Mem-elements for Neuromorphic Circuits with Artificial Intelligence Applications

Chapter 14: Guidelines for benchmarking non-ideal analog memristive crossbars for neural networks

Alex James    Indian Institute of Information Technology and Management, Kerala, India

Abstract

The memristor crossbar configuration can emulate dot-product computations in the analog domain and provides a simplistic way to implement a layer of an analog artificial neural network. Multiple crossbars collectively in a modular architecture can implement large neural networks. However, scaling of analog neural networks to numerous layers and a large number of inputs leads to practical design constraints imposed by non-idealities of the devices and crossbar arrays. This chapter highlights the critical aspects of benchmarking the memristive crossbar for ...