Acta Optica Sinica, Volume. 45, Issue 14, 1420005(2025)

Advances in Photonic Reservoir Computing (Invited)

Xingxing Guo1,2, Zhiwei Dai1, Shuiying Xiang1,2、*, Hanxu Zhou1, Yahui Zhang1,2, Yanan Han1,2, Changjian Xie1, Tao Wang1, and Yue Hao2
Author Affiliations
  • 1State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an 710071, Shaanxi , China
  • 2State Key Discipline Laboratory of Wide Bandgap Semiconductor Technology, Xidian University, Xi’an 710126, Shaanxi , China
  • show less
    Figures & Tables(16)
    Biological, digital, and physical RC[21]. (a) Biological RC structure; (b) digital RC structure; (c) physical RC structure
    Research progress of on-chip integrated RC system[41-44,47,49,53,57]
    Research progress on free space diffraction RC system[59-62]
    Integrated passive silicon 16-node photonic RC chip[43]
    Novel integrated photonic RC chip based on next-generation RC framework[53]
    RC system based on SLM[60]
    Research progress of delay RC systems for SLs[64-66,68-69,71-72,74-77,79-81]
    Architecture of deep reservoir photonic computer[77]
    Application of deep-latency photonic RC system in nonlinear channel tasks[77]
    Research progress of delay RC systems based on SL by team from Xidian University[82-92]
    Schematic diagram of photonic delay-based RC system using four-channel DFB array[91]
    Performance comparison between single-channel (blue) and four-channel (red) RC systems[91]
    Future application scenarios of photonic RC hardware[95]
    • Table 0. [in Chinese]

      View table

      Table 0. [in Chinese]

      Ref.TypeConstruction method of reservoir layerNumber of reservoir nodesTaskTask accuracy rate
      [59]Free-space diffraction reservoir systemSLM pixels2025 (virtual node)Time series prediction (Mackey-Glass sequence)NMSE≈0.013
      [60]Phase modulation SLM combined with scattering media16,384(virtual node)Human action recognition (KTH database)Classification accuracy 91.3%
      [62]24 diffractive-coupled VCSEL arrays24

      2 bit XOR task,

      3 bit head recognition task

      2 bit XOR BER is 0.008,

      3 bit header recognition BER is 0.007,

      DAC root mean square error is 0.067

    • Table 1. Related work on spatial reservoir

      View table

      Table 1. Related work on spatial reservoir

      Ref.TypeConstruction method of reservoir layerNumber of reservoir nodesTaskTask accuracy rate
      [43]Research progress of on-chip integrated RC systemPassive silicon photonic chip with a 16-node square grid16Boolean logical operation 5 bit header identificationArbitrary Boolean logic operations, 5 bit header identification quantity can reach 12.5 Gbit/s
      [44]Integrated coherent linear photonic processor512 (virtual node)MNIST image classificationAccuracy rate of MNIST image classification test reaches 91.3%
      [53]Silicon photonic chips based on NG-RC45

      Santa Fe time series prediction,

      Lorenz63 task,

      COVID-19 image classification

      Santa Fe task (NMSE is 0.029),

      Lorenz63 task (NMSE is 1.43×10-2),

      classification accuracy of COVID-19 is 92.1% (single wavelength)

      [57]Integrated photonic RC chip with an octagonal layout52Channel equalization capabilityAt 5 km, 5 orders of magnitude lower than rectangular layout, 10 orders lower than feed-forward equalizer; K-nearest neighbor regression BER is 7.012×10⁻¹⁴
    • Table 2. Related work on time-delay reservoir

      View table

      Table 2. Related work on time-delay reservoir

      Ref.Construction method of reservoir layerNumber of virtual nodes in reservoirTaskTask accuracy rate
      [64]SLs combined with fiber feedback loops and optical/electrical injection information388Speech digit chaotic time series prediction

      Speech digital recognition (0.014% error rate),

      time series prediction: 1 (0.6% error rate at 1.3×10⁷ s-1)

      [65]Optical feedback delay system of a single longitudinal mode SRL uses the two-direction modes (CW and CCW) to handle tasks separately.200Handle Santa Fe prediction and channel equalization simultaneously

      Santa Fe chaotic time series prediction (NMSE≈0.022),

      channel equalization (SER≈2.3×10⁻³)

      [82]Single VCSEL using XP and YP polarization modes200

      Santa Fe chaotic time series prediction (XP mode),

      waveform recognition (YP mode)

      Santa Fe chaotic time series prediction (NMSE is 0.0266),

      waveform recognition (NMSE is 0.0167)

      [77]Cascaded injection-locked SLs for all-optical interconnection between layers (4 layers)320Nonlinear signal equalization in optical fiber communication systems (compensating for fiber kerr nonlinear impairments)BER is 1.0×10⁻³ (3-layer photonic RC)
      [57]4-layer deep residual time-delay reservoir960Nonlinear channel equalization, chaotic time series prediction

      Nonlinear channel equalization (SER is 2.6×10-³),

      chaotic time series prediction (NMSE is 0.0169)

      [83]Four distributed feedback laser arrays400Iris flower classification task

      Experiment SER is 0.052,

      simulated SER is 0

      [92]Parallel processing using multiple longitudinal modes of a single Fabry-Perot laser256Santa Fe chaotic prediction (two longitudinal modes), channel equalization (two longitudinal modes)

      NMSE is 0.013,

      BER is 0.001

      [80]Introduction of heterogeneous reservoir with “imperfect” physical models50Chaotic dynamics time series

      Experimental NMSE is 0.0683,

      simulated NMSE is 0.000252

    Tools

    Get Citation

    Copy Citation Text

    Xingxing Guo, Zhiwei Dai, Shuiying Xiang, Hanxu Zhou, Yahui Zhang, Yanan Han, Changjian Xie, Tao Wang, Yue Hao. Advances in Photonic Reservoir Computing (Invited)[J]. Acta Optica Sinica, 2025, 45(14): 1420005

    Download Citation

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

    Category: Optics in Computing

    Received: Apr. 16, 2025

    Accepted: Jun. 19, 2025

    Published Online: Jul. 22, 2025

    The Author Email: Shuiying Xiang (syxiang@xidian.edu.cn)

    DOI:10.3788/AOS250938

    CSTR:32393.14.AOS250938

    Topics