Pith. sign in

REVIEW 4 major objections 5 minor 53 references

Fully analog end-to-end online training with real-time adaptibility on integrated photonic platform

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The paper demonstrates a silicon photonic chip that trains a neural network end-to-end in real time, using optically measured gradients so that all learning happens in the analog domain.

desk verdict Real integrated MGD training on silicon photonics, but 'fully analog' is contradicted by the paper's own Methods; worth refereeing for the hardware, not for the abstraction. read the letter →

arxiv 2506.18041 v1 pith:AXEZS476 submitted 2025-06-22 physics.optics

classification physics.optics
keywords analogtrainingonlinelearningphotonicneuromorphicprocessingmicroringresonatormultiplexedgradientdescentin-situadaptiveintegratedphotonics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that a silicon photonic chip can train a neural network fully in the analog domain, in real time, without digital matrix multiplications. The chip uses multiplexed gradient descent: small perturbations are added to microring-resonator weights, and the change in loss is read directly from on-chip balanced photodetectors, so gradients are measured optically. The authors report over 90% accuracy on linear classification, over 80% on a quadratic (nonlinear) task, and adaptive re-training that tracks a rotating decision boundary and recovers from temperature and crosstalk disturbances. If correct, this would be a step toward neuromorphic processors that learn continuously at gigabaud rates while staying robust to hardware drift.

What carries the argument

The load-bearing object is the multiplexed gradient descent (MGD) algorithm, a perturbative scheme in which each weight is jittered by a random binary sequence and the instantaneous loss variation is multiplied by that sequence to estimate the gradient without backpropagation. On the hardware side, the machinery is the monolithically integrated photonic processor: microring resonator (MRR) weight banks tuned by in-resonator photoconductive heaters, balanced photodetectors that sum weighted optical signals on-chip, and microring modulator neurons that supply the nonlinear activation. The MGD update rule lets the chip update weights using only locally measured optical signals rather than digital matrix multiplications.

What would settle it

Trace the training data path on the experimental setup: if every weight update requires the photodetector outputs to be digitized by the RFSoC ADC and the loss to be computed on the FPGA, then the 'fully analog' claim is falsified. A direct test would be to disable the FPGA's loss computation and show that no learning occurs, or to show that training continues with the ADC removed.

Watch

Extended reading notes

Core claim

The central discovery claimed is that multiplexed gradient descent turns the photonic processor itself into a trainer: because the gradient of the loss with respect to each weight is proportional to the product of the weight perturbation and the resulting loss variation, the chip can measure that product at the output photodetectors and update weights locally. The integrated device combines a wavelength-multiplexed microring weight bank, balanced photodetectors, and modulator neurons on one foundry-fabricated chip. Experimentally, the authors report linear classification accuracy above 90%, nonlinear (quadratic) classification above 80%, online tracking of thirteen rotated decision boundaries with recovery in under ten epochs, and stable operation under a ±2 °C temperature change and sustained thermal crosstalk. A simulation of a 4-30-3 network on the Yin-Yang dataset reaches 97.9% accuracy.

Load-bearing premise

The demonstration depends on the assumption that the loss and gradient can be measured and applied entirely in the analog domain, with no digital computation inside the training loop.

Editorial extensions

If this is right

  • Training no longer requires a separate digital computer for gradient computation, which removes a major latency and bandwidth bottleneck in photonic neural networks.
  • Because updates are local and perturbative, the same hardware can keep learning while doing inference, enabling continuous adaptation to changing data and drift.
  • The demonstrated recovery times (about 20 seconds after a task change) and the projected speedup with electro-optic tuning suggest online learning at rates useful for real-time signal processing.
  • The method is hardware-agnostic, so the same algorithm should transfer to memristors, analog CMOS, and spintronic platforms.
  • The 97.9% accuracy in the simulated 4-30-3 network indicates the approach scales beyond two-neuron demonstrations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's 'fully analog' claim is stronger than what the Methods section shows: the loss is computed on the RFSoC FPGA after the photodetector outputs are digitized, and the quadratic nonlinearity is programmed digitally, so a strict reading is that the demonstration is 'analog forward pass, digitally assisted training' until the loss function itself is computed by analog circuitry.
  • A natural test of the approach is to lower the ADC resolution and see where training accuracy collapses; this would quantify how much digital precision is actually required and whether the analog gradient estimate is the limiting factor.
  • The online-tracking experiment is essentially a continual-learning benchmark; connecting it to established continual-learning metrics such as forgetting and plasticity could sharpen what 'adaptivity' means here.
  • The authors note that PN-junction electro-optic tuning could cut convergence times by a factor of 1,000; if realized, this would open applications in high-speed adaptive filtering and control.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript reports a silicon photonic integrated circuit with a microring-resonator weight bank and balanced photodetectors, trained online by multiplexed gradient descent. The authors claim fully analog end-to-end training, demonstrating linear and nonlinear classification, online tracking of a rotating decision boundary, and real-time adaptation to temperature and crosstalk. A simulation of a 4-30-3 network on the Yin-Yang dataset is presented to support scalability.

Significance. The experimental system is a useful testbed: it is a foundry-manufactured chip with integrated MRR weights and balanced photodetectors, and the demonstrations that a perturbative weight-update rule can track a slowly rotating decision boundary and recover from temperature and crosstalk perturbations are concrete and potentially valuable. However, the headline claims of fully analog, end-to-end, on-chip training and on-chip nonlinearity are not supported by the Methods, so the significance of the work as presented is substantially lower than claimed. The paper would need to be reframed as a hybrid electronic-photonic system with an FPGA in the training loop, and even then the novelty relative to prior computer-in-the-loop perturbative training should be assessed.

major comments (4)
  1. [§2 and §6.2] The central claim that the entire training pipeline remains in the analog domain is contradicted by the experimental description. Section 6.2 states that the RFSoC FPGA modulates inputs and measures outputs, that the baseline loss is first measured, then after perturbing the weights the loss is measured again, and the difference is used as the gradient; this requires digital storage and subtraction of the sampled loss values. Section 2 also states that the weights are controlled by SMUs that communicate with the RFSoC FPGA, so weight updates are not generated or applied locally on the photonic chip. The abstract's 'fully analog end-to-end training' and the statement in §2 that 'the entire training pipeline remains within the analog domain' are therefore unsupported.
  2. [§3.1] The nonlinear classification experiment does not demonstrate a photonic nonlinearity. The text says the quadratic nonlinearity was 'programmed on the RFSoC FPGA,' and Fig. 2(a)'s vision of microring-modulator-based nonlinear activation is not the implemented setup. Consequently, the paper does not show end-to-end analog training with an on-chip nonlinearity; it shows a linear optical weighting operation followed by a digital nonlinearity.
  3. [§4, Fig. 6] The scalability simulation uses a ReLU nonlinearity that 'can be realized by appropriately biasing the microring modulator neurons' (Ref. [48]) but was not experimentally validated in this paper. Because the only experimental nonlinearity was digital, the simulation does not demonstrate that the fabricated hardware scales to the 4-30-3 network. The comparison with a PyTorch/ADAM baseline does not compensate for this missing experimental link.
  4. [§2, §3.2, §4] The speed claims are not supported by the reported data. The DAC runs at 1 GBaud with 8 samples per symbol, which corresponds to a 125 Mbaud symbol rate rather than 1 GBaud; the training updates in the online-tracking demonstration occur on a timescale of hundreds of seconds with a recovery time of about 20 s; and the 16 GBaud figure used in the energy-efficiency estimate in §4 is the photodetector bandwidth, not the demonstrated operational speed. These quantitative claims should be reconciled with the experimental parameters.
minor comments (5)
  1. [Title and Abstract] There are several typos, including 'adaptibility' in the title, 'anolog' in the abstract, and 'in-resonantor' in §6.1; these should be corrected.
  2. [§3.2 and §5] The conclusion reports 86.5% accuracy for online tracking, while §3.2 reports '>85% within a few epochs' and angle-specific best accuracies; the definition of the single 86.5% figure should be clarified.
  3. [§3.3] The 'rate of equivalent weight change' formula lacks a clear definition of t_T and the units of the reported values; please define the quantity precisely and state how it was measured.
  4. [Fig. 4 caption] The caption says 'three different angles' while the text describes thirteen shuffled angles ranging from 15° to 255°; the caption should be made consistent with the text and with the subplot.
  5. [§6.2] The statement that 'the difference between the perturbed and unperturbed loss function corresponds to the gradient' is an oversimplification of the MGD update rule; a precise equation with sign and normalization would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the experimental results are measured from the fabricated photonic hardware, and the MGD citation is independent support rather than a definitional loop.

full rationale

Walking the claimed derivation chain: the paper's central experimental claim is in-situ training of an MRR weight bank using the multiplexed gradient descent (MGD) algorithm. The MGD update rule is credited to prior work [38], but the paper does not rederive it from the target result, and the reported classification accuracies are direct measurements from the fabricated PIC under test, not values produced by fitting a parameter to the target and then renaming it a prediction. The nonlinear classification task uses a quadratic nonlinearity programmed on the RFSoC FPGA; this is a described experimental implementation choice, not a reduction of the claimed outcome to the task definition. The Yin-Yang scalability simulation is presented with its own Algorithm 1 and benchmarked against a PyTorch-ADAM network, so it is self-contained rather than an imported result. Citations to co-authored prior work [21,47,48] are contextual or supporting; none is invoked to forbid alternatives or as the sole justification of a central claim. The paper's own Methods contradict the headline 'fully analog' claim by describing ADC sampling, RFSoC FPGA-based loss readout, and FPGA-programmed nonlinearity, but that is a correctness/overclaim discrepancy, not a circular reduction: the gradient is not defined as the analog measurement, and no equation in the paper equates the output to the input by construction. Therefore no circular step is found.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central experimental claims depend on the MGD algorithm's behavior (from prior self-cited work), the assumed linearity of the photonic measurement chain, and the assumption that simulated nonlinear activations map onto the hardware. No new physical entities are introduced.

free parameters (3)
  • perturbation amplitude (Δw)
    The MGD algorithm requires choosing the amplitude of the random weight perturbations; the value is not given in the main text and is expected to affect convergence and final accuracy.
  • learning rate
    The weight update step size is not reported in the main text; it is a hand-chosen hyperparameter that determines convergence speed and stability.
  • bias-voltage ramp step = 25 mV/epoch
    Chosen by the authors for the thermal crosstalk experiment; not a fitted parameter but a controlled quantity.
assumptions (4)
  • domain assumption The MGD update rule converges for the tested tasks; convergence properties are assumed from the cited prior work [38,47] rather than re-derived.
    Invoked throughout Section 3; the paper does not analyze convergence conditions for the photonic implementation.
  • domain assumption The photodetector current is a linear weighted sum of the optical inputs, with negligible crosstalk apart from the thermal effects studied.
    Invoked in Section 2 and Methods 6.1; the incoherent WDM architecture avoids phase interference but presumes linearity of the BPD summation.
  • domain assumption The voltage-to-optical-weight mapping of the IRPH heaters is approximately linear over the [0,2] V range.
    Stated in Section 3.3; used to justify the interpretation of temperature changes as equivalent weight changes.
  • ad hoc to paper A ReLU nonlinearity can be realized by biasing microring modulator neurons as in Ref. [48]; this is assumed in the Yin-Yang simulation and not experimentally validated.
    Used in the Section 4 simulation to claim scalability of the photonic approach.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fully analog end-to-end online training with real-time adaptibility on integrated photonic platform." pith.science (2026). https://pith.science/paper/AXEZS476

@misc{pith2026250618041,
  author       = {Pith},
  title        = {Pith review of: Fully analog end-to-end online training with real-time adaptibility on integrated photonic platform},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AXEZS476}},
  note         = {Machine review of arXiv:2506.18041}
}
read the original abstract

Analog neuromorphic photonic processors are uniquely positioned to harness the ultrafast bandwidth and inherent parallelism of light, enabling scalability, on-chip integration and significant improvement in computational performance. However, major challenges remain unresolved especially in achieving real-time online training, efficient end-to-end anolog systems, and adaptive learning for dynamical environmental changes. Here, we demonstrate an on-chip photonic analog end-to-end adaptive learning system realized on a foundry-manufactured silicon photonic integrated circuit. Our platform leverages a multiplexed gradient descent algorithm to perform in-situ, on-the-fly training, while maintaining robustness in online tracking and real-time adaptation. At its core, the processor features a monolithic integration of a microring resonator weight bank array and on-chip photodetectors, enabling direct optical measurement of gradient signals. This eliminates the need for high-precision digital matrix multiplications, significantly reducing computational overhead and latency, an essential requirement for effective online training. We experimentally demonstrate real-time, end-to-end analog training for both linear and nonlinear classification tasks at gigabaud rates, achieving accuracies of over 90\% and 80\%, respectively. Our analog neuromorphic processor introduces self-learning capabilities that dynamically adjust training parameters, setting the stage for truly autonomous neuromorphic architectures capable of efficient, real-time processing in unpredictable real-world environments. As a result, we showcase adaptive online tracking of dynamically changing input datasets and achieve over 90\% accuracy, alongside robustness to external temperature fluctuations and internal thermal crosstalk.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 33 canonical work pages

  1. [38]

    APL Machine Learning1(2), 026118 (2023) https://doi.org/10.1063/5.0157645 https://pubs.aip.org/aip/aml/article- pdf/doi/10.1063/5.0157645/18017061/026118 1 5.0157645.pdf

    McCaughan, A.N., Oripov, B.G., Ganesh, N., Nam, S.W., Dienstfrey, A., Buck- ley, S.M.: Multiplexed gradient descent: Fast online training of modern datasets on hardware neural networks without backpropagation. APL Machine Learning1(2), 026118 (2023) https://doi.org/10.1063/5.0157645 https://pubs.aip.org/aip/aml/article- pdf/doi/10.1063/5.0157645/18017061/...

  2. [48]

    Tait, A.N., Lima, T., Nahmias, M.A., Miller, H.B., Peng, H.-T., Shastri, B.J., Prucnal, P.R.: Sil- icon photonic modulator neuron. Phys. Rev. Appl.11, 064043 (2019) https://doi.org/10.1103/ PhysRevApplied.11.064043

  3. [1]

    Nature Precedings (2010) https://doi.org/10

    Bywater, R.: Solving the protein folding problems. Nature Precedings (2010) https://doi.org/10. 1038/npre.2010.4730.1

  4. [2]

    Nature634(8032), 61–68 (2024) https://doi.org/10.1038/s41586-024-07930-y

    Zhou, L., Schellaert, W., Mart ´ ınez-Plumed, F., Moros-Daval, Y., Ferri, C., Hern´ andez-Orallo, J.: Larger and more instructable language models become less reliable. Nature634(8032), 61–68 (2024) https://doi.org/10.1038/s41586-024-07930-y

  5. [3]

    Nature623(7985), 48–57 (2023) https://doi.org/10.1038/s41586-023-06558-8

    Chen, Y., Nazhamaiti, M., Xu, H., Meng, Y., Zhou, T., Li, G., Fan, J., Wei, Q., Wu, J., Qiao, F., Fang, L., Dai, Q.: All-analog photoelectronic chip for high-speed vision tasks. Nature623(7985), 48–57 (2023) https://doi.org/10.1038/s41586-023-06558-8

  6. [4]

    Review of Scientific Instruments 65(6), 1803–1832 (1994) https://doi.org/10.1063/1.1144830 https://pubs.aip.org/aip/rsi/article- pdf/65/6/1803/19141715/1803 1 online.pdf

    Bishop, C.M.: Neural networks and their applications. Review of Scientific Instruments 65(6), 1803–1832 (1994) https://doi.org/10.1063/1.1144830 https://pubs.aip.org/aip/rsi/article- pdf/65/6/1803/19141715/1803 1 online.pdf

  7. [5]

    Nature521(7553), 436–444 (2015) https://doi

    LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. Nature521(7553), 436–444 (2015) https://doi. org/10.1038/nature14539

  8. [6]

    In: 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, pp

    Bengio, Y., Boulanger-Lewandowski, N., Pascanu, R.: Advances in optimizing recurrent networks. In: 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 8624–8628 (2013). https://doi.org/10.1109/ICASSP.2013.6639349

Show all 53 references
  1. [7]

    Krogh, A.: What are artificial neural networks? Nature Biotechnology26(2), 195–197 (2008) https: //doi.org/10.1038/nbt1386

  2. [8]

    Frontiers in Computational Neuroscience11(2017) https://doi.org/10.3389/ fncom.2017.00024

    Scellier, B., Bengio, Y.: Equilibrium propagation: Bridging the gap between energy-based models and backpropagation. Frontiers in Computational Neuroscience11(2017) https://doi.org/10.3389/ fncom.2017.00024

  3. [9]

    Optica11(8), 1039–1049 (2024) https://doi.org/10.1364/OPTICA.523225

    Xu, T., Zhang, W., Zhang, J., Luo, Z., Xiao, Q., Wang, B., Luo, M., Xu, X., Shastri, B.J., Prucnal, P.R., Huang, C.: Control-free and efficient integrated photonic neural networks via hardware-aware training and pruning. Optica11(8), 1039–1049 (2024) https://doi.org/10.1364/OP...

  4. [10]

    IEEE transactions on pattern analysis and machine intelligencePP(2024) https://doi.org/10.1109/TPAMI.2024.3466853

    Zhao, G., Shu, X., Zhou, R.: High-performance real-world optical computing trained by in situ gradient-based model-free optimization. IEEE transactions on pattern analysis and machine intelligencePP(2024) https://doi.org/10.1109/TPAMI.2024.3466853

  5. [11]

    Nature Communications16(1), 421 (2025) https: //doi.org/10.1038/s41467-024-55508-z

    Lin, H., Ou, J., Fan, Z., Yan, X., Hu, W., Cui, B., Xu, J., Li, W., Chen, Z., Yang, B., Liu, K., Mo, L., Li, M., Lu, X., Zhou, G., Gao, X., Liu, J.-M.: In situ training of an in-sensor artificial neural network based on ferroelectric photosensors. Nature Communications16(1), 4...

  6. [12]

    Nature601(7894), 549–555 (2022) https: //doi.org/10.1038/s41586-021-04223-6

    Wright, L.G., Onodera, T., Stein, M.M., Wang, T., Schachter, D.T., Hu, Z., McMahon, P.L.: Deep physical neural networks trained with backpropagation. Nature601(7894), 549–555 (2022) https: //doi.org/10.1038/s41586-021-04223-6

  7. [13]

    Light: Science & Applications13(1), 263 (2024) https://doi.org/10.1038/ s41377-024-01590-3

    Fu, T., Zhang, J., Sun, R., Huang, Y., Xu, W., Yang, S., Zhu, Z., Chen, H.: Optical neural networks: progress and challenges. Light: Science & Applications13(1), 263 (2024) https://doi.org/10.1038/ s41377-024-01590-3

  8. [14]

    Nature588(7836), 39–47 (2020) https://doi.org/10.1038/s41586-020-2973-6

    Wetzstein, G., Ozcan, A., Gigan, S., Fan, S., Englund, D., Soljaˇ ci´ c, M., Denz, C., Miller, D.A.B., Psaltis, D.: Inference in artificial intelligence with deep optics and photonics. Nature588(7836), 39–47 (2020) https://doi.org/10.1038/s41586-020-2973-6

  9. [15]

    Optica9(12), 1323–1332 (2022) https://doi.org/10.1364/OPTICA.475493

    Filipovich, M.J., Guo, Z., Al-Qadasi, M., Marquez, B.A., Morison, H.D., Sorger, V.J., Prucnal, P.R., Shekhar, S., Shastri, B.J.: Silicon photonic architecture for training deep neural networks with direct feedback alignment. Optica9(12), 1323–1332 (2022) https://doi.org/10.136...

  10. [16]

    Nature Electronics 4(2), 151–161 (2021) https://doi.org/10.1038/s41928-020-00523-3

    Dalgaty, T., Castellani, N., Turck, C., Harabi, K.-E., Querlioz, D., Vianello, E.: In situ learning using intrinsic memristor variability via markov chain monte carlo sampling. Nature Electronics 4(2), 151–161 (2021) https://doi.org/10.1038/s41928-020-00523-3

  11. [17]

    Nature Communications15(1), 9081 (2024) https://doi.org/10.1038/s41467-024-53261-x

    Lin, Z., Shastri, B.J., Yu, S., Song, J., Zhu, Y., Safarnejadian, A., Cai, W., Lin, Y., Ke, W., Ham- mood, M., Wang, T., Xu, M., Zheng, Z., Al-Qadasi, M., Esmaeeli, O., Rahim, M., Pakulski, G., 10 Schmid, J., Barrios, P., Jiang, W., Morison, H., Mitchell, M., Guan, X., Jaeger,...

  12. [18]

    Nature632(8026), 768–774 (2024) https://doi.org/10.1038/ s41586-024-07711-7

    Dohare, S., Hernandez-Garcia, J.F., Lan, Q., Rahman, P., Mahmood, A.R., Sutton, R.S.: Loss of plasticity in deep continual learning. Nature632(8026), 768–774 (2024) https://doi.org/10.1038/ s41586-024-07711-7

  13. [19]

    Nature Reviews Neuroscience21(6), 335–346 (2020) https://doi.org/10.1038/s41583-020-0277-3

    Lillicrap, T.P., Santoro, A., Marris, L., Akerman, C.J., Hinton, G.: Backpropagation and the brain. Nature Reviews Neuroscience21(6), 335–346 (2020) https://doi.org/10.1038/s41583-020-0277-3

  14. [20]

    https://arxiv.org/abs/2406.03372

    Momeni, A., Rahmani, B., Scellier, B., Wright, L.G., McMahon, P.L., Wanjura, C.C., Li, Y., Skalli, A., Berloff, N.G., Onodera, T., Oguz, I., Morichetti, F., Hougne, P., Gallo, M.L., Sebastian, A., Mirhoseini, A., Zhang, C., Markovi´ c, D., Brunner, D., Moser, C., Gigan, S., Ma...

  15. [21]

    Nanophotonics12(5), 833–845 (2023) https://doi.org/10.1515/nanoph-2022-0553

    Buckley, S.M., Tait, A.N., McCaughan, A.N., Shastri, B.J.: Photonic online learning: a perspective. Nanophotonics12(5), 833–845 (2023) https://doi.org/10.1515/nanoph-2022-0553

  16. [22]

    https: //arxiv.org/abs/2409.12965

    Wang, Z., M¨ uller, K., Filipovich, M., Launay, J., Ohana, R., Pariente, G., Mokaadi, S., Brossollet, C., Moreau, F., Cappelli, A., Poli, I., Carron, I., Daudet, L., Krzakala, F., Gigan, S.: Optical training of large-scale Transformers and deep neural networks with direct feed...

  17. [23]

    Science Advances9(42), 9127 (2023) https://doi.org/10.1126/sciadv

    Br¨ uckerhoff-Pl¨ uckelmann, F., Bente, I., Becker, M., Vollmar, N., Farmakidis, N., Lomonte, E., Lenzini, F., Wright, C.D., Bhaskaran, H., Salinga, M., Risse, B., Pernice, W.H.P.: Event-driven adap- tive optical neural network. Science Advances9(42), 9127 (2023) https://doi.o...

  18. [24]

    Nature Photonics13(2), 80–90 (2019) https://doi.org/10.1038/s41566-018-0310-5

    Marpaung, D., Yao, J., Capmany, J.: Integrated microwave photonics. Nature Photonics13(2), 80–90 (2019) https://doi.org/10.1038/s41566-018-0310-5

  19. [25]

    Nature Photonics15(2), 102–114 (2021) https://doi.org/10.1038/s41566-020-00754-y

    Shastri, B.J., Tait, A.N., Lima, T., Pernice, W.H.P., Bhaskaran, H., Wright, C.D., Prucnal, P.R.: Photonics for artificial intelligence and neuromorphic computing. Nature Photonics15(2), 102–114 (2021) https://doi.org/10.1038/s41566-020-00754-y

  20. [26]

    Nature (London) 632(8024), 264–265 (2024)

    Querlioz, D.: Physics solves a training problem for artificial neural networks. Nature (London) 632(8024), 264–265 (2024)

  21. [27]

    Nature Electronics4(11), 837–844 (2021) https://doi.org/10.1038/s41928-021-00661-2

    Huang, C., Fujisawa, S., Lima, T.F., Tait, A.N., Blow, E.C., Tian, Y., Bilodeau, S., Jha, A., Yaman, F., Peng, H.-T., Batshon, H.G., Shastri, B.J., Inada, Y., Wang, T., Prucnal, P.R.: A silicon photonic– electronic neural network for fibre nonlinearity compensation. Nature Ele...

  22. [28]

    Nature Communications14(1), 1107 (2023) https://doi.org/10.1038/ s41467-023-36814-4

    Zhang, W., Tait, A., Huang, C., Lima, T., Bilodeau, S., Blow, E.C., Jha, A., Shastri, B.J., Prucnal, P.: Broadband physical layer cognitive radio with an integrated photonic processor for blind source separation. Nature Communications14(1), 1107 (2023) https://doi.org/10.1038/...

  23. [29]

    Nature Communications14(1), 8197 (2023) https: //doi.org/10.1038/s41467-023-43982-w

    Lederman, J.C., Zhang, W., Lima, T.F., Blow, E.C., Bilodeau, S., Shastri, B.J., Prucnal, P.R.: Real- time photonic blind interference cancellation. Nature Communications14(1), 8197 (2023) https: //doi.org/10.1038/s41467-023-43982-w

  24. [30]

    Light: Science & Applications13(1), 14 (2024) https://doi.org/10

    Zhang, W., Lederman, J.C., Lima, T., Zhang, J., Bilodeau, S., Hudson, L., Tait, A., Shastri, B.J., Prucnal, P.R.: A system-on-chip microwave photonic processor solves dynamic rf interference in real time with picosecond latency. Light: Science & Applications13(1), 14 (2024) ht...

  25. [31]

    Scientific Reports12(1), 5605 (2022) https://doi.org/10.1038/s41598-022-09370-y

    Totovic, A., Giamougiannis, G., Tsakyridis, A., Lazovsky, D., Pleros, N.: Programmable photonic neural networks combining wdm with coherent linear optics. Scientific Reports12(1), 5605 (2022) https://doi.org/10.1038/s41598-022-09370-y

  26. [32]

    Nature589(7840), 44–51 (2021) https://doi.org/10.1038/s41586-020-03063-0

    Xu, X., Tan, M., Corcoran, B., Wu, J., Boes, A., Nguyen, T.G., Chu, S.T., Little, B.E., Hicks, D.G., Morandotti, R., Mitchell, A., Moss, D.J.: 11 tops photonic convolutional accelerator for optical neural networks. Nature589(7840), 44–51 (2021) https://doi.org/10.1038/s41586-0...

  27. [33]

    Nature589(7840), 52–58 (2021) https://doi.org/10.1038/s41586-020-03070-1

    Feldmann, J., Youngblood, N., Karpov, M., Gehring, H., Li, X., Stappers, M., Le Gallo, M., Fu, X., Lukashchuk, A., Raja, A.S., Liu, J., Wright, C.D., Sebastian, A., Kippenberg, T.J., Pernice, W.H.P., Bhaskaran, H.: Parallel convolutional processing using an integrated photonic...

  28. [34]

    Science380(6643), 398–404 (2023) https://doi.org/10.1126/science.ade8450 https://www.science.org/doi/pdf/10.1126/science.ade8450

    Pai, S., Sun, Z., Hughes, T.W., Park, T., Bartlett, B., Williamson, I.A.D., Minkov, M., 11 Milanizadeh, M., Abebe, N., Morichetti, F., Melloni, A., Fan, S., Solgaard, O., Miller, D.A.B.: Experimentally realized in situ backpropagation for deep learning in photonic neural netwo...

  29. [35]

    Nature Machine Intelligence5(10), 1119–1129 (2023) https: //doi.org/10.1038/s42256-023-00723-4

    Zheng, Z., Duan, Z., Chen, H., Yang, R., Gao, S., Zhang, H., Xiong, H., Lin, X.: Dual adaptive training of photonic neural networks. Nature Machine Intelligence5(10), 1119–1129 (2023) https: //doi.org/10.1038/s42256-023-00723-4

  30. [36]

    Nature632(8024), 280–286 (2024) https://doi.org/10.1038/s41586-024-07687-4

    Xue, Z., Zhou, T., Xu, Z., Yu, S., Dai, Q., Fang, L.: Fully forward mode training for optical neural networks. Nature632(8024), 280–286 (2024) https://doi.org/10.1038/s41586-024-07687-4

  31. [37]

    Nature Communications13(1), 5572 (2022) https://doi.org/10.1038/s41467-022-33259-z

    Mourgias-Alexandris, G., Moralis-Pegios, M., Tsakyridis, A., Simos, S., Dabos, G., Totovic, A., Passalis, N., Kirtas, M., Rutirawut, T., Gardes, F.Y., Tefas, A., Pleros, N.: Noise-resilient and high- speed deep learning with coherent silicon photonics. Nature Communications13(...

  32. [39]

    https://arxiv.org/abs/2102

    Kriener, L., G¨ oltz, J., Petrovici, M.A.: The Yin-Yang dataset (2022). https://arxiv.org/abs/2102. 08211

  33. [40]

    IEEE Journal of Selected Topics in Quantum Electronics28(6: High Density Integr

    Guo, Z., Tait, A.N., Marquez, B.A., Filipovich, M., Morison, H., Prucnal, P.R., Chrostowski, L., Shekhar, S., Shastri, B.J.: Multi-level encoding and decoding in a scalable photonic tensor processor with a photonic general matrix multiply (gemm) compiler. IEEE Journal of Selec...

  34. [41]

    Jayatilleka, H., Murray, K., Guill´ en-Torres, M., Caverley, M., Hu, R., Jaeger, N.A.F., Chrostowski, L., Shekhar, S.: Wavelength tuning and stabilization of microring-based filters using silicon in- resonator photoconductive heaters. Opt. Express23(19), 25084–25097 (2015) htt...

  35. [42]

    IEEE Transactions on Neural Networks 1(1), 58–70 (1990) https://doi.org/10.1109/72.80205

    Dembo, A., Kailath, T.: Model-free distributed learning. IEEE Transactions on Neural Networks 1(1), 58–70 (1990) https://doi.org/10.1109/72.80205

  36. [43]

    Nature Chemistry (2024) https://doi.org/10.1038/s41557-024-01670-2

    Yang, C., Li, Y., Zhou, S., Guo, Y., Jia, C., Liu, Z., Houk, K.N., Dubi, Y., Guo, X.: Author correction: Real-time monitoring of reaction stereochemistry through single-molecule observations of chirality- induced spin selectivity. Nature Chemistry (2024) https://doi.org/10.103...

  37. [44]

    Nature 408(6809), 196–199 (2000) https://doi.org/10.1038/35041567

    Blaser, E., Pylyshyn, Z.W., Holcombe, A.O.: Tracking an object through feature space. Nature 408(6809), 196–199 (2000) https://doi.org/10.1038/35041567

  38. [45]

    npj Robotics2(1), 6 (2024) https://doi.org/10.1038/s44182-024-00013-0

    Lale, S., Renn, P.I., Azizzadenesheli, K., Hassibi, B., Gharib, M., Anandkumar, A.: Falcon: Fourier adaptive learning and control for disturbance rejection under extreme turbulence. npj Robotics2(1), 6 (2024) https://doi.org/10.1038/s44182-024-00013-0

  39. [46]

    Optica 9(5), 579–584 (2022) https://doi.org/10.1364/OPTICA.446100

    Zhang, W., Huang, C., Peng, H.-T., Bilodeau, S., Jha, A., Blow, E., Lima, T.F., Shastri, B.J., Prucnal, P.: Silicon microring synapses enable photonic deep learning beyond 9-bit precision. Optica 9(5), 579–584 (2022) https://doi.org/10.1364/OPTICA.446100

  40. [47]

    APL Machine Learning3(2), 026107 (2025) https://doi.org/10.1063/5.0258271 https://pubs.aip.org/aip/aml/article- pdf/doi/10.1063/5.0258271/20492805/026107 1 5.0258271.pdf

    Oripov, B.G., Dienstfrey, A., McCaughan, A.N., Buckley, S.M.: Scaling of hardware-compatible perturbative training algorithms. APL Machine Learning3(2), 026107 (2025) https://doi.org/10.1063/5.0258271 https://pubs.aip.org/aip/aml/article- pdf/doi/10.1063/5.0258271/20492805/026...

  41. [49]

    Optics express28(11), 16845–16856 (2020)

    Hagan, D.E., Ye, M., Wang, P., Cartledge, J.C., Knights, A.P.: High-speed performance of a tdfa- band micro-ring resonator modulator and detector. Optics express28(11), 16845–16856 (2020)

  42. [50]

    Science382(6676), 1297–1303 (2023) https://doi.org/10.1126/ science.adi8474 https://www.science.org/doi/pdf/10.1126/science.adi8474

    Momeni, A., Rahmani, B., Mall´ ejac, M., Hougne, P., Fleury, R.: Backpropagation-free training of deep physical neural networks. Science382(6676), 1297–1303 (2023) https://doi.org/10.1126/ science.adi8474 https://www.science.org/doi/pdf/10.1126/science.adi8474

  43. [51]

    Nature Photonics (2024) https://doi.org/10.1038/s41566-024-01567-z 12

    Bandyopadhyay, S., Sludds, A., Krastanov, S., Hamerly, R., Harris, N., Bunandar, D., Streshin- sky, M., Hochberg, M., Englund, D.: Single-chip photonic deep neural network with forward-only training. Nature Photonics (2024) https://doi.org/10.1038/s41566-024-01567-z 12

  44. [52]

    Nature606(7914), 501–506 (2022) https://doi.org/10.1038/s41586-022-04714-0

    Ashtiani, F., Geers, A.J., Aflatouni, F.: An on-chip photonic deep neural network for image classification. Nature606(7914), 501–506 (2022) https://doi.org/10.1038/s41586-022-04714-0

  45. [53]

    Hai, M.S., Sakib, M.N., Liboiron-Ladouceur, O.: A 16 GHz silicon-based monolithic balanced photodetector with on-chip capacitors for 25 GBaud front-end receivers. Opt. Express21(26), 32680–32689 (2013) https://doi.org/10.1364/OE.21.032680 13

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.