Pith. sign in

REVIEW 4 major objections 5 minor 66 references

Miniaturized spectrometer enabled by end-to-end deep learning on large-scale radiative cavity array

T0 review · 4 major / 5 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 demonstrates a compact on-chip spectrometer in which a 36x30 array of high-Q radiative cavities turns an incident spectrum into a camera image that an end-to-end deep network decodes, resolving 0.048 nm features across an 80 nm…

desk verdict A credible mini-BIC cavity array spectrometer with deep learning reconstruction, but the arbitrary-spectrum claim needs out-of-distribution proof before I'd trust it. read the letter →

arxiv 2411.13353 v1 pith:7XFCZK2I submitted 2024-11-20 physics.optics

Xinyi Zhou , Cheng Zhang , Xiaoyu Zhang , Yi Zuo , Zixuan Zhang , Feifan Wang , Zihao Chen , Hongbin Li
show 1 more author
Chao Peng
This is my paper · ORCID
classification physics.optics
keywords miniaturizedspectrometerboundstatesincontinuumphotoniccrystalcavityarrayend-to-enddeeplearningspectralreconstructionnear-infraredspectroscopyhigh-Qopticalcomputational
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 reports a miniaturized spectrometer that reconstructs unknown incident spectra from a single image of a 36x30 array of photonic-crystal cavities. The cavities are miniaturized bound states in the continuum, each supporting high-Q modes with distinctive out-of-plane radiation patterns, so the whole array turns a spectrum into a spatially encoded camera image. An end-to-end deep network trained on 83,000 waveform-shaper-generated spectra learns to invert that encoding directly, bypassing per-cavity calibration. The authors demonstrate 0.048 nm resolution over an 80 nm near-infrared band, with mean reconstruction fidelity above 95% on 12,450 blind spectra. If the result holds, it offers a compact, scan-free route to high-resolution spectroscopy for portable diagnostics.

What carries the argument

The load-bearing element is the mini-BIC cavity, a photonic-crystal slab resonator whose out-of-plane radiation is suppressed by topological charges, giving it high-Q modes with distinct radiation patterns. One thousand eighty such cavities, each slightly detuned so their resonances cover 1525-1605 nm, convert an incident spectrum into a 1280x1024 pixel image; an EfficientNet convolutional network then maps that image to 8,000 spectral intensity samples at 0.01 nm spacing. The network's training procedure uses pre-training on 65% of the data followed by fine-tuning on the remainder, with 2% Gaussian noise added to 40% of images for generalization, and a composite loss combining mean square error with cosine similarity.

What would settle it

Take the trained spectrometer and present it with spectra not generated by the same waveform-shaper recipe, such as gas-cell absorption lines or doublets narrower than 0.04 nm, and compare reconstructions against a commercial reference; a systematic fidelity drop outside the training family would show that the 0.048 nm claim is distribution-specific rather than general.

Watch

Extended reading notes

Core claim

The central claim is that a large radiative cavity array, combined with end-to-end deep learning, resolves the usual trade-off between detection range and resolution in on-chip spectrometers. The device uses 1,080 mini-BIC cavities with quality factors above $10^4$ and deliberately intertwined resonance wavelengths spanning 1525-1605 nm; each cavity radiates a distinct spatial pattern when the incident spectrum overlaps its modes. The camera image of the whole array is fed into an EfficientNet network with 8,000 outputs at 0.01 nm spacing, trained with a composite MSE-plus-cosine-similarity loss and a two-stage transfer-learning schedule. In blind tests the reconstructed spectra match a commercial reference spectrometer, resolving single peaks of 0.048 nm FWHM, distinguishing two 0.04 nm peaks separated by 0.08 nm, and keeping mean fidelity above 95% with 0.008 nm mean peak-position deviation across the band.

Load-bearing premise

The load-bearing premise is that the 83,000 training spectra produced by the waveform shaper cover the full variety of arbitrary spectra the device will meet in real use; if real spectra fall outside that family, the claimed 0.048 nm resolution and greater-than-95% fidelity are not established for them.

Editorial extensions

If this is right

  • A spectrometer on a 505 x 606 um chip can resolve 0.048 nm spectral features across 1525-1605 nm from a single camera frame, without moving parts or wavelength scanning.
  • Calibration of individual cavities is unnecessary; the end-to-end network absorbs fabrication deviations, so the array can be scaled up without per-resonator characterization.
  • The detection range can be extended beyond 80 nm by adding cavities at other resonance wavelengths and training on correspondingly broader spectra, limited in the current demonstration by the training source bandwidth.
  • Out-of-plane excitation avoids on-chip light routing, cutting insertion loss to about 14 dB and enabling 12.5 uW/nm sensitivity without an optical amplifier.
  • Blind-test fidelity over 12,450 spectra exceeds 95% on average, with 0.008 nm mean peak-position deviation across the full detection band.

Reading between the lines

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

  • A natural extension, not stated in the paper, is to test the device on out-of-distribution spectra such as gas-cell absorption lines or molecular emission features that the waveform-shaper training family did not generate, to map where the 95% fidelity claim actually holds.
  • The same array-plus-network recipe should transfer to other wavelength windows by rescaling lattice periods and retraining, making the architecture a generic template for compact spectrometers rather than a near-infrared-specific device.
  • Because resolution is set by cavity Q while sensitivity drops as Q increases, an explicit trade-off curve would tell whether one array can simultaneously reach picometer-level resolution and microwatt-level detection, a question the current demonstration leaves open.
  • The two-dimensional radiative pattern acts as an optical encoder; coupling the cavities to analyte-responsive materials could turn the spectrometer into a targeted chemical or biological sensor.
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 paper reports a miniaturized spectrometer based on a 36×30 array of mini-BIC cavities on a photonic-crystal slab, with out-of-plane excitation and a CMOS camera recording the array's radiative response. An EfficientNet-B0 network is trained end-to-end on 83,000 randomly generated spectra (from a Finisar WaveShaper 4000S) to map the camera image to an 8000-point spectrum over 1525–1605 nm. The authors report a resolution of 0.048 nm, a mean reconstruction fidelity exceeding 95% over 12,450 validation spectra, and a detection sensitivity of 12.5 µW/nm, and they compare selected reconstructions with a commercial spectrometer.

Significance. If the claimed performance transfers to real-world spectra, this is a valuable advance: a 505 µm × 606 µm on-chip device with 80 nm bandwidth, sub-0.05 nm resolution, and no moving parts or in-plane optical routing would be very attractive for portable spectroscopy. The paper's strengths are the large-scale fabrication and characterization of 1080 high-Q cavities, the use of an end-to-end learning approach that avoids per-cavity calibration, the systematic validation on a large held-out set with bootstrap statistics, and direct comparisons to a commercial instrument. However, all training, validation, and 'blind' test spectra are generated by the same waveform shaper, so the central claim of solving arbitrary unknown spectra rests on the untested assumption that the WaveShaper's output family is representative of the full target distribution.

major comments (4)
  1. [Setup and algorithms; Experimental results] All spectra used for training, validation, and blind testing (83,000 training, 12,450 validation, and the Fig. 5 tests) are generated by the same Finisar WaveShaper 4000S. The network therefore has the opportunity to learn the WaveShaper's transfer-function artifacts, grid quantization, and drift rather than only the physical encoding of the cavity array. The Introduction's claim that 'the trained network can solve arbitrary unknown spectra' is not supported by tests drawn from the same generation family. Please add an independent out-of-distribution test, for example using a calibrated gas cell, an etalon, a different tunable laser with known lines, or an independently characterized source, and report the reconstruction error on that test. This is load-bearing because the headline capability is arbitrary-spectrum reconstruction.
  2. [Experimental results; Fig. 5(g)] The 'fidelity' metric underpins the headline claim 'fidelity exceeding 95%', but the paper never defines it. The reader cannot tell whether fidelity is cosine similarity, normalized mean-square error, or another quantity, nor what value constitutes an acceptable reconstruction. Please provide the exact formula, the distribution (mean, standard deviation, confidence interval) for the 12,450 validation spectra, and the same statistics for the proposed out-of-distribution test.
  3. [Experimental results, resolution claim] The 0.048 nm resolution is demonstrated on single and double peaks with nominal FWHM of 0.04 nm generated by the same WaveShaper used for training. Because the training set contains rich spectral features generated by the same device, the network may be reproducing the WaveShaper's line shapes rather than resolving the physical limit of the cavity array. To substantiate the resolution claim, please provide a resolution test with a source whose line shape and bandwidth are independent of the training generator (e.g., a narrow-linewidth laser with calibrated frequency comb, or a molecular absorption line), or provide an information-theoretic analysis of the array's resolution limit and show that 0.048 nm is consistent with the physical encoding without relying on the training prior.
  4. [Experimental results, sensitivity] The reported 'detection sensitivity of 12.5 µW/nm' is not defined or measured as a minimum detectable power. The text states that a single narrow peak with FWHM of 0.048 nm was 'accurately solved at a low incident power of 0 dBm', corresponding to 12.5 µW/nm, but this appears to be 1 mW divided by 80 nm rather than a measured power-detection limit. Since sensitivity is listed in the abstract and conclusion, please define the metric, describe the measurement procedure (e.g., a power sweep), and report the actual minimum detectable power for a defined signal-to-noise ratio.
minor comments (5)
  1. [Setup and algorithms] The text describes 'migration learning' and 'leveraging generic features learned from pre-trained models on large-scale datasets', but the described procedure uses 65% of the same training dataset for the first stage, not an external large-scale dataset. Please correct the terminology and clarify the actual transfer-learning procedure.
  2. [Setup and algorithms, Data processing] The output layer is described as '8000 nodes to present the light intensities ... with a spectral resolution of 0.01 nm'. This is more precisely an output grid spacing of 0.01 nm (80 nm / 8000); the resolution is an experimentally demonstrated quantity, not an inherent property of the output layer. Please revise the wording.
  3. [Experimental results, Fig. 5(f)] The inset legend in Fig. 5(f) states 'Mean FWHM = 0.06 nm', while the text reports a mean peak deviation of 0.008 nm and a standard deviation of the FWHM error of 0.027 nm. Please clarify which quantity is plotted in the inset and ensure the figure and text are consistent.
  4. [Discussion] The statement that the detection range can be 'easily extended with additional broadband training data' should be qualified by the physical spectral coverage of the cavity array, which is designed for 1525–1605 nm; the training-data bandwidth is not the only limitation.
  5. [Experimental results, Fig. 5] Please specify how the 801 peaks in the Fig. 5(f) sweep were generated and confirm that they are WaveShaper outputs, so the reader can assess the independence of this test from the training distribution.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the end-to-end reconstruction is a supervised fit on held-out data, and self-citations to prior mini-BIC work are not load-bearing.

full rationale

The paper's central claim is an empirical demonstration, not a first-principles derivation: a neural network maps camera images of a mini-BIC cavity array to spectra. The mapping is trained on 83,000 WaveShaper-generated spectra (70,550 training / 12,450 validation) and evaluated on held-out spectra from the same generator, with a subset cross-checked against a commercial spectrometer (AQ6374). This is a standard supervised-learning evaluation. The held-out validation is genuinely unseen, so the 95% fidelity and 0.048 nm resolution are not equivalent by construction to the training data; they are interpolation results on the training distribution. The main limitation is external validity: 'arbitrary unknown spectra' (Introduction, Experimental results) is stronger than what can be shown by testing only on spectra drawn from the same WaveShaper family. That is a generalization/validation concern, not a circular derivation chain. The paper's self-citations ([54] for the mini-BIC principle; [61,62] for cross-filtering techniques) are not load-bearing because the current work independently measures Q ~ 7.9e4 and the cavity modes on its own sample, and the cited prior work is published and externally falsifiable. No equation, fitted parameter, or generated dataset is renamed as a prediction; the reconstruction is openly trained end-to-end. Therefore, no circular step meeting the required evidentiary standard is present.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claims depend on the trained network weights (a fitted model), on the mini-BIC cavity behavior taken from prior work, and on the representativeness of the training spectra. No new physical entities are introduced.

free parameters (2)
  • EfficientNet-B0 model weights = trainable parameters (millions)
    The mapping from array images to spectra is entirely learned from 70,550 training samples; the reported resolution and fidelity are properties of this fitted model, not of a derived physical principle.
  • Output wavelength grid spacing = 0.01 nm
    The 8,000 output nodes correspond to a chosen 0.01 nm grid; this discretization influences the achievable reported resolution, though the physical limit is set by cavity Q.
assumptions (4)
  • domain assumption mini-BIC cavities, as described in prior work [54], possess discrete high-Q modes with the radiative patterns and Q values used here.
    The device design and expected resolution rely on the existence and robustness of these modes; the paper does not re-derive the BIC physics.
  • domain assumption The waveform shaper's programmed spectra are accurate ground truth labels for training.
    The network learns to reproduce the programmed spectra; any discrepancy between programmed and actual optical spectra becomes systematic label error that limits achievable fidelity.
  • ad hoc to paper Randomly generated spectra from the waveshaper are representative of the 'arbitrary' spectra the device will encounter.
    This is assumed when claiming the solution of arbitrary unknown spectra; no evidence is provided that real-world spectra (e.g., chemical absorption lines) lie within the training distribution.
  • domain assumption The cavity array's spatially resolved response is a one-to-one encoding of the spectrum over 1525-1605 nm.
    If two different spectra produced identical camera images, the network could not distinguish them; the diversity of resonances is argued but injectivity is not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Miniaturized spectrometer enabled by end-to-end deep learning on large-scale radiative cavity array." pith.science (2026). https://pith.science/paper/7XFCZK2I

@misc{pith2026241113353,
  author       = {Pith},
  title        = {Pith review of: Miniaturized spectrometer enabled by end-to-end deep learning on large-scale radiative cavity array},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7XFCZK2I}},
  note         = {Machine review of arXiv:2411.13353}
}
abstract

Miniaturized (mini-) spectrometers are highly desirable tools for chemical, biological, and medical diagnostics because of their potential for portable and in situ spectral detection. In this work, we propose and demonstrate a mini-spectrometer that combines a large-scale radiative cavity array with end-to-end deep learning networks. Specifically, we utilize high-Q bound states in continuum cavities with distinct radiation characteristics as the fundamental units to achieve parallel spectral detection. We realize a 36 $\times$ 30 cavity array that spans a wide spectral range from 1525 to 1605 nm with quality factors above 10^4. We further train a deep network with 8000 outputs to directly map arbitrary spectra to array responses excited by the out-of-plane incident. Experimental results demonstrate that the proposed mini-spectrometer can resolve unknown spectra with a resolution of 0.048 nm in a bandwidth of 80 nm and fidelity exceeding 95%, thus offering a promising method for compact, high resolution, and broadband spectroscopy.

Figures

Figures reproduced from arXiv: 2411.13353 by the authors.

Figure 1
Figure 1. Schematic and principle of mini-spectrometer. (a) Schematic of mini-spectrometer. The cavities radiate in the vertical direction (blue arrow) under out-of-plane excitation of inci￾dent light (purple arrow), each cavity displays a unique radiation pattern when the excitation wavelength matches the cavity resonance. A CMOS camera records the radiation pattern, which contains the spectral characteristics of the inciden… view at source ↗
Figure 2
Figure 2. Fabricated sample and array arrangement strategy. (a) The scanning electron microscope (SEM) images of the fabricated sample show each mini-BIC cavity consisting of region A and region B from top and side views. FIB is used to cleave the PhC for a side view of the etched air holes. (b) Thermogram of wavelength distribution, which contains 30 rows × 36 columns of array cells of 1080 cavities, aligning with the center… view at source ↗
Figure 3
Figure 3. Experimental setup and results. (a) Schematic of the measurement setup. Light Source: an amplified spontaneous emission (ASE) or a narrow line-width laser, BS: beam split￾ter, RFP: rear focal plane, PD: photodetector. (b) Measured reflection spectrum by scanning the wavelength from 1525 to 1605 nm using the narrow line-width laser, showing the mini-BIC cavity supports 4 modes in such a range. (c) A typical CMOS imag… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: End-to-end training process network. (a) The generalization ability of the deep￾learning model is enhanced by introducing trace noise to 40% of the dataset (83,000 images) before feeding it into the network. The dataset is split into training and validation sets. The m…
Figure 5
Figure 5. Figure 5: Blind tests of spectral reconstruction. The outputs of our mini-spectrometer are compared with a commercial spectrometer as reference values for a series of representative spectral features. The reconstructed spectra (purple solid line) agree well with the reference va…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 63 canonical work pages

  1. [1]

    P., Mattley, Y

    Bacon, C. P., Mattley, Y . & DeFrece, R. Miniature spectroscopic instrumentation: Appli- cations to biology and chemistry. Review of Scientific Instruments 75, 1–16 (2004)

    work page 2004

  2. [2]

    Hank, T. B. et al. Spaceborne imaging spectroscopy for sustainable agriculture: Contribu- tions and challenges. Surveys in Geophysics 40, 515–551 (2018)

    work page 2018

  3. [3]

    S., Guedes, S., Caixeta, D

    Vitorino, R., Barros, A. S., Guedes, S., Caixeta, D. C. & Sabino-Silva, R. Diagnostic and monitoring applications using near infrared (nir) spectroscopy in cancer and other diseases. Photodiagnosis and Photodynamic Therapy 42, 103633 (2023)

    work page 2023

  4. [4]

    Cutshaw, G. et al. The emerging role of raman spectroscopy as an omics approach for metabolic profiling and biomarker detection toward precision medicine. Chemical Reviews 123, 8297–8346 (2023)

    work page 2023

  5. [5]

    A., Asvany, O., Br ¨unken, S

    McGuire, B. A., Asvany, O., Br ¨unken, S. & Schlemmer, S. Laboratory spectroscopy techniques to enable observations of interstellar ion chemistry. Nature Reviews Physics 2, 402–410 (2020)

    work page 2020

  6. [6]

    & Hasan, T

    Yang, Z., Albrow-Owen, T., Cai, W. & Hasan, T. Miniaturization of optical spectrometers. Science 371 (2021)

    work page 2021

  7. [7]

    Li, A. et al. Advances in cost-effective integrated spectrometers. Light: Science & Applications 11 (2022)

    work page 2022

  8. [8]

    & Crozier, K

    Tan, H., Li, B. & Crozier, K. B. Optical fiber speckle spectrometer based on reversed-lens smartphone microscope. Scientific Reports 13 (2023)

    work page 2023

Show all 66 references
  1. [9]

    Kim, U. J. et al. Drug classification with a spectral barcode obtained with a smartphone raman spectrometer. Nature Communications 14 (2023). 19

    2023

  2. [10]

    Cheben, P. et al. A high-resolution silicon-on-insulator arrayed waveguide grating mi- crospectrometer with sub-micrometer aperture waveguides. Optics Express 15, 2299 (2007)

    2007

  3. [11]

    & Boyd, R

    Gao, B., Shi, Z. & Boyd, R. W. Design of flat-band superprism structures for on-chip spectroscopy. Optics Express 23, 6491 (2015)

    2015

  4. [12]

    Cheng, R. et al. Broadband on-chip single-photon spectrometer. Nature Communications 10 (2019)

    2019

  5. [13]

    & Kar, S

    Hejazi, D., Kari Rezapour, N., Ferrier, J., Ostadabbas, S. & Kar, S. Dispersion-free highly accurate color recognition using excitonic 2d materials and machine learning. Materials Today 59, 18–24 (2022)

    2022

  6. [15]

    Zhang, Z. et al. Folded digital meta-lenses for on-chip spectrometer. Nano Letters 23, 3459–3466 (2023)

    2023

  7. [16]

    & Wolffenbuttel, R

    Correia, J., de Graaf, G., Kong, S., Bartek, M. & Wolffenbuttel, R. Single-chip cmos optical microspectrometer. Sensors and Actuators A: Physical 82, 191–197 (2000)

    2000

  8. [17]

    Pervez, N. K. et al. Photonic crystal spectrometer. Optics Express 18, 8277 (2010)

    2010

  9. [18]

    Ebermann, M. et al. Tunable mems fabry-p´erot filters for infrared microspectrometers: a review. In Piyawattanametha, W. & Park, Y .-H. (eds.)MOEMS and Miniaturized Systems XV, vol. 9760, 97600H (SPIE, 2016)

    2016

  10. [19]

    Zhou, X. et al. Ultra-compact on-chip spectrometer based on thermally tuned topological miniaturized bound states in the continuum cavity. Optics Letters 48, 4993 (2023). 20

    2023

  11. [20]

    & Xia, F

    Yuan, S., Naveh, D., Watanabe, K., Taniguchi, T. & Xia, F. A wavelength-scale black phosphorus spectrometer. Nature Photonics 15, 601–607 (2021)

    2021

  12. [21]

    Zheng, B. et al. Single-detector black phosphorus monolithic spectrometer with high spec- tral and temporal resolution. Applied Physics Letters 120 (2022)

    2022

  13. [22]

    & Chen, X

    Wang, H., Chen, S. & Chen, X. Room-temperature self-powered infrared spectrometer based on a single black phosphorus heterojunction diode.Nano Letters 24, 326–330 (2023)

    2023

  14. [23]

    Deng, W. et al. Electrically tunable two-dimensional heterojunctions for miniaturized near- infrared spectrometers. Nature Communications 13 (2022)

    2022

  15. [24]

    Yoon, H. H. et al. Miniaturized spectrometers with a tunable van der waals junction. Sci- ence 378, 296–299 (2022)

    2022

  16. [25]

    Podmore, H. et al. Demonstration of a compressive-sensing fourier-transform on-chip spec- trometer. Optics Letters 42, 1440 (2017)

    2017

  17. [26]

    Souza, M. C. M. M., Grieco, A., Frateschi, N. C. & Fainman, Y . Fourier transform spec- trometer on silicon with thermo-optic non-linearity and dispersion correction.Nature Com- munications 9 (2018)

    2018

  18. [27]

    Kita, D. M. et al. High-performance and scalable on-chip digital fourier transform spec- troscopy. Nature Communications 9 (2018)

    2018

  19. [29]

    Montesinos-Ballester, M. et al. On-chip fourier-transform spectrometer based on spatial heterodyning tuned by thermo-optic effect. Scientific Reports 9 (2019). 21

    2019

  20. [30]

    & Rose, T

    Paudel, U. & Rose, T. Ultra-high resolution and broadband chip-scale speckle enhanced fourier-transform spectrometer. Optics Express 28, 16469 (2020)

    2020

  21. [31]

    & Tsang, H

    Xu, H., Qin, Y ., Hu, G. & Tsang, H. K. Scalable integrated two-dimensional fourier- transform spectrometry. Nature Communications 15 (2024)

    2024

  22. [32]

    Wang, Z. et al. Single-shot on-chip spectral sensors based on photonic crystal slabs.Nature Communications 10 (2019)

    2019

  23. [33]

    Yang, J. et al. Deep-learning based on-chip rapid spectral imaging with high spatial reso- lution. Chip 2, 100045 (2023)

    2023

  24. [34]

    Yang, Z. et al. Single-nanowire spectrometers. Science 365, 1017–1020 (2019)

    2019

  25. [35]

    Cheng, Z. et al. Generalized modular spectrometers combining a compact nanobeam mi- crocavity and computational reconstruction. ACS Photonics 9, 74–81 (2021)

    2021

  26. [36]

    Li, H. et al. A near-infrared miniature quantum dot spectrometer. Advanced Optical Mate- rials 9 (2021)

    2021

  27. [37]

    Gao, L., Qu, Y ., Wang, L. & Yu, Z. Computational spectrometers enabled by nanophotonics and deep learning. Nanophotonics 11, 2507–2529 (2022)

    2022

  28. [38]

    Bian, L. et al. A broadband hyperspectral image sensor with high spatio-temporal resolu- tion. Nature 635, 73–81 (2024)

    2024

  29. [39]

    Wang, S.-W. et al. Concept of a high-resolution miniature spectrometer using an integrated filter array. Optics Letters 32, 632 (2007)

    2007

  30. [40]

    & Englund, D

    Gan, X., Pervez, N., Kymissis, I., Hatami, F. & Englund, D. A high-resolution spectrometer based on a compact planar two dimensional photonic crystal cavity array. Applied Physics Letters 100 (2012). 22

    2012

  31. [41]

    & Tsang, H

    Xu, H., Qin, Y ., Hu, G. & Tsang, H. K. Integrated single-resonator spectrometer beyond the free-spectral-range limit. ACS Photonics 10, 654–666 (2023)

    2023

  32. [42]

    Zheng, S. N. et al. Microring resonator-assisted fourier transform spectrometer with en- hanced resolution and large bandwidth in single chip solution. Nature Communications 10 (2019)

    2019

  33. [43]

    Yao, C. et al. Broadband picometer-scale resolution on-chip spectrometer with reconfig- urable photonics. Light: Science & Applications 12 (2023)

    2023

  34. [44]

    Zhang, L. et al. Ultrahigh-resolution on-chip spectrometer with silicon photonic resonators. Opto-Electronic Advances 5, 210100–210100 (2022)

    2022

  35. [45]

    Kong, L. et al. Single-detector spectrometer using a superconducting nanowire. Nano Letters 21, 9625–9632 (2021)

    2021

  36. [46]

    & Bawendi, M

    Bao, J. & Bawendi, M. G. A colloidal quantum dot spectrometer. Nature 523, 67–70 (2015)

    2015

  37. [47]

    Zhu, X. et al. Broadband perovskite quantum dot spectrometer beyond human visual reso- lution. Light: Science & Applications 9 (2020)

    2020

  38. [48]

    & Lipson, M

    Nitkowski, A., Chen, L. & Lipson, M. Cavity-enhanced on-chip absorption spectroscopy using microring resonators. Optics Express 16, 11930 (2008)

    2008

  39. [49]

    Lee, H.-S. et al. Design of mid-infrared filter array based on plasmonic metal nanodiscs array and its application to on-chip spectrometer. Scientific Reports 11 (2021)

    2021

  40. [50]

    Brown, C. et al. Neural network-based on-chip spectroscopy using a scalable plasmonic encoder. ACS Nano 15, 6305–6315 (2021). 23

    2021

  41. [51]

    Tua, D. et al. Imaging-based intelligent spectrometer on a plasmonic rainbow chip. Nature Communications 14 (2023)

    2023

  42. [52]

    Tang, F. et al. Metasurface spectrometers beyond resolution-sensitivity constraints (2024)

    2024

  43. [53]

    Zhang, Y . et al. Miniature computational spectrometer with a plasmonic nanoparticles-in- cavity microfilter array. Nature Communications 15 (2024)

    2024

  44. [54]

    Chen, Z. et al. Observation of miniaturized bound states in the continuum with ultra-high quality factors. Science Bulletin 67, 359–366 (2022)

    2022

  45. [55]

    & Ying, Y

    Zhang, X., Lin, T., Xu, J., Luo, X. & Ying, Y . Deepspectra: An end-to-end deep learning approach for quantitative spectral analysis. Analytica Chimica Acta 1058, 48–57 (2019)

    2019

  46. [56]

    Wang, X. et al. End-to-end analysis modeling of vibrational spectroscopy based on deep learning approach. Journal of Chemometrics 34 (2020)

    2020

  47. [57]

    & Heged˝us, P

    Sun, Y ., Brockhauser, S. & Heged˝us, P. Comparing end-to-end machine learning methods for spectra classification. Applied Sciences 11, 11520 (2021)

    2021

  48. [58]

    & Wang, Z

    Chen, Y . & Wang, Z. End-to-end quantitative analysis modeling of near-infrared spec- troscopy based on convolutional neural network. Journal of Chemometrics 33 (2019)

    2019

  49. [59]

    & Bao, J

    Zhang, J., Zhu, X. & Bao, J. Solver-informed neural networks for spectrum reconstruction of colloidal quantum dot spectrometers. Optics Express 28, 33656 (2020)

    2020

  50. [60]

    Galli, M. et al. Light scattering and fano resonances in high-q photonic crystal nanocavities. Applied Physics Letters 94 (2009)

    2009

  51. [61]

    Jin, J. et al. Topologically enabled ultrahigh-q guided resonances robust to out-of-plane scattering. Nature 574, 501–504 (2019). 24

    2019

  52. [62]

    & Zhen, B

    Yin, X., Jin, J., Solja ˇci´c, M., Peng, C. & Zhen, B. Observation of topologically enabled unidirectional guided resonances. Nature 580, 467–471 (2020)

    2020

  53. [63]

    EfficientNet, 109–123 (Apress, 2021)

    Koonce, B. EfficientNet, 109–123 (Apress, 2021)

    2021

  54. [64]

    & Hinkley, D

    Davison, A. & Hinkley, D. Bootstrap Methods and Their Application. Bootstrap Methods and Their Application (Cambridge University Press, 1997)

    1997

  55. [65]

    Peng, J., Nie, W., Li, T. & Xu, J. An end-to-end doa estimation method based on deep learning for underwater acoustic array. In OCEANS 2022, Hampton Roads , 1–6 (IEEE, 2022)

    2022

  56. [66]

    & Zhang, Y

    Gupta, A., Mishra, R. & Zhang, Y . Senglean: An end-to-end deep learning approach for super-resolution of sentinel-2 multiresolution multispectral images. IEEE Transactions on Geoscience and Remote Sensing 62, 1–19 (2024)

    2024

  57. [67]

    & Benini, S

    Signoroni, A., Savardi, M., Baronio, A. & Benini, S. Deep learning meets hyperspectral image analysis: A multidisciplinary review. Journal of Imaging 5, 52 (2019)

    2019

  58. [68]

    & Noda, S

    Liang, Y ., Peng, C., Sakai, K., Iwahashi, S. & Noda, S. Three-dimensional coupled-wave analysis for square-lattice photonic crystal surface emitting lasers with transverse-electric polarization: finite-size effects. Optics Express 20, 15945 (2012). 25 Input spectrum Dataset N...

    2012

Pith tools

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