A Self-Rectifying Memristor for High Accuracy Neuromorphic Pattern Recognition

Abstract The development of self-rectifying memristors with reliable synaptic functionalities is critical for high-density neuromorphic computing systems. In this work, a memristor based on a WO3/SiO2 heterojunction was fabricated to achieve both intrinsic rectification behavior and biosynaptic characteristics. It is found that the device demonstrates reliable bipolar resistive switching featuring a pronounced asymmetric I−V response, enabling effective suppression of sneak current without the need for external selector devices. In particular, the incorporation of the SiO2 interfacial layer regulates oxygen vacancy distribution and conductive filament dynamics, significantly improving switching uniformity, endurance, and retention performance. Detailed mechanism analysis reveals multiple conduction regimes, including Poole−Frenkel emission, Schottky emission, Hopping conduction, Fowler−Nordheim emission, direct tunneling, and Ohmic conduction, governed by electric field modulation. Furthermore, the memristor successfully emulates key synaptic behaviors, such as long-term potentiation or depression, paired-pulse facilitation, and spike-dependent plasticity through controllable conductance modulation under pulse stimuli. Neuromorphic simulations based on crossbar arrays demonstrate high pattern recognition accuracies on MNIST and Fashion MNIST datasets. These findings demonstrate the feasibility of the proposed memristor for energy-efficient and densely integrated neuromorphic computing hardware.