Photonics Research, Volume. 13, Issue 6, 1647(2025)

High-speed and versatile ONN through parametric-based nonlinear computation

Xin Dong1、†, Yuanjia Wang1、†, Xiaoxiao Wen1, Yi Zhou1,2,3、*, and Kenneth K. Y. Wong1,2,4、*
Author Affiliations
  • 1Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China
  • 2Advanced Biomedical Instrumentation Centre, Hong Kong, China
  • 3e-mail: zhouyi08@hku.hk
  • 4e-mail: kywong@eee.hku.hk
  • show less
    Figures & Tables(6)
    Experimental setup of parametric process-based ONN. DCF, dispersion compensation fiber; HNLF, highly nonlinear fiber; MLL, mode-locked laser; OSA, optical spectrum analyzer; PC, polarization controller; PD, photodetector.
    Simulation coding block diagram. Self-implemented simulation functions: SMF_simu( ), HNLF_simu( ), and DCF_simu( ).
    (a) Simulated data sample (flattened, “1”) with MNIST-digit database before nonlinear interaction in HNLF (red, frequency domain) and after HNLF (blue, time domain); (b) tendency of accuracy in relation of input power decay.
    (a) Encoded optical spectrum with MNIST-digit database before and after nonlinear interaction in HNLF; (b) confusion matrices with and without nonlinearity, 1200 samples; (c) confusion matrices with and without nonlinearity, 6000 samples.
    (a) Confusion matrices with and without nonlinearity, 1200 samples; (b) linear discriminant analysis with and without nonlinearity, 1200 samples.
    • Table 1. Classification Reports of MNIST-Digit Dataseta

      View table
      View in Article

      Table 1. Classification Reports of MNIST-Digit Dataseta

      With Nonlinearity
      Input dataOpticalOpticalOpticalOpticalOptical
      ArchitectureRidge regression HNLF1-FC layer HNLF1-FC layer ReLU1-FC layer GeLU1-FC layer leaky-ReLU
      Accuracy87.8%87.3%88.5%88.2%89.4%
      Parameters784784784784784
      Without Nonlinearity
      Input dataOpticalDigitalDigitalDigitalDigital
      ArchitectureRidge regression1-FC layerLogistic regressionLDA (1200 samples)LDA (6000 samples)1-FC layer
      Accuracy81.5%80.3%81.4%65.8%82.6%87.5%
      Parameters78478478478450,890
    Tools

    Get Citation

    Copy Citation Text

    Xin Dong, Yuanjia Wang, Xiaoxiao Wen, Yi Zhou, Kenneth K. Y. Wong, "High-speed and versatile ONN through parametric-based nonlinear computation," Photonics Res. 13, 1647 (2025)

    Download Citation

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

    Category: Fiber Optics and Optical Communications

    Received: Dec. 18, 2024

    Accepted: Mar. 24, 2025

    Published Online: May. 30, 2025

    The Author Email: Yi Zhou (zhouyi08@hku.hk), Kenneth K. Y. Wong (kywong@eee.hku.hk)

    DOI:10.1364/PRJ.553388

    CSTR:32188.14.PRJ.553388

    Topics