Effect of memristor’s potentiation-depression curves peculiarities in the convergence of physical perceptrons

Abstract

Neuromorphic computing aims to emulate the architecture and information processing mechanismsof the mammalian brain. This includes the implementation by hardware of neural networks. Oxide based memristor arrays with cross-bar architecture appear as a possible physical implementation ofneural networks. In this paper, we obtain experimental potentiation-depression (P-D) curves ondifferent manganite-based memristive systems and simulate the learning process of perceptrons forcharacter recognition. Weanalyze how the specific characteristics of the P-D curves affect theconvergence time -characterized by the EPOCHs-to-convergence (ETC) parameter- of the network.Our work shows that ETC is reduced for systems displaying P-D curves with relatively low granularityand non-linear and asymmetric response. In addition, we also show that noise injection during thesynaptic weight actualization further reduces the ETC. The results obtained here are expected tocontribute to the optimization of hardware neural networks based on memristors cross-bar arrays.