Nanolaminate Ferroelectric Transistor Enabling Wide-Reservoir In Sensor Neuromorphic Vision

Abstract

This work reports a hardware-oriented hybrid reservoir computing (HRC) system based on a nanolaminate ferroelectric thin-film transistor (FeTFT) that unifies volatile and nonvolatile functions in a single three-terminal device. The HZO/HfO2/HZO gate stack modulates grain size and suppresses ferroelectric variability, enabling precise multilevel control and highly linear weight updates via the incremental step pulse with verify algorithm (ISPVA). Electrical input induces long-term memory, while optical excitation yields short-term memory, allowing dual-mode operation. Light-driven 4-bit reservoirs operate at picoampere currents (similar to 10 pW/device) and emulate nociceptive neuron behavior. Combining three wavelength-dependent reservoirs (405, 450, 532 nm) expands the feature space and improves classification accuracy. Using ISPVA-linearized readout, the system achieves 93.1% and 85.1% accuracies on MNIST and Fashion-MNIST, respectively exceeding prior FeTFT/memristor-based RC systems. This approach establishes a scalable, energy-efficient route toward multifunctional in-sensor neuromorphic computing based on a unified ferroelectric platform.