Controlling Ion Dynamics at Nanoscale for Memristor-Based Neuromorphic Computing

The physical separation of memory and processing in conventional architectures results in significant energy dissipation during data movement. This has motivated research into brain-inspired information processing, where memory and learning are realized in a single unit using water and ions. Herein, we report biomimetic memristive and synaptic-like ion dynamics in an aqueous environment using ion-track etched smart nanochannels. Our experiments demonstrate that driving ions through asymmetric bipolar surface charges generates a memristive effect capable of withstanding hours of endurance stress, and that the memristor type can be dynamically changed by the local chemical environment due to counterion over-screening. We identify ion accumulation and depletion as the single unifying mechanism underlying both memristive switching and synaptic plasticity. These controllable ion dynamics emulate a broad spectrum of plasticity: from short-term plasticity to long-term potentiation and depression (LTP/LTD) with near-linear, low-asymmetry conductance modulation and a low energy consumption of 13.2 pJ per synaptic event. Implementing these reversible weight updates in artificial neural network (ANN) simulations yields a recognition accuracy of 94.54% for handwritten digit recognition (small-digit MNIST), rivaling solid-state memristors. These findings demonstrate that systematic control of ion interactions within nanochannels provides a high-performance, energy-efficient foundation for neuromorphic computing.