CoChat

Machine learning powered ellipsometry

Open access
Liu, Jinchao, Zhang, Di, Yu, Dianqiang, Ren, Mengxin, Xu, Jingjun · 2021 · Light: Science & Applications

The DOI matches its Crossref record; 57 of the 59 references it cites were found in a registry or an index.

Abstract

Ellipsometry is a powerful method for determining both the optical constants and thickness of thin films. For decades, solutions to ill-posed inverse ellipsometric problems require substantial human–expert intervention and have become essentially human-in-the-loop trial-and-error processes that are not only tedious and time-consuming but also limit the applicability of ellipsometry. Here, we demonstrate a machine learning based approach for solving ellipsometric problems in an unambiguous and fully automatic manner while showing superior performance. The proposed approach is experimentally validated by using a broad range of films covering categories of metals, semiconductors, and dielectrics. This method is compatible with existing ellipsometers and paves the way for realizing the automatic, rapid, high-throughput optical characterization of films.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41377-021-00482-0

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Principle of R, T assisted ellipsometry. a In ellipsometry, light Ei is obliquely incident onto a thin film with unknown (n, κ, d), which is on a substrate with known optical constants. Multiple beam interference occurs in the film, as shown in the inset. The orthogonal polarization components (Ep and Es) have unequal reflection coefficients rp and rs, leading to elliptically polarized reflection Er. p and s refer to polarizations with electric fields parallel and normal to the plane of incidence, respectively. b The ellipsometric angles Ψ and Δ are defined by, tan Ψ eiΔ = rp/rs, as denoted by F. The total transmittance T and reflectance R are calculated by averaging the contributions from the p- and s-polarizations, denoted by G. The physical process from (n, κ, d) to (Ψ, Δ, R, T) has been well studied and can be modeled accurately as functions F and G. On the other hand, the inverse, i.e., analytically inferring (n, κ, d) from (Ψ, Δ, R, T), is generally impossible, and iterative fitting techniques have therefore been developed to numerically search a set of (n, κ, d) that best fits the experimental data. Solid lines denote experimentally measured (Ψ, Δ, R, T), dashed lines present results recovered by the tentative (n, κ, d) solution
Fig. 1 Principle of R, T assisted ellipsometry. a In ellipsometry, light Ei is obliquely incident onto a thin film with unknown (n, κ, d), which is on a substrate with known optical constants. Multiple beam interference occurs in the film, as shown in the inset. The orthogonal polarization components (Ep and Es) have unequal reflection coefficients rp and rs, leading to elliptically polarized reflection Er. p and s refer to polarizations with electric fields parallel and normal to the plane of incidence, respectively. b The ellipsometric angles Ψ and Δ are defined by, tan Ψ eiΔ = rp/rs, as denoted by F. The total transmittance T and reflectance R are calculated by averaging the contributions from the p- and s-polarizations, denoted by G. The physical process from (n, κ, d) to (Ψ, Δ, R, T) has been well studied and can be modeled accurately as functions F and G. On the other hand, the inverse, i.e., analytically inferring (n, κ, d) from (Ψ, Δ, R, T), is generally impossible, and iterative fitting techniques have therefore been developed to numerically search a set of (n, κ, d) that best fits the experimental data. Solid lines denote experimentally measured (Ψ, Δ, R, T), dashed lines present results recovered by the tentative (n, κ, d) solutionCC BY · PubMed Central
Fig. 2 SUNDIAL for ellipsometric analysis. This method consists of inverse and forward neural modules (encircled by rectangles). The inverse module is trained to perform γF⁻¹+(1−γ)G⁻¹, and the forward module is trained to be surrogates for the functions F and G. The modules are joined together in a loop manner, as indicated by the gray arrows. A novel iterative inference strategy based on stochastic gradient descent, indicated by the loops enclosed by gray and yellow arrows, is proposed to allow training on analytically generated simulated data and inference on real-world experimental data
Fig. 2 SUNDIAL for ellipsometric analysis. This method consists of inverse and forward neural modules (encircled by rectangles). The inverse module is trained to perform γF⁻¹+(1−γ)G⁻¹, and the forward module is trained to be surrogates for the functions F and G. The modules are joined together in a loop manner, as indicated by the gray arrows. A novel iterative inference strategy based on stochastic gradient descent, indicated by the loops enclosed by gray and yellow arrows, is proposed to allow training on analytically generated simulated data and inference on real-world experimental dataCC BY · PubMed Central
Fig. 3 Results of SUNDIAL on Au, TiO2, and Si. a, b Measured Ψ, Δ, R, and T spectra are presented by empty circles. Model-generated spectra, drawn as solid lines, are calculated using (n, κ, d) by the SUNDIAL method and forward functions F and G. Each column corresponds to a material. c The (n, κ, d) retrieved by SUNDIAL
Fig. 3 Results of SUNDIAL on Au, TiO2, and Si. a, b Measured Ψ, Δ, R, and T spectra are presented by empty circles. Model-generated spectra, drawn as solid lines, are calculated using (n, κ, d) by the SUNDIAL method and forward functions F and G. Each column corresponds to a material. c The (n, κ, d) retrieved by SUNDIALCC BY · PubMed Central
Fig. 4 Comparison of SUNDIAL with the traditional fitting technique. a Residual on three typical materials of SUNDIAL (solid lines) and the traditional fitting technique (dashed lines). Both methods fit (Ψ, Δ) very well, while the traditional fitting method shows a significant error in fitting (R, T). b RMSE of the compared methods on more tested materials. SUNDIAL (solid dots) significantly outperformed the traditional fitting method (empty diamonds) in terms of overall accuracy. RMSEΨΔ = RMSEΨ + RMSEΔ, and RMSERT = RMSER + RMSET. c Effect of γ. Experiments with different γ values are conducted, and the RMSEs are shown here. Tungsten (W) is used here as an example
Fig. 4 Comparison of SUNDIAL with the traditional fitting technique. a Residual on three typical materials of SUNDIAL (solid lines) and the traditional fitting technique (dashed lines). Both methods fit (Ψ, Δ) very well, while the traditional fitting method shows a significant error in fitting (R, T). b RMSE of the compared methods on more tested materials. SUNDIAL (solid dots) significantly outperformed the traditional fitting method (empty diamonds) in terms of overall accuracy. RMSEΨΔ = RMSEΨ + RMSEΔ, and RMSERT = RMSER + RMSET. c Effect of γ. Experiments with different γ values are conducted, and the RMSEs are shown here. Tungsten (W) is used here as an exampleCC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 57 references

Research works printed but not linked: 1

  1. Born, M. & Wolf, E. Principles of Optics. 7th edn. (Cambridge University Press, Cambridge, 2013).

Cited by

See all 112

Metadata

Authors
Liu, J., Zhang, D., Yu, D., Ren, M., & Xu, J.
Affiliations
  • Nankai University (CN)
  • Shanxi University (CN)
Corresponding author
  • Ren, Mengxin
  • Xu, Jingjun
Venue
Light: Science & Applications
Published
2021-03-12
Volume
10
Issue
1
Pages
55-55
ISSN
2047-7538 (ISSN-L)
OpenAlex
W3135793694
Cited by
102
References
59
Citations by year
2026: 27 · 2025: 22 · 2024: 22 · 2023: 15 · 2022: 12 · 2021: 4
FWCI
4.03
Type
journal article
Language
en
Open access
Licence
Locations
  • Light Science & Applications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Natural Science Foundation of China (2020B0301030009, 111 Project, 12074200, 11904182, 92050114, 11374006, B07013, 61775106, 61633012, 2017YFA0303800, 11774185, 11711530205, 2019YFA0705000, 62076140, 91750204)
  • Tsinghua University
  • Nankai University
  • Higher Education Discipline Innovation Project (B07013, IRT0149)
  • Fundamental Research Funds for the Central Universities (010-63201008, 010-63201009, 2019YFA0705000, 010‐63201003, 2017YFA0303800)
Topics
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Optical Polarization and Ellipsometry
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
  • Physical Sciences › Materials Science › Electronic, Optical and Magnetic Materials › Metamaterials and Metasurfaces Applications
Keywords
ellipsometry, machine learning, inverse problems, optical constants, thin film thickness, high-throughput characterization, metals, semiconductors, dielectric materials

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Machine learning powered ellipsometry. https://app.cochat.ai/papers/10.1038/s41377-021-00482-0
The paper
Liu, J., Zhang, D., Yu, D., Ren, M., & Xu, J. (2021). Machine learning powered ellipsometry. Light: Science & Applications. https://doi.org/10.1038/s41377-021-00482-0

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.