Pith. sign in

REVIEW 3 major objections 4 minor 19 references

Deep Learning-Optimized, Fabrication Error-Tolerant Photonic Crystal Nanobeam Cavities for Scalable On-Chip Diamond Quantum Systems

T0 review · 3 major / 4 minor · reviewed 2026-08-08 · 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 a CNN surrogate trained on FDTD data containing modeled surface roughness and sidewall slant can optimize diamond nanobeam cavities so that Q-factor degradation under fabrication errors drops by up to 52% relative to…

desk verdict Plausible DL workflow for fabrication-tolerant cavity design, but the roughness model's symmetric boundary conditions and the abstract's accuracy claims are bigger than the evidence supports. read the letter →

arxiv 2502.03987 v1 pith:TPH5J54P submitted 2025-02-06 physics.optics physics.app-phquant-ph

Sander van Haagen , Salahuddin Nur , Ryoichi Ishihara This is my paper · ORCID
classification physics.opticsphysics.app-phquant-ph
keywords photoniccrystalnanobeamcavitiesdiamondcolorcentersfabricationerrortolerancedeeplearningsurrogatemodelconvolutionalneuralnetworkQ-factoroptimizationsurfaceroughnesssidewallslant
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

This paper argues that optimizing nanobeam cavities for maximum Q-factor under perfect geometry produces devices that are unnecessarily fragile once real fabrication errors appear, and that a convolutional neural network trained on imperfect geometries can instead find designs whose performance degrades far less. For two diamond nanobeam cavity families—the elliptical-hole L2 cavity and the corrugated fishbone cavity—the authors inject modeled surface roughness ($\pm 2.5$ nm on flat surfaces, $\pm 4.0$ nm on curved surfaces) and a $5^\circ$ sidewall slant into the FDTD training data, then use the trained network as a fast surrogate to optimize the 13 or 16 geometric parameters. The resulting real-world-optimized cavities have lower ideal Q-factors than the ideal-optimized controls, but under the targeted imperfection they keep comparable or higher Q: the L2 slant-optimized cavity degrades $52.42\%$ less and holds $Q \approx 2 \times 10^4$ at $5^\circ$ slant where the ideal-optimized cavity falls to about $9.4 \times 10^3$, and the fishbone slant-optimized cavity reaches $Q \approx 1.08 \times 10^5$ at $5^\circ$ slant. The paper claims this design-based tolerance is a path to scalable diamond quantum photonics, since the CNN predicts Q up to a million times faster than FDTD with test prediction errors as low as $3.99\%$ and correlations up to $0.988$. No device is fabricated; the claim is that this simulation-level promise transfers to real thin-film diamond manufacturing.

What carries the argument

The load-bearing machinery is a convolutional neural network surrogate trained on finite-difference time-domain data that already contains the fabrication imperfection being optimized against. The network takes the 13 (L2) or 16 (fishbone) geometric parameter displacements as a zero-padded input matrix, passes them through one convolutional layer with 50 kernels of size $2 \times 3$ and three fully connected layers (ReLU activations, dropout between the second and third), and predicts $\log_{10}(Q)$; training minimizes mean-squared error plus an $L_2$ weight-decay penalty. Once trained, the network is interrogated by two optimizers—gradient ascent and CMA-ES—each augmented with an $L_2$ penalty that keeps parameter displacements inside the training distribution, and the resulting candidates are confirmed with FDTD. The new step relative to earlier deep-learning nanocavity optimization is inserting surface roughness ($\pm 2.5$ nm flat and $\pm 4.0$ nm curved) and a $5^\circ$ sidewall slant into the dataset-generation stage, so the surrogate learns Q as a function of geometry under realistic error conditions rather than under perfect geometry.

What would settle it

Fabricate at least the L2 ideal-optimized cavity and the L2 slant-optimized cavity in thin-film diamond, measure their Q-factors under the actual process's roughness and sidewall slant, and compare the degradation ratio. If the slant-optimized cavity does not keep a substantially higher Q than the ideal-optimized one at the realized slant angle, or if electron-microscopy characterization shows roughness and slant statistics far outside the modeled $\pm 2.5/\pm 4.0$ nm and $5^\circ$ values, the claimed $52\%$ reduction in Q-factor degradation would not transfer to practice.

Watch

Extended reading notes

Core claim

The central discovery is that a nanophotonic cavity optimized against modeled fabrication errors keeps its Q-factor far better than a cavity optimized for a perfect geometry, even though the tolerant design has a lower Q under ideal conditions. In the paper's comparisons, the four real-world-optimized cavities consistently showed less Q-factor degradation than the two ideal-optimized controls under the imperfection they were trained against: the L2 slant-optimized cavity degrades $52.42\%$ less at $5^\circ$ slant, holding $Q \approx 1.96 \times 10^4$ versus $9.4 \times 10^3$; the fishbone slant-optimized cavity reaches $Q \approx 1.08 \times 10^5$ at $5^\circ$ slant, more than double its ideal-optimized counterpart's $5.1 \times 10^4$; the fishbone roughness-optimized cavity averages $Q \approx 1.65 \times 10^4$ under roughness, $25.72\%$ above the ideal-optimized cavity; and the L2 roughness-optimized cavity holds $Q \approx 5.0 \times 10^4$ with $16.24\%$ less degradation than the control. The paper also reports that the L2 base cavity's electric-field mode is approximately twice as broad in the $x$-direction as conventional nanobeam designs, which relaxes the ion-implantation alignment accuracy needed to couple an emitter to the cavity mode.

Load-bearing premise

The gain rests on the assumption that the modeled imperfections—surface roughness of $\pm 2.5$ nm on flat surfaces, $\pm 4.0$ nm on curved surfaces, and a $5^\circ$ sidewall slant—faithfully represent real thin-film diamond fabrication errors, and that resilience seen in simulation will transfer to fabricated devices; the paper reports no fabrication, and its own conclusion names nanofabrication as the most critical next step.

Editorial extensions

If this is right

  • A real-world-optimized cavity can beat an ideal-optimized cavity under the imperfection it was trained for: at $5^\circ$ sidewall slant the L2 slant-optimized cavity degrades $52.42\%$ less and keeps $Q \approx 1.96 \times 10^4$, while the fishbone slant-optimized cavity reaches $Q \approx 1.08 \times 10^5$, more than twice the ideal-optimized control.
  • Fabrication tolerance can be engineered at the design stage rather than only at the fabrication-process stage, because the CNN evaluates candidate geometries up to a million times faster than FDTD and makes 13- and 16-parameter searches practical.
  • Choosing a base cavity with a deliberately broad mode—the L2 cavity's mode is about twice as wide in $x$ as conventional designs—relaxes the ion-implantation alignment requirement, improving the chance of emitter-cavity coupling without adding fabrication steps.
  • The train-on-imperfections loop is presented as system-agnostic: the same methodology can be applied to other nanophotonic structures beyond diamond nanobeam cavities.
  • Realistic-condition Q-factors projected by the method are around $5 \times 10^4$ for the L2 roughness-tolerant cavities, substantially narrowing the documented order-of-magnitude gap between simulated and experimental Q-factors for diamond cavities.

Reading between the lines

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

  • The paper leaves implicit a design-philosophy shift: rather than maximizing the ideal Q-factor, one should optimize the worst-case Q-factor over the expected fabrication-error distribution, accepting a lower ideal Q in exchange for a flatter performance drop across manufacturing variability.
  • Because the trained networks systematically underpredict high-Q designs and perform best inside the training parameter range, an iterative active-learning loop—adding FDTD-validated high-Q candidates to the training set—is a natural next step for pushing the approach toward fabricated devices.
  • A testable extension would replace the generic roughness amplitudes and slant angle with statistics measured from a specific fabrication line, which would reveal whether the $52\%$ figure survives contact with a real process.
  • The broadened L2 mode suggests a yield-versus-Purcell trade-off worth quantifying: spreading the field to ease implantation tolerances lowers peak enhancement per emitter but may raise the fraction of devices that couple usefully to the cavity.
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 proposes a deep-learning workflow for designing diamond nanobeam cavities that are robust to fabrication imperfections. Two base designs (L2 and fishbone) are optimized under ideal, surface-roughness, and sidewall-slant conditions using a CNN surrogate trained on FDTD data; the resulting designs are then re-verified with FDTD. The authors report that the CNN predicts Q-factors with test errors as low as 3.99% and correlations up to 0.988, and claim that roughness- and slant-optimized designs show reduced Q-factor degradation, with a best case of 52.42% less degradation under sidewall slant. The paper is simulation-only, and the authors explicitly state that nanofabrication is the most critical next step.

Significance. If the simulation results were reliable, the paper would be a useful demonstration of a practical design strategy: training a surrogate on imperfect structures and selecting candidates with FDTD validation is a sensible way to search a high-dimensional fabrication-tolerance space. The manuscript is honest in several respects: Table S6 reports the large prediction errors (18–68%) of the surrogate on the actual optimized designs, and the text acknowledges that the fishbone slant case shifts the ideal operating point rather than providing intrinsic robustness. The main limitation is the absence of experimental validation and the questionable treatment of roughness under symmetric boundary conditions, which is load-bearing for the roughness-robustness claims. The paper does not provide code or data, but the methodology is described in sufficient detail that the critical simulations could be reproduced.

major comments (3)
  1. [Supplementary Information D1 and C1] The roughness simulations use symmetric boundary conditions in all three directions even though the text acknowledges that roughness breaks the structural symmetry (“Although roughness disrupts the symmetry of the structures, SBCs are applied in all three directions to ensure the simulations remain computationally manageable”). In a finite-difference solver, SBCs mirror the geometry across the symmetry planes, so each “random roughness” realization is actually a spatially correlated, mirror-symmetrized version of one random surface rather than an independent rough structure. This can systematically alter scattering losses, especially for the fishbone design with its narrow ridges. Because the roughness training labels and all roughness comparisons in Figures 8 and S13 and Tables S8/S10 are produced by these SBC simulations, the reported robustness ordering and degradation reductions may be artifacts of the boundary condition. The authors should re-run a subset of roughness cases in the full domain or with non-symmetric boundary conditions and report how the Q-values and the cavity-1/cavity-2 ordering change. This is a necessary check for the central roughness-robustness claim.
  2. [Abstract and Table S6] The abstract and conclusions state that the CNNs achieve prediction errors below 3.99% and correlation coefficients up to 0.988. These numbers are test-set averages over the in-distribution dataset (Table S4), but the six FDTD-validated optimized designs have NN prediction errors of 18.09%, 28.11%, 34.02%, 39.35%, 57.41%, and 68.19% (Table S6). The optimization procedure operates precisely in the sparse high-Q tail where the surrogate is least accurate, so the headline accuracy is not representative of the accuracy at the designs that the method actually recommends. The manuscript should report the prediction-error distribution for the optimized candidates or for a high-Q test subset, and the abstract and conclusions should be reworded so that the 3.99% figure is not presented as the accuracy of the optimization pipeline.
  3. [Results & Discussion, Figures 9 and S14, Tables S8/S10] The abstract's “52% reduction in Q-factor degradation” is the result of only one favorable comparison (L2 cavity 3 under sidewall slant, Figure 9). The other three comparisons show smaller or even absent robustness gains: fishbone cavity 2 under roughness shows a 15.55% smaller average degradation (Figure 8), L2 cavity 2 under roughness shows a 16.24% reduction (Figure S13), and fishbone cavity 3 under slant is explicitly described as having its ideal configuration shifted to about 5° rather than being inherently more robust (Figure S14 and surrounding text). The abstract should state the range of observed degradation reductions and identify that 52% is the single best case, or the central claim should be rephrased as “up to 52% in one of the four comparisons.”
minor comments (4)
  1. [Table S2] The minimum value for lm is listed as 10 nm, equal to its maximum; this should presumably be −10 nm.
  2. [Conclusions, first paragraph] The word “affects” should be “effects” in “counteracting the affects of such structural imperfections.”
  3. [Abstract] The claim of a “two-fold expansion in field distribution” is supported in the main text only for the base L2 design relative to conventional nanobeams (FWHMx ≈ 84 nm vs. ≈ 40 nm), not as a property of the optimized structures; the abstract should attribute this to the chosen base designs.
  4. [Supplementary Information G3] The selection of the best roughness-optimized candidate uses only 30 seeds for 10 candidates and then 100 seeds for two finalists; reporting the standard error of the Q difference would make it clear whether the claimed robustness advantage exceeds seed-to-seed variability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the CNN surrogate is trained and tested against FDTD, and every optimized candidate is re-verified with FDTD before selection.

full rationale

The paper's claimed derivation chain is self-contained: FDTD simulations generate Q-factors for random designs under ideal, rough, and slanted conditions; the CNN is trained on an 80/20 split, evaluated on held-out FDTD data (Eq. 4 and Tables S4-S5), and used only as a fast surrogate during optimization. All final candidates are then re-simulated with FDTD (Section IX and Supplementary G1, G3), so the reported Q-factors, including the 52% degradation reduction for L2 cavity 3, are FDTD outputs rather than CNN predictions relabeled as results. No fitted parameter is renamed as a prediction, and no equation reduces a claimed output to a training constant. The citations to Asano and Noda (refs 37 and 38) are used only as architectural inspiration for the CNN and for the gradient-ascent optimization procedure; they are not invoked as a uniqueness theorem, nor do they force the specific optimized designs. The skeptical concern about symmetric boundary conditions applied to roughness realizations (Supplementary D1) is a modeling-fidelity or correctness risk, not a circularity: it questions whether the simulated robustness transfers to fabricated devices, but the Q-values are still computed rather than assumed. The paper also explicitly states that nanofabrication is the most critical next step, avoiding the overclaim that experimental robustness has been demonstrated. Overall, no load-bearing step reduces to its own inputs by construction.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central robustness claim rests on hand-chosen imperfection amplitudes, an unvalidated assumption that SBCs remain valid for rough structures, and a surrogate NN whose predictions on the selected high-Q designs carry 18 to 68% error. These are assumptions the reader pays for, not results established by external data.

free parameters (8)
  • Roughness amplitude (flat and curved surfaces) = ±2.5 nm flat, ±4.0 nm curved
    Defines the perturbation level in all roughness training data and directly determines the reported real-world Q values (Section D1).
  • Sidewall slant angle = 5 degrees
    Defines the slant used for training the slant-robust NN and for comparing cavity 1 versus cavity 3 (Section D2, Figure 9).
  • NN learning rate alpha = 0.001
    Selected after empirical study of hyperparameters; affects convergence and final surrogate accuracy.
  • NN momentum gamma = 0.9
    Selected empirically along with alpha; the paper reports momentum-based SGD.
  • Weight decay lambda = 0 or 0.001
    L2-regularization on NN weights; shown to suppress predictions at high Q, so it affects which designs survive the optimization.
  • Optimization L2 penalties (lambda_GA, lambda_ES) = Varied sets, e.g., 0 to 0.25 for GA, 0 to 0.1 for CMA-ES
    Regularization keeps candidate designs near the base or best structure within the training distribution; the final choice depends on FDTD validation.
  • GA learning rate and momentum = alpha_GA=1e-5, gamma_GA=0.9
    Chosen empirically; controls how far the local optimizer moves per iteration.
  • CMA-ES initial standard deviation sigma_0 = 1
    Sets the initial exploration spread for the global optimizer.
assumptions (4)
  • domain assumption FDTD simulations with the stated grid, domain, and boundary conditions accurately compute Q-factors for nanobeam cavities.
    All Q-factors, including the ground truth for NN training, come from Rsoft FullWAVE. The paper does not benchmark the simulation against experimental measurements.
  • domain assumption Symmetric boundary conditions remain valid for roughness-perturbed structures, despite the text noting roughness disrupts symmetry.
    Supplementary Information D1 states SBCs are applied in all three directions for roughness simulations to ensure computational manageability; this can suppress radiation losses and inflate Q.
  • domain assumption The CNN trained on 1250 samples generalizes to the optimization landscape, and the regularization keeps designs in distribution.
    The paper notes NN predictions worsen for high Q and outside the training range; the optimization relies on the NN gradient direction, not just on final FDTD verification.
  • ad hoc to paper Surface roughness amplitudes of ±2.5 nm on flat and ±4.0 nm on curved surfaces, and a 5 degree sidewall slant, represent realistic thin-film diamond fabrication errors.
    These values are chosen without citing a measured roughness or slant distribution from diamond etching processes; the robustness claim depends directly on these magnitudes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Learning-Optimized, Fabrication Error-Tolerant Photonic Crystal Nanobeam Cavities for Scalable On-Chip Diamond Quantum Systems." pith.science (2026). https://pith.science/paper/TPH5J54P

@misc{pith2026250203987,
  author       = {Pith},
  title        = {Pith review of: Deep Learning-Optimized, Fabrication Error-Tolerant Photonic Crystal Nanobeam Cavities for Scalable On-Chip Diamond Quantum Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TPH5J54P}},
  note         = {Machine review of arXiv:2502.03987}
}
read the original abstract

Cavity-enhanced diamond color center qubits can be initialized, manipulated, entangled, and read individually with high fidelity, which makes them ideal for large-scale, modular quantum computers, quantum networks, and distributed quantum sensing systems. However, diamond's unique material properties pose significant challenges in manufacturing nanophotonic devices, leading to fabrication-induced structural imperfections and inaccuracies in defect implantation, which hinder reproducibility, degrade optical properties and compromise the spatial coupling of color centers to small mode-volume cavities. A cavity design tolerant to fabrication imperfections, such as surface roughness, sidewall slant, and non-optimal emitter positioning, can improve coupling efficiency while simplifying fabrication. To address this challenge, a deep learning-based optimization methodology is developed to enhance the fabrication error tolerance of nanophotonic devices. Convolutional neural networks (CNNs) are applied to promising designs, such as L2 and fishbone nanobeam cavities, predicting Q-factors up to one million times faster than traditional finite-difference time-domain (FDTD) simulations, enabling efficient optimization of complex, high-dimensional parameter spaces. The CNNs achieve prediction errors below 3.99% and correlation coefficients up to 0.988. Optimized structures demonstrate a 52% reduction in Q-factor degradation, achieving quality factors of 5e4 under real-world conditions and a two-fold expansion in field distribution, enabling efficient coupling of non-optimally positioned emitters. This methodology enables scalable, high-yield manufacturing of robust nanophotonic devices, including the cavity-enhanced diamond quantum systems developed in this study.

Figures

Figures reproduced from arXiv: 2502.03987 by the authors.

Figure 1
Figure 1. A flowchart illustrating the process for optimizing nanocavity designs using DL, [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. A schematic representation of the CNN architecture employed to optimize [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. An overview showcasing an example of the dataset, training, and correlation [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: A visual representation of the geometric properties of the base design for the [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]
Figure 5
Figure 5. Figure 5: A visual representation of the geometric properties of the base design for the [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]
Figure 6
Figure 6. Figure 6: Optical properties of the L2 nanobeam cavity. [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: Optical properties of the fishbone nanobeam cavity. [PITH_FULL_IMAGE:figures/full_fig_p023_7.png]
Figure 8
Figure 8. Figure 8: The distribution of Q-factors of fishbone cavity 1 and cavity 2 under different [PITH_FULL_IMAGE:figures/full_fig_p026_8.png]
Figure 9
Figure 9. Figure 9: A graph presenting the Q-factor of L2 cavity 1 and cavity 3 under various slant [PITH_FULL_IMAGE:figures/full_fig_p027_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 17 canonical work pages

  1. [1]

    Y.; Li, Y.; Nur, S.; Iwai, T.; Miyatake, T.; Kawaguchi, K.; Doi, Y.; Sato, S

    (1) Ishihara, R.; Hermias, J.; Yu, S.; Yu, K. Y.; Li, Y.; Nur, S.; Iwai, T.; Miyatake, T.; Kawaguchi, K.; Doi, Y.; Sato, S. 3D Integration Technology for Quantum Computer based on Diamond Spin Qubits. 2021 IEEE International Electron Devices Meeting (IEDM). 2021; pp 14.5.1–14.5.4. (2) Degen, C. L.; Reinhard, F.; Cappellaro, P . Quantum sensing.Reviews of ...

    work page 2021

  2. [9]

    Coupling of a single tin-vacancy center to a photonic crystal cavity in diamond

    (17) Kuruma, K.; Pingault, B.; Chia, C.; Renaud, D.; Hoffmann, P .; Iwamoto, S.; Ronning, C.; Lonˇcar, M. Coupling of a single tin-vacancy center to a photonic crystal cavity in diamond. Applied Physics Letters 2021,

    work page 2021

  3. [16]

    Activation functions in neural networks

    (49) Sharma, S.; Sharma, S.; Athaiya, A. Activation functions in neural networks. Towards Data Sci 2017, 6, 310–316. (50) Girosi, F.; Jones, M.; Poggio, T. Regularization theory and neural networks architec- tures. Neural computation 1995, 7, 219–269. (51) Sutskever, I.; Martens, J.; Dahl, G.; Hinton, G. On the importance of initialization and momentum in...

    work page 2017

  4. [29]

    Q-Factor

    (60) Kajfez, D. Q-Factor. Encyclopedia of RF and Microwave Engineering 2005, (61) Lalanne, P . Mode volume of electromagnetic resonators: let us try giving credit where it is due. arXiv preprint arXiv:2011.00218 2020, (62) Nazarov, Y. V .; Danon, J.Advanced quantum mechanics: a practical guide ; Cambridge University Press,

    arXiv 2005

  5. [88]

    Fabrication of 15NV- centers in diamond using a deterministic single ion implanter.New Journal of Physics 2021, 23, 063067

    (44) Groot-Berning, K.; Jacob, G.; Osterkamp, C.; Jelezko, F.; Schmidt-Kaler, F. Fabrication of 15NV- centers in diamond using a deterministic single ion implanter.New Journal of Physics 2021, 23, 063067. (45) Asano, T.; Song, B.-S.; Noda, S. Analysis of the experimental Q factors (˜ 1 million) of photonic crystal nanocavities. Optics express 2006, 14, 19...

    work page 2021

  6. [104]

    Automated optimization of photonic crystal slab cavities

    (31) Minkov, M.; Savona, V . Automated optimization of photonic crystal slab cavities. Scientific reports 2014, 4,

    work page 2014

  7. [105]

    M.; Markham, M.; M ¨ucklich, F.; others Spin measurements of NV centers coupled to a photonic crystal cavity

    (23) Jung, T.; G¨orlitz, J.; Kambs, B.; Pauly, C.; Raatz, N.; Nelz, R.; Neu, E.; Edmonds, A. M.; Markham, M.; M ¨ucklich, F.; others Spin measurements of NV centers coupled to a photonic crystal cavity. Apl Photonics 2019,

    work page 2019

  8. [109]

    (30) Lai, Y.; Pirotta, S.; Urbinati, G.; Gerace, D.; Minkov, M.; Savona, V .; Badolato, A.; Galli, M

    Journal of Lightwave Technology 2008, 26, 1532–1539. (30) Lai, Y.; Pirotta, S.; Urbinati, G.; Gerace, D.; Minkov, M.; Savona, V .; Badolato, A.; Galli, M. Genetically designed L3 photonic crystal nanocavities with measured quality factor exceeding one million. Applied Physics Letters 2014,

    work page 2008

Show all 19 references
  1. [111]

    W.; Haas, M.; Guo, X.; Kuruma, K.; Jin, C.; Li, Z.; Awschalom, D

    32 (20) Ding, S. W.; Haas, M.; Guo, X.; Kuruma, K.; Jin, C.; Li, Z.; Awschalom, D. D.; Dele- gan, N.; Heremans, F. J.; High, A.; others High-Q Cavity Interface for Color Centers in Thin Film Diamond. arXiv preprint arXiv:2402.05811 2024, (21) Regan, B.; Trycz, A.; Fr ¨och, J. ...

    2024 arXiv

  2. [118]

    E.; Aghaeimeibodi, S.; Riedel, D.; Dory, C.; Lu, H.; McQuade, P

    (18) Rugar, A. E.; Aghaeimeibodi, S.; Riedel, D.; Dory, C.; Lu, H.; McQuade, P . J.; Shen, Z.- X.; Melosh, N. A.; Vuˇckovi´c, J. Quantum photonic interface for tin-vacancy centers in diamond. Physical Review X 2021, 11, 031021. (19) Mouradian, S.; Wan, N. H.; Schr¨oder, T.; En...

    2021

  3. [1372]

    Optimization of photonic crystal nanocavities based on deep learning

    (37) Asano, T.; Noda, S. Optimization of photonic crystal nanocavities based on deep learning. Optics express 2018, 26, 32704–32717. (38) Asano, T.; Noda, S. Iterative optimization of photonic crystal nanocavity designs by using deep neural networks. Nanophotonics 2019, 8, 224...

    2018

  4. [2009]

    K.; Riedinger, R.; Machielse, B.; Levonian, D

    (14) Bhaskar, M. K.; Riedinger, R.; Machielse, B.; Levonian, D. S.; Nguyen, C. T.; Knall, E. N.; Park, H.; Englund, D.; Lonˇcar, M.; Sukachev, D. D.; others Experimental demonstra- tion of memory-enhanced quantum communication. Nature 2020, 580, 60–64. (15) Nguyen, C.; Sukache...

    2020

  5. [2013]

    http://qce-it-infra.ewi.tudelft.nl/qce_ servers.html

    (63) TUDelft Q&CE IT Infrastructure. http://qce-it-infra.ewi.tudelft.nl/qce_ servers.html. (64) Hendrycks, D.; Gimpel, K. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415 2016, Acknowledgement The authors thank Ir. M. van Beusekom (Q&CE) for essential IT su...

    2016 arXiv

  6. [2021]

    Improvement in the qual- ity factors for photonic crystal nanocavities via visualization of the leaky components

    (25) Nakamura, T.; Takahashi, Y.; Tanaka, Y.; Asano, T.; Noda, S. Improvement in the qual- ity factors for photonic crystal nanocavities via visualization of the leaky components. Optics express 2016, 24, 9541–9549. (26) Maeno, K.; Takahashi, Y.; Nakamura, T.; Asano, T.; Noda,...

    2016

  7. [2022]

    S.; Robledo, L.; Taminiau, T

    (8) Bernien, H.; Hensen, B.; Pfaff, W.; Koolstra, G.; Blok, M. S.; Robledo, L.; Taminiau, T. H.; Markham, M.; Twitchen, D. J.; Childress, L.; others Heralded entanglement between solid-state qubits separated by three metres. Nature 2013, 497, 86–90. (9) Janitz, E.; Bhaskar, M....

    2013

  8. [4401]

    Deep reinforcement learning empowers automated inverse design and optimization of photonic crystals for nanoscale laser cavities

    (40) Li, R.; Zhang, C.; Xie, W.; Gong, Y.; Ding, F.; Dai, H.; Chen, Z.; Yin, F.; Zhang, Z. Deep reinforcement learning empowers automated inverse design and optimization of photonic crystals for nanoscale laser cavities. Nanophotonics 2023, 12, 319–334. (41) Wiecha, P . R.; Ar...

    2023

  9. [5124]

    Optimization of a photonic crystal nanocavity using covariance matrix adaptation evolution strategy

    (32) Takahashi, K.; Baba, T. Optimization of a photonic crystal nanocavity using covariance matrix adaptation evolution strategy. IEEE Photonics Journal 2022, 14, 1–5. (33) Abe, R.; Takeda, T.; Shiratori, R.; Shirakawa, S.; Saito, S.; Baba, T. Optimization of an H0 photonic cr...

    2022

  10. [5625]

    31 (11) Fox, A. M. Quantum optics: an introduction; Oxford university press, 2006; Vol

    2006

  11. [7080]

    Photonic switching devices based on semiconductor nano- structures

    (58) Jin, C.-Y.; Wada, O. Photonic switching devices based on semiconductor nano- structures. Journal of Physics D: Applied Physics 2014, 47, 133001. 36 (59) Pfeifer, H.; Ratschbacher, L.; Gallego, J.; Saavedra, C.; Faßbender, A.; von Haaren, A.; Alt, W.; Hofferberth, S.; K¨oh...

    2014

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.