Dual-ion-channel synaptic transistor with weight fusion of distinct relaxation for boosted reservoir computing

Inspired by the synergistic Na+/K+ flux regulation intrinsic to glutamatergic synapses, we present a monolithic dual-ion-channel synaptic transistor (DICST) in which a biodegradable α-cellulose ionic gel selectively doped with spatially separated Na+ and K+ ions is conformally interfaced with a defect-engineered ITZO channel, thereby reproducing complementary short- and long-term plasticity within a single device footprint. By dynamically alternating the stimulated gates while modulating pulse width and frequency, the device leverages Na+-mediated slow weight updates and long-term potentiation for low-frame-rate visual persistence, and K+-mediated rapid weight updates and short-term facilitation for high-pass temporal filtering, thus enabling functional switching with enhanced real-time processing capability. We demonstrate an “ion-fusion” readout scheme that coalesces the Na+- and K+-modulated reservoir states, which can enrich the feature representation by adaptively weighting and fusing the distinct relaxation dynamics of the two ions. In fingerprint classification, the fused-state convolutional neural network (CNN) attained the highest mean accuracy across the training, validation, and test partitions, reaching 99.552 ± 0.321% on the training set over five random seeds. This work demonstrates that this fusion strategy significantly enhances classification discriminability, affording a scalable hardware paradigm for real-time neuromorphic computation.