Self-rectifying memristors with high rectification ratio and dynamic linearity for in-memory computing

• Published in: Applied physics letters 2024-09, Vol.125 (13)
• Main Authors: Zhang, Guobin, Wang, Zijian, Fan, Xuemeng, Wang, Zhen, Li, Pengtao, Luo, Qi, Gao, Dawei, Wan, Qing, Zhang, Yishu
• Format: Article
• Language: English
• Subjects: Arrays; Computer memory; Hafnium oxide; Hardware; Large scale integration; Linearity; Memristors; Synapses; Thickness; Ultra Large Scale Integration
• ISSN: 0003-6951; 1077-3118
• DOI: 10.1063/5.0225833

In the era of big data, the necessity for in-memory computing has become increasingly pressing, rendering memristors a crucial component in next-generation computing architectures. The self-rectifying memristor (SRM), in particular, has emerged as a promising solution to mitigate the sneak path current issue in crossbar architectures. In this work, a Pt/HfO2/WO3−x/TiN SRM structure is reported with an impressive rectification ratio above 106. To elucidate the underlying mechanisms, we systematically investigate the impact of the WO3−x resistive layer thickness modulation on the device's conductive behavior. Our findings reveal that the abundant traps in the WO3−x resistive layer and the excellent insulating property of HfO2 synergistically suppress negative current while promoting positive current. According to the simulation, the crossbar array based on the proposed SRMs can realize an array scale of over 21 Gbit. Furthermore, artificial synapses fabricated using these SRMs demonstrate a remarkable linearity of 0.9973. In conclusion, our results underscore the great potential of these SRMs for the ultra-large-scale integration of neuromorphic hardware, providing a guide for future ultra-high-energy efficiency hardware with minimal circuit overhead.