強誘電体酸化物の巨大抵抗変化を利用して脳型素子を実現～強誘電体の電気分極を用いてシナプスの機能を模倣する～

2025-09-09 東京大学

(a) PbTiO₃薄膜の電圧–電流特性とそのニューラルネットワークへの応用。(b) 理想的なニューラルネットワークおよび(c) PbTiO₃メモリスタにおける画像認識実験の混合行列。

＜関連情報＞

• https://www.t.u-tokyo.ac.jp/press/pr2025-09-09-001
• https://www.t.u-tokyo.ac.jp/hubfs/press-release/2025/0909/001/text.pdf
• https://advanced.onlinelibrary.wiley.com/doi/10.1002/adfm.202510715

単結晶PbTiO3強誘電性メムリストにおけるスイッチング性能向上：シナプス可塑性の再現に向けて Enhanced Switching Performance in Single-Crystalline PbTiO3 Ferroelectric Memristors for Replicating Synaptic Plasticity

Haining Li, Zhiqiang Liao, Risa Kataoka, Md Sarker Shamim, Takeshi Kijima, Hiroyasu Yamahara, Hitoshi Tabata, Munetoshi Seki
Advanced Functional Materials  Published: 04 September 2025
DOI:https://doi.org/10.1002/adfm.202510715

Abstract

A large ON/OFF ratio in a memristor provides reliable state distinction, enabling precise weight updates to emulate synaptic plasticity. A well-reproduced ferroelectric polarization in the perovskite oxide single layer obtained by growing high-quality single crystals plays an important role in elevating the ON/OFF ratio. Herein, a single-crystalline PbTiO₃-based ferroelectric memristor is demonstrated, and structural investigations confirm its extremely sharp interface, well-ordered lattice structure, and epitaxial growth. Pt/PbTiO₃/Nb:SrTiO₃ metal–ferroelectric–semiconductor memristors exhibit promising resistive switching properties, including high repeatability, good endurance, long retention, and a larger ON/OFF ratio >10⁵ (stable over 1200 for retention), which is larger than that of most single-layer BaTiO₃ and BiFeO₃ memristors. PbTiO₃-based memristors effectively mimic key synaptic plasticity, including spike-amplitude-dependent plasticity, paired-pulse facilitation/depression, spike-rate-dependent plasticity, short-term memory, transition from short-term memory to long-term memory, long-term memory, and spike-timing-dependent plasticity. These have been systematically investigated based on stable pulse training on resistance modulations. Simulations of neuromorphic computing for different neuron network structures achieved pattern recognition rates of approximately 92%–96%, indicating high accuracy and versatility. This paper introduces an effective and straightforward strategy for enhancing the ON/OFF ratio of ferroelectric PbTiO₃ memristors, reinforcing their potential for use in hardware-based neural networks.