Enhancing memristor-based neural network accuracy using a variability reduction technique
Please login to view abstract download link
An analysis of a hardware neural network is performed by simulation to study the reduction of variability that is obtained by using several memristors in parallel. A reduction is achived by increasing the replication factor (number of memristor in parallel that are used to implement each synaptic weight in a fully-connected neural network). However, a saturation effect is obtained when the replication factor is increased, depending on the variability of the memristors.