Acta Optica Sinica, Volume. 44, Issue 21, 2110001(2024)

Conductance Constraint-Based High-Accuracy Image Recognition Network

Lihua Xu1,2, Yibo Zhao1, and Chengdong Yang2、*
Author Affiliations
  • 1School of Electronics & Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, Jiangsu , China
  • 2School of Electronic and Information Engineering, Wuxi University, Wuxi 214105, Jiangsu , China
  • show less
    Figures & Tables(6)
    Schematic diagrams of device structure and electrical test. (a) Schematic diagram of device structure and correspondence between device and biological synapses; (b) schematic diagram of electrical test process
    Optical synaptic plasticity of B-B SJ synaptic devices. (a) I-V curves measured under dark, illumination (9 μW/cm2) and post-illumination conditions; (b) transition from STP to LTP at different pulse numbers; (c) PPF index of device; (d) measured learning-relearning behavior; (e) decay of EPSC under different pulse numbers (inset is relationship among degree of learning, rate of forgetting, and number of pulses)
    CNN model
    Schematic diagram of mapping of conductivity to weight and variations of conductivity and EPSC of B-B SJ devices. (a) Schematic diagram of mapping process of conductivity to weight in memristive array; (b) characteristic plot of first convolutional layer; (c) EPSCs excited by different pulse numbers (3, 6, and 10); (d) variation in conductivity for different pulse numbers (3, 6, and 10)
    Recognition accuracy confusion matrices and accuracy distributions of M-CNNs corresponding to different pulse numbers. (a) Confusion matrix corresponding to 3 pulses; (b) accuracies of training set and test set corresponding to 3 pulses; (c) confusion matrix corresponding to 6 pulses; (d) accuracies of training set and test set corresponding to 6 pulses; (e) confusion matrix corresponding to 10 pulses; (f) accuracies of training set and test set corresponding to 10 pulses
    • Table 1. Comparison of image recognition accuracies for different memristors as synapses

      View table

      Table 1. Comparison of image recognition accuracies for different memristors as synapses

      Issued yearNeural networkMemristor structureAccuracy /%Reference
      2020CNNTi/HfO2/TiN92.79[26]
      2021FCNW/HfO2/TiN90.10[27]
      2021CNNPt/HfO2∶Cu/Cu96.75[25]
      2022ANNPt/Li4Ti5O12/TiO2/Pt87.00[28]
      2022ANNPDMS/N2200/PU96.26[29]
      2022M-CNNPt/Pr0.7Ca0.3MnO3/TiN96.16[16]
      2023ANNIGZO94.60[30]
      2024ANNPt/BTO/NSTO94.50[31]
      2024CNNAg/BA0.15MA0.85PbI3/FTO94.80[32]
      2024M-CNNB-B SJ95.12Ours
    Tools

    Get Citation

    Copy Citation Text

    Lihua Xu, Yibo Zhao, Chengdong Yang. Conductance Constraint-Based High-Accuracy Image Recognition Network[J]. Acta Optica Sinica, 2024, 44(21): 2110001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: May. 27, 2024

    Accepted: Jul. 11, 2024

    Published Online: Nov. 20, 2024

    The Author Email: Yang Chengdong (860118@cwxu.edu.cn)

    DOI:10.3788/AOS241074

    CSTR:32393.14.AOS241074

    Topics