基于磁性隧道结的群体编码实现无监督聚类

张亚君; 蔡佳林; 乔亚; 曾中明; 袁喆; 夏钶

doi:10.7498/aps.71.20220252

摘要

利用新型材料器件发展类脑计算硬件研究的关键问题是发展出合适的算法, 能够发挥新器件的特点和优势. 群体编码是生物神经系统常见的编码方式, 能够有效去除噪音, 实现短时程记忆及复杂的非线性映射功能. 本文选择自旋电子学器件中研究较多、工艺较成熟的磁性隧道结, 应用其可调控的随机动力学实现群体编码. 作为一个应用的例子, 超顺磁隧道结构建的二层脉冲神经网络成功完成了鸢尾花数据集的无监督聚类. 数值仿真表明基于磁性隧道结的群体编码可以有效对抗器件的非均一性, 为类脑计算硬件研究提供重要的参考.

关键词:

Abstract

Developing suitable algorithms that utilize the natural advantages of the corresponding devices is a key issue in the hardware research of brain-inspired computing. Population coding is one of the computational schemes in biological neural systems and it contains the mechanisms for noise reduction, short-term memory and implementation of complex nonlinear functions. Here we show the controllable stochastic dynamical behaviors for the technically mature spintronic device, magnetic tunnel junctions, which can be used as the basis of population coding. As an example, we construct a two-layer spiking neural network, in which groups of magnetic tunnel junctions are used to code input data. After unsupervised learning, this spiking neural network successfully classifies the iris data set. Numerical simulation demonstrates that the population coding is robust enough against the nonuniform dispersion in devices, which is inevitable in fabrication and integration of hardware devices.

Keywords:

作者及机构信息

1.
北京师范大学物理学系, 高等量子研究中心, 北京　100875

2.
中国科学院苏州纳米技术与纳米仿生研究所, 苏州　215123

3.
北京计算科学研究中心, 北京　100193

通信作者: 曾中明, zmzeng2012@sinano.ac.cn ; 袁喆, zyuan@bnu.edu.cn ;

基金项目: 国家自然科学基金(批准号: 11734004, 12174028)资助的课题.

Authors and contacts

1.
Center for Advanced Quantum Studies, Department of Physics, Beijing Normal University, Beijing 100875, China

2.
Suzhou Institute of Nano-Tech and Nano-Bionics, Chinese Academy of Sciences, Suzhou 215123, China

3.
Beijing Computational Science Research Center, Beijing 100193, China

Corresponding author: Zeng Zhong-Ming, zmzeng2012@sinano.ac.cn ; Yuan Zhe, zyuan@bnu.edu.cn ;

Funds: Project supported by the National Natural Science Foundation of China (Grant Nos. 11734004, 12174028)

文章全文

参考文献

[1]	LeCun Y, Bengio Y, Hinton G 2015 Nature 521 436 Google Scholar
[2]	Roy K, Jaiswal A, Panda P 2019 Nature 575 607 Google Scholar
[3]	Davies M, Srinivasa N, Lin T H, et al. 2018 IEEE Micro. 38 82 Google Scholar
[4]	Pei J, Deng L, Song S, et al. 2019 Nature 572 106 Google Scholar
[5]	Ambrogio S, Narayanan P, Tsai H, Shelby R M, Boybat I, di Nolfo C, Sidler S, Giordano M, Bodini M, Farinha N C P, Killeen B, Cheng C, Jaoudi Y, Burr G W 2018 Nature 558 60 Google Scholar
[6]	Torrejon J, Riou M, Araujo F A, et al. 2017 Nature 547 428 Google Scholar
[7]	Yao P, Wu H Q, Gao B, Tang J S, Zhang Q T, Zhang W Q, Yang J J, Qian H 2020 Nature 577 641 Google Scholar
[8]	Zhang X M, Zhuo Y, Luo Q, et al. 2020 Nat. Commun. 11 51 Google Scholar
[9]	Zhang Y, Wang Z R, Zhu J D, Yang Y C, Rao M Y, Song W H, Zhuo Y, Zhang X M, Cui M L, Shen L L, Huang R, Yang J J 2020 Appl. Phys. Rev. 7 011308 Google Scholar
[10]	Jung S, Lee H, Myung S, et al. 2022 Nature 601 211 Google Scholar
[11]	Grollier J, Querlioz D, Camsari K Y, Everschor-Sitte K, Fukami S, Stiles M D 2020 Nat. Electron. 3 360 Google Scholar
[12]	Lan X K, Cao Y, Liu X Y, Xu K J, Liu C, Zheng H Z, Wang K Y 2021 Adv. Intell. Syst. 3 2000182 Google Scholar
[13]	Jiang W C, Chen L N, Zhou K Y, Li L Y, Fu Q W, Du Y W, Liu R H 2019 Appl. Phys. Lett. 115 192403 Google Scholar
[14]	Zhang Y J, Zheng Q, Zhu X R, Yuan Z, Xia K 2020 Sci. China Phys. Mech. Astron. 63 277531 Google Scholar
[15]	Zheng Q, Mi Y Y, Zhu X R, Yuan Z, Xia K 2020 Phys. Rev. Appl. 14 044060 Google Scholar
[16]	Zheng Q, Zhu X R, Mi Y Y, Yuan Z, Xia K 2020 AIP Adv. 10 025116 Google Scholar
[17]	Sengupta A, Roy K 2016 Phys. Rev. Appl. 5 024012 Google Scholar
[18]	Yu W C, Xiao J, Bauer G E W 2021 Phys. Rev. B 104 L180405 Google Scholar
[19]	Song K M, Jeong J S, Pan B, et al. 2020 Nat. Electron. 3 148 Google Scholar
[20]	Kurenkov A, DuttaGupta S, Zhang C, Fukami S, Horio Y, Ohno H 2019 Adv. Mater. 31 e1900636 Google Scholar
[21]	Romera M, Talatchian P, Tsunegi S, et al. 2018 Nature 563 230 Google Scholar
[22]	Banino A, Barry C, Uria B, et al. 2018 Nature 557 429 Google Scholar
[23]	Mi Y, Katkov M, Tsodyks M 2017 Neuron 93 323 Google Scholar
[24]	Pouget A, Dayan P, Zemel R 2000 Nat. Rev. Neurosci. 1 125 Google Scholar
[25]	Thakur C S, Wang R, Hamilton T J, Tapson J, van Schaik A 2016 IEEE Transactions on Circuits and Systems I: Regular Papers 63 211 Google Scholar
[26]	Tuma T, Pantazi A, Le Gallo M, Sebastian A, Eleftheriou E 2016 Nat. Nanotechnol. 11 693 Google Scholar
[27]	Mizrahi A, Hirtzlin T, Fukushima A, Kubota H, Yuasa S, Grollier J, Querlioz D 2018 Nat. Commun. 9 1533 Google Scholar
[28]	Cai J L, Fang B, Zhang L K, Lv W X, Zhang B S, Zhou T J, Finocchio G, Zeng Z M 2019 Phys. Rev. Appl. 11 034015 Google Scholar
[29]	Cai K M, Yang M Y, Ju H L, et al. 2017 Nat. Mater. 16 712 Google Scholar
[30]	Li Z, Zhang S 2004 Phys. Rev. B 69 134416 Google Scholar
[31]	Diehl P U, Cook M 2015 Front. Comput. Neurosci. 9 99 Google Scholar
[32]	Fisher R A 1936 Annals of Eugenics 7 179 Google Scholar
[33]	Dayan P, Abbott L F 2001 Theoretical Neuroscience (Cambridge, MA: MIT Press) pp108–112
[34]	Biswas A, Prasad S, Lashkare S, Ganguly U 2016 arXiv: 1612.02233
[35]	Hayakawa K, Kanai S, Funatsu T, Igarashi J, Jinnai B, Borders W A, Ohno H, Fukami S 2021 Phys. Rev. Lett. 126 117202 Google Scholar

施引文献

图 1 (a) 磁性隧道结示意图; (b) 不同电流下磁性隧道结电阻随时间的变化; (c) 磁性隧道结翻转频率与电流的函数关系; (d) 一组存在不同偏置电流的隧道结实现群体编码 (散点为Néel-Brown理论仿真结果, 实线对应隧道结的调谐曲线)

Fig. 1. (a) Schematic of a magnetic tunnel junction; (b) measured electrical resistance of a magnetic tunnel junction as a function of time under different electrical currents; (c) switching frequency of a magnetic tunnel junction as a function of electrical current; (d) population coding using a group of magnetic tunnel junctions with different bias currents (The dots are simulated data using the Néel-Brown theory and the solid lines are the corresponding tuning curves).

下载: 全尺寸图片幻灯片

图 2 (a) 群体编码脉冲神经网络示意图; (b) 网络训练过程示意图; (c) 鸢尾花数据集无监督聚类测试结果; (d) 用于编码一个数据的隧道结数目与输出神经元数量对网络聚类正确率的影响

Fig. 2. (a) Schematic of spiking neural network; (b) schematic illustration of the network learning process; (c) test results of the unsupervised classification of the iris data set; (d) the influence of number of magnetic tunnel junctions used in the population coding and number of output neurons.

下载: 全尺寸图片幻灯片

图 3 磁性隧道结(a)势垒和(b)翻转临界电流的不均一性对网络性能的影响

Fig. 3. Population coding using the magnetic tunnel junctions with a fluctuation in their (a) energy barriers and (b) critical currents for switching.

下载: 全尺寸图片幻灯片

图 4 不同观测时间(以采样时间dt为单位)对网络执行无监督聚类正确率的影响(上轴表示每个观测时间对应的能耗)

Fig. 4. Classification accuracy of unsupervised clustering performed by the network as a function of the different observation time (The upper axis shows the energy consumption corresponding to each observation time).

下载: 全尺寸图片幻灯片

map

[1]	LeCun Y, Bengio Y, Hinton G 2015 Nature 521 436 Google Scholar
[2]	Roy K, Jaiswal A, Panda P 2019 Nature 575 607 Google Scholar
[3]	Davies M, Srinivasa N, Lin T H, et al. 2018 IEEE Micro. 38 82 Google Scholar
[4]	Pei J, Deng L, Song S, et al. 2019 Nature 572 106 Google Scholar
[5]	Ambrogio S, Narayanan P, Tsai H, Shelby R M, Boybat I, di Nolfo C, Sidler S, Giordano M, Bodini M, Farinha N C P, Killeen B, Cheng C, Jaoudi Y, Burr G W 2018 Nature 558 60 Google Scholar
[6]	Torrejon J, Riou M, Araujo F A, et al. 2017 Nature 547 428 Google Scholar
[7]	Yao P, Wu H Q, Gao B, Tang J S, Zhang Q T, Zhang W Q, Yang J J, Qian H 2020 Nature 577 641 Google Scholar
[8]	Zhang X M, Zhuo Y, Luo Q, et al. 2020 Nat. Commun. 11 51 Google Scholar
[9]	Zhang Y, Wang Z R, Zhu J D, Yang Y C, Rao M Y, Song W H, Zhuo Y, Zhang X M, Cui M L, Shen L L, Huang R, Yang J J 2020 Appl. Phys. Rev. 7 011308 Google Scholar
[10]	Jung S, Lee H, Myung S, et al. 2022 Nature 601 211 Google Scholar
[11]	Grollier J, Querlioz D, Camsari K Y, Everschor-Sitte K, Fukami S, Stiles M D 2020 Nat. Electron. 3 360 Google Scholar
[12]	Lan X K, Cao Y, Liu X Y, Xu K J, Liu C, Zheng H Z, Wang K Y 2021 Adv. Intell. Syst. 3 2000182 Google Scholar
[13]	Jiang W C, Chen L N, Zhou K Y, Li L Y, Fu Q W, Du Y W, Liu R H 2019 Appl. Phys. Lett. 115 192403 Google Scholar
[14]	Zhang Y J, Zheng Q, Zhu X R, Yuan Z, Xia K 2020 Sci. China Phys. Mech. Astron. 63 277531 Google Scholar
[15]	Zheng Q, Mi Y Y, Zhu X R, Yuan Z, Xia K 2020 Phys. Rev. Appl. 14 044060 Google Scholar
[16]	Zheng Q, Zhu X R, Mi Y Y, Yuan Z, Xia K 2020 AIP Adv. 10 025116 Google Scholar
[17]	Sengupta A, Roy K 2016 Phys. Rev. Appl. 5 024012 Google Scholar
[18]	Yu W C, Xiao J, Bauer G E W 2021 Phys. Rev. B 104 L180405 Google Scholar
[19]	Song K M, Jeong J S, Pan B, et al. 2020 Nat. Electron. 3 148 Google Scholar
[20]	Kurenkov A, DuttaGupta S, Zhang C, Fukami S, Horio Y, Ohno H 2019 Adv. Mater. 31 e1900636 Google Scholar
[21]	Romera M, Talatchian P, Tsunegi S, et al. 2018 Nature 563 230 Google Scholar
[22]	Banino A, Barry C, Uria B, et al. 2018 Nature 557 429 Google Scholar
[23]	Mi Y, Katkov M, Tsodyks M 2017 Neuron 93 323 Google Scholar
[24]	Pouget A, Dayan P, Zemel R 2000 Nat. Rev. Neurosci. 1 125 Google Scholar
[25]	Thakur C S, Wang R, Hamilton T J, Tapson J, van Schaik A 2016 IEEE Transactions on Circuits and Systems I: Regular Papers 63 211 Google Scholar
[26]	Tuma T, Pantazi A, Le Gallo M, Sebastian A, Eleftheriou E 2016 Nat. Nanotechnol. 11 693 Google Scholar
[27]	Mizrahi A, Hirtzlin T, Fukushima A, Kubota H, Yuasa S, Grollier J, Querlioz D 2018 Nat. Commun. 9 1533 Google Scholar
[28]	Cai J L, Fang B, Zhang L K, Lv W X, Zhang B S, Zhou T J, Finocchio G, Zeng Z M 2019 Phys. Rev. Appl. 11 034015 Google Scholar
[29]	Cai K M, Yang M Y, Ju H L, et al. 2017 Nat. Mater. 16 712 Google Scholar
[30]	Li Z, Zhang S 2004 Phys. Rev. B 69 134416 Google Scholar
[31]	Diehl P U, Cook M 2015 Front. Comput. Neurosci. 9 99 Google Scholar
[32]	Fisher R A 1936 Annals of Eugenics 7 179 Google Scholar
[33]	Dayan P, Abbott L F 2001 Theoretical Neuroscience (Cambridge, MA: MIT Press) pp108–112
[34]	Biswas A, Prasad S, Lashkare S, Ganguly U 2016 arXiv: 1612.02233
[35]	Hayakawa K, Kanai S, Funatsu T, Igarashi J, Jinnai B, Borders W A, Ohno H, Fukami S 2021 Phys. Rev. Lett. 126 117202 Google Scholar

计量

文章访问数: 7095
PDF下载量: 277
被引次数: 0

姓名
邮箱
手机号码
标题
留言内容
验证码

搜索

留言板

基于磁性隧道结的群体编码实现无监督聚类