Ferroelectric tunnel junctions with 5 nm-thick HZO for tunable synaptic plasticity and neuromorphic computing

Abstract

Hafnium oxide-based ferroelectric tunnel junctions (FTJs) have emerged as promising candidates for next-generation neuromorphic computing due to their ability to function as memristive devices. Their CMOS compatibility and low power consumption make them attractive for synaptic applications in artificial neural networks. The electrical properties of FTJs are significantly influenced by the thickness of the ferroelectric layer. In this study, we investigated the electrical characteristics, including remanent polarization (P-r) and tunneling electroresistance (TER) ratio, of FTJs with hafnium zirconium oxide (HZO) thicknesses of 5 nm, 7 nm, and 10 nm. Among these, the device with a 5 nm HZO layer exhibited the best performance, achieving a maximum 2 P-r of similar to 47.33 mu C/cm(2) and a maximum TER of similar to 2974.44 %. Furthermore, we explored the short-term memory characteristics and synaptic properties of this device, demonstrating its potential for neuromorphic computing applications. Our findings confirm the transition from short-term to long-term memory, mimicking human brain functionality under varying input pulse conditions. Finally, integrating this device as the reservoir layer in a reservoir computing system enabled high classification accuracies of 98.24 % on MNIST and 87.01 % on Fashion MNIST, highlighting its feasibility for neuromorphic systems.