Lithium niobate volatile memristors for efficient reservoir computing

Lithium niobate (LiNbO3) thin films, serving as next-generation optoelectronic integration platforms, enable nonvolatile memristors with exceptional convolutional computing capabilities through inherent ferroelectricity. Although volatile memristors offer distinct advantages for temporal information processing due to their transient dynamics, such devices remain underexplored in LiNbO3 systems. Critically, grain boundary-associated conductive channel formation provides an effective approach for regulating volatile resistive switching behavior. Herein, we report a polycrystalline LiNbO3-based volatile memristor fabricated via pulsed laser deposition. Oxygen-vacancy-related dynamics contribute to the volatile switching behavior, while the polycrystalline grain-boundary structure provides favorable environments for localized conductive-channel evolution, which is essential for emulating the transient nature of biological neural signals. This memristor is capable of accurately emulating biorealistic synaptic functions, including paired-pulse facilitation (PPF; τ1 = 1.2 ms and τ2 = 11.6 ms) and spike-dependent synaptic plasticity. As a seven-bit physical reservoir computing unit, it achieves accuracy of 94.11% in MNIST digit classification by leveraging its intrinsic nonlinear dynamics. The memristor’s simplified structure and fabrication process ensure excellent scalability, while its inherent optical/piezoelectric multimodality enables interaction with optical and piezoelectric signals. This work lays the foundation for advanced intelligent sensing and neuromorphic computing systems.