A modelling method of wire resistance in memristor crossbar array for artificial neural network

Abstract

Memristor crossbar arrays are potential for realizing artificial neural networks. It is due to the fact that memristor crossbars are low power consumption and small area occupation. However, the performance of crossbar array has limited by the wire resistance. The presence of wire resistance makes the crossbar circuit more complicated for analyzing because the number of circuit elements increases remarkably. In this work, we propose a method for modelling wire resistance in crossbar-based circuits. Wire resistance is modeled by using a proposed equivalent wire resistance which is obtained by analyzing the crossbar circuit using superposition method. To verify the accuracy of the proposed method, the crossbar circuit was tested for character recognition. The simulation result illustrated that the discrepancy of the output voltage between using the conventional simulation method and the proposed method is as low as 1.7% on average when wire resistance is varied from 0.5 to 2.5Ω. The advantage of the proposed method is the reduction of the simulation time. For the crossbar size of 64×26, the proposed method takes 11.7s for simulation whereas the conventional method takes 108.92s.