pith. sign in

arxiv: 2606.07050 · v1 · pith:YOODO7FCnew · submitted 2026-06-05 · 📡 eess.SP

Optimized Sampling of Angle-Resolved Scatterometry Data Using End-to-End Compressed Learning Model for Nanograss Deficiency Detection

Pith reviewed 2026-06-27 21:19 UTC · model grok-4.3

classification 📡 eess.SP
keywords angle-resolved scatterometrycompressed learningnanograss deficiency detectionsampling optimizationend-to-end trainingconvolutional neural networkzinc oxide nanostructures
0
0 comments X

The pith

A learnable latitude-based sampling layer lets a neural network classify five levels of nanograss deficiency from angle-resolved scatterometry data using up to 90 percent fewer angular points.

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

This paper establishes that an end-to-end framework jointly optimizes a sampling layer and convolutional classifier to detect vacancy deficiencies in zinc oxide nanograss from ARS images. The sampling layer learns which latitudinal regions carry the needed information by exploiting the physical structure of the scatter patterns, cutting the required measurements sharply while keeping classification stable under noise. A sympathetic reader would care because dense angular sampling currently limits fast, in-line quality inspection during nanostructure manufacturing. The reported results show that performance stays close to the full-image baseline even at extreme reductions, with GAN augmentation aiding training on small datasets.

Core claim

The proposed framework integrates a learnable latitude-based sampling layer with a convolutional neural network, allowing sampling and classification to be jointly optimized during training. The sampling layer exploits the physical structure of ARS patterns and learns informative latitudinal regions, which reduces the sampling search space and improves convergence. Evaluation results show that the proposed approach achieves high and stable deficiency-level classification performance under different noise conditions, matching full-image performance with up to 90% fewer angular sampling points while the accuracy drop stays below 10 percentage points even at 99.7% reduction.

What carries the argument

learnable latitude-based sampling layer integrated with a CNN that selects informative angular regions from ARS patterns and trains jointly with the classifier

If this is right

  • The model reaches 94.2% accuracy for five-level deficiency classification and 98.6% for binary deficient versus non-deficient separation on full images.
  • Classification performance remains high and stable under added noise even after large reductions in the number of angular sampling points.
  • Pretraining on GAN-generated data produces fast convergence after only a few fine-tuning epochs on limited real samples.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same latitude-based sampling idea could be applied to other rotationally symmetric measurement domains where acquisition time is a bottleneck.
  • Hardware implementations that physically steer the detector to the learned angles might cut inspection time in manufacturing lines.
  • Retraining the sampling layer on data from different nanostructure materials would test whether the learned regions depend mainly on geometry or on material-specific scattering.

Load-bearing premise

The latitude-based sampling layer can learn informative regions from the physical structure of ARS patterns that generalize to new samples under varying noise conditions.

What would settle it

Testing the trained sampling points on a fresh set of ARS images from different nanograss samples and observing more than a 15-point accuracy drop relative to full sampling for five-level classification would falsify the generalization result.

Figures

Figures reproduced from arXiv: 2606.07050 by Armin Dekorsy, Carsten Bockelmann, Mehdi Abdollahpour.

Figure 2
Figure 2. Figure 2: Vertical angle (𝛼) and horizontal angle (𝛽) in ARS sphere B. End-to-End Classification and Sampling Model Neural networks are well-suited for ARS deficiency detection because scattering patterns are high-dimensional and contain complex, non-linear features that are difficult to model with hand-crafted methods. Therefore, we use a neural network to develop an end-to-end model that jointly optimizes sampling… view at source ↗
Figure 3
Figure 3. Figure 3: Workflow of the end-to-end compressed learning framework [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mean intensity distribution across different latitudes in ARS images. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Straight-Through Estimator (STE) applied to the sampling layer in [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Conditional WGAN-GP architecture for generating ARS images corresponding to five nanosurface deficiency levels. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Classification accuracy for different numbers of sampling points under [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Confusion matrices for deficiency level classification on the noiseless dataset using different maximum numbers of sampling points (50, 100, 200, and [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison between ADDA and GAN-generated scatterometry images for the five deficiency classes (0%, 10%, 20%, 40%, and 60%). For each [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Test accuracy over 500 fine-tuning epochs, evaluated across five folds of the ADDA-generated dataset for different maximum numbers of sampling points. [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Heatmap of Frequently Sampled Regions from extremely sparse sampling toward moderate sampling [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
read the original abstract

Reliable inspection of nanosurfaces is essential to ensure the quality of nanostructure manufacturing. Angle-resolved scatterometry provides a non-invasive inspection method that can be used in-line but often suffers from long acquisition times due to dense angular sampling. This paper addresses the data acquisition challenge by proposing an end-to-end compressed learning framework for 5-level vacancy deficiency detection in zinc oxide nanograss using ARS images. The proposed framework integrates a learnable latitude-based sampling layer with a convolutional neural network, allowing sampling and classification to be jointly optimized during training. The sampling layer exploits the physical structure of ARS patterns and learns informative latitudinal regions, which reduces the sampling search space and improves convergence. Evaluation results show that the proposed approach achieves high and stable deficiency-level classification performance under different noise conditions. Using full ARS images, the model achieves 94.2% accuracy for five-level deficiency classification and 98.6% accuracy for separating deficient from non-deficient nanosurfaces. The proposed sampling model matches full-image performance while using up to 90% fewer angular sampling points. Even when sampling points are reduced by 99.7%, the classification accuracy decreases by less than 10 percentage points. To further improve training with limited data, we also studied a GAN-based augmentation approach and used GAN-generated data for model pretraining. Augmented data resulted in fast convergence within only a few fine-tuning epochs.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The manuscript proposes an end-to-end compressed learning framework that jointly optimizes a learnable latitude-based sampling layer with a CNN classifier for 5-level vacancy deficiency detection in ZnO nanograss using angle-resolved scatterometry (ARS) images. It reports 94.2% accuracy for 5-class deficiency classification and 98.6% for deficient vs. non-deficient on full images, with the sampling model matching full-image performance at up to 90% fewer points and <10 pp accuracy drop at 99.7% reduction under varying noise; GAN augmentation is used to address limited data.

Significance. If the empirical results hold under proper validation, the joint optimization of physically motivated latitude sampling with classification could enable substantially faster in-line ARS inspection while preserving accuracy, addressing a key practical limitation in nanomanufacturing quality control. The end-to-end training and use of GAN pretraining for data scarcity are methodological strengths worth noting.

major comments (3)
  1. [Abstract / Experimental evaluation] Abstract and experimental evaluation section: the reported accuracies (94.2%, 98.6%) and stability claims under noise lack any mention of dataset size, train/validation/test split, cross-validation protocol, or number of independent runs, which are load-bearing for assessing whether the <10 pp drop at 99.7% reduction is statistically reliable or generalizable.
  2. [Results] Results on sampling reduction: the claim that the model 'matches full-image performance' while using 90% fewer points is presented without baseline comparisons (e.g., random sampling, fixed uniform latitudes, or non-learnable compressed sensing methods) or error bars, making it impossible to determine if the latitude layer contributes beyond what simpler strategies achieve.
  3. [Method / Training details] Training procedure for the latitude sampling layer: no indication is given whether the noise conditions used in the reported 'different noise conditions' tests were held out from the joint optimization of the sampler and CNN; if test noise was seen during training, the robustness result does not demonstrate out-of-distribution generalization as claimed.
minor comments (2)
  1. [Method] Notation for the latitude sampling layer should be defined more explicitly (e.g., how the learned latitudes map to physical ARS angles) to aid reproducibility.
  2. [Abstract] The 5 deficiency levels are referenced but not enumerated or justified in the abstract; a brief definition or reference to the physical meaning would improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and will revise the manuscript accordingly to improve clarity and rigor.

read point-by-point responses
  1. Referee: [Abstract / Experimental evaluation] Abstract and experimental evaluation section: the reported accuracies (94.2%, 98.6%) and stability claims under noise lack any mention of dataset size, train/validation/test split, cross-validation protocol, or number of independent runs, which are load-bearing for assessing whether the <10 pp drop at 99.7% reduction is statistically reliable or generalizable.

    Authors: We agree these details are necessary. In the revision we will add the dataset size, train/validation/test split ratios, cross-validation protocol, and results from multiple independent runs (with standard deviations) to support the reported accuracies and noise robustness claims. revision: yes

  2. Referee: [Results] Results on sampling reduction: the claim that the model 'matches full-image performance' while using 90% fewer points is presented without baseline comparisons (e.g., random sampling, fixed uniform latitudes, or non-learnable compressed sensing methods) or error bars, making it impossible to determine if the latitude layer contributes beyond what simpler strategies achieve.

    Authors: We accept this point. The revised manuscript will include comparisons against random sampling, fixed uniform latitudes, and non-learnable compressed sensing baselines, along with error bars from repeated runs to demonstrate the contribution of the learnable latitude layer. revision: yes

  3. Referee: [Method / Training details] Training procedure for the latitude sampling layer: no indication is given whether the noise conditions used in the reported 'different noise conditions' tests were held out from the joint optimization of the sampler and CNN; if test noise was seen during training, the robustness result does not demonstrate out-of-distribution generalization as claimed.

    Authors: We will add explicit clarification in the methods section that the noise levels used for the reported robustness tests were held out from training. The joint optimization used either clean data or a disjoint set of noise conditions, enabling the out-of-distribution evaluation. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical end-to-end training with independent evaluation

full rationale

The paper describes a jointly optimized latitude sampling layer plus CNN trained on ARS images for deficiency classification. Performance numbers (94.2% full-image accuracy, <10 pp drop at 99.7% reduction) are reported as outcomes of that training process, not as quantities derived from the model's own equations or from self-citations that close the loop. No load-bearing step equates a claimed prediction to a fitted input by construction, imports uniqueness from prior author work, or renames a known result. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that ARS patterns possess exploitable latitudinal structure allowing reduced sampling without loss of classification power, plus standard supervised learning assumptions such as representative training data.

axioms (1)
  • domain assumption ARS patterns have physical latitudinal structure that a learnable sampling layer can exploit to select informative regions
    Invoked in the description of the sampling layer that reduces search space by exploiting ARS pattern structure

pith-pipeline@v0.9.1-grok · 5795 in / 1295 out tokens · 23447 ms · 2026-06-27T21:19:26.974650+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

45 extracted references · 6 linked inside Pith

  1. [1]

    Echlin,Handbook of sample preparation for scanning electron mi- croscopy and X-ray microanalysis

    P. Echlin,Handbook of sample preparation for scanning electron mi- croscopy and X-ray microanalysis. Springer Science & Business Media, 2011

  2. [2]

    Characterization of polymer surfaces with atomic force microscopy,

    S. N. Magonov and D. H. Reneker, “Characterization of polymer surfaces with atomic force microscopy,”Annual Review of Materials Science, vol. 27, no. 1, pp. 175–222, 1997

  3. [3]

    Evaluation of a novel ultra small target technology supporting on-product overlay measurements,

    H.-J. H. Smilde, A. den Boef, M. Kubis, M. Jak, M. van Schijndel, A. Fuchs, M. van der Schaar, S. Meyer, S. Morgan, J. Wuet al., “Evaluation of a novel ultra small target technology supporting on-product overlay measurements,” inMetrology, Inspection, and Process Control for Microlithography XXVI, vol. 8324. SPIE, 2012, pp. 431–438

  4. [4]

    Fast characterization of moving samples with nano-textured surfaces,

    M. H. Madsen, P.-E. Hansen, M. Zalkovskij, M. Karamehmedovi ´c, and J. Garnæs, “Fast characterization of moving samples with nano-textured surfaces,”Optica, vol. 2, no. 4, pp. 301–306, 2015

  5. [5]

    Model-assisted measuring method for periodical sub-wavelength nanostructures,

    G. Alexe, A. Tausendfreund, D. St¨obener, and A. Fischer, “Model-assisted measuring method for periodical sub-wavelength nanostructures,”Applied Optics, vol. 57, no. 1, pp. 92–101, 2018

  6. [6]

    Overview of scatterometry applications in high volume silicon manufacturing,

    C. Raymond, “Overview of scatterometry applications in high volume silicon manufacturing,” inAIP Conference Proceedings, vol. 788, no. 1. American Institute of Physics, 2005, pp. 394–402

  7. [7]

    Overlay measurement using angular scatterometer for the capability of integrated metrology,

    C.-H. Ko and Y.-S. Ku, “Overlay measurement using angular scatterometer for the capability of integrated metrology,”Optics Express, vol. 14, no. 13, pp. 6001–6010, 2006

  8. [8]

    Scatterometry for advanced process control in semiconductor device manufacturing,

    A. den Boef, H. Cramer, S. Petra, B. O. F. Auer, J. Schmetz-Schagen, A. Koolen, O. van Loon, G. de Gersem, P. Klandermans, and E. Bakker, “Scatterometry for advanced process control in semiconductor device manufacturing,” inFifth International Conference on Optical and Pho- tonics Engineering, vol. 10449. SPIE, 2017, pp. 201–208

  9. [9]

    In situ characterization of nanowire dimensions and growth dynamics by optical reflectance,

    M. Heurlin, N. Anttu, C. Camus, L. Samuelson, and M. T. Borgstrom, “In situ characterization of nanowire dimensions and growth dynamics by optical reflectance,”Nano letters, vol. 15, no. 5, pp. 3597–3602, 2015

  10. [10]

    Specular spectroscopic scatterometry,

    X. Niu, N. Jakatdar, J. Bao, and C. J. Spanos, “Specular spectroscopic scatterometry,”IEEE Transactions on Semiconductor Manufacturing, vol. 14, no. 2, pp. 97–111, 2001

  11. [11]

    Angle resolved optical metrology,

    R. M. Silver, B. M. Barnes, A. Heckert, R. Attota, R. Dixson, and J. Jun, “Angle resolved optical metrology,” inMetrology, Inspection, and Process Control for Microlithography XXII, vol. 6922. SPIE, 2008, pp. 590–601

  12. [12]

    Imaging scatterometry for flexible measurements of patterned areas,

    M. H. Madsen and P.-E. Hansen, “Imaging scatterometry for flexible measurements of patterned areas,”Optics express, vol. 24, no. 2, pp. 1109–1117, 2016

  13. [13]

    Scatterometry—fast and robust measurements of nano-textured surfaces,

    ——, “Scatterometry—fast and robust measurements of nano-textured surfaces,”Surface Topography: Metrology and Properties, vol. 4, no. 2, p. 023003, 2016

  14. [14]

    In-process measuring procedure for sub-100 nm structures,

    M. Zimmermann, A. Tausendfreund, S. Patzelt, G. Goch, S. Kieß, M. Shaikh, M. Gr´egoire, and S. Simon, “In-process measuring procedure for sub-100 nm structures,”Journal of Laser Applications, vol. 24, no. 4, 2012. ABDOLLAHPOURet al.: OPTIMIZED SCATTEROMETRY SAMPLING FOR NANOGRASS DEFICIENCY DETECTION 13

  15. [15]

    Robust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information,

    E. J. Cand `es, J. Romberg, and T. Tao, “Robust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information,” IEEE Transactions on information theory, vol. 52, no. 2, pp. 489–509, 2006

  16. [16]

    Compressed sensing,

    D. L. Donoho, “Compressed sensing,”IEEE Transactions on information theory, vol. 52, no. 4, pp. 1289–1306, 2006

  17. [17]

    Learned primal-dual reconstruction,

    J. Adler and O. ¨Oktem, “Learned primal-dual reconstruction,”IEEE transactions on medical imaging, vol. 37, no. 6, pp. 1322–1332, 2018

  18. [18]

    Compressed learning: Uni- versal sparse dimensionality reduction and learning in the measurement domain,

    R. Calderbank, S. Jafarpour, and R. Schapire, “Compressed learning: Uni- versal sparse dimensionality reduction and learning in the measurement domain,”preprint, 2009

  19. [19]

    Compressed learning: A deep neural network approach,

    A. Adler, M. Elad, and M. Zibulevsky, “Compressed learning: A deep neural network approach,”arXiv preprint arXiv:1610.09615, 2016

  20. [20]

    Compressed learning for tactile object recognition,

    B. Hollis, S. Patterson, and J. Trinkle, “Compressed learning for tactile object recognition,”IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 1616–1623, 2018

  21. [21]

    Direct inference on compressive measurements using convolutional neural networks,

    S. Lohit, K. Kulkarni, and P. Turaga, “Direct inference on compressive measurements using convolutional neural networks,” in2016 IEEE international conference on image processing (ICIP). IEEE, 2016, pp. 1913–1917

  22. [22]

    Compressed learning for image classification: A deep neural network approach,

    E. Zisselman, A. Adler, and M. Elad, “Compressed learning for image classification: A deep neural network approach,” inHandbook of Numer- ical Analysis. Elsevier, 2018, vol. 19, pp. 3–17

  23. [23]

    Enabling on-device classification of ecg with compressed learning for health iot,

    W. Li, H. Chu, B. Huang, Y. Huan, L. Zheng, and Z. Zou, “Enabling on-device classification of ecg with compressed learning for health iot,” Microelectronics Journal, vol. 115, p. 105188, 2021

  24. [24]

    Spatial control of defect creation in graphene at the nanoscale,

    A. W. Robertson, C. S. Allen, Y. A. Wu, K. He, J. Olivier, J. Neethling, A. I. Kirkland, and J. H. Warner, “Spatial control of defect creation in graphene at the nanoscale,”Nature communications, vol. 3, no. 1, p. 1144, 2012

  25. [25]

    The discrete-dipole-approximation code adda: Capabilities and known limitations,

    M. A. Yurkin and A. G. Hoekstra, “The discrete-dipole-approximation code adda: Capabilities and known limitations,”Journal of Quantitative Spectroscopy and Radiative Transfer, vol. 112, no. 13, pp. 2234–2247, 2011

  26. [26]

    Scatterometric defect measurements–uncertainty assessment by means of a virtual instrument and a statistical analysis,

    T. M. H. Rahman, D. St ¨obener, and A. Fischer, “Scatterometric defect measurements–uncertainty assessment by means of a virtual instrument and a statistical analysis,”Surface Topography: Metrology and Properties, vol. 12, no. 3, p. 035019, 2024

  27. [27]

    Compressed learning for nanosurface deficiency recognition using angle-resolved scatterometry data,

    M. Abdollahpour, C. Bockelmann, T. M. H. Rahman, A. Dekorsy, and A. Fischer, “Compressed learning for nanosurface deficiency recognition using angle-resolved scatterometry data,”IEEE Access, pp. 1–1, 2026

  28. [28]

    Estimating or propagating gradients through stochastic neurons for conditional computation,

    Y. Bengio, N. L ´eonard, and A. Courville, “Estimating or propagating gradients through stochastic neurons for conditional computation,”arXiv preprint arXiv:1308.3432, 2013

  29. [29]

    Machine learning techniques applied for the detection of nanoparticles on surfaces using coherent fourier scatterome- try,

    D. Kolenov and S. Pereira, “Machine learning techniques applied for the detection of nanoparticles on surfaces using coherent fourier scatterome- try,”Optics Express, vol. 28, no. 13, pp. 19 163–19 186, 2020

  30. [30]

    Advanced euv resist characterization using scatterometry and machine learning,

    D. Schmidt, K. Petrillo, M. Breton, J. Fullam, R. Koret, I. Turovets, and A. Cepler, “Advanced euv resist characterization using scatterometry and machine learning,” in2021 32nd Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC). IEEE, 2021, pp. 1–4

  31. [31]

    Machine learning for scattering data: strategies, perspec- tives and applications to surface scattering,

    A. Hinderhofer, A. Greco, V. Starostin, V. Munteanu, L. Pithan, A. Gerlach, and F. Schreiber, “Machine learning for scattering data: strategies, perspec- tives and applications to surface scattering,”Applied Crystallography, vol. 56, no. 1, pp. 3–11, 2023

  32. [32]

    Scaling laws for neural language models,

    J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei, “Scaling laws for neural language models,”arXiv preprint arXiv:2001.08361, 2020

  33. [33]

    A survey on image data augmentation for deep learning,

    C. Shorten and T. M. Khoshgoftaar, “A survey on image data augmentation for deep learning,”Journal of big data, vol. 6, no. 1, pp. 1–48, 2019

  34. [34]

    Discrete-dipole approximation for scattering calculations,

    B. T. Draine and P. J. Flatau, “Discrete-dipole approximation for scattering calculations,”JOSA A, vol. 11, no. 4, pp. 1491–1499, 1994

  35. [35]

    The discrete dipole approximation for simulation of light scattering by particles much larger than the wave- length,

    M. A. Yurkin and A. G. Hoekstra, “The discrete dipole approximation for simulation of light scattering by particles much larger than the wave- length,”Journal of Quantitative Spectroscopy and Radiative Transfer, vol. 106, no. 1-3, pp. 558–589, 2007

  36. [36]

    Accurate computation of electric field enhancement factors for metallic nanoparticles using the discrete dipole approximation,

    A. E. DePrince and R. J. Hinde, “Accurate computation of electric field enhancement factors for metallic nanoparticles using the discrete dipole approximation,”Nanoscale research letters, vol. 5, pp. 592–596, 2010

  37. [37]

    Generative adversarial networks,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial networks,” Communications of the ACM, vol. 63, no. 11, pp. 139–144, 2020

  38. [38]

    Generative adversarial nets,

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial nets,” in Advances in Neural Information Processing Systems, Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, and K. Weinberger, Eds., vol. 27. Curran Associates, Inc., 2014

  39. [39]

    Conditional generative adversarial nets,

    M. Mirza and S. Osindero, “Conditional generative adversarial nets,”arXiv preprint arXiv:1411.1784, 2014

  40. [40]

    Improved training of wasserstein gans,

    I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, and A. C. Courville, “Improved training of wasserstein gans,”Advances in neural information processing systems, vol. 30, 2017

  41. [41]

    Wasserstein gan,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein gan,” 2017. [Online]. Available: https://arxiv.org/abs/1701.07875

  42. [42]

    Adam: A method for stochastic optimization,

    D. P. Kingma, “Adam: A method for stochastic optimization,”arXiv preprint arXiv:1412.6980, 2014

  43. [43]

    Angle-resolved optical characterization of a plasmonic trian- gular array of elliptical holes in a gold layer,

    M. Angelini, K. Jefimovs, P. Pellacani, D. Kazazis, F. Marabelli, and F. Floris, “Angle-resolved optical characterization of a plasmonic trian- gular array of elliptical holes in a gold layer,”Optics, vol. 5, no. 1, pp. 195–206, 2024

  44. [44]

    Hyperspectral imag- ing for high-throughput, spatially resolved spectroscopic scatterometry of silicon nanopillar arrays,

    B. Gawlik, C. Barrera, E. T. Yu, and S. Sreenivasan, “Hyperspectral imag- ing for high-throughput, spatially resolved spectroscopic scatterometry of silicon nanopillar arrays,”Optics express, vol. 28, no. 10, pp. 14 209– 14 221, 2020

  45. [45]

    Efficient rigorous coupled-wave analysis simulation of mueller matrix ellipsometry of three-dimensional multilayer nanostructures,

    H.-L. Pham, T. Alcaire, S. Soulan, D. Le Cunff, and J.-H. Tortai, “Efficient rigorous coupled-wave analysis simulation of mueller matrix ellipsometry of three-dimensional multilayer nanostructures,”Nanomaterials, vol. 12, no. 22, p. 3951, 2022. Mehdi Abdollahpourreceived the B.Sc. degree in Electrical Engineering (Electronics) and the M.Sc. degree in Biom...