Journal: Light, Science & Applications

Article Title: Optical neural networks: progress and challenges

doi: 10.1038/s41377-024-01590-3

Figure Lengend Snippet: ONNs constructed by MZIs. a On-chip ONN based on 56 MZIs . b Mathematical inference diagram of the training process for ONN that supports in-situ online training . c On-chip ONNs based on amplitude and phase modulation . d ONN chip designed and fabricated based on MZIs and diffractive units . a Reproduced from ref. with permission of Springer Nature: Nature Photonics. b Reproduced with permission from ref. from © Optical Society of America. c Reproduced from ref. with permission of Springer Nature: Nature Communications. d Reproduced from ref. with permission of Springer Nature: Nature Communications

Article Snippet: Fig. 10 ONNs constructed by MZIs. a On-chip ONN based on 56 MZIs . b Mathematical inference diagram of the training process for ONN that supports in-situ online training . c On-chip ONNs based on amplitude and phase modulation . d ONN chip designed and fabricated based on MZIs and diffractive units . a Reproduced from ref. with permission of Springer Nature: Nature Photonics. b Reproduced with permission from ref. from © Optical Society of America. c Reproduced from ref. with permission of Springer Nature: Nature Communications. d Reproduced from ref. with permission of Springer Nature: Nature Communications In 2021, Zhang et al. pointed out that most research on ONNs still only uses traditional real-value frameworks designed for digital computers.

Techniques: Construct, In Situ

Journal: Light, Science & Applications

Article Title: Optical neural networks: progress and challenges

doi: 10.1038/s41377-024-01590-3

Figure Lengend Snippet: ONNs implementation by MRR weight banks. a ONN for nonlinear compensation in long-distance fiber optic links . b The photonic integrated circuit system designed based on the crossbar framework, which supports training the structural parameters of ONNs online . c ONN system with the light source, data loading area, and data processing units on a single chip . d Convolution operations for human emotion recognition by a microcomb-enabled integrated ONN . a Reproduced from ref. with permission of Springer Nature: Nature Electronics. b Reprinted from ref. with the permission of ACS Publishing. c Reproduced from ref. with permission of Springer Nature: Nature Communications. d Reproduced from ref. with permission of De Gruyter Publishing

Article Snippet: Fig. 10 ONNs constructed by MZIs. a On-chip ONN based on 56 MZIs . b Mathematical inference diagram of the training process for ONN that supports in-situ online training . c On-chip ONNs based on amplitude and phase modulation . d ONN chip designed and fabricated based on MZIs and diffractive units . a Reproduced from ref. with permission of Springer Nature: Nature Photonics. b Reproduced with permission from ref. from © Optical Society of America. c Reproduced from ref. with permission of Springer Nature: Nature Communications. d Reproduced from ref. with permission of Springer Nature: Nature Communications In 2021, Zhang et al. pointed out that most research on ONNs still only uses traditional real-value frameworks designed for digital computers.

Techniques:

Journal: Light, Science & Applications

Article Title: Optical neural networks: progress and challenges

doi: 10.1038/s41377-024-01590-3

Figure Lengend Snippet: Various integrated optical components for constructing on-chip ONNs. a Incoherent optical scattering unit and optical stochastic matrix . b Three-dimensional interconnect waveguides of constructing ONNs . c Architecture and implementation diagram of the ONN chip . d Data architecture and working principle of a photonic tensor core . a Reproduced from ref. with the permission of Science China Press Publishing. b Reproduced with the permission from ref. from © Optical Society of America. c Reproduced from ref. with permission of Springer Nature: Nature. d Reproduced from ref. with permission of Springer Nature: Nature Photonics

Article Snippet: Fig. 10 ONNs constructed by MZIs. a On-chip ONN based on 56 MZIs . b Mathematical inference diagram of the training process for ONN that supports in-situ online training . c On-chip ONNs based on amplitude and phase modulation . d ONN chip designed and fabricated based on MZIs and diffractive units . a Reproduced from ref. with permission of Springer Nature: Nature Photonics. b Reproduced with permission from ref. from © Optical Society of America. c Reproduced from ref. with permission of Springer Nature: Nature Communications. d Reproduced from ref. with permission of Springer Nature: Nature Communications In 2021, Zhang et al. pointed out that most research on ONNs still only uses traditional real-value frameworks designed for digital computers.

Techniques: