High-Temperature LiPON-Gated Synaptic Transistors for Reservoir Computing

Abstract

Conventional neuromorphic devices often suffer from thermal degradation and performance deterioration, limiting their operation in harsh, high-temperature environments. Here, we report a high-temperature synaptic transistor (HTST) based on a lithium phosphorus oxynitride (LiPON) solid-state electrolyte that enables reliable neuromorphic computing up to 423 K. The device exhibits a subthreshold swing (SS) of 240 mV/dec and maintains dynamically tunable hysteresis windows and temporal responses over a wide temperature range (298–423 K). Temperature-dependent low-frequency noise measurements show a 1/f-dominated power spectral density (PSD) across all conditions, with the noise intensity coefficient Kf (a measure of effective interfacial trap density) decreasing monotonically as temperature increases. Arrhenius analysis yields consistent activation energies of ~0.50 eV for both Kf and the synaptic relaxation time $\tau $ , suggesting that a thermally activated mechanism of Li+ trapping and detrapping at the LiPON/ITO interface may serve as a plausible microscopic link between low-frequency noise and synaptic plasticity. Exploiting this temperature-adaptive ion–electron coupling, a physical reservoir computing system is implemented using four transistors, integrating temperature-programmable nodes and virtual expansion to generate 160 reservoir states, achieving a normalized mean square error (NMSE) as low as 0.0246 in the NARMA2 time-series prediction task. These results highlight the potential of LiPON-based synaptic transistors for energy-efficient neuromorphic hardware and robust sensing in extreme environments.