Digital Realization of Associative Memory Neural Network Based on Memristor Crossbar Array

Abstract

Pavlov’s associative neural network defines the conditional reflex in which a biologically potent stimulus is paired with a previously neutral stimulus. To date, memristors have shown potential in the hardware implementation of Pavlov’s associate network. In this work, the Pavlov associative memory network has been implemented in a memristor crossbar array-based digital circuit consisting of a 4-bit shift register and 16 memristors in a 12×12 Ag/TiO2/Pt memristor crossbar array. This memristor exhibits a low threshold operating voltage of ~0.43 V and a good fit to the voltage threshold adaptive memristor (VTEAM) model. Compared to analogue methods, this digital circuit features ultra-low power consumption (22.05 μW), high frequency (7.6 MHz), and cheap cost (160 devices). The application of a crossbar memristor array in this digital circuit realizes 16 times less power than its counterpart without memristors, making this design promising for further development of brain-inspired associative learning systems.

Keywords: Memristor, digital circuit implementation, Pavlov, associative learning rules, crossbar, VTEAM.