Design of Multilayer Perceptron Network Based on Metal-Oxide Memristive Devices

Abstract

A key problem at hardware implementation of artificial neural networks based on memristors (ANNM) is to ensure the required accuracy of their operation at the transition from models to real fabricated memristive devices. Due to a number of factors, such as the imperfections in stateof- the-art memristors and memristive arrays, ANNM design and tuning methods, additional computation errors occur during the process of ANNM hardware implementation. The article proposes a general approach to the simulation and design of a multilayer perceptron (MLP) network implemented with original cross-bar arrays of metal-oxide memristive devices. The proposed approach is based on the theory of engineering tolerances, simulation and the design of experiments. The authors present the research results for the ANNM trained to solve the problem of nonlinear classification for a bidirectional adaptive neural interface.