期刊
ADVANCED ELECTRONIC MATERIALS
卷 8, 期 12, 页码 -出版社
WILEY
DOI: 10.1002/aelm.202200656
关键词
crossbar arrays; neural networks; nonlinear selectors; one selector-one resistive switching memories
资金
- SK Hynix Inc.
- New Architecture Research (NAR) program [NRF-2020R1A3B2079882]
- National Research Foundation of Korea (NRF) of the Republic of Korea
This paper presents a crossbar array using resistive switching random-access memory, along with a novel selector device to prevent leakage current. By using ruthenium dioxide as the selector electrode, a nonlinear selector with high nonlinearity and sufficient endurance is achieved. Experimental results show that the proposed selector device achieves high accuracy in classification tasks.
A crossbar array using resistive switching random-access memory requires a selector device to prevent leakage current. However, the high current flow during the electroforming and first reset process (switching from a low resistance state to a high resistance state) can degrade the selector device. Ruthenium dioxide, a conducting oxide electrode with low oxygen affinity preventing excess oxygen vacancy in a dielectric, is used as a selector electrode to acquire a TiO2-based nonlinear selector that endures a high current flow. The selector shows high nonlinearity (approximate to 7 x 10(4)), high forward and reverse current ratio (approximate to 5.5 x 10(2)), and sufficient endurance (>10(7)) in the one selector-one resistive switching memory (1S1R) structure, where the HfO2 comprises the resistive switching memory. 9 x 9 crossbar array composed of the 1S1R device is used as neuromorphic hardware to classify simple characters by offline supervised learning. Further classification simulation of the rescaled Modified National Institute of Standards and Technology dataset shows 95.77% accuracy with achievable array structure.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据