Interface-Engineered Proton-Gated Oxide Transistors with Multitimescale Synaptic Plasticity

Ion-gated transistors that mimic biological synapses are essential for energy-efficient neuromorphic hardware capable of adaptive learning and sensory processing. However, achieving both high-speed operation and stable memory in silicon-compatible platforms remains challenging due to the trade-off between ionic mobility and retention characteristics. In this work, we developed proton-gated synaptic transistors composed of an electron-beam (EB)-deposited SiO2 electrolyte and a sputtered WO3 channel. A low film density with the oxygen deficiency of EB-deposited SiO2 enables gate-controlled proton intercalation and deintercalation, producing versatile synaptic plasticity of continuously modulating the WO3 channel conductance. The devices exhibit a wide time response range (pulse widths from 100 ms to 50 μs) and a large dynamic range (>103) with a retention time of ∼18 s, successfully emulating short-term plasticity and nociceptor-like sensory behaviors such as threshold spiking, relaxation, and sensitization. They also achieve ∼90% image recognition accuracy and a low power consumption of ∼200 fJ for a single synaptic event. Furthermore, an introduction of a thin Al2O3 layer between the electrolyte and channel effectively suppresses proton back-diffusion, extending the retention time to ∼860 s. This also results in high recognition accuracy across a wide range of pulse widths, from subseconds to submilliseconds. By integrating two transistors with different retention times into a "one-transistor-one-synapse" architecture, we further demonstrated hardware-level temporal control of synaptic plasticity that enables event-driven learning, forgetting, and memory without altering input stimuli. These results indicate that all-oxide proton-gated transistors incorporating interface engineering provide a neuromorphic platform that enables highly flexible designs with excellent synaptic characteristics.