Nitrogen-doped titanium dioxide nanorod array memristors with synaptic features and tunable memory lifetime for neuromorphic computing

JOURNAL OF ALLOYS AND COMPOUNDS

卷:868

Article

Yu, Yantao^1,2,3  Wang, Chunqi^1,2,3  Jiang, Chao^1,2,3  Wang, Lanzhi^1,2,3  Wang, Ze^1,2,3  Du, Zuojuan^1,2,3  Sun, Jia⁵  Abrahams, Isaac⁴  Huang, Xiaozhong^1,2,3 

[1] Cent South Univ, Powder Met Res Inst, Changsha 410083, Peoples R China

[2] Hunan Key Lab Adv Fibers & Composites, Changsha 410012, Peoples R China

[3] Cent South Univ, State Key Lab Powder Met, Changsha 410083, Peoples R China

[4] Queen Mary Univ London, Sch Biol & Chem Sci, London E1 4NS, England

[5] Cent South Univ, Sch Phys & Elect, Hunan Key Lab Super Microstruct & Ultrafast Proc, Changsha 410083, Hunan, Peoples R China

关键词: Memristor;    Titanium dioxide;    Nitrogen-doped;    Tunable memory lifetime;   

DOI  :  10.1016/j.jallcom.2021.159194

来源: Elsevier

【 摘 要 】

Titanium dioxide memristors are promising next generation devices for neuromorphic computing, which can mimic synapses of human brain. With respect to meet satisfactory performance, gradual conductance changes and stability of cycling are essential. In this work, nitrogen-doped titanium dioxide nanorod arrays (N-TiO2 NARs) memristors with high performance have been fabricated using a hydrothermal process. The X-ray photo electron spectra confirm the existence of Ti.N bonds, with uniform distribution of nitrogen confirmed by elemental mapping. The bidirectional gradually changing conductance and endurance of over 8000 cycles have been achieved. The ratio of high resistance state (HRS) and low resistance state (LRS) exceeds 16. The synaptic features, such as potentiation, depression, paired pulse facilitation (PPF), experience-learning and STDP have been achieved using continuous sweeping voltage and pulse trains respectively. It is possible to emulate the learning processes by stimulating the memristor using pulses of different amplitude, duration and interval. By manipulating the stimulation pulse duration, voltage and current compliance, memory lifetimes (MLTs) can be tuned from seconds to tens of thousands of seconds. These transitional processes are associated with anion vacancy creation/accumulation and diffusion. Those synaptic features and tunable MLTs provide a promising approach for neuromorphic computing. (C) 2021 Published by Elsevier B.V.

【 授权许可】

Free