Flexible Laser-Induced Graphene Memristor: Fabrication and SPICE based Emulation of an Artificial Neural Network

Abstract

This work demonstrates laser-induced graphene memristors, fabricated using a patterning-free, low cost and simple process directly on a flexible polyimide substrate. The fabricated memristors show repeatable non-volatile bipolar resistive switching with state retention up to 103 seconds. A simple perceptron network for the classification of black-and-white images is later implemented using an experimentally extracted compact model. Successful training of the network by integrating SPICE model with MATLAB shows the possibility to emulate the on-chip learning process. Further, by properly modulating the applied voltage pulse amplitude and period, a reduction in the energy consumed by training the neural network is achieved.