhibit 74x and 12x reduction in power-area-product values for the proposed SOT-MRAM based neuron. To verify the functionally of the proposed neuron within larger scale designs, we have implemented a circuit realization of a 784x16x10 SOT-MRAM based multiplayer perceptron (MLP) for MNIST pattern recognition application using SPICE circuit simulation tool. The results obtained exhibit that the proposed SOT-MRAM based MLP can achieve accuracies comparable to an ideal binarized MLP architecture implemented on GPU, while realizing orders of magnitude increase in processing speed.