Learning in Memristive Neural Networks

TL;DR

This paper explores the implementation and control of memristor-based neural networks, addressing how to evaluate, measure, and steer memristor resistances in hardware neural network circuits, demonstrated through proof-of-concept and MNIST examples.

Contribution

It provides a detailed circuit implementation of memristive neural networks and solutions for resistance evaluation, measurement, and control within such systems.

Findings

Successful evaluation of memristor states from external measurements

Methods to steer memristor resistances using external voltages

Demonstration with proof-of-concept and MNIST-trained neural network

Abstract

Memristors are nonlinear two-terminal circuit elements whose resistance at a given time depends on past electrical stimuli. Recently, networks of memristors have received attention in neuromorphic computing since they can be used to implement Artificial Neural Networks (ANNs) in hardware. For this, one can use a class of memristive circuits called crossbar arrays. In this paper, we describe a circuit implementation of an ANN and resolve three questions concerning such an implementation. In particular, we show (1) how to evaluate the implementation at an input, (2) how the resistance values of the memristors at a given time can be determined from external (current) measurements, and (3) how the resistances can be steered to desired values by applying suitable external voltages to the network. The results will be applied to two examples: an academic example to show proof of concept and an ANN that was trained using the MNIST dataset.