Parasitic Extraction and Signal Integrity Analysis of Memristor-Based Crossbar Arrays for Neuromorphic Computing

The demand for AI applications has created the need to create a low-power solution to train neural networks. Neuromorphic computing has the potential to reduce this power consumption with the usage of a non-volatile in-memory computing element called the memristor. However, the neuromorphic system designed with memristors is susceptible to parasitic effects. The accuracy of the computation of the analog neural network mapped into a memristive crossbar array is significantly affected by the parasitic elements of the circuit. In this paper, the parasitic equivalent circuit model of the memristor-based crossbar array has been proposed. The proposed circuit model is further analyzed for IR drop, signal distortion, change in memristive value, etc. The paper also explores the crossbar array's signal-integrity (SI) issue. Adding parasitic elements to the circuit and evaluating its signal integrity have made the memristor-based crossbar array more realistic and accurate, contributing to a more effective and practical design of the neuromorphic system.