MEMRISTOR – ANN FOR ENERGY EFFICIENT PATTERN RECOGNITION

Abstract

This work explores the application of Spiking Neural Network (SNN), Memristor (MR) crossbar arrays for efficient digit recognition (0-9). Leak Integrate and Fire (LIF) Neuron is also implemented to generate spikes which can be fed into memristive crossbar arrays 4X4, final decision is made based on the highest current obtained from a particular column designated for the digit. For circuit level implementation TSMC gpdk180nm library file is used and LT Spice is used to carry out the simulation work. The MNIST Dataset is used and accuracy of 88% is obtained when simulated the same in Pycharm using Keras from Tensor flow library which is used to map activation functions of the Artificial Neural Network (ANN).