Multi-Terminal Reconfigurable Silver Nanowire Network for Associative-Learning-Like Response and Path Optimization

Emulating associative-learning-like responses and hardware-state-guided path optimization using reconfigurable physical networks offers a promising route toward adaptive neuromorphic hardware. Herein, a multiterminal reconfigurable Ag nanowire network is presented. This network exhibits a honeybee-inspired associative learning response and provides experimentally measured conductance states for ant colony pathfinding simulations. By employing a multielectrode array to deliver synchronized odor-like and nectar-like electrical stimuli, the network dynamically reconfigures its conductance pathways in a training-dependent manner. Following training, the postsynaptic current increases by 2 orders of magnitude after the first learning cycle and reaches a sustained suprathreshold level after three cycles, demonstrating that the device can simulate long-term memory consolidation. Software simulations further show that the experimentally measured Ag NW learning states can modulate a handwritten digit recognition model through a hardware-derived memory gate, enabling conditioned-stimulus-only recognition to improve progressively during associative learning. Furthermore, specific stimulation protocols reconfigure the conductance pathways of the Ag NW network, and the resulting hardware conductance states are used as weighting factors in ant-colony-inspired software simulations. With these hardware-derived weights, the simulated optimal path length is reduced from 1643.54 units to 1415.18 units, corresponding to an improvement of approximately 14% in path-search efficiency. This study highlights the potential of integrating reconfigurable physical networks with bioinspired algorithms, offering a hardware-software codesign platform for future adaptive neuromorphic systems.