Dynamics of Dual Memristors-Based Neuron Circuit for Pattern Recognition

Abstract

The implementation of large-scale, low-power hard-ware neuromorphic computing systems with memristors is an emerging way to break through the limitations of traditional computers and improve performance. This paper proposes a dual memristors-based neuron circuit based on N-type locally-active memristors (N-type LAMs). The admittance characteristics of N-type LAMs in different operating regions are studied using a small signal analysis method, determining the possibility of oscillation in this neuron circuit. Under the different input signals, significant neuromorphic behaviors of biological neurons such as all-or-none, periodic spiking, multi-spiking firing, spiking bursting, multi-period oscillation and chaos are successfully simulated. Meanwhile, the spiking generation mechanism of the dual memristors-based neuron circuit and the impact of different memristors series connections on neuron dynamics are investigated by using the local activity theory. Then, a dual memristors-based neural network is designed to realize pattern recognition, which can be widely applied in the consumer electronics field. Finally, floating memristor emulators are applied to implement the neuron circuit in hardware. The experimental results are consistent with the simulation results and theoretical analysis, verifying the practicality of the dual memristors-based neuron circuit and the validity of the theoretical analysis.