SPICE analysis of dense memristor crossbars for low power neuromorphic processor designs

Abstract

This paper provides an analysis of neuromorphic circuits that are capable of learning logic functions. In this paper the training simulations are carried out in SPICE. Our simulations capture low level circuit functionality within the memristor crossbars as well as wire resistances between memristors. This is essential when properly modeling crossbar circuits. Wire resistances, wire capacitances, output comparators, and the number of data inputs are all investigated in this paper to show how these may impact a larger neuromorphic crossbar. Furthermore, it was shown that neural networks can properly train the passive memristor-based crossbars without having to use virtual ground mode operational amplifiers as suggested in previous work. This reduces the number of transistors required by the circuit by about 3 times and reduces the circuit power consumption by about 50 times when compared to the virtual ground design. The key impact of this study is the demonstration through low level circuit simulations that dense memristor crossbars can be effectively utilized to build neuromorphic processors.