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.