Back-Propagation Neural Network based on Analog Memristive Synapse

Due to the low power consumption and its feasibility of highly scalable crossbar arrays, memristors have recently arouse a great interest to mimic the principles and operations of biological brains in neuromorphic computing. Two kinds of memristors (TiO2-based memristor and$\mathrm {T}\mathrm {a}_{2}\mathrm {O}_{5-\mathrm {x}}$-based memristor) are separately combined into synapses to simulate a three-layer perceptron neural network operated by back-propagation algorithms on MATLAB platform. The device to device variations are considered in terms of the model's accuracy. Moreover, the effect of the target values and the initial weights on the final weight distribution are discussed.