Pith. sign in

REVIEW 4 major objections 6 minor 41 references

In-situ and Non-contact Etch Depth Prediction in Plasma Etching via Machine Learning (ANN & BNN) and Digital Image Colorimetry

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · 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 plasma etch depth can be read from the RGB values of a digital photograph with a validation RMSE around 3.2 nm, and that a Bayesian version of the network supplies trustworthy uncertainty intervals for each prediction.

desk verdict A plausible proof-of-concept that RGB photos can predict SiO2 etch depth, undercut by an unsupported 'in-situ' claim and a very thin statistical basis. read the letter →

arxiv 2505.03826 v1 pith:372NZVSK submitted 2025-05-03 cs.CV cs.AI

Minji Kang , Seongho Kim , Eunseo Go , Donghyeon Paek , Geon Lim , Muyoung Kim , Soyeun Kim , Sung Kyu Jang
show 5 more authors
Min Sup Choi Woo Seok Kang Jaehyun Kim Jaekwang Kim Hyeong-U Kim
This is my paper · ORCID
classification cs.CVcs.AI
keywords plasmaetchingdigitalimagecolorimetryetchdepthpredictionartificialneuralnetworkBayesianuncertaintyquantificationsilicondioxideopticalemissionspectroscopy
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 tries to establish that plasma etch depth can be monitored without touching the wafer or stopping the chamber, using only the color of the wafer's surface. The authors train an ANN to map RGB values extracted from photographs of etched SiO2 coupons to the etch depth recorded by ellipsometry, and report a held-out MSE of 10.38 nm² (about 3.2 nm RMSE) on seven validation conditions. They also train a Bayesian neural network with Monte Carlo dropout so that each prediction comes with an uncertainty interval; for the image-based model, true values fall inside the ±1σ interval 63.16% of the time and outside ±2σ only 1.97% of the time. If the claim holds, thickness monitoring in plasma etching could be done with a camera and a trained network instead of a separate ex-situ measurement step.

What carries the argument

Digital image colorimetry (DIC) is the central object: a photograph of the etched coupon on a white background is reduced to average red, green, and blue intensities between 0 and 255 over a representative rectangle, and those three numbers are the ANN's input features. The RGB-to-etch-depth mapping is carried by a feedforward ANN with one 32-neuron ReLU hidden layer and 20% dropout, trained with the Adam optimizer and weight decay. The BNN variant is the same network with Monte Carlo dropout: 50 stochastic forward passes per input produce a predictive distribution whose mean is the point prediction and whose standard deviation defines the uncertainty interval. Coverage analysis, which counts how often true etch depths fall in the ±1σ and ±2σ bands, is the mechanism the paper uses to argue that the uncertainty estimates are reliable.

What would settle it

Photograph a wafer of known etch depth both on a white background and through a plasma-chamber viewport under representative process illumination, and compare the RGB triplets for the same physical state. If the two RGB distributions differ by more than the pixel-to-pixel variation in the training images, the current model cannot support the claimed in-situ use; alternatively, capture the 84-condition image set through the viewport and check whether the validation MSE stays near 10.38 nm².

Watch

Extended reading notes

Core claim

The central claim is that the RGB triplet of a digital image of a dielectric surface carries enough information about ellipsometric etch depth that an ANN can predict that depth without knowing pressure, gas flow, or RF power. With 84 etched coupon conditions and a seven-condition validation split, the image-based ANN achieves a validation MSE of 10.38 nm², versus 113.0 nm² for a linear RGB model, while the process-parameter ANN achieves 7.33 nm² versus 33.94 nm² for a linear parameter model. The BNN extension, implemented by treating each of nine measurement points as a separate sample and applying Monte Carlo dropout at inference, gives coverage of 63.16% within ±1σ and 34.87% within ±2σ for image inputs, and 68.25% and 23.81% for process-parameter inputs, with 7.94% outside ±2σ in the latter case. The paper reads these numbers as evidence that non-contact RGB monitoring is a viable, cost-effective alternative to conventional ex-situ analysis.

Load-bearing premise

The load-bearing premise is that the color-to-etch-depth relationship learned from ex-situ photos taken on a white background under fixed lighting still holds when the method is called in-situ, meaning images taken through a chamber viewport during plasma processing; if viewport transmission, plasma emission, or viewing geometry changes the RGB values for the same physical etch state, the reported accuracy and uncertainty coverage will not transfer.

Editorial extensions

If this is right

  • A camera-based monitor could flag etch depth in real time without transferring the wafer out of vacuum, reducing measurement delay and contamination risk.
  • Because RGB inputs contain no explicit process parameters, a trained network may transfer across plasma recipes that produce the same surface colors.
  • The BNN's uncertainty intervals give an operator a criterion for trusting or rejecting a prediction, such as flagging drift when coverage degrades.
  • The same ANN/BNN pipeline could be retrained for other insulating films, such as silicon nitride, using the same ellipsometer-plus-photograph data collection.

Reading between the lines

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

  • The claimed independence of the RGB-to-etch-depth mapping from pressure, CF4 flow, and RF power rests on one 84-condition grid; a stronger test would train on some pressure/flow/power combinations and test on others, since the seven validation points all come from the same grid.
  • Inside a real chamber, plasma optical emission contaminates the image signal; a natural extension the paper does not test is to subtract or normalize plasma emission lines before extracting RGB values.
  • If the color signal is dominated by thin-film interference, the same mapping should work for other transparent dielectrics, and could be checked on a staircase etch-depth calibration coupon with known thickness steps.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The manuscript reports machine-learning models for predicting the remaining SiO2 thickness after plasma etching. Eighty-four coupon wafers were etched in an ICP-RIE tool over a grid of chamber pressure, CF4 flow, and top power; thickness was measured by spectroscopic ellipsometry and surface images were captured with an iPhone 12 on a white background. The authors train an ANN to predict etch depth from the three process parameters, obtaining a held-out MSE of 7.33 nm^2 versus 33.94 nm^2 for a linear model, and an ANN trained on RGB values from the digital images, obtaining an MSE of 10.38 nm^2 versus 113.0 nm^2. They also apply a Monte Carlo Dropout BNN to the repeated measurements and report coverage statistics for predictive intervals. The paper concludes that this integration offers a viable, cost-effective alternative for real-time, in-situ, and non-invasive monitoring of plasma etching.

Significance. If the claims are supported, the RGB-based ANN is a genuinely interesting low-cost alternative to ellipsometric thickness measurement, and the BNN uncertainty quantification would strengthen its practical value for process monitoring. The experimental design is a clear strength: 84 process conditions with ellipsometry mapping, OES-based plasma stability checks, held-out validation, and direct comparison against a linear baseline. The held-out MSEs genuinely support the narrower claim that etch depth can be predicted from ex-situ smartphone photographs of coupons on a white background. However, the paper's headline claim of in-situ monitoring is not supported by the experimental protocol, and the validation statistics rest on a single small split with an unspecified BNN data-splitting procedure. The central feasibility result is defensible if reframed as ex-situ DIC-based prediction, but the current manuscript overclaims.

major comments (4)
  1. [Abstract, Fig. 1, §2, §3.1] The central 'in-situ' claim is not supported by the experimental protocol. The abstract and Figure 1 state that RGB data are collected 'during the RIE process,' but Section 3.1 is titled 'Training set generation with ex-situ analysis' and Section 2 reports that 'coupon wafers were placed on a white background during image acquisition' with an iPhone 12 after etching. Plasma emission, viewport window transmission, chamber lighting, and viewing geometry all change RGB values for a fixed thickness, and the paper reports no illumination normalization, spectral calibration, or through-viewport images. Consequently, the held-out MSE of 10.38 nm^2 in Table 4 demonstrates interpolation among ex-situ white-background photos, not that the mapping transfers in-situ. Please either add in-chamber validation data or revise the abstract, title, and Figure 1 to claim ex-situ feasibility and explicitly discuss the transfer assumptions.
  2. [§3.4.1, Table 3, Table 4] The validation protocol rests on a single random split of only seven samples for the process-parameter ANN and for the RGB ANN. With n=7, the reported MSEs (7.33 versus 33.94 and 10.38 versus 113.0) are sensitive to the particular split, training seed, and network initialization. No confidence intervals, repeated-seed results, or cross-validation results are given. Please report results over multiple splits or with a leave-one-condition-out scheme, including the mean and standard deviation of the MSE.
  3. [§3.4.1 and §3.4.2, BNN coverage analysis] For the BNN experiments, 756 samples are formed by treating nine repeated ellipsometer measurements as independent, and 152 are held out, but the text does not specify whether the split is at the coupon level or the individual-measurement level. If measurements from the same coupon appear in both training and validation, the quoted coverage (68.25%/23.81%/7.94% and 63.16%/34.87%/1.97%) is inflated by leakage. The DIC-BNN section similarly does not state how repeated RGB measurements were obtained or how the 152 validation samples were selected. Please specify the split, use coupon-level grouping (for example, GroupKFold), and report the average predictive standard deviation alongside the coverage percentages.
  4. [§3.4.2, last paragraph] The statement that the RGB-based model can generalize across plasma processes because it does not use explicit process parameters is too strong. The training data cover one tool, one gas chemistry, and one 300-nm SiO2 film, and RGB values may also encode process-induced surface conditions. Please temper this generalization claim or support it with data from a different tool, chemistry, or film stack.
minor comments (6)
  1. [Section numbering] The manuscript jumps from Section 3.2 to Section 3.4; Section 3.3 is missing and should be renumbered.
  2. [§3.4.2, figure references] In Section 3.4.2, the text refers to 'surface features shown in Figure 5' and 'thickness illustrated in Figure 6,' but the relevant panels appear to be Figures 2 and 3; the cross-references should be corrected.
  3. [Table 2] Table 2 has inconsistent formatting: rows for the 30 mTorr conditions contain only eight entries and the column alignment is unclear; please reformat so all nine thickness columns are visible.
  4. [Eq. (4)] Equation (4) defines the linear model without a bias term even though the surrounding text introduces a bias b_l; this should be reconciled.
  5. [§2, last paragraph] The sentence describing 'in-situ RGB data obtained from DIC' conflicts with the ex-situ image acquisition described earlier in the same section; use 'ex-situ' consistently unless in-chamber images are actually acquired.
  6. [§3.4.1, BNN implementation] The dropout probability and the number of hidden units for the BNN/MC Dropout model are not specified; these settings affect the reported coverage and should be stated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: predictions are evaluated on held-out validation data and no derivation reduces to its own fitted inputs.

full rationale

The paper's derivation chain is a supervised regression: process parameters or RGB triplets are used as inputs, ellipsometer thickness as targets, and all reported errors (process-parameter ANN MSE 7.33, RGB ANN MSE 10.38, BNN coverage fractions of 68.25% and 63.16% inside ±1 sigma) are computed on validation samples excluded from training. Nothing in the equations (Eqs. 1-4) or the experimental protocol defines the target in terms of the fitted model, and no fitted parameter is renamed as a prediction. The self-citations (references 12, 13, 28-30) appear only as background literature for ex-situ analysis and DIC+ML; no central claim or uniqueness argument is imported from those papers. The BNN coverage analysis is an internal consistency check based on MC-dropout posterior samples compared with held-out true values, which is a legitimate model evaluation rather than circular reasoning. The most serious weakness is external validity: RGB images were acquired ex-situ on a white background (Section 2), while the abstract and Figure 1 assert in-situ collection during RIE; no through-viewport or in-chamber RGB data support the in-situ claim. That is an unsupported generalization and a correctness risk, not a circularity, so it does not raise the circularity score.

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

The paper introduces no new physical entities. Its contribution is empirical, and the entries above capture the hand-chosen modeling choices and unstated assumptions on which the reported prediction errors and uncertainty coverage rest.

free parameters (3)
  • ANN architecture and MC Dropout settings = 32 hidden neurons; 20% dropout; 50 stochastic forward passes
    The hidden layer size, dropout probability, and number of MC forward passes are hand-selected without sensitivity analysis; reported MSE and coverage depend on them.
  • Single random train/validation split = 77/7 for the ANN scenarios; 604/152 for the BNN scenario
    All error and coverage metrics come from one arbitrary split with no seed or repeated splits; other splits would yield different numbers.
  • RGB extraction region = unstated rectangle in each image
    Color features come from a manually chosen rectangular region; the location and size are not specified, and no criterion for selection is given.
assumptions (4)
  • domain assumption The 84 experimental conditions are representative of the etch process space of interest.
    The models are trained and tested only within this grid of pressure, CF4 flow, and top power; generalization to other conditions, materials, or etchers is assumed in the discussion.
  • domain assumption MC Dropout with 50 forward passes provides a sufficiently accurate approximation to Bayesian inference.
    The paper adopts Gal and Ghahramani's approximation (ref 32) as the BNN implementation; the uncertainty quality depends on this approximation holding for this model and dataset.
  • domain assumption The nine ellipsometer measurements per wafer are independent samples of measurement noise.
    The BNN treats all 756 thickness measurements as independent data points; if the nine points are spatially correlated, effective sample size is smaller and coverage statistics are overconfident.
  • domain assumption The single RGB triplet from one rectangular region represents the wafer-wide average thickness.
    Etch depth is measured at nine locations, but color is extracted from one region; any spatial non-uniformity is neglected by the DIC model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of In-situ and Non-contact Etch Depth Prediction in Plasma Etching via Machine Learning (ANN & BNN) and Digital Image Colorimetry." pith.science (2026). https://pith.science/paper/372NZVSK

@misc{pith2026250503826,
  author       = {Pith},
  title        = {Pith review of: In-situ and Non-contact Etch Depth Prediction in Plasma Etching via Machine Learning (ANN & BNN) and Digital Image Colorimetry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/372NZVSK}},
  note         = {Machine review of arXiv:2505.03826}
}
read the original abstract

Precise monitoring of etch depth and the thickness of insulating materials, such as Silicon dioxide and silicon nitride, is critical to ensuring device performance and yield in semiconductor manufacturing. While conventional ex-situ analysis methods are accurate, they are constrained by time delays and contamination risks. To address these limitations, this study proposes a non-contact, in-situ etch depth prediction framework based on machine learning (ML) techniques. Two scenarios are explored. In the first scenario, an artificial neural network (ANN) is trained to predict average etch depth from process parameters, achieving a significantly lower mean squared error (MSE) compared to a linear baseline model. The approach is then extended to incorporate variability from repeated measurements using a Bayesian Neural Network (BNN) to capture both aleatoric and epistemic uncertainty. Coverage analysis confirms the BNN's capability to provide reliable uncertainty estimates. In the second scenario, we demonstrate the feasibility of using RGB data from digital image colorimetry (DIC) as input for etch depth prediction, achieving strong performance even in the absence of explicit process parameters. These results suggest that the integration of DIC and ML offers a viable, cost-effective alternative for real-time, in-situ, and non-invasive monitoring in plasma etching processes, contributing to enhanced process stability, and manufacturing efficiency.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages

  1. [1]

    Romero, M.; Sanz, M.; Tanarro, I.; Jiménez, A.; Muñoz, E., J. Phys. D: Appl. Phys. 2010, 43 (49), 495202

    work page 2010

  2. [2]

    H.; Miao, Y.; Zhou, Z.; Wu, Z.; Kong, Z.; Gao, J.; Yang, H.; Ren, Y.; Zhang, Y.; Shi, J., Nanomater

    Radamson, H. H.; Miao, Y.; Zhou, Z.; Wu, Z.; Kong, Z.; Gao, J.; Yang, H.; Ren, Y.; Zhang, Y.; Shi, J., Nanomater. 2024, 14 (10), 837

    work page 2024

  3. [3]

    J.; Kim, K

    Ji, Y. J.; Kim, K. S.; Kim, K. H.; Byun, J. Y.; Yeom, G. Y., Appl. Sci. Converg. Technol 2019, 28 (5), 142-147

    work page 2019

  4. [4]

    Tsai, S.-J.; Wang, C.-L.; Lee, H.-C.; Lin, C.-Y.; Chen, J.-W.; Shiu, H.-W.; Chang, L.-Y.; Hsueh, H.-T.; Chen, H.-Y.; Tsai, J.-Y., Sci. Rep. 2016, 6 (1), 28326

    work page 2016

  5. [5]

    Geng, K.; Chen, D.; Zhou, Q.; Wang, H., Electronics 2018, 7 (12), 416

    work page 2018

  6. [6]

    J.; Franklin, A

    Das, S.; Sebastian, A.; Pop, E.; McClellan, C. J.; Franklin, A. D.; Grasser, T.; Knobloch, T.; Illarionov, Y.; Penumatcha, A. V.; Appenzeller, J., Nat. Electron. 2021, 4 (11), 786- 799

    work page 2021

  7. [7]

    China Inf

    Yan, S.; Cong, Z.; Lu, N.; Yue, J.; Luo, Q., Sci. China Inf. Sci. 2023, 66 (10), 200404

    work page 2023

  8. [8]

    Lim, N.; Efremov, A.; Kwon, K.-H., Thin Solid Films 2019, 685, 97-107

    work page 2019

Show all 41 references
  1. [9]

    J.; Zhang, Z.; Wang, S.; Hudson, E

    Gasvoda, R. J.; Zhang, Z.; Wang, S.; Hudson, E. A.; Agarwal, S., J. Vac. Sci. Technol. A 2020, 38 (5)

    2020

  2. [10]

    Huff, M., Micromachines 2021, 12 (8), 991

    2021

  3. [11]

    Pasquariello, D.; Hedlund, C.; Hjort, K., J.Electrochem. Soc. 2000, 147 (7), 2699

    2000

  4. [12]

    S.; Kim, T.; Kim, H.-U., Appl

    Kim, C.; Lee, S.; Kim, M.; Choi, M. S.; Kim, T.; Kim, H.-U., Appl. Sci. Converg. Technol. 2023, 32 (5), 106-109

    2023

  5. [13]

    K.; Seok, H.; Bak, M

    Kim, C.; Kim, M.; Cho, D.; Kanade, C. K.; Seok, H.; Bak, M. S.; Kim, D.; Kang, W. S.; Kim, T.; Kim, H.-U., Chem. Mater. 2023, 35 (3), 1016-1028

    2023

  6. [14]

    Dragoi, V.; Lindner, P., ECS transactions 2006, 3 (6), 147

    2006

  7. [15]

    W.; Smith, L

    King, S. W.; Smith, L. L.; Barnak, J. P.; Ku, J. -H.; Christman, J. A.; Benjamin, M. C.; Bremser, M. D.; Nemanich, R. J.; Davis, R. F., MRS Online Proceedings Library (OPL) 1995, 395, 739

    1995

  8. [16]

    Fan, Y.; Li, J.; Guo, Y.; Xie, L.; Zhang, G., Measurement 2021, 171, 108829

    2021

  9. [17]

    Pounds, K.; Bao, H.; Luo, Y.; De, J.; Schneider, K.; Correll, M.; Tong, Z., ACS Food Sci. Technol. 2022, 2 (7), 1123-1134

    2022

  10. [18]

    V.; Maldonado -Flores, C.; Gómez -Navarro, C

    Granados-Vega, B. V.; Maldonado -Flores, C.; Gómez -Navarro, C. S.; Warren -Vega, W. M.; Campos-Rodrí guez, A.; Romero-Cano, L. A., Foods 2023, 12 (20), 3829

    2023

  11. [19]

    2015, 184, 23-29

    Masawat, P.; Harfield, A.; Namwong, A., Food Chem. 2015, 184, 23-29

    2015

  12. [20]

    R.; Ding, X., Food Control 2020, 108, 106885

    Huang, J.; Sun, J.; Warden, A. R.; Ding, X., Food Control 2020, 108, 106885

    2020

  13. [21]

    T.; Santos, V

    de Oliveira Krambeck Franco, M.; Suarez, W. T.; Santos, V. B. d., Food Anal. Methods 2017, 10, 508-515

    2017

  14. [22]

    D.; Sakti, L

    Permana, M. D.; Sakti, L. K.; Luthfiah, A.; Firdaus, M. L.; Takei, T.; Eddy, D. R.; Rahayu, I., Trends Sci. 2023, 20 (4), 5149-5149

    2023

  15. [23]

    Bang-iam, N.; Udnan, Y.; Masawat, P., Microchem. J. 2013, 106, 270-275

    2013

  16. [24]

    F.; Aydın, M.; Demir, M

    Aydın, Ö. F.; Aydın, M.; Demir, M. C.; Kahraman, S., Chemometrics Intellig. Lab. Syst. 2025, 257, 105310

    2025

  17. [25]

    L.; Alwi, W.; Trinoveldi, F.; Rahayu, I.; Rahmidar, L.; Warsito, K., Procedia Environ

    Firdaus, M. L.; Alwi, W.; Trinoveldi, F.; Rahayu, I.; Rahmidar, L.; Warsito, K., Procedia Environ. Sci. 2014, 20, 298-304

    2014

  18. [26]

    K.; Datta, S

    Bhatt, S.; Kumar, S.; Gupta, M. K.; Datta, S. K.; Dubey, S. K., Meas. Sci. Technol. 2023, 35 (1), 015030

    2023

  19. [27]

    -A.; Song, J.; Jung, H

    Choi, W.; Shin, J.; Hyun, K. -A.; Song, J.; Jung, H. -I., Biosens. Bioelectron. 2019, 130, 414-419. 20

    2019

  20. [28]

    Lee, S.; Park, Y.; Liu, P.; Kim, M.; Kim, H.-U.; Kim, T., Sensors 2023, 23 (19), 8226

    2023

  21. [29]

    K.; Kim, J.; Kim, S.; Kim, C.; Lee, H.-C.; Kang, W.; Choi, M

    Kang, M.; Jang, S. K.; Kim, J.; Kim, S.; Kim, C.; Lee, H.-C.; Kang, W.; Choi, M. S.; Kim, H.; Kim, H.-U., J. Sens. Actuator Netw. 2024, 13 (6), 75

    2024

  22. [30]

    -U.; Kim, T., Sensors 2023, 23 (12), 5563

    Park, Y.; Liu, P.; Lee, S.; Cho, J.; Joo, E.; Kim, H. -U.; Kim, T., Sensors 2023, 23 (12), 5563

    2023

  23. [31]

    R.; LeSar, R., Annu

    Chernatynskiy, A.; Phillpot, S. R.; LeSar, R., Annu. Rev. Mater. Res. 2013, 43 (1), 157- 182

    2013

  24. [32]

    In Dropout as a bayesian approximation: Representing model uncertainty in deep learning, international conference on machine learning, PMLR: 2016; pp 1050-1059

    Gal, Y.; Ghahramani, Z. In Dropout as a bayesian approximation: Representing model uncertainty in deep learning, international conference on machine learning, PMLR: 2016; pp 1050-1059

    2016

  25. [33]

    Xue, Y.; Cheng, S.; Li, Y.; Tian, L., Optica 2019, 6 (5), 618-629

    2019

  26. [34]

    In Study of plasma system by OES (optical emission spectroscopy), Proc

    Kolpaková, A.; Kudrna, P.; Tichý, M. In Study of plasma system by OES (optical emission spectroscopy), Proc. of 20th Annual Conference of Doctoral Students, 2011; pp 180-185

    2011

  27. [35]

    J., Perkin Elmer Corp 1997, 3 (2), 115

    Charles, B.; Fredeen, K. J., Perkin Elmer Corp 1997, 3 (2), 115

    1997

  28. [36]

    A.; Lichtenberg, A

    Lieberman, M. A.; Lichtenberg, A. J., MRS Bulletin 1994, 30 (12), 899-901

    1994

  29. [37]

    -N.; Bogaerts, A., Plasma Sources Sci

    Zhao, S.-X.; Gao, F.; Wang, Y. -N.; Bogaerts, A., Plasma Sources Sci. Technol. 2012, 22 (1), 015017

    2012

  30. [38]

    W.; Hershkowitz, N., Plasma Sources Sci

    Jenq, J.-S.; Ding, J.; Taylor, J. W.; Hershkowitz, N., Plasma Sources Sci. Technol. 1994, 3 (2), 154

    1994

  31. [39]

    P.; Durocher -Jean, A.; Gangwar, R

    Sainct, F. P.; Durocher -Jean, A.; Gangwar, R. K.; Mendoza Gonzalez, N. Y.; Coulomb e, S.; Stafford, L., plasma 2020, 3 (2), 38-53

    2020

  32. [40]

    Xu, W.; Mao, X.; Zhou, N.; Zhang, Q.-Y.; Peng, B.; Shen, Y., Vacuum 2022, 196, 110785

    2022

  33. [41]

    Suto, M.; Washida, N., J. Chem. Phys. 1983, 78 (3), 1012-1018

    1983

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.