Acta Optica Sinica, Volume. 45, Issue 14, 1420011(2025)

Principles and Applications of Nonlinearity in Optical Neural Networks (Invited)

Yufei Wang1,2, Yumeng Chen1, Yongzheng Yang1, Kun Liao1、*, Xiaoyong Hu1,2、**, and Qihuang Gong1,2
Author Affiliations
  • 1State Key Laboratory of Artificial Microstructure and Mesoscopic Physics, School of Physics, Peking University, Beijing 100871, China
  • 2Yangtze Delta Institute of Optoelectronics, Peking University, Nantong 226010, Jiangsu , China
  • show less
    Figures & Tables(10)
    All optical nonlinear computing schemes based on second-order nonlinear frequency doubling effect. (a) Nonlinear scheme based on parametric amplification enabled by second-order nonlinearity in PPLN[11]; (b) degenerate optical parametric oscillator formed by antisymmetric coupling between fundamental and second harmonic waves in PPLN[13]; (c) nonlinear mapping of image data via interaction between random scattering and second harmonic generation in LN crystal[14]
    All optical nonlinear schemes based on material saturable and reverse saturable absorption effects. (a) Nonlinear scheme based on rubidium atomic saturable absorption for controlling forward and backward light propagation[15]; (b) nonlinear scheme based on saturable and reverse saturable absorption curves in metal carbides[17]; (c) nonlinear scheme based on two-dimensional MoTe2 flake transferred onto waveguide facet[18]; (d) nonlinear scheme based on Bi2Te3 thin film saturable absorption on Si3N4 waveguide[20]
    All-optical nonlinear schemes based on other third-order nonlinear effects. (a) Nonlinear scheme based on free-carrier dispersion in resonator[7]; (b) tunable nonlinear scheme based on combined effects of free-carrier dispersion and thermo-optic modulation in resonator[8]; (c) nonlinear scheme using four-wave mixing in LN to enable pump-pulse-induced modulation of input pulses[21]
    All-optical nonlinear schemes based on PCM modulation. (a) On-chip optical signals alter refractive index of PCM to modulate resonator transmission spectrum and generate nonlinear activation[26]; (b) off-chip optical pumping induces a phase transition in PCM, enabling nonlinear modulation of on-chip optical signals[28]
    Opto-electro-optical nonlinear schemes based on direct modulation in on-chip waveguides. (a) Nonlinear modulation scheme based on MZI under thermal effect[29]; (b) nonlinear modulation scheme using resonator with forward-biased voltage inducing free-carrier dispersion[31]; (c) nonlinear modulation scheme using resonator with reverse-biased voltage inducing free-carrier dispersion[32]; (d) nonlinear modulation scheme based on MZM exploiting electro-optic effect[35]
    Opto-electro-optical nonlinear schemes based on on-chip two-dimensional material modulation. (a) Nonlinear modulation scheme based on MoS₂ opto-resistive random access memory switch[36]; (b) nonlinear modulation scheme based on graphene/silicon heterojunction[37]
    Opto-electronic nonlinear schemes based on photodetection. (a) Nonlinear scheme based on complex modulation using on-chip diffractive units and balanced photodetectors[46]; (b) Nonlinear modulation scheme based on vertical-cavity surface-emitting lasers with homodyne detection[47]
    Applications of nonlinear activation functions in feedforward neural network architectures. (a) Complex-valued photonic nonlinear activation function demonstrated on MNIST and CIFAR-10 image classification tasks using complex photonic neural networks[37]; (b) nonlinear response of laser array output performs Fashion-MNIST and EMNIST image recognition tasks within biological neural network structure[47]; (c) Nonlinear photonic computing unit used in coherent optical neural network architecture for vowel classification task[33]; (d) nonlinear activation function used to simulate evolution of SSH model in photonic neural network[20]
    Applications of nonlinear neurons in reservoir computing neural networks. (a) Large-scale reservoir computing neural network constructed with phase-type spatial light modulator based nonlinear neurons demonstrated on human action recognition tasks[44]; (b) reservoir computing network based on multimode spiral waveguides applied to time series prediction[49]; (c) photonic layered neurons based on integrated quantum dot lasers used to construct reservoir networks for image recognition tasks[55]; (d) deep reservoir computing network built with cascaded injection-locked semiconductor laser chips and optical feedback loops for optical module nonlinear equalization[56]
    Applications of nonlinear optical pulse activation in optical spiking neural networks. (a) Nonlinear optical pulse activation using FP-SA laser for digital pattern recognition[60]; (b) nonlinear optical pulse activation using resonator structure incorporating PCM for alphabet pattern recognition[26]
    Tools

    Get Citation

    Copy Citation Text

    Yufei Wang, Yumeng Chen, Yongzheng Yang, Kun Liao, Xiaoyong Hu, Qihuang Gong. Principles and Applications of Nonlinearity in Optical Neural Networks (Invited)[J]. Acta Optica Sinica, 2025, 45(14): 1420011

    Download Citation

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

    Category: Optics in Computing

    Received: Apr. 15, 2025

    Accepted: Jun. 23, 2025

    Published Online: Jul. 18, 2025

    The Author Email: Kun Liao (kunliao@pku.edu.cn), Xiaoyong Hu (xiaoyonghu@pku.edu.cn)

    DOI:10.3788/AOS250924

    CSTR:32393.14.AOS250924

    Topics