Towards Hardware Implementation of Double-Layer Perceptron Based on Metal-Oxide Memristive Nanostructures

TL;DR

This paper demonstrates a neural network built with metal-oxide memristive nanostructures, showing potential for scalable, nonlinear classification despite device variability, advancing hardware neural network implementation.

Contribution

It introduces a double-layer perceptron using memristive nanostructures with a training algorithm that accounts for device variations, enabling scalable hardware neural networks.

Findings

Successfully implemented a memristive neural network circuit.

The network can solve nonlinear classification problems.

The model is scalable despite device variability.

Abstract

Construction and training principles have been proposed and tested for an artificial neural network based on metal-oxide thin-film nanostructures possessing bipolar resistive switching (memristive) effect. Experimental electronic circuit of neural network is implemented as a double-layer perceptron with a weight matrix composed of 32 memristive devices. The network training algorithm takes into account technological variations of the parameters of memristive nanostructures. Despite the limited size of weight matrix the developed neural network model is well scalable and capable of solving nonlinear classification problems.