pith. sign in

arxiv: 2605.14690 · v2 · pith:LQB2KRA6new · submitted 2026-05-14 · ⚛️ physics.optics

Integrated photonic computing: towards high-dimensional information processing

Ji Qin , Zhi-Kai Pong , Xuke Qiu , Liangyu Deng , Runchen Zhang , Yunqi Zhang , Jinge Guo , Yifei Ma
show 24 more authors
Zimo Zhao Yuanxing Shen Patrick Salter Martin Booth Stephen Morris Honghui He Min Gu Bowei Dong Chao He Chao He (1) ((1) Department of Engineering Science University of Oxford UK (2) Tsinghua Shenzhen International Graduate School Tsinghua University China (3) School of Artificial Intelligence Science Technology University of Shanghai for Science (4) Institute of Photonic Chips (5) Institute of Microelectronics Agency for Science Research A*STAR Singapore)
This is my paper

Pith reviewed 2026-05-20 21:19 UTC · model grok-4.3

classification ⚛️ physics.optics
keywords integrated photonicshigh-dimensional computingoptical computingtopological photonicsenergy efficiencyartificial intelligenceMach-Zehnder interferometersoptical skyrmions
0
0 comments X

The pith

By embracing the higher dimensionality of light, photonic computing offers a new paradigm for high-performance and energy-efficient information processing.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This review examines the shift in integrated photonic computing from low-dimensional to high-dimensional architectures that leverage the phase, amplitude, spatial modes, wavelength channels, and polarization of light. The authors show how basic devices such as Mach-Zehnder interferometers and microring resonators can be extended to process multiple data streams in a single waveguide for greater throughput. They discuss system-level methods like time-wavelength interleaving and hardware-aware training to improve efficiency and co-design with algorithms. Five key challenges are identified, with emerging topological structures like optical skyrmions proposed to enhance robustness through fault-tolerant encoding. A sympathetic reader would care because this could provide a scalable alternative to traditional electronics strained by AI demands and Moore's law limits.

Core claim

The paper claims that high-dimensional architectures exploiting spatial modes and wavelength channels to carry multiple independent data streams through a single waveguide achieve higher throughput with moderate hardware overhead, and that combining these with system-level techniques and topological structures can overcome current challenges to establish a new paradigm beyond incremental improvements.

What carries the argument

High-dimensional architectures that use spatial modes and wavelength channels to process multiple independent data streams in a single waveguide, building on low-dimensional elements like Mach-Zehnder interferometers and microring resonators.

Load-bearing premise

That system-level techniques such as time-wavelength interleaving and hardware-aware training, together with emerging topological structures, will be sufficient to overcome the challenges of electro-optic conversion efficiency, computing parallelism, spatial integration, reconfigurability, and robustness.

What would settle it

An experiment showing that high-dimensional photonic processors do not achieve net improvements in energy efficiency or throughput compared to optimized electronic systems after including all conversion and integration overheads would falsify the proposed paradigm shift.

read the original abstract

The rapid growth of artificial intelligence, coupled with the slowing of Moore's law, is straining computing infrastructure, as CMOS electronics face inherent limits in bandwidth, energy efficiency, and parallelism. Integrated photonic computing encodes and processes information using the phase, amplitude, spatial modes, wavelength channels, and polarisation of guided optical fields, offering a scalable and energy-efficient route beyond charge-based signalling. Here, we review on-chip photonic computing, emphasising the progression from low-dimensional to high-dimensional architectures. At the foundational level, low-dimensional approaches manipulate the phase and amplitude of guided light through Mach-Zehnder interferometers, diffractive structures, microring resonators, and absorptive elements, forming a programmable basis for optical matrix-vector multiplication. Crucially, high-dimensional architectures exploit spatial modes and wavelength channels to carry multiple independent data streams through a single waveguide, achieving higher throughput with moderate hardware overhead. Practical deployment, however, demands more than device innovation. We examine how system-level techniques, from time-wavelength interleaving to hardware-aware training, address energy efficiency, precision, and algorithm-hardware co-design. Five challenges nevertheless remain: electro-optic conversion efficiency, computing parallelism, spatial integration, reconfigurability, and robustness. We highlight emerging topological structures, such as optical skyrmions, as a promising route to fault-tolerant, topologically protected encoding that exploits the largely untapped polarisation degree of freedom. We argue that, by embracing the higher dimensionality of light, photonic computing can offer not merely an incremental improvement but a new paradigm for high-performance, energy-efficient information processing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. This review surveys integrated photonic computing, progressing from low-dimensional devices (Mach-Zehnder interferometers, microring resonators, diffractive structures) that manipulate phase and amplitude for optical matrix-vector multiplication to high-dimensional architectures that exploit spatial modes, wavelength channels, and polarization to increase throughput with moderate hardware overhead. It covers system-level techniques including time-wavelength interleaving and hardware-aware training, identifies five persistent challenges (electro-optic conversion efficiency, computing parallelism, spatial integration, reconfigurability, and robustness), and highlights emerging topological structures such as optical skyrmions for topologically protected encoding. The paper concludes that embracing higher dimensionality of light can deliver a new paradigm for high-performance, energy-efficient information processing beyond incremental gains.

Significance. If the synthesis of the literature is accurate and complete, the review could help frame research priorities by clearly delineating the transition to high-dimensional photonic systems and by flagging concrete engineering bottlenecks. The explicit listing of the five challenges and the pointer to topological encodings provide a useful roadmap; credit is due for organizing the discussion around both device-level and system-level considerations.

major comments (1)
  1. [Abstract and challenges discussion] Abstract and the section enumerating the five challenges: the central claim that high-dimensional architectures constitute a 'new paradigm' for energy-efficient computing is in tension with the explicit listing of electro-optic conversion efficiency as a remaining challenge. High-dimensional encoding (spatial modes, wavelength multiplexing) occurs after the E/O conversion step; the manuscript does not cite or derive a concrete mechanism by which additional optical degrees of freedom reduce the energy cost or loss of that conversion itself. If conversion remains the dominant bottleneck, the system-level efficiency gain required to justify the paradigm-shift language is not yet secured by the dimensionality argument alone.
minor comments (2)
  1. [Section on emerging topological structures] Ensure that the discussion of optical skyrmions includes sufficient primary references and a brief explanation of how their topological protection maps onto integrated waveguide platforms, to avoid the impression that this is an unsubstantiated suggestion.
  2. [Throughout the review of device approaches] Verify that all cited performance metrics (energy per operation, parallelism figures) are drawn from comparable experimental conditions and clearly distinguished from simulation results.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback and for recognizing the review's organization of low- to high-dimensional photonic architectures along with the explicit roadmap of challenges. We address the single major comment below and outline a targeted revision.

read point-by-point responses

  1. Referee: [Abstract and challenges discussion] Abstract and the section enumerating the five challenges: the central claim that high-dimensional architectures constitute a 'new paradigm' for energy-efficient computing is in tension with the explicit listing of electro-optic conversion efficiency as a remaining challenge. High-dimensional encoding (spatial modes, wavelength multiplexing) occurs after the E/O conversion step; the manuscript does not cite or derive a concrete mechanism by which additional optical degrees of freedom reduce the energy cost or loss of that conversion itself. If conversion remains the dominant bottleneck, the system-level efficiency gain required to justify the paradigm-shift language is not yet secured by the dimensionality argument alone.

    Authors: We appreciate the referee's identification of this tension. The manuscript explicitly lists electro-optic conversion efficiency among the five persistent challenges and does not claim or derive a mechanism by which spatial modes or wavelength multiplexing directly lower conversion losses, as these operations occur after the E/O interface. Our use of 'new paradigm' refers to the shift from single-degree-of-freedom (phase/amplitude) matrix-vector multiplication to multi-dimensional optical processing that increases throughput and parallelism with moderate hardware overhead, as supported by the reviewed system-level techniques such as time-wavelength interleaving and hardware-aware training. These can yield net system efficiency gains even when conversion remains a separate bottleneck. To eliminate ambiguity, we will revise the abstract and the challenges discussion to distinguish the optical-domain benefits of dimensionality from the unresolved conversion challenge, replacing 'new paradigm' with more precise language that frames high-dimensional approaches as enabling scalable information processing beyond incremental low-dimensional improvements. revision: yes

Circularity Check

0 steps flagged

No significant circularity: review paper with no internal derivations

full rationale

This manuscript is a review article surveying low- and high-dimensional photonic computing approaches drawn from external literature. It presents no original equations, derivations, fitted parameters, or predictions that could reduce to the paper's own inputs by construction. The central argument for high-dimensional encoding as a new paradigm is framed as a synthesis of reviewed device techniques and system-level methods, while explicitly listing five open challenges (including electro-optic conversion efficiency) without claiming they are resolved via dimensionality in a self-referential way. All load-bearing claims rest on cited prior work rather than internal self-citation chains or ansatzes.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The review rests on standard domain assumptions about photonic device behavior and the limitations of CMOS; it introduces optical skyrmions as an emerging concept without independent verification in the text.

axioms (2)
  • domain assumption CMOS electronics face inherent limits in bandwidth, energy efficiency, and parallelism.
    Opening motivation stated in the abstract.
  • domain assumption High-dimensional architectures achieve higher throughput with moderate hardware overhead.
    Presented as a central advantage of exploiting spatial modes and wavelength channels.
invented entities (1)
  • optical skyrmions no independent evidence
    purpose: Fault-tolerant, topologically protected encoding that exploits the polarization degree of freedom.
    Highlighted as a promising route in the final section, but no specific data, derivation, or falsifiable prediction is supplied in the abstract.

pith-pipeline@v0.9.0 · 5943 in / 1287 out tokens · 47745 ms · 2026-05-20T21:19:04.624658+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

142 extracted references · 142 canonical work pages · 2 internal anchors

  1. [1]

    Introduction The rapid advancement of artificial intelligence (AI), driven by ever -larger models and data -intensive workloads, is placing unprecedented demands on computing infrastructure. Graphics processing unit (GPU) -based systems, still fundamentally reliant on complementary metal-oxide-semiconductor (CMOS) electronics, are struggling to deliver su...

    work page

  2. [2]

    Fundamental properties of light The optical properties of light most directly exploited in photonic computing include phase, amplitude, polarisation, spatial mode structure, and spectral content, which together form the basis for optical information encoding and computation. Depending on whether optical phase relationships are preserved and accessible, th...

    work page

  3. [3]

    Implementation of photonic computing This section outlines how the optical degrees of freedom introduced in Section 2 are harnessed for practical computing, bridging physical principles with functional architectures at both the device and system levels. Following the framework established in Section 1, we examine low -dimensional computing approaches in S...

    work page

  4. [4]

    Challenges and perspectives Enhancing the computational capabilities of optical chips hinges on scaling matrix sizes and improving parallelism 80,106 . Larger matrices help amortise the fixed energy cost of electro -optic conversions, while high-dimensional techniques further boost throughput (see Supplementary Information for detailed analysis). Scaling ...

    work page

  5. [5]

    These constraints have motivated the search for alternative computing paradigms capable of sustaining next-generation AI workloads

    Conclusion The rapid growth of AI has intensified demands on computing infrastructure, yet conventional CMOS electronics face fundamental limits in bandwidth, energy efficiency, and parallelism. These constraints have motivated the search for alternative computing paradigms capable of sustaining next-generation AI workloads. This review has surveyed the f...

    work page

  6. [6]

    Ning, S. et al. Photonic-electronic integrated circuits for high-performance computing and AI accelerators. J. Light. Technol. 42, 7834–7859 (2024)

    work page 2024

  7. [7]

    Moore’s Law forever? Science 299, 210–211 (2003)

    Lundstrom, M. Moore’s Law forever? Science 299, 210–211 (2003)

    work page 2003

  8. [8]

    Baker, R. J. CMOS Circuit Design, Layout, and Simulation. (IEEE Press, Piscataway, NJ, 2019)

    work page 2019

  9. [9]

    Radamson, H. H. et al. State of the art and future perspectives in advanced CMOS technology. Nanomaterials 10, 1555 (2020)

    work page 2020

  10. [10]

    Horowitz, M. et al. Scaling, power, and the future of CMOS. in IEEE International Electron Devices Meeting,

    work page

  11. [11]

    9–15 (IEEE, Tempe, Arizona, USA, 2005)

    IEDM Technical Digest. 9–15 (IEEE, Tempe, Arizona, USA, 2005). doi:10.1109/IEDM.2005.1609253

    work page doi:10.1109/iedm.2005.1609253 2005

  12. [12]

    Zhao, Y., Gobbi, M., Hueso, L. E. & Samorì , P. Molecular approach to engineer two-dimensional devices for CMOS and beyond-CMOS applications. Chem. Rev. 122, 50–131 (2022)

    work page 2022

  13. [13]

    A., Dyuzhev, N

    Bespalov, V. A., Dyuzhev, N. A. & Kireev, V. Yu. Possibilities and limitations of CMOS technology for the production of various microelectronic systems and devices. Nanobiotechnology Rep. 17, 24–38 (2022)

    work page 2022

  14. [14]

    H., Lee, C

    Krambeck, R. H., Lee, C. M. & Law, H.-F. S. High-speed compact circuits with CMOS. IEEE J. Solid-State Circuits 17, 614–619 (1982)

    work page 1982

  15. [15]

    Hill, I., Chanawala, P., Singh, R., Sheikholeslam, S. A. & Ivanov, A. CMOS reliability from past to future: a survey of requirements, trends, and prediction methods. IEEE Trans. Device Mater. Reliab. 22, 1–18 (2022)

    work page 2022

  16. [16]

    Teng, Q. et al. Reliability challenges in CMOS technology: a manufacturing process perspective. Microelectron. Eng. 281, 112086 (2023)

    work page 2023

  17. [17]

    Leiserson, C. E. et al. There’s plenty of room at the top: What will drive computer performance after Moore’s Law? Science 368, eaam9744 (2020)

    work page 2020

  18. [18]

    The future of computing beyond Moore’s Law

    Shalf, J. The future of computing beyond Moore’s Law. Philos. Trans. R. Soc. Math. Phys. Eng. Sci. 378, 20190061 (2020)

    work page 2020

  19. [19]

    Lattner, C. et al. MLIR: a compiler infrastructure for the end of Moore’s Law. Preprint at https://doi.org/10.48550/arXiv.2002.11054 (2020)

    work page doi:10.48550/arxiv.2002.11054 2002

  20. [20]

    & Netzer, G

    Johnsson, L. & Netzer, G. The impact of Moore’s Law and loss of Dennard scaling: Are DSP SoCs an energy efficient alternative to x86 SoCs? J. Phys. Conf. Ser. 762, 012022 (2016)

    work page 2016

  21. [21]

    & Shen, Y

    Huang, Y., Guo, B. & Shen, Y. GPU energy consumption optimization with a global-based neural network method. IEEE Access 7, 64303–64314 (2019)

    work page 2019

  22. [22]

    & Vetter, J

    Mittal, S. & Vetter, J. S. A survey of methods for analyzing and improving GPU energy efficiency. ACM Comput. Surv. 47, 1–23 (2015)

    work page 2015

  23. [23]

    & Chowdhury, M

    You, J., Chung, J.-W. & Chowdhury, M. Zeus: Understanding and optimizing GPU energy consumption of DNN training. in 20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23) 119–139 (USENIX Association, Boston, MA, 2023)

    work page 2023

  24. [24]

    Patterson, D. et al. Carbon emissions and large neural network training. Preprint at https://doi.org/10.48550/arXiv.2104.10350 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2104.10350 2021

  25. [25]

    Brown, T. B. et al. Language models are few-shot learners. Preprint at https://doi.org/10.48550/arXiv.2005.14165 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2005.14165 2005

  26. [26]

    & Gupta, N

    Jot Singh, J., Dhawan, D. & Gupta, N. All-optical photonic crystal logic gates for optical computing: an extensive review. Opt. Eng. 59, 110901 (2020)

    work page 2020

  27. [27]

    G., Liu, Y

    Kibebe, C. G., Liu, Y. & Tang, J. Harnessing optical advantages in computing: a review of current and future trends. Front. Phys. 12, 1379051 (2024)

    work page 2024

  28. [28]

    McMahon, P. L. The physics of optical computing. Nat. Rev. Phys. 5, 717–734 (2023)

    work page 2023

  29. [29]

    & Fefer, Y

    Rysin, A., Livshits, P., Sofer, S. & Fefer, Y. Impact of increased resistive losses of metal interconnects upon ULSI devices reliability and functionality. Microelectron. Eng. 92, 119–122 (2012)

    work page 2012

  30. [30]

    Al Aziz, M. et al. An overview of achieving energy efficiency in on-chip networks. Int. J. Commun. Netw. Distrib. Syst. 5, 444–458 (2010)

    work page 2010

  31. [31]

    Burger, D., Goodman, J. R. & Kägi, A. Memory bandwidth limitations of future microprocessors. in Proceedings of the 23rd annual international symposium on Computer architecture 78–89 (ACM, Philadelphia Pennsylvania USA, 1996). doi:10.1145/232973.232983

    work page doi:10.1145/232973.232983 1996

  32. [32]

    Zou, P. et al. Optical computing accelerators: Principle, application, and perspective. Front. Phys. 20, 032302 (2025)

    work page 2025

  33. [33]

    Li, R. et al. Photonics for neuromorphic computing: Fundamentals, devices, and opportunities. Adv. Mater. 37, 2312825 (2025)

    work page 2025

  34. [34]

    & Dong, X

    Li, C., Zhang, X., Li, J., Fang, T. & Dong, X. The challenges of modern computing and new opportunities for optics. PhotoniX 2, 20 (2021)

    work page 2021

  35. [35]

    Pintus, P. et al. Integrated non-reciprocal magneto-optics with ultra-high endurance for photonic in-memory computing. Nat. Photonics 19, 54–62 (2025)

    work page 2025

  36. [36]

    Feldmann, J. et al. Parallel convolutional processing using an integrated photonic tensor core. Nature 589, 52–58 (2021)

    work page 2021

  37. [37]

    Aghaee Rad, H. et al. Scaling and networking a modular photonic quantum computer. Nature 638, 912–919 (2025)

    work page 2025

  38. [38]

    Feng, C. et al. Wavelength-division-multiplexing-based electronic-photonic network for high-speed computing (conference presentation). in Smart Photonic and Optoelectronic Integrated Circuits XXII (eds He, S. & Vivien, L.) 15 (SPIE, San Francisco, United States, 2020). doi:10.1117/12.2551323

    work page doi:10.1117/12.2551323 2020

  39. [39]

    Sotirova, A. S. et al. Low cross-talk optical addressing of trapped-ion qubits using a novel integrated photonic chip. Light Sci. Appl. 13, 199 (2024)

    work page 2024

  40. [40]

    Zuo, Y. et al. All-optical neural network with nonlinear activation functions. Optica 6, 1132 (2019)

    work page 2019

  41. [41]

    Sui, X., Wu, Q., Liu, J., Chen, Q. & Gu, G. A review of optical neural networks. IEEE Access 8, 70773–70783 (2020)

    work page 2020

  42. [42]

    Zhang, Y. et al. Memory-less scattering imaging with ultrafast convolutional optical neural networks. Sci. Adv. 10, eadn2205 (2024)

    work page 2024

  43. [43]

    Li, G. H. Y. et al. All-optical ultrafast ReLU function for energy-efficient nanophotonic deep learning. Nanophotonics 12, 847–855 (2023)

    work page 2023

  44. [44]

    Ho, R., Mai, K. W. & Horowitz, M. A. The future of wires. Proc. IEEE 89, 490–504 (2001)

    work page 2001

  45. [45]

    Corcoran, B. et al. Ultra-dense optical data transmission over standard fibre with a single chip source. Nat. Commun. 11, 2568 (2020)

    work page 2020

  46. [46]

    & Lončar, M

    Desiatov, B., Shams-Ansari, A., Zhang, M., Wang, C. & Lončar, M. Ultra-low-loss integrated visible photonics using thin-film lithium niobate. Optica 6, 380 (2019)

    work page 2019

  47. [47]

    Miller, D. A. B. Attojoule optoelectronics for low-energy information processing and communications. J. Light. Technol. 35, 346–396 (2017)

    work page 2017

  48. [48]

    & Chen, Q

    Zuo, C. & Chen, Q. Exploiting optical degrees of freedom for information multiplexing in diffractive neural networks. Light Sci. Appl. 11, 208 (2022)

    work page 2022

  49. [49]

    & Ozcan, A

    Li, J., Hung, Y.-C., Kulce, O., Mengu, D. & Ozcan, A. Polarization multiplexed diffractive computing: all-optical implementation of a group of linear transformations through a polarization-encoded diffractive network. Light Sci. Appl. 11, 153 (2022)

    work page 2022

  50. [50]

    Xiang, S. et al. A review: Photonics devices, architectures, and algorithms for optical neural computing. J. Semicond. 42, 023105 (2021)

    work page 2021

  51. [51]

    Wetzstein, G. et al. Inference in artificial intelligence with deep optics and photonics. Nature 588, 39–47 (2020)

    work page 2020

  52. [52]

    Bogaerts, W. et al. Programmable photonic circuits. Nature 586, 207–216 (2020)

    work page 2020

  53. [53]

    Du, Y. et al. Implementation of optical neural network based on Mach–Zehnder interferometer array. IET Optoelectron. 17, 1–11 (2023)

    work page 2023

  54. [54]

    Shen, Y. et al. Deep learning with coherent nanophotonic circuits. Nat. Photonics 11, 441–446 (2017)

    work page 2017

  55. [55]

    Harris, N. C. et al. Linear programmable nanophotonic processors. Optica 5, 1623 (2018)

    work page 2018

  56. [56]

    Miller, D. A. B. Self-configuring universal linear optical component [invited]. Photon. Res. 1, 1 (2013)

    work page 2013

  57. [57]

    Miller, D. A. B. Self-aligning universal beam coupler. Opt. Express 21, 6360 (2013)

    work page 2013

  58. [58]

    & Liboiron-Ladouceur, O

    Shokraneh, F., Geoffroy-Gagnon, S. & Liboiron-Ladouceur, O. The diamond mesh, a phase-error- and loss- tolerant field-programmable MZI-based optical processor for optical neural networks. Opt. Express 28, 23495 (2020)

    work page 2020

  59. [59]

    Zhou, H. et al. Photonic matrix multiplication lights up photonic accelerator and beyond. Light Sci. Appl. 11, 30 (2022)

    work page 2022

  60. [60]

    Hua, S. et al. An integrated large-scale photonic accelerator with ultralow latency. Nature 640, 361–367 (2025)

    work page 2025

  61. [61]

    Lin, Z. et al. 120 GOPS photonic tensor core in thin-film lithium niobate for inference and in situ training. Nat. Commun. 15, 9081 (2024)

    work page 2024

  62. [62]

    Ahmed, S. R. et al. Universal photonic artificial intelligence acceleration. Nature 640, 368–374 (2025)

    work page 2025

  63. [63]

    & Dong, J

    Cheng, J., Zhou, H. & Dong, J. Photonic matrix computing: From fundamentals to applications. Nanomaterials 11, 1683 (2021)

    work page 2021

  64. [64]

    Guo, X., Xiang, J., Zhang, Y. & Su, Y. Integrated neuromorphic photonics: Synapses, neurons, and neural networks. Adv. Photonics Res. 2, 2000212 (2021)

    work page 2021

  65. [65]

    Tian, Y. et al. Scalable and compact photonic neural chip with low learning-capability-loss. Nanophotonics 11, 329–344 (2022)

    work page 2022

  66. [66]

    Zhu, H. H. et al. Space-efficient optical computing with an integrated chip diffractive neural network. Nat. Commun. 13, 1044 (2022)

    work page 2022

  67. [67]

    Fu, T. et al. Photonic machine learning with on-chip diffractive optics. Nat. Commun. 14, 70 (2023)

    work page 2023

  68. [68]

    Cheng, J. et al. Multimodal deep learning using on-chip diffractive optics with in situ training capability. Nat. Commun. 15, 6189 (2024)

    work page 2024

  69. [69]

    & Feng, L

    Wu, T., Menarini, M., Gao, Z. & Feng, L. Lithography-free reconfigurable integrated photonic processor. Nat. Photonics 17, 710–716 (2023)

    work page 2023

  70. [70]

    Programmable diffractive deep neural networks enabled by integrated rewritable metasurfaces

    Zarei, S. Programmable diffractive deep neural networks enabled by integrated rewritable metasurfaces. Sci. Rep. 15, 35624 (2025)

    work page 2025

  71. [71]

    Xu, Z. et al. Large-scale photonic chiplet taichi empowers 160-TOPS/W artificial general intelligence. Science 384, 202–209 (2024)

    work page 2024

  72. [72]

    E., Chu, S

    Little, B. E., Chu, S. T., Haus, H. A., Foresi, J. & Laine, J.-P. Microring resonator channel dropping filters. J. Light. Technol. 15, 998–1005 (1997)

    work page 1997

  73. [73]

    Universal relations for coupling of optical power between microresonators and dielectric waveguides

    Yariv, A. Universal relations for coupling of optical power between microresonators and dielectric waveguides. Electron. Lett. 36, 321–322 (2000)

    work page 2000

  74. [74]

    Bogaerts, W. et al. Silicon microring resonators. Laser Photonics Rev. 6, 47–73 (2012)

    work page 2012

  75. [75]

    Tait, A. N. et al. Microring weight banks. IEEE J. Sel. Top. Quantum Electron. 22, 312–325 (2016)

    work page 2016

  76. [76]

    Tait, A. N. et al. Neuromorphic photonic networks using silicon photonic weight banks. Sci. Rep. 7, 7430 (2017)

    work page 2017

  77. [77]

    N., Nahmias, M

    Tait, A. N., Nahmias, M. A., Shastri, B. J. & Prucnal, P. R. Broadcast and weight: An integrated network for scalable photonic spike processing. J. Light. Technol. 32, 4029–4041 (2014)

    work page 2014

  78. [78]

    & Takenaka, M

    Ohno, S., Tang, R., Toprasertpong, K., Takagi, S. & Takenaka, M. Si microring resonator crossbar array for on- chip inference and training of the optical neural network. ACS Photonics 9, 2614–2622 (2022)

    work page 2022

  79. [79]

    Shao, L. et al. Integrated microwave acousto-optic frequency shifter on thin-film lithium niobate. Opt. Express 28, 23728 (2020)

    work page 2020

  80. [80]

    Tang, R. et al. Symmetric silicon microring resonator optical crossbar array for accelerated inference and training in deep learning. https://doi.org/10.48550/ARXIV.2401.16072 (2024) doi:10.48550/ARXIV.2401.16072

    work page doi:10.48550/arxiv.2401.16072 2024

Showing first 80 references.