Design of Memristor-Based Binarized Multi-layer Neural Network with High Robustness

Xiaoyang Liu, Zhigang Zeng, Rusheng Ju

Abstract: Memristor-based neural networks are promising to alleviate the bottleneck of neuromorphic computing devices based on the von Neumann architecture. Various memristor-based neural networks, which are built with different memristor-based layers, have been proposed in recent years. But the memristor-based neural networks with full precision weight values are affected by memristor conductance variations which have negative impacts on networks’ performance. However, binarized neural networks only have two kinds of weight states, so the binarized neural networks built by memristors suffer little from the conductance variations. In this paper, a memristor-based binarized fully connected layer and a memristor-based batch normalization layer are designed. Then based on the proposed layers, the memristor-based binarized multi-layer neural network is built. The effectiveness of the network is substantiated through simulation experiments on pattern classification tasks. The robustness of the network is also explored and the results show that the network has high robustness to conductance variations.