A Compact Dual-Memristor Neuron and Its Neural Network for Binary Image Recognition

Abstract

Existing neural networks suffer from CMOS-based hybrid architectures and non-integrable reactive components in neurons, limiting integration density, scalability, and biological plausibility. This paper presents a simple dual-memristor neuron built with two low-cost locally active memristors, and integrates it with a memristive synaptic array to form a 4×3 fully memristive neural network for binary image recognition, validating the effectiveness of the neuron in network-level applications. Theoretical analysis and numerical simulations indicate that, under different external voltages, the neuron successfully generates various types of biphasic action potentials, demonstrating the capability of memristors to replace traditional capacitors and inductors. Low-cost hardware experiments verify the neuromorphic dynamics of the neuron, confirming the correctness of the theoretical analysis and simulations. Finally, a random binary image is employed to validate the effectiveness of the neural network, and the temporal coding mechanism in image recognition. This study offers a viable pathway toward highly integrated, fully memristive hardware systems for brain-inspired computing.