Memory Effects With Broken Symmetry in Nanofluidic Memristor for Neuromorphic Computing

ABSTRACT Memory and learning in biological systems arise from ion transport across nanoscale synaptic junctions in neural networks. These junctions act as a natural memristors and thus, reproducing this effect in artificial aqueous systems is crucial for mimicking neural functions and advancing neuromorphic computing. Herein, we successfully demonstrated the memristive effects through the spatial confinement of water and ions within a biomimetic nanochannel, using two distinct stimulation mechanisms (i) divalent‐ion screening and (ii) pH‐driven deprotonation. In both cases, broken symmetry within the medium coupled with surface effects, lead to hysteretic ion transport. This nanofluidic memristor also emulated biological memory features, including both short/long‐term potentiation and key synaptic functionalities, such as paired‐pulse facilitation (PPF) and paired‐pulse depression (PPD). The reversible modulation of ionic conductance of our nanofluidic device enabled dynamic encoding of synaptic weights, a key mechanism underlying adaptive learning behavior in neuromorphic systems. Leveraging this property, a three‐layer artificial neural network for pattern recognition is trained and recognition accuracy of 94.6% on the small‐digit MNIST dataset, which can compete with the performance of many solid‐state memristive synapses. The memory effect resemblance between our single‐channel system to biological counterparts, paves the way for elucidating the origin of memory in biological systems and advancing nanofluidic memristor‐based neuromorphic computing.