搜索

x

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

NbOx忆阻神经元的设计及其在尖峰神经网络中的应用

古亚娜 梁燕 王光义 夏晨阳

引用本文:
Citation:

NbOx忆阻神经元的设计及其在尖峰神经网络中的应用

古亚娜, 梁燕, 王光义, 夏晨阳

Design of NbOx memristive neuron and its application in spiking neural networks

Gu Ya-Na, Liang Yan, Wang Guang-Yi, Xia Chen-Yang
PDF
HTML
导出引用
  • NbOx忆阻器凭借其纳米尺寸、阈值切换及局部有源特性在神经形态计算领域展现出巨大的应用前景. 对NbOx忆阻器动力学特性的深入分析和研究有利于忆阻神经元电路的设计和优化. 本文基于局部有源理论, 采用小信号分析方法对NbOx忆阻器物理模型展开了研究, 定量分析了产生尖峰振荡的区域和条件, 并确定了激励信号幅值和尖峰频率之间的定量关系. 基于上述理论分析, 进一步设计了NbOx忆阻器神经元, 并结合忆阻突触十字交叉阵列, 构建了25×10的尖峰神经网络(spiking neuron network, SNN). 最后, 分别利用频率编码和时间编码两种方式, 有效地实现了数字0到9模式的识别功能.
    NbOx memristors show great application prospect in neuromorphic computing due to its nanoscale size, threshold switching, and locally active properties. The in-depth analysis and study of NbOx memristors’s dynamic properties are beneficial to the design and optimization of memristive neuron circuits. In this paper, based on the local active theory, the physical model of NbOx memristor is studied by using the small signal analysis method, and the region and conditions of the peak oscillation are quantitatively analyzed, and the quantitative relationship between the excitation signal amplitude and the peak frequency is determined. Based on the above theoretical analysis, NbOx memristor neurons are further designed and combined with the memristive synaptic crisscross array in order to construct a 25×10 spiking neural network (SNN). Finally, the recognitional function of digital 0 to 9 patterns is effectively realized by using frequency coding and time coding respectively.
      通信作者: 梁燕, liangyan@hdu.edu.cn
    • 基金项目: 国家自然科学基金(批准号: 62171173)和浙江省自然科学基金(批准号: LY20F010008)资助的课题.
      Corresponding author: Liang Yan, liangyan@hdu.edu.cn
    • Funds: Project supported by the National Natural Science Foundation of China (Grant No. 62171173) and the Natural Science Foundation of Zhejiang Province, China (Grant No. LY20F010008).
    [1]

    Chua L O 1971 IEEE Trans. Circuits Syst. 18 5

    [2]

    Williams R S 2008 IEEE Spectr. 45 12

    [3]

    Zhou J, Cai F, Wang Q, Chen B, S Gaba, W D Lu 2016 IEEE Electron Device Lett. 37 4Google Scholar

    [4]

    王春华, 蔺海荣, 孙晶如, 周玲, 周超, 邓全利 2020 电子与信息学报 42 795Google Scholar

    Wang C H, Lin H R, Sun R J, Zhou L, Zhou C, Deng Q L 2020 J. Electron. Inf. Technol. 42 795Google Scholar

    [5]

    Yang J J, Strukov D B, Stewart D R 2013 Nat. Nanotechnol. 8 1Google Scholar

    [6]

    Wu H, Zhou J, Chen M, Xu Q, Bao B 2021 Chaos, Solitons Fractals. 154 2022

    [7]

    Kim S, Du C, Sheridan P, Ma W, Choi S H, Lu W D 2015 ACS Nano 15 3

    [8]

    Weiher M, Herzig M, Tetzlaff R, Ascoli A, Mikolajick T, Slesazeck S 2019 IEEE Trans. Circuits Syst. 66 7

    [9]

    Strukov D B 2016 Appl. Phys. A 122 4Google Scholar

    [10]

    Kvatinsky S, Ramadan M, Friedman E G, Kolodny A 2015 IEEE Trans. Circuits Syst. Express Briefs. 62 8

    [11]

    Chua L O 2005 Int. J. Bifurcation Chaos 15 11

    [12]

    Ruan J Y, Sun K H, Mou J, He S B, Zhang L M 2018 Eur. Phys. J. Plus 133 3Google Scholar

    [13]

    Weiher M, Herzig M, Tetzlaff R, Ascoli A, Mikolajick T, Slesazeck S 2019 IEEE Trans. Circuits Syst. Regul. Pap. 66 7

    [14]

    Liang Y, Wang G Y, Chen G R, Dong Y J, Yu D S, Iu H H C 2020 IEEE Trans Circuits Syst. 67 5139Google Scholar

    [15]

    Mannan Z I, Choi H, Kim H, Chua L O 2016 Int. J. Bifurcation Chaos 26 1630009Google Scholar

    [16]

    Jin P P, Wang G Y, Liang Y, Iu H H C, Chua L O 2021 IEEE Trans. Circuits Syst. 68 11

    [17]

    Yi W, Tsang K K, Lam S K, Bai X, Crowell J A, Flores E A 2018 Nat. Commun. 7 9

    [18]

    Lin H R, Wang C H, Sun Y C, Yao W 2020 Nonlinear Dyn. 100 4

    [19]

    Wei Q M, Tang J S, Li X Y, Zhong Y N, Gao B, Qian H, Wu H Q 2021 5th IEEE Electron Devices Technology & Manufacturing Conference (EDTM) Chengdu, China, April 8–11, 2021 pp1–3

    [20]

    Frank D J, Dennard R H, Nowak E, Solomon P M, Taur Y, Wong H S P 2001 Proc. IEEE. 89 3Google Scholar

    [21]

    Yeo I, Chu M, Gi S, Hwang H, Lee B 2019 IEEE Trans. Electron Devices 66 7Google Scholar

    [22]

    Zhang X M, Wu Z H, Lu J K, et al. 2020 IEEE International Electron Devices Meeting(IEDM) Electr Network, December 12–18, 2020

    [23]

    Wang Z R, Joshi S, Savel’ ev S, et al. 2018 Nat. Electron. 1 2Google Scholar

    [24]

    Sheridan P, Ma W, Lu W 2014 IEEE International Symposium on Circuits and Systems (ISCAS) Melbourne, June 1–5, 2014 pp1078–1081

    [25]

    Kumar S, Strachan J P, Williams R S 2017 Nature 548 7667

    [26]

    Lottermoser T, Lonkai T, Amann U, Hohlwein D, Ihringer J, Fiebig M 2004 Nature 430 6999

    [27]

    Sawicki M, Chiba D, Korbecka A, Nishitani Y, Majewski J A, Matsukura F, Dietl T, Ohno H 2008 Nature 455 7212

    [28]

    Ascoli1 A, Demirkol1 A S, Tetzlaff R, Slesazeck S, Mikolajick T, Chua L O 2021 Front. Neurosci. 15 651452Google Scholar

    [29]

    Liang Y, Zhu Q, Wang G Y, Nath S K, Iu H H C, Nandi S K, Elliman R G 2021 IEEE Trans. Circuits Syst. Regul. Pap. 68 1278Google Scholar

    [30]

    徐泠风, 李传东, 陈玲 2016 物理学报 65 240701Google Scholar

    Xu L F, Li C D, Chen L 2016 Acta Phys. Sin. 65 240701Google Scholar

    [31]

    Zhang Y, Wang X, Li Y, Friedman E G 2017 IEEE Trans. Circuits Syst. Express Briefs. 64 7

    [32]

    洪庆辉 2019 博士学位论文 (武汉: 华中科技大学)

    Hong Q H 2019 Ph.D. Dissertation (Wuhan: Huazhong University of Science and Technology) (in Chinese)

  • 图 1  S型NbOx LAM响应于1 MHz, 10 MHz, 1 GHz的正弦信号的捏滞曲线

    Fig. 1.  Pinch hysteresis curves of S-type NbOx LAM in response to sinusoidal signals at 1 MHz, 10 MHz and 1 GHz.

    图 2  NbOx LAM的直流V-I图 (a) 浅蓝色部分是忆阻器的NDR区; (b) 红色线是负载线, 插图为电压偏置电路

    Fig. 2.  DC V-I plot of NbOx LAM: (a) The light blue part is the NDR region of the memristor; (b) the red line is the load line and the inset is the bias circuit with DC voltage supply.

    图 3  h (V0 , T)与温度T的关系图

    Fig. 3.  Relationship between h (V0 , T) and temperature T.

    图 4  (a) NbOx-Mott忆阻器在工作点Q (0.008 A, 0.3003 V)处的小信号等效电路模型; (b) Rx对工作点的依赖性; (c) Lx对工作点的依赖性; (d) Ry对工作点的依赖性

    Fig. 4.  (a) Small-signal equivalent circuit model of NbOx-Mott memristor at the operating point Q (0.008 A, 0.3003 V); (b) the dependence of Rx on the operating point; (c) the dependence of Lx on the operating point; (d) the dependence of Ry on the operating point.

    图 5  (a) I = 0.008 A时的实部虚部的频率响应; (b) 奈奎斯特图

    Fig. 5.  (a) Frequency responses of the real and imaginary parts at I = 0.008 A; (b) Nyquist plot.

    图 6  二阶振荡电路

    Fig. 6.  Second-order oscillator circuit.

    图 7  雅可比矩阵的特征值在0.042 nF < C < 10 nF范围内的变化

    Fig. 7.  Variations of the eigenvalues of the Jacobian matrix for 0.042 nF < C < 10 nF.

    图 8  I = 0.008 A, C = 0.3 nF时, NbOx LAM的二阶振荡器的仿真结果 (a)电压vm、状态变量T和电流im的瞬态波形; (b)稳定点的im-T相图; 当I = 0.008 A, C = 0.8 nF时, NbOx LAM的二阶振荡器的仿真结果: (c)电压vm、状态变量T和电流im的瞬态波形; (d) 振荡状态的im-T相图

    Fig. 8.  Simulation results of the NbOx LAM second-order oscillator: (a) The transient waveforms of vm, T and im at I = 0.008 A and C = 0.3 nF; (b) the stable equilibrium on im-T phase plane at I = 0.008 A and C = 0.3 nF; (c) the transient waveforms of vm, T and im at I = 0.008 A and C = 0.8 nF; (d) the limit cycle on the im-T phase plane at I = 0.008 A and C = 0.8 nF.

    图 9  (a) 输入的直流电流激励取10, 30和 50 mA时, 忆阻器电流的时域图; (b) 不同电流激励对应的尖峰数量关系图

    Fig. 9.  (a) The time-domain waveforms of im at different input DC current excitations of 10, 30 and 50 mA; (b) the number of spikes corresponding to different current excitations.

    图 10  基于NbOx忆阻器的神经元电路

    Fig. 10.  Neuron circuit based on NbOx memristor.

    图 11  (a) 忆阻突触处于ON状态时, Vi1vm时域图; (b) 忆阻突触处于ON状态时, im时域图; (c) 忆阻突触处于OFF状态时, Vi1vm时域图; (d) 忆阻突触处于OFF状态时, im时域图

    Fig. 11.  (a) The time-domain waveforms of Vi1 and vm when the memristive synapse is at ON state; (b) the time-domain waveforms of im at ON state; (c) the time-domain waveforms of Vi1 and vm when the memristive synapse is at OFF state; (d) the time-domain waveforms of im at OFF state.

    图 12  5×5的10种数字模式

    Fig. 12.  10 digital patterns of 5×5.

    图 13  电压突触电路实现

    Fig. 13.  Implementation of voltage synapse circuit.

    图 14  由25×10的突触阵列以及10个输出神经元构成的尖峰神经网络

    Fig. 14.  A spiking neural network consisting of a 25×10 synaptic array and 10 output neurons.

    图 15  (a) 数字2输入尖峰神经网络, 10个输出神经元的电流im输出时域图; (b) 10种模式输入尖峰神经网络时各输出神经元输出电流频率的情况

    Fig. 15.  (a) The time-domain waveforms of im of 10 output neurons when “2” mode is input to SNN; (b) the output current frequencies of each output neuron when ten modes are input to the spiking neural network.

    图 16  基于TC SNN的神经元电路

    Fig. 16.  Neuron circuit based on TC SNN.

    图 17  (a) 单尖峰电路的仿真结果图; (b) 未应用单尖峰电路方案的仿真结果

    Fig. 17.  (a) The simulation results with the single-spike circuit; (b) the simulation results without the single-spike circuit.

    图 18  (a)“2”模式输入SNN时, 10个输出神经元的电流时域图; (b) 不同输入模式对应的神经元输出首个脉冲的时间

    Fig. 18.  (a) The time-domain waveforms of im of 10 output neurons when “2” mode is input to SNN; (b) the time of outputting the first pulse of neurons corresponding to different input modes.

  • [1]

    Chua L O 1971 IEEE Trans. Circuits Syst. 18 5

    [2]

    Williams R S 2008 IEEE Spectr. 45 12

    [3]

    Zhou J, Cai F, Wang Q, Chen B, S Gaba, W D Lu 2016 IEEE Electron Device Lett. 37 4Google Scholar

    [4]

    王春华, 蔺海荣, 孙晶如, 周玲, 周超, 邓全利 2020 电子与信息学报 42 795Google Scholar

    Wang C H, Lin H R, Sun R J, Zhou L, Zhou C, Deng Q L 2020 J. Electron. Inf. Technol. 42 795Google Scholar

    [5]

    Yang J J, Strukov D B, Stewart D R 2013 Nat. Nanotechnol. 8 1Google Scholar

    [6]

    Wu H, Zhou J, Chen M, Xu Q, Bao B 2021 Chaos, Solitons Fractals. 154 2022

    [7]

    Kim S, Du C, Sheridan P, Ma W, Choi S H, Lu W D 2015 ACS Nano 15 3

    [8]

    Weiher M, Herzig M, Tetzlaff R, Ascoli A, Mikolajick T, Slesazeck S 2019 IEEE Trans. Circuits Syst. 66 7

    [9]

    Strukov D B 2016 Appl. Phys. A 122 4Google Scholar

    [10]

    Kvatinsky S, Ramadan M, Friedman E G, Kolodny A 2015 IEEE Trans. Circuits Syst. Express Briefs. 62 8

    [11]

    Chua L O 2005 Int. J. Bifurcation Chaos 15 11

    [12]

    Ruan J Y, Sun K H, Mou J, He S B, Zhang L M 2018 Eur. Phys. J. Plus 133 3Google Scholar

    [13]

    Weiher M, Herzig M, Tetzlaff R, Ascoli A, Mikolajick T, Slesazeck S 2019 IEEE Trans. Circuits Syst. Regul. Pap. 66 7

    [14]

    Liang Y, Wang G Y, Chen G R, Dong Y J, Yu D S, Iu H H C 2020 IEEE Trans Circuits Syst. 67 5139Google Scholar

    [15]

    Mannan Z I, Choi H, Kim H, Chua L O 2016 Int. J. Bifurcation Chaos 26 1630009Google Scholar

    [16]

    Jin P P, Wang G Y, Liang Y, Iu H H C, Chua L O 2021 IEEE Trans. Circuits Syst. 68 11

    [17]

    Yi W, Tsang K K, Lam S K, Bai X, Crowell J A, Flores E A 2018 Nat. Commun. 7 9

    [18]

    Lin H R, Wang C H, Sun Y C, Yao W 2020 Nonlinear Dyn. 100 4

    [19]

    Wei Q M, Tang J S, Li X Y, Zhong Y N, Gao B, Qian H, Wu H Q 2021 5th IEEE Electron Devices Technology & Manufacturing Conference (EDTM) Chengdu, China, April 8–11, 2021 pp1–3

    [20]

    Frank D J, Dennard R H, Nowak E, Solomon P M, Taur Y, Wong H S P 2001 Proc. IEEE. 89 3Google Scholar

    [21]

    Yeo I, Chu M, Gi S, Hwang H, Lee B 2019 IEEE Trans. Electron Devices 66 7Google Scholar

    [22]

    Zhang X M, Wu Z H, Lu J K, et al. 2020 IEEE International Electron Devices Meeting(IEDM) Electr Network, December 12–18, 2020

    [23]

    Wang Z R, Joshi S, Savel’ ev S, et al. 2018 Nat. Electron. 1 2Google Scholar

    [24]

    Sheridan P, Ma W, Lu W 2014 IEEE International Symposium on Circuits and Systems (ISCAS) Melbourne, June 1–5, 2014 pp1078–1081

    [25]

    Kumar S, Strachan J P, Williams R S 2017 Nature 548 7667

    [26]

    Lottermoser T, Lonkai T, Amann U, Hohlwein D, Ihringer J, Fiebig M 2004 Nature 430 6999

    [27]

    Sawicki M, Chiba D, Korbecka A, Nishitani Y, Majewski J A, Matsukura F, Dietl T, Ohno H 2008 Nature 455 7212

    [28]

    Ascoli1 A, Demirkol1 A S, Tetzlaff R, Slesazeck S, Mikolajick T, Chua L O 2021 Front. Neurosci. 15 651452Google Scholar

    [29]

    Liang Y, Zhu Q, Wang G Y, Nath S K, Iu H H C, Nandi S K, Elliman R G 2021 IEEE Trans. Circuits Syst. Regul. Pap. 68 1278Google Scholar

    [30]

    徐泠风, 李传东, 陈玲 2016 物理学报 65 240701Google Scholar

    Xu L F, Li C D, Chen L 2016 Acta Phys. Sin. 65 240701Google Scholar

    [31]

    Zhang Y, Wang X, Li Y, Friedman E G 2017 IEEE Trans. Circuits Syst. Express Briefs. 64 7

    [32]

    洪庆辉 2019 博士学位论文 (武汉: 华中科技大学)

    Hong Q H 2019 Ph.D. Dissertation (Wuhan: Huazhong University of Science and Technology) (in Chinese)

  • [1] 郭慧朦, 梁燕, 董玉姣, 王光义. 蔡氏结型忆阻器的简化及其神经元电路的硬件实现. 物理学报, 2023, 72(7): 070501. doi: 10.7498/aps.72.20222013
    [2] 黄颖, 顾长贵, 杨会杰. 神经网络超参数优化的删除垃圾神经元策略. 物理学报, 2022, 71(16): 160501. doi: 10.7498/aps.71.20220436
    [3] 丁大为, 卢小齐, 胡永兵, 杨宗立, 王威, 张红伟. 分数阶忆阻耦合异质神经元的多稳态及硬件实现. 物理学报, 2022, 71(23): 230501. doi: 10.7498/aps.71.20221525
    [4] 武长春, 周莆钧, 王俊杰, 李国, 胡绍刚, 于奇, 刘洋. 基于忆阻器的脉冲神经网络硬件加速器架构设计. 物理学报, 2022, 71(14): 148401. doi: 10.7498/aps.71.20220098
    [5] 朱佳雪, 张续猛, 王睿, 刘琦. 面向神经形态感知和计算的柔性忆阻器基脉冲神经元. 物理学报, 2022, 71(14): 148503. doi: 10.7498/aps.71.20212323
    [6] 温新宇, 王亚赛, 何毓辉, 缪向水. 忆阻类脑计算. 物理学报, 2022, 71(14): 140501. doi: 10.7498/aps.71.20220666
    [7] 王世场, 卢振洲, 梁燕, 王光义. N型局部有源忆阻器的神经形态行为. 物理学报, 2022, 71(5): 050502. doi: 10.7498/aps.71.20212017
    [8] 白婧, 关富荣, 唐国宁. 神经元网络中局部同步引发的各种效应. 物理学报, 2021, 70(17): 170502. doi: 10.7498/aps.70.20210142
    [9] 黄志精, 李倩昀, 白婧, 唐国宁. 在具有排斥耦合的神经元网络中有序斑图的熵测量. 物理学报, 2019, 68(11): 110503. doi: 10.7498/aps.68.20190231
    [10] 汪芃, 李倩昀, 黄志精, 唐国宁. 在兴奋-抑制混沌神经元网络中有序波的自发形成. 物理学报, 2018, 67(17): 170501. doi: 10.7498/aps.67.20180506
    [11] 刘玉东, 王连明. 基于忆阻器的spiking神经网络在图像边缘提取中的应用. 物理学报, 2014, 63(8): 080503. doi: 10.7498/aps.63.080503
    [12] 李志军, 曾以成, 李志斌. 改进型细胞神经网络实现的忆阻器混沌电路. 物理学报, 2014, 63(1): 010502. doi: 10.7498/aps.63.010502
    [13] 吴望生, 唐国宁. 不同耦合下混沌神经元网络的同步. 物理学报, 2012, 61(7): 070505. doi: 10.7498/aps.61.070505
    [14] 陈军, 李春光. 具有自适应反馈突触的神经元模型中的混沌:电路设计. 物理学报, 2011, 60(5): 050503. doi: 10.7498/aps.60.050503
    [15] 董继扬, 张军英, 陈 忠. 自动波竞争神经网络及其在单源最短路问题中的应用. 物理学报, 2007, 56(9): 5013-5020. doi: 10.7498/aps.56.5013
    [16] 王瑞敏, 赵 鸿. 神经元传输函数对人工神经网络动力学特性的影响. 物理学报, 2007, 56(2): 730-739. doi: 10.7498/aps.56.730
    [17] 吴忠强, 谭拂晓, 王绍仙. 基于无源化的细胞神经网络超混沌系统同步. 物理学报, 2006, 55(4): 1651-1658. doi: 10.7498/aps.55.1651
    [18] 何国光, 曹志彤. 混沌神经网络的控制. 物理学报, 2001, 50(11): 2103-2107. doi: 10.7498/aps.50.2103
    [19] 于丽娟, 朱长纯. 用人工神经网络预测场发射开启电场. 物理学报, 2000, 49(1): 170-173. doi: 10.7498/aps.49.170
    [20] 常胜江, 刘, 张文伟, 申金媛, 翟宏琛, 张延. 适用于神经元状态非等概率分布的神经网络模型及其光学实现. 物理学报, 1998, 47(7): 1101-1109. doi: 10.7498/aps.47.1101
计量
  • 文章访问数:  3713
  • PDF下载量:  189
  • 被引次数: 0
出版历程
  • 收稿日期:  2022-01-20
  • 修回日期:  2022-02-13
  • 上网日期:  2022-03-04
  • 刊出日期:  2022-06-05

/

返回文章
返回