High-Reliability Volatile Self-Rectifying Memristors for Reservoir Computing

Emerging neural network architectures, such as reservoir computing, require devices with short-term plasticity to perform dynamic information processing, motivating the exploration of volatile memristors. In this work, we develop a highly reliable self-rectifying volatile memristor fabricated entirely via room-temperature physical vapor deposition (PVD) without additional treatment. The device demonstrates superior rectification and endurance, with a rectification ratio exceeding 105, stable operation over 104programming cycles, and uniform decay time constants. Exploiting its volatile and nonlinear dynamics, a memristor-based reservoir computing system with high-dimensional feature projection is implemented, achieving 96% accuracy in spoken digit recognition tasks. These results demonstrate the potential of volatile self-rectifying memristors for dynamic information processing and scalable neuromorphic computing applications.