-
非晶三氧化钨（a-WO3）材料因其高浓度氧空位、适中的带隙以及与CMOS技术的兼容性，成为非易失性电阻开关（RS）忆阻器的理想候选材料。然而，a-WO3忆阻器在模拟开关行为上的低效表现，阻碍了其在高效神经计算领域的应用，而神经计算方式能够满足人工智能等数据密集型应用日益增长的需求。本研究通过在介质层中设计氧空位（Vo）梯度分布结构，成功实现了a-WO3忆阻器的模拟开关功能。采用Vo梯度分布结构的器件展现出高度可靠的模拟开关特性，其特点包括：低周期间波动性、较高的线性增强/抑制过程、超长数据保持时间（>104秒）以及自限流特性。基于该结构忆阻器的人工神经网络实现了97.64%的手写数字识别准确率。此外，我们提出了一种导电细丝（CFs）演化方案，通过形成非锥形导电细丝防止其突然形成和断裂，实现了可控的渐进式多级电导调制。这些研究成果确定了Vo梯度分布的a-WO3忆阻器作为高密度神经计算平台的潜力，为设计优化实验方案提供了宝贵思路和实用策略。Amorphous tungsten trioxide (a-WO3) has emerged as an ideal candidate material for non-volatile memristors, attributed to its high concentration of oxygen vacancies, moderate band gap, and compatibility with CMOS technology. This renders it broad application prospects in neuromorphic computing systems. However, its poor analog switching performance severely hinders its practical application in high-efficiency artificial intelligence data processing. To enhance the analog switching performance of WO3 memristors, this study adopts radio frequency (RF) magnetron sputtering technology to deposit a five-layer amorphous tungsten trioxide (a-WO3) thin film with a gradient distribution of oxygen vacancy concentration on a platinum/silicon (Pt/Si) substrate. X-ray Photoelectron Spectroscopy (XPS) analysis confirms that the oxygen vacancy (Vo) concentration decreases gradually from the bottom to the top layer,verifying the successful fabrication of the five-layer a-WO3 thin film with a gradient distribution of oxygen vacancies. Compared with a-WO3 memristors with a uniform Vo concentration, the device with the Vo gradient distribution exhibits highly reliable analog switching characteristics (low cycle-to-cycle variability, high linearity in potentiation/depression processes), ultra-long data retention (>104 s), and self-current-limiting behavior. An artificial neural network (ANN) based on this structured memristor achieves a handwritten digit recognition accuracy of 97.64%. The RS essence of a-WO3 memristors with Vo concentration gradient distribution lies in the formation/rupture of VOdominated conductive filaments (CFs). The Vo gradient distribution enables controllable evolution of CFs by modifying the electric field and ion migration rules. During CF formation, oxygen ions migrate toward the top electrode, and Vo accumulates gradually first in the bottom electrode region; meanwhile, the electric field induced by Vo gradient suppresses the local abrupt growth of CFs, leading to the formation of uniform nonconical structures and avoiding resistance mutation. During CF rupture, ions migrate toward the bottom electrode, and non-conical CFs can rupture synchronously and progressively, ultimately achieving precise regulation of multi-level conductance. The conduction mechanism shows that the low-voltage region of the high-resistance state (HRS) exhibits an I-V linear relationship, corresponding to the ohmic conduction mechanism. In thehigh-voltage region of HRS, I has a linear relationship with both V2 and V2.5, which conforms to the space-charge-limited current (SCLC) theory. The gradient distribution of oxygen vacancies (VO) regulates the formation and rupture of conductive filaments (CFs), thereby solving the core issue of poor analog switching performance in traditional WO3 memristors. This provides a critical “Vo gradient regulation” design strategy for highdensity neuromorphic computing. It is expected to play a significant role in fields such as image recognition, speech recognition, and intelligent robots.
-
Keywords:
- memristor /
- analog resistive switching /
- amorphous tungsten trioxide /
- oxygen vacancy /
- neuromorphic computing
-
[1] Pan J Y, Kan H, Liu Z R, Gao S, Wu E X, Li Y, Zhang C W 2024 npj Flex. Electron. 8 70
[2] Yan X B, Bai J H, Zhang Y X, Wang H, Zhao J H, Zhou Z Y, Sun Y, Wang Z R, Guo Z Q, Zhao Z, Niu J Z 2024 Mater. Today. Nano 25 100458
[3] Feng C Y, Wu W W, Liu H D, Wang J K, Wan H Z, Ma G K, Wang H 2023 Nanomaterials 13 2720
[4] Jang B, Kim J, Lee J, Jang J, Kwon H J 2024 J. Mater. Sci. Technol. 189 68
[5] Ren J W, Shen H Z, Liu Z Y, Xu M, Li D H 2022 ACS Appl. Mater. Interfaces 14 21141
[6] Park S O, Jeong H, Park J, Bae J, Choi S 2022 Nat. Commun. 13 2888
[7] Wang Y, Chen S D, Cheng X H, Chen W, Xiong Z Y, Lv Z Y, Wu C Y, Wang L, Zhang G H, Zhu X B, Luo L B, Han S T 2023 Adv. Funct. Mater. 34 2309807
[8] Abbas H, Li J Y, Ali A, Hussain S, Jung J W, Ang D S 2025 J. Mater. Sci. Technol. 216 99
[9] Gong Y C, Ming J Y, Wu S Q, Yi M D, Xie L H, Huang W, Ling H F 2024 Acta Phys. Sin. 73 207302(in Chinese)[贡以纯,明建宇,吴思齐,仪明东,解令 海,黄维,凌海峰2024 73 207302]
[10] Panda D, Hui Y F, Tseng T Y 2024 Nanoscale 16 16148
[11] Ju D, Kim S 2024 Adv. Funct. Mater. 34 2409436
[12] Ganaie M M, Kumar A, Shringi A K, Sahu S, Saliba M, Kumar M 2024 Adv. Funct. Mater. 34 2405080
[13] Elshekh H, Wang H Y, Yang C, Zhu S H 2024 J. Appl. Phys. 135 135107
[14] Hsu C C, Chuang H, Jhang W C 2021 J. Alloy. Compd. 882 160758
[15] Mahata C, Pyo J, Jeon B, Ismail M, Moon J, Kim S 2023 Adv. Compos. Hybrid Mater. 6 144
[16] Zhang K J, Ren K, Qin X Z, Zhu S H, Yang F, Zhao Y, Zhang Y 2021 IEEE Trans. Electron Devices 68 3807
[17] Li C, Hsieh J, Hung M T, Huang B 2015 Vacuum 118 125
[18] Li Z N, Tian B Y, Xue K H, Wang B, Xu M, Lu H, Sun H J, Miao X S 2019 IEEE Trans. Electron Devices 40 1068
[19] Dai X H, Hua Q L, Jiang C S, Long Y, Dong Z L, Shi Y H, Huang T C, Li H T, Meng H X, Yang Y, Wei R L, Shen G Z, Hu W G 2024 Nano Energy 124 109473
[20] Rudrapal K, Bhattacharya G, Adyam V, Chaudhuri A R 2022 Adv. Electron. Mater. 8 2200250
[21] Cho Y, Kim J, Kang M, Kim S 2023 Materials 16 1687
[22] Huang R M, Yan X Z, Ye S, Kashtiban R, Beanland R, Morgan K A, Charlton M D, Groot C D 2017 Nanoscale Res. Lett. 12 384
[23] Zou L Q, Peng Z R, Sun H J, Yi Y H, Zhu C Q, Xu J Y, Zhang J M, Hu X B, Dang Y P, Ye L, Miao X S 2024 Adv. Funct. Mater. 35 2416325
[24] Yue J L, Zou L Q, Bai N, Zhu C Q, Yi Y H, Xue F, Sun H J, Hu S, Cheng W M, He Q, Lu H, Ye L, Miao X S 2024 Small Methods 8 2301657
[25] Huang W Y, Nie L H, Lai X C, Fang J L, Chen Z L, Chen J Y, Jiang Y P, Tang X G 2024 ACS Appl. Mater. Interfaces 16 23615
[26] Qu Z Z, Zhang B L, Li C F, Peng Y T, Wang L P, Li Q X, Zeng Z H, Dong J H 2021 Electrochim. Acta 377 138123
[27] Liu Y H, Wang C X, Li Z Y Wang Y S, Lu W Q, Huang H L 2021 Physica E 127 114515
[28] Rajkumari R, Alam M W, Souayeh B, Singh N K 2024 J. Mater. Sci. 59 3270
[29] Zhang L, Tang Z H, Yao D J, Fan Z Y, Hu S C, Sun Q J, Tang X G, Jiang Y P, Guo X B, Huang M Q 2022 Mater. Today Phy. 23 100650
[30] Guo D, Wu Z, An Y, Li P, Wang P, Chu X, Guo X, Zhi Y, Lei M, Li L 2015 Appl. Phys. Lett. 106 042105
[31] Kumari A, Shanbogh S M, Udachyan I, Kandaiah S, Roy A, Varade V, A. Ponnam 2020 ACS Appl. Mater. Interfaces 12 56568
[32] Chabungbam A S, Kim M, Thakre A, Kim D E, Park H H 2025 J. Mater. Sci. Technol. 213 125
[33] He X L, Chen P 2025 Acta Phys. Sin. 74 028501(in Chinese)[何小龙,陈鹏 2025 74 028501]
[34] Chen T, Zhang T, Yin Y X, Xie Y S, Qiu X Y 2023 Acta Phys. Sin.72 148401(in Chinese)[陈涛,张涛,殷元祥,谢雨莎,邱晓燕2023 72 148401]
[35] Ilyas N, Wang J Y, Li C M, Fu H, Li D Y, Jiang X D, Gu D, Jiang Y D, Li W 2022 J. Mater. Sci. Technol. 97 254
[36] Wang Y C, Yan Y, Wang C, Chen Y T, Li J Y, Zhao J S, Hwang C S 2018 Appl. Phys. Lett. 113 072902
[37] Chauhan M, Choudhary S, Sharma S K 2024 Adv. Electron. Mater. 10 2300724
[38] Kim D, Shin J, Kim S 2022 Appl. Surf. Sci. 599 153876
[39] Wang Y, Huang H X, Huang X L, Guo T T 2023 Acta Phys. Sin. 72 197201(in Chinese)[王英,黄慧香,黄香林,郭婷婷2023 72 197201]
[40] Liu Y, Zhou X F, Yan H, Shi X, Chen K, Zhou J Y, Meng J L, Wang T Y, Ai Y L, Wu J X 2023 Adv. Mater. 35 2301321
[41] Yang W H, Kan H, Shen G Z, Li Y 2024 Adv. Funct. Mater. 34 2312885
[42] Kundale S S, Pawar P S, Kumbhar D D, Devara I K G, Sharma I, Patil P R, Lestari W A, Shim S, Park J, Dongale T D 2024 Adv. Sci. 11 2405251
[43] Xiong X Y, Wu F, Yang Y O, Liu Y M, Wang Z G, Tian H, Dong M D 2023 Adv. Funct. Mater. 34 2213348
[44] Huang J H, Yang S D, Tang X, Yang L L, Chen W J, Chen Z B, Li X M, Zeng Z P, Tang Z K, Gui X C 2023 Adv. Mater. 35 2303737[45] Zhang G B, Wang Z J, Fan X M, Wang Z, Li PT, Luo Q, Gao D W, Wan Q, Zhang Y S 2024 Appl. Phys. Lett. 125 133501
计量
- 文章访问数: 35
- PDF下载量: 2
- 被引次数: 0
下载:
