Abstract

Although speech recognition has been widely implemented in software neural networks, a hardware implementation based on energy efficient computing architecture is still missing. In this study, we have fabricated W/MgO/SiO2/Mo memristor array with multilevel resistance states, where the weights of the artificial synapses in the memristor array can be tuned precisely by voltage pulses. Based on the array, we have performed speech recognition in memristive spiking neural network (SNN) with improved supervised tempotron algorithm on TIDGITS dataset , demonstrating software-comparable accuracy for speech recognition in the memristive SNN. We envision that such memristive SNN can pave the way to building a bio-inspired spike-based neuromorphic system for audio learning.

ToC Figure