Pith. sign in

REVIEW 3 major objections 4 minor 31 references

Optoelectronic recurrent neural network using optical-electrical-optical converters with RC delay

T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that the RC delay of OEO converters can be tuned to keep the effective recurrent matrix near-unitary, restoring trainability of lossy optoelectronic RNNs up to $32\times32$ scale.

desk verdict A clean but incremental modeling paper: RC delay in OEO converters acts as a leaky RNN term, giving a useful design rule for photonic recurrent hardware, but the math is known and the simulations validate only the model's own assumptions. read the letter →

arxiv 2411.16186 v1 pith:7KVKK6DF submitted 2024-11-25 physics.optics

Masaya Arahata , Shota Kita , Kazuo Aoyama , Akihiko Shinya , Hiroshi Sawada , Masaya Notomi This is my paper · ORCID
classification physics.optics
keywords opticalneuralnetworksrecurrentoptoelectronicOEOconvertersRCdelayloop-losscompensationgradientvanishingandexplosionunitarymatricessequentialimageclassification
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

Optical recurrent networks lose information as light circulates through a lossy loop, and integrating amplifiers to restore the signal is hard. This paper argues that the resistive-capacitive (RC) delay of an optical–electrical–optical (OEO) converter—usually a speed limitation—can instead provide a built-in memory that compensates for the loop loss. With the converter modeled as $z_t = \beta y_t + \alpha z_{t-1}$, where $\alpha = \exp(-\Delta t/\tau_{\mathrm{RC}})$, the loop's effective matrix is $S = \beta W + \alpha I$; the paper derives the band $-\alpha+1 \le \beta \le \alpha+1$ within which the eigenvalues of $S$ stay near magnitude one. Simulated time-series classification of MNIST and Fashion-MNIST on $4\times4$ through $32\times32$ circuits shows higher accuracy inside this band and degraded accuracy outside it. The paper's practical message is that choosing the RC constant is a design knob that improves trainability of lossy OE-RNNs rather than a parasitic effect to be eliminated.

What carries the argument

The object that carries the argument is the discrete-time RC memory $\alpha = \exp(-\Delta t/\tau_{\mathrm{RC}})$ entering the recurrence $z_t = \beta y_t + \alpha z_{t-1}$. This single exponential memory converts the lossy unitary loop $\beta W$ into the effective matrix $S = \beta W + \alpha I$; the diagonal term $\alpha I$ is the mathematical embodiment of the RC delay. The load-bearing manipulation is the eigenvalue shift $\lambda_h \mapsto \beta \lambda_h + \alpha$, which replaces exact unitarity with an annulus bound $|\alpha-\beta| \le |\lambda_s| \le \alpha+\beta$ and yields the design inequality Eq. 7. The same machinery also handles finite pulse widths, where a rise-time factor $\gamma$ appears in $z(t) = (1-\gamma)\beta x(t) + \alpha z(t-1)$, and it extends to real-valued orthonormal weight banks unchanged.

What would settle it

Measure the impulse response of a fabricated OEO converter and check whether it is a single exponential with the assumed $\tau_{\mathrm{RC}}$; if it shows two distinct decay rates or a power-dependent decay, the model $z_t = \beta y_t + \alpha z_{t-1}$ fails. Alternatively, on a chip with independently tunable $\beta$ and $\alpha$, classify a time-series task at points inside and outside the band and check that test accuracy peaks inside the band—a flat accuracy map would falsify the compensation claim.

Watch

Extended reading notes

Core claim

The central claim is that the RC delay of an OEO converter can compensate an optical recurrent loop's loss or excess gain. Treating the converter as a first-order linear system gives $z_t = \beta y_t + \alpha z_{t-1}$, and the full recurrence becomes $h_t = \beta x_t + (\beta W + \alpha I) h_{t-1}$. Because $W$ is unitary, each eigenvalue $\lambda_h = e^{i\theta}$ of $W$ is shifted to $\beta e^{i\theta} + \alpha$, so the magnitudes of the effective eigenvalues lie between $|\alpha-\beta|$ and $\alpha+\beta$. The paper shows that the necessary condition for all effective eigenvalues to have magnitude one is $-\alpha+1 \le \beta \le \alpha+1$, and it identifies this band as the region where gradient explosion and vanishing are relaxed. Sequential MNIST and Fashion-MNIST simulations on $4\times4$, $8\times8$, $16\times16$, and $32\times32$ OE-RNNs confirm that test accuracy is high inside the band and degraded outside it.

Load-bearing premise

The analysis assumes that the OEO converter's response to an input pulse is a single-exponential RC decay, summarized by one number $\alpha$; if a real converter has multiple time constants, nonlinear amplitude response, or phase distortion, the eigenvalue band in Eq. 7 would not be the right predictor of trainability.

Editorial extensions

If this is right

  • For a lossy loop ($\beta < 1$), increasing the RC memory $\alpha$ restores accuracy that would otherwise be lost to gradient vanishing; the paper's 4×4 simulations show accuracy rising with $\alpha$ at $\beta = 0.6$ and $\beta = 0.2$.
  • For a gainy loop ($\beta > 1$), the same inequality $-\alpha+1 \le \beta \le \alpha+1$ marks where gradient explosion is avoided, so both loss and gain can be compensated by the same diagonal memory term.
  • Inside the predicted band, larger circuits (8×8, 16×16, 32×32) achieve high test accuracy on both MNIST and Fashion-MNIST; outside the band accuracy drops, confirming the band as a trainability design rule.
  • The analysis extends to real-valued intensity-based optical weight banks, where the same inequality holds with orthonormal matrices whose eigenvalues are $\pm 1$.
  • Because larger RC constants are generally easier to fabricate, the compensation effect means designers can intentionally relax RC specifications without degrading RNN performance; finite-pulse rise-time loss can be recovered by increasing $\beta$.

Reading between the lines

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

  • Editorial inference: the same $\alpha I$ diagonal shift could be used as a physical regularization principle for training recurrent models in other noisy analog hardware; any platform with a tunable decaying self-feedback can add a diagonal memory to a lossy weight matrix to keep effective eigenvalues near the unit circle.
  • Editorial inference: a two-pole OEO model (separate detector and modulator time constants) would replace the single $\alpha$ with two memory constants; the compensation band would shift, but the qualitative mechanism—diagonal memory balancing radial eigenvalue shrinkage—should survive, so testing Eq. 7 against a two-pole model would show how robust the design rule is.
  • Editorial inference: the paper's simulation uses compressed FFT features as sequential inputs; a direct experimental check could feed real high-rate time-series data through a chip with tunable bias and measure the accuracy map over $(\alpha, \beta)$, comparing it to the Eq. 7 band.
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

3 major / 4 minor

Summary. The paper studies an optoelectronic recurrent neural network (OE-RNN) in which each OEO converter is modeled as a first-order linear recurrence, Eq. (2): z_t = beta y_t + alpha z_{t-1}, with beta the loop gain and alpha = exp(-Delta t / tau_RC) the RC delayed feedback. The authors derive the effective recurrent matrix S = beta W + alpha I for a unitary hidden matrix W, bound its eigenvalues by |alpha - beta| <= |lambda_s| <= alpha + beta, and obtain the necessary condition Eq. (7), -alpha + 1 <= beta <= alpha + 1, for |lambda_s| = 1. They argue that adding RC delay can compensate for loop loss or gain. Simulations of sequential MNIST and Fashion-MNIST classification with Clements-circuit hidden units from 4x4 to 32x32 show test accuracy increasing in the region defined by Eq. (7). Section IV discusses device feasibility and introduces a finite-pulse-response extension in Eq. (8).

Significance. If the claims hold, the paper identifies a simple and useful design rule: RC delay, normally considered an impairment, can be chosen so that its diagonal contribution alpha I restores the effective spectral radius of beta W + alpha I toward unity, mitigating gradient vanishing and explosion in optical RNNs. The analytical derivation is transparent and contains no free parameters, the design inequality is easy to apply, and the simulations cover nontrivial circuit sizes (up to 32x32) and two datasets. The significance is moderated by the fact that both the analysis and the simulations use the same idealized linear single-pole model; the paper's own finite-pulse discussion (Eq. 8) is not incorporated into either the theory or the simulations, and Eq. (7) is only a necessary condition.

major comments (3)
  1. [Section II, Eq. (7)] The condition -alpha + 1 <= beta <= alpha + 1 is derived as a necessary condition for |lambda_s| = 1, but the text immediately above Eq. (7) states that 'the unitary condition can be recovered by appropriately adding the diagonal term of RC delay.' This overstates the result: satisfying Eq. (7) does not imply all eigenvalues of S have unit magnitude. For example, with beta = 0.6 and alpha = 0.4, an eigenvalue direction lambda_h = -1 gives |lambda_s| = 0.2, so the gradient can still vanish for that component. The paper later uses the softer wording 'can be relaxed,' and the simulation evidence is consistent with a relaxation effect, but the central theoretical claim should be reworded and the distinction between necessary and sufficient conditions should be made explicit.
  2. [Section II, Eq. (1) and Section IV, Eq. (8)] The compensation effect rests entirely on the linear recurrence Eq. (2), which assumes a small-angle MZM linearization, impulse-response excitation (t_p << tau_RC), and a single exponential memory. Section IV itself introduces the finite-pulse-response variant Eq. (8), z(t) = (1 - gamma) beta x(t) + alpha z(t-1), and notes that the rise time acts as an additional loss that must be compensated by increasing beta, but this variant is not used in the gradient analysis or in the simulations. If gamma is non-negligible, the effective loop gain changes and Eq. (7) is no longer the correct boundary; if the MZM is driven outside the small-angle region, the nonlinearity makes the spectral-radius argument inapplicable. The authors should either include simulations of Eq. (8) and of a nonlinear MZM transfer function, or state the parameter range over which Eq. (7) remains a valid predictor of trainability.
  3. [Section III] The agreement between the theoretical analysis and the simulations is expected by construction: the simulations implement the same recurrence Eq. (2) used to derive Eq. (7). This is internal consistency, not an independent test of the model. The paper would be substantially stronger if at least one experiment or simulation used a different or more realistic forward model, such as the finite-pulse model of Eq. (8), a measured OEO impulse response, or a nonlinear MZM characteristic, to show that the compensation effect survives beyond the idealized model used in both the analysis and the main simulations.
minor comments (4)
  1. [Fig. 3] The caption of Fig. 3(b) says 'test accuracy' while the text in Section III.A describes the same panel as 'training accuracy'; please make the labels consistent.
  2. [Section IV.A] There is a typo in 'with the large road resistance R'; it should read 'load resistance.'
  3. [Abstract and Section III.A] The phrase 'degrade RNN performance' should be 'degraded RNN performance' in the abstract and in the discussion of Fig. 3.
  4. [Section III] The simulation section does not report hyperparameters (learning rate, number of epochs, optimization algorithm, number of runs) or run-to-run variability; adding these details would support the quantitative heatmap comparisons in Fig. 3(f) and Fig. 5.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Eq. (7) is derived from the stated model, simulations check the same model, and self-citations are not load-bearing.

full rationale

The paper's central result, Eq. (7), is a direct algebraic consequence of its explicitly stated model z_t = β y_t + α z_{t−1} (Eq. 2) and the recurrence h_t = β x_t + (βW + αI)h_{t−1} (Eq. 3). For unitary W, the eigenvalues of S = βW + αI are βλ_h + α, so |λ_s| lies between |α − β| and α + β, and requiring |λ_s| = 1 yields −α + 1 ≤ β ≤ α + 1. No parameter is fitted to simulation data, and the simulations implement the same model, so the agreement is an internal-consistency check rather than a circular reduction of a prediction to its input. The self-citations (e.g., Refs. [18], [19], [21], with overlapping authors) support OEO device feasibility and measured conversion efficiency; they are not load-bearing for the eigenvalue derivation. The paper explicitly assumes a linear-region MZM and short-pulse impulse response, and Section IV introduces a finite-pulse-response variant (Eq. 8) not incorporated into the theory; these are modeling limitations and correctness risks, not circular steps. No step of the derivation is equivalent by construction to its own output.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the linear model in Eq. 2 and the unitary hidden unit. No external data benchmarks are used; the simulations evaluate the same model from which the theory is derived.

free parameters (3)
  • beta (loop gain) = scanned from 0 to 1 in steps of 0.1 or 0.2 in simulations
    Controls the effective recurrent matrix S = beta W + alpha I and is the target of the compensation condition. Not fitted, but scanned by hand.
  • alpha (RC delayed feedback) = scanned from 0 to 1 in steps of 0.1 or 0.2
    Represents exp(-Delta t / tau_RC); the central variable of the paper. Scanned, not fitted.
  • branch power ratio = 0.96:0.04 loop to output
    Chosen to reduce additional loop losses; affects the actual loop gain.
assumptions (6)
  • domain assumption OEO converter is a first-order linear time-invariant system described by z_t = beta y_t + alpha z_{t-1} (Eq. 2)
    Invoked in Section II to derive the recurrence and the eigenvalue condition. Real OEO converters may have more complex dynamics.
  • domain assumption MZM is operated in the linear region, so sin(theta) is approximated by theta in Eq. 1
    Used to linearize the OEO transfer function. Nonlinearity is mentioned as an advantage but not included in the model.
  • domain assumption Hidden unit W is a unitary matrix (Clements circuit), so |lambda_h| = 1
    Required for the eigenvalue range calculation in Eq. 6 and the condition Eq. 7.
  • domain assumption The recurrent loop has no activation function
    Stated in Section II: 'an output unit and activation function are not included'. This makes the network linear and the eigenvalue analysis exact, but unrealistic for deep nonlinear RNNs.
  • domain assumption Input pulses are short compared to the RC time constant (t_p << tau_RC)
    Used to justify the impulse response and the discrete-time recurrence Eq. 2. Finite pulse width is discussed in Section IV but not in the main analysis.
  • ad hoc to paper Gradient stability is determined by the necessary condition Eq. 7 on eigenvalue magnitudes
    The paper uses this necessary condition as a predictor of the high-accuracy region in simulations, without proving sufficiency or accounting for input and output weights and loss function.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optoelectronic recurrent neural network using optical-electrical-optical converters with RC delay." pith.science (2026). https://pith.science/paper/7KVKK6DF

@misc{pith2026241116186,
  author       = {Pith},
  title        = {Pith review of: Optoelectronic recurrent neural network using optical-electrical-optical converters with RC delay},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7KVKK6DF}},
  note         = {Machine review of arXiv:2411.16186}
}
abstract

Optical neural network (ONN) has been attracting intense attention owing to their low latency and low-power consumption. Among the ONNs, optical recurrent neural network (RNN) enables low-power and high-speed time-series data processing using a compact loop structure. The loop losses need to be efficiently compensated so that the time-series information is maintained in the RNN operation. For this purpose, we focus on the optoelectronic RNN (OE-RNN) with optical-electrical-optical (OEO) converters to compensate for the loop losses. However, the effect of resistive-capacitive (RC) delay of OEO converters on the RNN performance is unclear. Here, we study in simulation an OE-RNN equipped with OEO converters with RC delay. We confirm that our modeled OE-RNN achieves the high training accuracy of time-series data classification even when RC delay is comparably large to the time interval of time-series data. Our analyses reveal that the accumulation of time-series data by RC delay does not degrade the RNN performance but rather can compensate for the degraded RNN performance due to loop losses. From the theoretical analysis referring to the gradient explosion and vanishing problems, we find the region related to loss and RC delay where the high training accuracy can be achieved. In simulation, we confirm this compensation effect in the large OE-RNN circuit up to 32$\times$32 scale. Our proposed scheme opens a new way of time-series data processing by utilizing RC delay for the optical computing and optical communication.

Figures

Figures reproduced from arXiv: 2411.16186 by the authors.

Figure 1
Figure 1. (a) Schematic diagram of Elman-type OE-RNN with the R [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Schematic diagram of Elman-type 4 × 4 OE-RNN circuit. The input, hidden, and branch units are composed of MZIs with 50/50 BSs and PSs. After FFT, time-series input signals for each MNIST image are generated from its selected FFT signals. The relation between the input and output of OEO unit is expressed by Eq. 2 at time t and t − 1 with the loop gain β and RC delayed feedback α. The output signal of branch unit is d… view at source ↗
Figure 3
Figure 3. Simulation results of Elman-type 4 × 4 OE-RNN circuit. (a) Loop gain β dependence of training accuracy with the RC delayed feedback α = 0. (b) RC delayed feedback dependence of test accuracy with each loop gain. RC delayed feedback dependence of training accuracy with the loop gain (c) β = 1.0, (d) β = 0.6, and (e) β = 0.2, respectively. (f) RC delayed feedback α and loop gain β dependence of test accuracy, where th… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Schematic diagram of Elman-type N × N OE-RNN circuit. The one dimensional time-series input data is distributed into the N port by using the branch structure in the input unit. The OEO unit has the loop gain β and RC delayed feedback α. The output of hidden unit compos…
Figure 5
Figure 5. Figure 5: RC delayed feedback α and loop gain β dependence of test accuracy of sequential (a-c) MNIST and (d-f) Fashion-MNIST task with l = 20 of (a,d) 8 × 8, (b,e) 16 × 16, and (c,f) 32 × 32 OE-RNN circuit, respectively. The area surrounded by red lines corresponds to Eq. 7. Th…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 27 canonical work pages

  1. [1]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, pp. 436–444, May 2015

    work page 2015

  2. [2]

    Restoring and attribut- ing ancient texts using deep neural networks,

    Y . Assael, T. Sommerschield, B. Shillingford, M. Bor- dbar, J. Pavlopoulos, M. Chatzipanagiotou, I. Androut- sopoulos, J. Prag, and N. Freitas, “Restoring and attribut- ing ancient texts using deep neural networks,” Nature, vol. 603, pp. 280–283, March 2022

    work page 2022

  3. [3]

    Effi- cient processing of deep neural networks: a tutorial and survey,

    V . Sze, Y .-H. Chen, T.-J. Y ang, and J. S. Emer, “Effi- cient processing of deep neural networks: a tutorial and survey,” Proceedings of the IEEE , vol. 105, no. 12, pp. 2295–2329, 2017

    work page 2017

  4. [4]

    Carbon emissions and large neural network training,

    D. Patterson, J. Gonzalez, Q. Le, C. Liang, L.-M. Munguia, D. Rothchild, D. So, M. Texier, and J. Dean, “Carbon emissions and large neural network training,” arXiv preprint , p. arXiv:2104.10350, 2021

    arXiv 2021

  5. [5]

    Deep learning with coherent nanophotonic circuits,

    Y . Shen, N. C. Harris, S. Skirlo, M. Prabhu, T. B.-Jones, M. Hochberg, X. Sun, S. Zhao, H. Larochelle, D. En- glund, and M. Soljaˇ ci´ c, “Deep learning with coherent nanophotonic circuits,” Nature Photonics , vol. 11, pp. 441–446, June 2017

    work page 2017

  6. [6]

    Neuromorphic photonic net- works using silicon photonic weight banks,

    A. N. Tait, T. Lima, E. Zhou, A. Wu, M. Nahmias, B. Shastri, and P . Prucnal, “Neuromorphic photonic net- works using silicon photonic weight banks,” Scientific Reports, vol. 7, no. 7430, August 2017

    work page 2017

  7. [7]

    Paral- lel convolutional processing using an integrated photonic tensor core,

    J. Feldmann, N. Y oungblood, M. Karpov, H. Gehring, X. Li, M. Stappers, M. L. Gallo, X. Fu, A. Lukashchuk, A. S. Raja, J. Liu, C. D. Wright, A. Sebastian, T. J. Kippenberg, W . H. P . Pernice, and H. Bhaskaran, “Paral- lel convolutional processing using an integrated photonic tensor core,” Nature, vol. 589, pp. 52–58, January 2021

    work page 2021

  8. [8]

    16 ×16 MZI-based photonic accelerator toward space and wavelength division multiplexing,

    K. Ikeda, S. Kita, M. Nakajima, K. Takata, K. Aoyama, K. Suzuki, Y . Maegami, M. Ohno, G. Gong, N. Y amamoto, K. Y amada, A. Shinya, H. Sawada, 8 TABLE I THE COMPARISON OF DEVICE PARAMETERS OF OEO CONVERTERS . Vπ C R τRC tp Ix Bandwidth MZM 5 V 1 pF 5 k Ω 5 ns 100 ps 50 mW ∼ 100 MHz Slow-light MZM [30] 3-5 V 100 fF 5 k Ω 500 ps 10 ps 30-50 mW ∼ GHz MRR-EO...

    work page 2024

Show all 31 references
  1. [9]

    Scaling on-chip photonic neural processors using arbitrarily programmable wave propa- gation,

    T. Onodera, M. M. Stein, B. A. Ash, M. M. Sohoni, M. Bosch, R. Y anagimoto, M. Jankowski, T. P . McKenna, T. Wang, G. Shvets, M. R. Shcherbakov, L. G. Wright, and P . L. McMahon, “Scaling on-chip photonic neural processors using arbitrarily programmable wave propa- gation,” ar...

    2024

  2. [10]

    11 tops photonic convolutional accelerator for optical neural networks,

    X. Xu, M. Tan, B. Corcoran, J. Wu, A. Boes, T. G. Nguyen, S. T. Chu, B. E. Little, D. G. Hicks, R. Moran- dotti, A. Mitchell, and D. J. Moss, “11 tops photonic convolutional accelerator for optical neural networks,” Nature, vol. 589, pp. 44–51, January 2021

    2021

  3. [11]

    Neuromorphic computing based on wavelength-division multiplexing,

    X. Xu, W . Han, M. Tan, Y . Sun, Y . Li, J. Wu, R. Moran- dotti, A. Mitchell, K. Xu, and D. J. Moss, “Neuromorphic computing based on wavelength-division multiplexing,” IEEE J. Sel. T op. Quantum Electron., vol. 29, no. 2, April 2023

    2023

  4. [12]

    Reinforcement learning in a large-scale photonic recurrent neural network,

    J. Bueno, S. Maktoobi, L. Froehly, I. Fischer, M. Jacquo t, L. Larger, and D. Brunner, “Reinforcement learning in a large-scale photonic recurrent neural network,” Optica, vol. 5, no. 6, pp. 756–760, June 2018

    2018

  5. [13]

    Higher- dimensional processing using a photonic tensor core with continuous-time data,

    B. Dong, S. Aggarwal, W . Zhou, U. E. Ali, N. Far- makidis, J. S. Lee, Y . He, X. Li, D.-L. Kwong, C. D. Wright, W . H. P . Pernice, and H. Bhaskaran, “Higher- dimensional processing using a photonic tensor core with continuous-time data,” Nature Photonics , vol. 17, pp. 1080–...

    2023

  6. [14]

    All-Optical WDM Recurrent Neural Networks With Gating,

    G. M.-Alexandris, G. Dabos, N. Passalis, A. Totovi´ c, A. Tefas, and N. Pleros, “All-Optical WDM Recurrent Neural Networks With Gating,” IEEE Journal of Selected T opics in Quantum Electronics , vol. 26, no. 5, October 2020

    2020

  7. [15]

    Compact Design of on-Chip Elman Optical Recurrent Neural Network,

    C. Feng, Z. Zhao, Z. Ying, J. Gu, D. Z. Pan, and R. T. Chen, “Compact Design of on-Chip Elman Optical Recurrent Neural Network,” 2020 Conference on Lasers and Electro-Optics (CLEO) , no. JTh2B.8, May 2020

    2020

  8. [16]

    I/O-efficient iterative matrix inversion with photonic integrated circuits,

    M. Chen, Y . Wang, C. Y ao, A. Wonfor, S. Y ang, R. Penty, and Q. Cheng, “I/O-efficient iterative matrix inversion with photonic integrated circuits,” Nature Communica- tions, vol. 15, no. 5926, July 2024

    2024

  9. [17]

    Chip-scale all-optical complex-valued matrix inverter,

    X. Liu, J. Cheng, H. Zhou, J. Dong, and X. Zhang, “Chip-scale all-optical complex-valued matrix inverter, ” APL Photonics , vol. 9, no. 056106, May 2024

    2024

  10. [18]

    Photonic- crystal nano-photodetector with ultrasmall capacitance for on-chip light-to-voltage conversion without an am- plifier,

    K. Nozaki, S. Matsuo, T. Fujii, K. Takeda, M. Ono, A. Shakoor, E. Kuramochi, and M. Notomi, “Photonic- crystal nano-photodetector with ultrasmall capacitance for on-chip light-to-voltage conversion without an am- plifier,” Optica, vol. 3, no. 5, May 2016

    2016

  11. [19]

    Forward-biased nanophotonic detector for ultralow-energy dissipation receiver,

    K. Nozaki, S. Matsuo, T. Fujii, K. Takeda, A. Shinya, E. Kuramochi, and M. Notomi, “Forward-biased nanophotonic detector for ultralow-energy dissipation receiver,” APL Photonics, vol. 3, no. 046101, April 2018

    2018

  12. [20]

    Single chip photonic deep neural network with accelerated training,

    S. Bandyopadhyay, A. Sludds, S. Krastanov, R. Hamerly, N. Harris, D. Bunandar, M. Streshinsky, M. Hochberg, and D. Englund, “Single chip photonic deep neural network with accelerated training,” arXiv preprint , p. arXiv:2208.01623, August 2022

    2022 arXiv

  13. [21]

    Femtofarad optoelec- tronic integration demonstrating energy-saving signal conversion and nonlinear functions,

    K. Nozaki, S. Matsuo, T. Fujii, K. Takeda, A. Shinya, E. Kuramochi, and M. Notomi, “Femtofarad optoelec- tronic integration demonstrating energy-saving signal conversion and nonlinear functions,” Nature Photonics , vol. 13, pp. 454–459, April 2019

    2019

  14. [22]

    Review of Silicon Photonics Technology and Platform Development,

    S. Y . Siew, B. Li, F. Gao, H. Y . Zheng, W . Zhang, P . Guo, S. W . Xie, A. Song, B. Dong, L. W . Luo, C. Li, X. Luo, and G.-Q. Lo, “Review of Silicon Photonics Technology and Platform Development,” J. Light. T echnol., vol. 39, no. 13, July 2021

    2021

  15. [23]

    Optimal design for universal multiport interferometers,

    W . R. Clements, P . C. Humphreys, B. J. Metcalf, W . S. Kolthammer, and I. A. Walmsley, “Optimal design for universal multiport interferometers,” Optica, vol. 3, no. 12, pp. 1460–1465, December 2016

    2016

  16. [24]

    Unitary Evolution Recurrent Neural Networks,

    M. Arjovsky, A. Shah, and Y . Bengio, “Unitary Evolution Recurrent Neural Networks,” arXiv, no. 1511.06464v4, May 2016

    2016 arXiv

  17. [25]

    Full-Capacity Unitary Recurrent Neural Networks,

    S. Wisdom, T. Powers, J. R. Hershey, J. L. Roux, and L. Atlas, “Full-Capacity Unitary Recurrent Neural Networks,” arXiv, no. 1611.00035v1, October 2016

    2016 arXiv

  18. [26]

    Tunable Efficient Unitary Neural Networks (EUNN) and their application to RNN,

    L. Jing, Y . Shen, T. Dubˇ cek, J. Peurifoy, S. Skirlo, M. Tegmark, and M. Soljaˇ ci´ c, “Tunable Efficient Unitary Neural Networks (EUNN) and their application to RNN,” arXiv, no. 1612.05231v1, December 2016

    2016 arXiv

  19. [27]

    Broadcast and weight: An Integrated Network For Scalable Pho- tonic Spike Processing,

    A. N. Tait, M. A. Nahmias, and B. J. Shastri, “Broadcast and weight: An Integrated Network For Scalable Pho- tonic Spike Processing,” IEEE J. Light. T echnol., vol. 32, no. 21, November 2014

    2014

  20. [28]

    Efficient low-loss InGaAsP/Si hybrid MOS optical modulator,

    G. Sinatkas, T. Christopoulos, O. Tsilipakos, and E. E. Kriezis, “Efficient low-loss InGaAsP/Si hybrid MOS optical modulator,” J. Appl. Phys. , vol. 130, no. 010901, July 2021

    2021

  21. [29]

    Efficient low-loss InGaAsP/Si hybrid MOS optical modulator,

    J.-H. Han, F. Boeuf, J. Fujikata, S. Takahashi, S. Takag i, and M. Takenaka, “Efficient low-loss InGaAsP/Si hybrid MOS optical modulator,” Nature Photonics , vol. 11, pp. 486–490, July 2017

    2017

  22. [30]

    Slow-light 9 mach–zehnder modulators based on si photonic crystals,

    T. Baba, H. C. Nguyen, N. Y azawa, Y . Ter- ada, S. Hashimoto, and T. Watanabe, “Slow-light 9 mach–zehnder modulators based on si photonic crystals,” Sci. T echnol. Adv. Mater . , vol. 15, no. 024602, April 2014

    2014

  23. [31]

    A 100 gb/s pam4 two-segment silicon microring resonator modulator using a standard foundry process,

    Y . Y uan, W . V . Sorin, Z. Huang, X. Zeng, D. Liang, A. Kumar, S. Palermo, M. Fiorentino, and R. G. Beau- soleil, “A 100 gb/s pam4 two-segment silicon microring resonator modulator using a standard foundry process,” ACS Photonics , vol. 9, pp. 1165–1171, March 2022

    2022

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.