• Advanced Photonics Nexus
  • Vol. 4, Issue 1, 016010 (2025)
Jianan Feng1、2, Chang Li1、2, Dahai Yang3、4, Yang Liu1、2, Jianyang Hu1、2, Chen Chen5, Yiqun Wang5, Jie Lin1、6、*, Lei Wang2, and Peng Jin1、2、*
Author Affiliations
  • 1Harbin Institute of Technology, Ministry of Education, Key Laboratory of Micro-systems and Micro-structures Manufacturing, Harbin, China
  • 2Harbin Institute of Technology, School of Instrumentation Science and Engineering, Harbin, China
  • 3Great Bay University, School of Physical Sciences, Dongguan, China
  • 4Great Bay University, Great Bay Institute for Advanced Study, Dongguan, China
  • 5Chinese Academy of Sciences, Suzhou Institute of Nano-Tech and Nano-Bionics, Suzhou, China
  • 6Harbin Institute of Technology, School of Physics, Harbin, China
  • show less
    DOI: 10.1117/1.APN.4.1.016010 Cite this Article Set citation alerts
    Jianan Feng, Chang Li, Dahai Yang, Yang Liu, Jianyang Hu, Chen Chen, Yiqun Wang, Jie Lin, Lei Wang, Peng Jin, "Compact planar-waveguide integrated diffractive optical neural network chip," Adv. Photon. Nexus 4, 016010 (2025) Copy Citation Text show less
    Schemes of (a) existing DNNs and (b) the proposed chip.
    Fig. 1. Schemes of (a) existing DNNs and (b) the proposed chip.
    Simulation classification for the designed chip. (a) Input digits. (b) Simulation results. (c) Intensity distributions.
    Fig. 2. Simulation classification for the designed chip. (a) Input digits. (b) Simulation results. (c) Intensity distributions.
    Schemes of the experimental setup and fabricated chip. (a) Schematic diagram of the experimental setup. (b) Photo of the experimental setup. (c) The fabricated chip. (d) Partial enlarged view of the chip.
    Fig. 3. Schemes of the experimental setup and fabricated chip. (a) Schematic diagram of the experimental setup. (b) Photo of the experimental setup. (c) The fabricated chip. (d) Partial enlarged view of the chip.
    Experimental classification for the designed chip. (a) Input digits. (b) Experimental results. (c) Intensity distributions.
    Fig. 4. Experimental classification for the designed chip. (a) Input digits. (b) Experimental results. (c) Intensity distributions.
    Cycle-test intensity results. (a) Intensity distribution of digit “1”. (b) Intensity distribution of digit “8”. (c) Intensity distribution of digit “9.”
    Fig. 5. Cycle-test intensity results. (a) Intensity distribution of digit “1”. (b) Intensity distribution of digit “8”. (c) Intensity distribution of digit “9.”
    Cycle-test consistency results. (a1) Simulation and (a2) experimental confusion matrices. (b) Accuracy of the 10-cycle test. (c) Consistency of the 10-cycle test.
    Fig. 6. Cycle-test consistency results. (a1) Simulation and (a2) experimental confusion matrices. (b) Accuracy of the 10-cycle test. (c) Consistency of the 10-cycle test.
    Fabrication steps for the three-layer chip.
    Fig. 7. Fabrication steps for the three-layer chip.
    Handwritten digit “0 to 4” classification for a three-layer chip. (a) Input digits. (b) Simulation results. (c) Experimental results. (d) Intensity distributions.
    Fig. 8. Handwritten digit “0 to 4” classification for a three-layer chip. (a) Input digits. (b) Simulation results. (c) Experimental results. (d) Intensity distributions.
    Handwritten digit “5 to 9” classification for a three-layer chip. (a) Input digits. (b) Simulation results. (c) Experimental results. (d) Intensity distributions.
    Fig. 9. Handwritten digit “5 to 9” classification for a three-layer chip. (a) Input digits. (b) Simulation results. (c) Experimental results. (d) Intensity distributions.
    Simulation classification accuracy for different numbers of layers for 10,000 test targets in the MNIST test data set.
    Fig. 10. Simulation classification accuracy for different numbers of layers for 10,000 test targets in the MNIST test data set.
    3D microscope characterization of the step thickness of the proposed diffractive neural networks.
    Fig. 11. 3D microscope characterization of the step thickness of the proposed diffractive neural networks.
    Amplitude-encoded experimental fabricated targets.
    Fig. 12. Amplitude-encoded experimental fabricated targets.
    DimensionSize of neuron (μm)Number of neurons in one layerPropagation distanceAccuracy (%)TOPS
    Ref. 332D separation400200 × 20012 cm91.751.6×107
    Ref. 402D separation41000 × 100020 cm846×109
    Ref. 271D integration2186500  μm86.71.38×104
    Our work2D integration4500 × 5001.2 cm73.43.1×109
    Table 1. Comparison between different architectures.
    Jianan Feng, Chang Li, Dahai Yang, Yang Liu, Jianyang Hu, Chen Chen, Yiqun Wang, Jie Lin, Lei Wang, Peng Jin, "Compact planar-waveguide integrated diffractive optical neural network chip," Adv. Photon. Nexus 4, 016010 (2025)
    Download Citation