Robust Neuromorphic Computing Enabled by Femtosecond Laser-Modulated Divergent Ion Dynamics in CuInP2S6

Inspired by the high efficiency and robustness of human brains, ion-based artificial synaptic devices have been widely explored for neuromorphic computing. However, achieving stability and reliable functionality without additional components remains challenging due to the inherent variability of ion dynamics. Herein, homeostatic plasticity is demonstrated in a CuInP₂S₆ (CIPS) artificial synapse via femtosecond laser treatment. The laser-induced modulation exhibits state-dependent effects, enhancing ionic mobility in the low conductance state (activation) and redistributing ions in the high conductance state (inhibition). The activation effect facilitates short-term plasticity by improving postsynaptic current, while the inhibition effect enhances long-term stability through reducing switching variability. Furthermore, tunable neuromorphic ion modulation in a two-terminal CIPS device is synergistically achieved through the combined influence of an electric field, optical illumination, and femtosecond laser treatment. Finally, a spiking neural network simulation based on a single representative device is demonstrated, achieving an improvement in the accuracy of Modified National Institute of Standards and Technology (MNIST) handwritten digit recognition from 92.5% to 98.2% via femtosecond laser modulation. This work presents a novel strategy for efficiently manipulating ion dynamics in 2D van der Waals ferroionic materials, contributing to the development of adaptive and robust neuromorphic computing systems.