Memristor-Based Spiking Neural Network Implementation Scheme

Memristors have a wide range of applications in fields of nonvolatile memory and biological neuromorphic computing. In this paper, we propose a memristor-based auto-update circuit, which can update the synaptic weights of the memristor completely and automatically. The circuit is combined with a spiking neural network to explore its hardware circuit design and character recognition is implemented. Compared with other networks, the proposed circuit in this paper has excellent robustness and can fulfill the learning task even under the harsh conditions of the circuit.