Memristor-based neural network circuit with weighted sum simultaneous perturbation training and its applications

In this work, a full circuit of memristor-based neural network with weighted sum simultaneous perturbation training is proposed. Firstly, synaptic designed by using pair memristors, which can represent negative, zero, and positive weights. Secondly, the designed, all operations being completed on without any computer aid. The trained algorithm. algorithm does not involve complex derivative calculation error back propagation, it only applies perturbations to sum, so implementation more simple. Finally, application simulations proposed are performed via PSpice. results simulation indicate that practical effective.