Ferroelectric memristors optimized in thickness for short-term memory-driven reservoir computing

Abstract

Optimized thickness ferroelectric memristors with short-term memory dynamics enabling efficient reservoir computingJunhyeok Park, a and Sungjun Kim a In this work, Mo/Hf0.5Zr0.5O2 (HZO)/n + Si ferroelectric memristors were fabricated, and thickness-dependent properties were systematically analyzed. The optimized 5 nm HZO device exhibited stabilized orthorhombic phase formation, yielding the highest remanent polarization (2Pr = 24.96 μC/cm 2 ), a tunneling electro-resistance (TER) ratio of 2474.05 %, and a rectifying ratio of 4530.78. These superior properties enabled reliable multilevel cell (MLC) operation, suppressed leakage currents, and minimized sneak paths in crossbar arrays. Short-term memory (STM) characteristics were experimentally confirmed through paired-pulse facilitation (PPF) and retention measurements, validating the device's ability to emulate synaptic dynamics. Furthermore, synaptic plasticity was successfully reproduced under spike-amplitude dependent plasticity (SADP), spike-duration dependent plasticity (SDDP), spike-rate dependent plasticity (SRDP), and spike-number dependent plasticity (SNDP) conditions. Leveraging these STM features, reservoir computing (RC) simulations achieved high recognition accuracy for both MNIST and Hand MNIST datasets, with the latter reaching 98.71 %, thereby demonstrating efficient processing of complex gesture data. These findings highlight the 5 nm HZO ferroelectric memristor as an optimized device for short-term memory based RC systems, offering strong potential for neuromorphic computing and next-generation memory technologies and computing.