Solution to alleviate the impact of line resistance on the crossbar array

Abstract

The crossbar array implementing the weighted sum computation and weight update operation is a promising hardware accelerator for neuromorphic computing. However, the voltage drop caused by the current flowing through the access lines could be aggressive for the resistive crossbar array in a fully parallel fashion. In this study, the authors analysed the impact of the line resistance on the crossbar array based on the SPICE simulation. It implies that the scale of the crossbar array and the ratio of line resistance to resistive random access memory resistance bring great influence on the performance of the crossbar array. Also an scheme of optimisation has been proposed to diminish its influence. Furthermore, considering line resistance, multi-layer perceptron based on crossbar array has been simulated with SPICE on Modified National Institute of Standards and Technology dataset. Those results could provide design guidelines for the practical hardware implementation of the neuromorphic accelerator.