Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Machine Learning Guided Polymorph Selection in Molecular Beam Epitaxy of In2Se3

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Bayesian optimization with a Gaussian-process surrogate guides molecular beam epitaxy to 91% phase-pure γ-In2Se3 in fewer than ten experiments.

desk verdict BO-guided MBE finds a genuinely new low-Se:In window for γ-In2Se3, and the 91% purity claim is credible but rests on a one-sample Raman-to-XRD calibration that needs strengthening. read the letter →

arxiv 2601.13156 v2 pith:UR5GM5JG submitted 2026-01-19 cond-mat.mtrl-sci

Ryan Trice , Mingyu Yu , Eric Welp , Morgan Applegate , Wesley Reinhart , Stephanie Law This is my paper
classification cond-mat.mtrl-sci PACS 81.15.Hi81.30.-t
keywords In2Se3polymorphselectionBayesianoptimizationmolecularbeamepitaxyGaussianprocessregressionRamanspectroscopyselenium-to-indiumfluxratiophase-purethinfilms
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

The paper tries to show that Bayesian optimization—a machine-learning method that proposes the next experiment by balancing uncertainty against predicted yield—can replace trial-and-error intuition in thin-film synthesis. Using a Gaussian-process surrogate trained on Raman-measured phase fractions, the authors navigated a four-parameter molecular beam epitaxy space for In2Se3 on sapphire and reached 91% phase-pure γ-In2Se3 within ten machine-selected runs. The optimizer's key move was to drop the conventional requirement that selenium flux exceed indium flux; it targeted the unexplored Se:In < 1 region, which human growers had avoided, and found that this selenium-poor window favors the γ polymorph. A second attempt to isolate α-In2Se3 stalled at 27% purity, with the model attributing the failure to amorphous film formation at low temperatures. If the phase-fraction labels are trustworthy, the result is a concrete demonstration that machine-guided search can accelerate phase-selective epitaxy in a multidimensional growth space.

What carries the argument

The central machinery is a Gaussian Process Regressor (GPR) surrogate model with a Matérn 2.5 kernel, used inside a Bayesian optimization loop with the Upper Confidence Bound acquisition function (μ + κσ, κ = 0.1) in a Kriging-believer batched ask-tell scheme. The surrogate is trained on phase fractions extracted from Voigt fits of Raman spectra (peak areas at 104, 110, 150, and 115 cm⁻¹ for α, β, γ, and InSe), with amorphous/polycrystalline films scored as 0% in every phase. A Sobol low-discrepancy sequence was used for initial space-filling exploration, and SHAP analysis was used to rank parameter influence. The mechanism works by turning each growth's Raman-derived phase fraction into a s

What would settle it

Grow films on c-plane Al2O3 at the paper's best reported conditions (≈470 °C, high indium flux, measured Se:In ≈ 0.3, cracker ≈ 840 °C) and measure phase fraction with quantitative XRD Rietveld or cross-sectional TEM; if the γ fraction is substantially below 84–91% or does not match the reported Raman-to-XRD ratio, the central claim fails.

Watch

Extended reading notes

Core claim

The paper reports that a Gaussian process regressor, trained sequentially on ~50 films and queried by an upper-confidence-bound acquisition function, identifies a previously unexplored region of low Se:In flux ratio (below 1:1) and moderately high substrate temperature (~470 °C) where γ-In2Se3 forms preferentially on c-plane Al2O3. The best film measured 84% γ by Raman peak-area fitting and 91% by XRD; the model improved γ fraction from 51% (best of the Sobol exploration) to 84–91% in fewer than ten Bayesian-optimized growths. The authors state this would not have been found without surrogate modeling, since conventional practice avoids Se:In < 1 to suppress InSe. They also conclude that sin

Load-bearing premise

The load-bearing premise is that Raman peak-area ratios give true phase fractions, a mapping the paper checks against XRD for only one sample; if the mapping is off, every training label and the 91% result collapse.

Editorial extensions

If this is right

  • If the central claim holds, phase-selective MBE of In2Se3 no longer requires a full grid search: a GPR-guided campaign can reach near-pure γ-In2Se3 in about ten samples, which is roughly the number of samples in a single DoE corner.
  • The discovery that Se:In < 1 favors γ-In2Se3 (with the caveat that the true flux ratio is systematically underestimated by QCM) reframes the growth recipe literature: selenium-deficient conditions may be the route to this polymorph.
  • The negative α-In2Se3 result implies that single-step codeposition on c-plane Al2O3 is not a viable route to crystalline α-In2Se3, at least within the explored temperature window (150–550 °C).
  • The same optimization loop—GPR + UCB with Raman-scored labels—should transfer to other chalcogenide systems with competing polymorphs, provided the characterization-to-phase-fraction mapping is reliable.

Reading between the lines

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

  • A reader might infer that the 84% (Raman) vs 91% (XRD) discrepancy for the best film is not merely noise: if Raman peak-area ratios systematically undercount γ in mixed films, the true phase purity could be higher than the training labels suggest, meaning the optimizer was working with conservative labels.
  • If the QCM underestimates true selenium flux, as the authors themselves suggest, then the 'Se:In < 1' region may actually be near-stoichiometric or slightly Se-rich; the more robust statement from the paper would then be about the control parameter (measured flux ratio), not the physical flux ratio.
  • A testable extension: apply the same GPR+UCB loop with an acquisition function that explicitly penalizes amorphous films (rather than scoring them as 0%) to see whether the α-In2Se3 search is genuinely plateaued or simply poorly rewarded.
  • The paper's reliance on a single Raman-to-XRD validation suggests a practical next step: calibrate the Voigt area-to-phase-fraction mapping on deliberately mixed standards before re-running BO on other polymorph targets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper reports an active-learning MBE campaign for In2Se3 on c-plane Al2O3. Four growth parameters (substrate temperature, In flux, Se flux, and Se cracker temperature) are used as inputs to a Gaussian-process surrogate; the target is the Raman-derived γ-In2Se3 phase fraction. After initial incorporation of historical data and a Sobol-sequence exploration phase, Bayesian optimization with a UCB acquisition function proposes experiments. Once the Se:In > 1 constraint is removed, the optimizer moves into the low-Se region and the γ fraction rises from 51% to 84% by Raman, with the best sample reaching 91% by XRD. A second optimization targeting α-In2Se3 fails to exceed 27% purity; the authors attribute this to amorphous/polycrystalline film formation at low temperature and conclude that single-step co-deposition on Al2O3 is unsuitable for crystalline α-In2Se3.

Significance. If the quantitative claims are established, the paper would provide a useful demonstration of BO for polymorph-selective thin-film synthesis, including the non-obvious finding that Se:In < 1 favors γ-In2Se3 on Al2O3. The work is transparent in its experimental tables and includes complete Raman fits in the SI, which is a strength. The negative result for α-In2Se3 is also potentially valuable. However, the central quantitative claims rest on Raman peak-area phase fractions that are validated against XRD for only one sample, with an unexplained 7-point discrepancy. The paper's significance is therefore conditional on additional calibration and uncertainty analysis.

major comments (3)
  1. [§2.2, Tables S1–S5, Fig. S19] The phase fractions used as training labels are computed from fitted areas of one main Raman peak per phase (104, 110, 150, 115 cm−1). This implicitly assumes equal scattering cross-sections and linear area-to-volume response across polymorphs and mixed-phase films. The only XRD validation, Fig. S19 for sample 48, gives 91% by XRD versus 84% by Raman, an unexplained 7-point discrepancy. Because all 60 training examples use this mapping, a composition-dependent bias could re-rank the GPR predictions and alter the BO trajectory. The qualitative finding is likely robust, but the quantitative '51% to 84%' trajectory and the headline '91% phase purity' require either a multi-sample Raman/XRD calibration, replicate uncertainties, or a clear statement of why the Raman area ratio is quantitative.
  2. [§III.A, Phase I historical data] The rescaling of historical Knudsen-cell Se fluxes to 25% of their original value is an ad hoc correction with no reported uncertainty. Six comparator samples were grown, but no quantitative comparison of their Raman-derived phase fractions against the historical samples is given in the main text. Since the GPR is initialized on these rescaled values, a miscalibrated rescaling factor could bias the initial model and influence subsequent BO suggestions. The authors should justify the 25% factor with the comparator data or show that the final conclusion is insensitive to this factor.
  3. [§IV.C, Phase V, Table S5] Amorphous and polycrystalline samples are assigned 0% for every phase in the α-optimization target. This encoding is reasonable for penalizing unusable films, but it creates a discontinuous response that may flatten gradients and distort the GPR near the low-temperature region. More importantly, the claim that BO 'increases efficiency' even in this negative result is not supported by any baseline or random-search comparison. The conclusion that single-step co-deposition is unsuitable for crystalline α-In2Se3 is plausible, but the efficiency claim should be softened or benchmarked.
minor comments (5)
  1. [Abstract / §III.B] The abstract says the 91% film was achieved in fewer than 10 BO run samples; the text reports five Phase III and five Phase IV BO samples, with the best sample being the seventh BO experiment. Please clarify the counting and use consistent phrasing.
  2. [Table S2 and other SI tables] The column labeled 'Se:In Cracker Ratio' appears to contain Se flux values with units cm−1s−1, not a ratio. Rename the column to avoid ambiguity.
  3. [SI figure numbering] Figure S2 appears twice with different content (uncertainty color plots and uncertainty PDPs). Renumber the SI figures.
  4. [§V Experimental] The experimental section numbering is inconsistent: Section V contains subsections numbered 2.1.1, 2.1.2, and 2.2, as if carried over from an earlier draft. Re-number to 5.1, 5.2, etc.
  5. [§IV.B] The phrase 'Matern 2.5 kernel' should be 'Matérn 5/2 kernel' for standard notation. Also, the BO package citation is to a GitHub repository; please add the specific version or DOI if available.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the BO loop is active learning, and the surrogate's predictions are tested against independently measured phase fractions.

full rationale

The paper's central claim is that Bayesian optimization guided MBE growth to a γ-In2Se3 film with 91% phase purity in fewer than ten BO-run samples. The GPR surrogate is trained on experimentally measured Raman-derived phase fractions; the BO acquisition function then proposes new growth conditions, which are grown and measured independently. This is a closed experimental loop, not a derivation that reduces to its inputs. The surrogate 'predictions' are not fitted outputs renamed as predictions: every proposed point was actually synthesized and characterized, and the final purity is supported by an XRD measurement (Figure S19), albeit with a 7-point discrepancy from the Raman value. There is no equation in which a target quantity is defined in terms of the quantity it claims to predict. The paper does cite prior work from the same groups, including [49] for initial β-In2Se3 growth conditions and [58], [62] for Se cracker and sticking-coefficient behavior, but those citations support experimental calibrations and initial parameter choices, not the central claim that the BO campaign succeeded. No uniqueness theorem or ansatz is imported from self-citations. The main weakness—that Raman peak-area ratios are used as quantitative phase fractions with only one XRD validation point—is a measurement-validity and calibration concern, not circularity. The derivation chain (measure → fit surrogate → suggest → measure) is self-consistent and externally testable, so the appropriate circularity score is 0.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The paper rests on standard experimental calibration and ML modeling choices rather than invented entities. The most consequential free parameters are the 25% Se-flux rescaling, the BO acquisition constant κ=0.1, the Sobol bounds, and the encoding of amorphous/polycrystalline samples as 0% for all phases. No new physical entities are postulated.

free parameters (6)
  • Se flux rescaling factor for Knudsen-cell data = 0.25
    Historical Phase I selenium fluxes from a Knudsen cell were multiplied by 0.25 to make them comparable to cracker-source flux readings (Section III.A); no quantitative validation of the factor is reported.
  • BO acquisition constant κ = 0.1
    Upper Confidence Bound acquisition in Eq. (1) uses κ=0.1, set by hand, controlling exploration/exploitation.
  • GPR kernel = Matern 2.5
    Kernel choice for the Gaussian process surrogate (Section 2.2); not optimized or compared at selection time.
  • Sobol parameter bounds = TG 200–625 °C, FIn 7e12–6e13, FSe 2e12–4.8e13, TSe 500–1000 °C
    Exploration space restricted by system limits and prior knowledge (Section III.A), affecting all BO suggestions.
  • Amorphous/polycrystalline target encoding = 0% for all polymorphs/phases
    For the α-In2Se3 objective, non-crystalline films were scored 0% in every phase (Section III.C), encoding a crystalline-only target.
  • Raman mode set for phase fractions = 104, 110, 150, 115 cm^-1
    Only the main peak of each phase is used to compute fractions (Section 2.2); no uncertainty model or cross-validation against XRD for most samples.
assumptions (5)
  • domain assumption Raman peak areas are linearly proportional to phase fractions in mixed-phase In2Se3 films.
    Used to create every training label and the reported phase purities (Section 2.2, Tables S1–S5). Not validated for most samples.
  • domain assumption QCM flux measurements at room-temperature sticking calibration are proportional to true fluxes; selenium re-evaporation causes only a systematic offset.
    The flux ratio is the key experimental variable, and the paper later argues FSe is underestimated (Section III.B).
  • domain assumption A GPR surrogate trained on ~29 to ~51 samples captures the relevant phase map over the four-parameter space.
    BO suggestions and all SHAP/PDP interpretations inherit this assumption (Section 2.2).
  • domain assumption Varying only substrate temperature and In/Se fluxes/cracker temperature is sufficient to control polymorph selection for fixed substrate and co-deposition.
    Scope deliberately fixed (Section III intro); other MBE controls are assumed irrelevant.
  • domain assumption Films scored as amorphous or polycrystalline contain zero of the target phase, and assigning 0% is appropriate for the optimization.
    Penalty applied in Section III.C; if amorphous films actually contain nanocrystalline α, the negative conclusion would need revision.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine Learning Guided Polymorph Selection in Molecular Beam Epitaxy of In2Se3." pith.science (2026). https://pith.science/paper/UR5GM5JG

@misc{pith2026260113156,
  author       = {Pith},
  title        = {Pith review of: Machine Learning Guided Polymorph Selection in Molecular Beam Epitaxy of In2Se3},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UR5GM5JG}},
  note         = {Machine review of arXiv:2601.13156}
}
read the original abstract

Indium selenide (In2Se3), a layered chalcogenide with multiple polymorphs, is a promising material for optoelectronic and ferroelectric applications. However, achieving polymorph-pure thin films remains a major challenge due to the complex growth space. In this work, Bayesian optimization (BO) is successfully leveraged to guide the molecular beam epitaxy growth of In2Se3 on Al2O3 substrates. By training a predictive Gaussian process regressor with sequential learning, we efficiently explored substrate temperature, indium flux, selenium flux, and cracker temperature, reducing experimental trials required for successful synthesis. A {\gamma}-In2Se3 film with 91% phase purity was achieved in fewer than 10 BO run samples. Attempts to isolate {\alpha}-In2Se3 were limited by amorphous film formation at low temperatures, indicating that single-step codeposition is unsuitable for crystalline {\alpha}-In2Se3 on Al2O3. Overall, this study validates BO as a powerful approach for phase-selective growth in complex material systems.

Figures

Figures reproduced from arXiv: 2601.13156 by the authors.

Figure 1
Figure 1. In addition to the multiple polymorphs, the In–Se phase diagram includes several competing phases, such as InSe and In4Se3, which must be avoided during growth[48]. Achieving phase-pure and polymorph-pure In2Se3 is further complicated by the sensitivity of polymorph formation to growth parameters like temperature, In to Se flux ratios, and substrate material[31], [49], [50], [51], [52], [53], [54], [55], [56], [57] … view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Lifetime Sample Tracking (LiST): A Data Platform for Materials Science

    cond-mat.mtrl-sci 2026-06 unverdicted novelty 3.0 of 10

    LiST is a data management platform that captures, curates, and disseminates synthesis, characterization, and modeling data from 2D materials research at a national user facility, enabling DOI-linked data packages.

Reference graph

Works this paper leans on

70 extracted references · 52 canonical work pages · cited by 1 Pith paper

  1. [1]

    History and current status of commercial pulsed laser deposition equipment,

    J. A. Greer, “History and current status of commercial pulsed laser deposition equipment,” J Phys D Appl Phys, vol. 47, no. 3, p. 034005, Jan. 2014, doi: 10.1088/0022-3727/47/3/034005

    work page doi:10.1088/0022-3727/47/3/034005 2014

  2. [2]

    Preparation of Y-Ba-Cu oxide superconductor thin films using pulsed laser evaporation from high T c bulk material,

    D. Dijkkamp et al., “Preparation of Y-Ba-Cu oxide superconductor thin films using pulsed laser evaporation from high T c bulk material,” Appl Phys Lett, vol. 51, no. 8, pp. 619–621, Aug. 1987, doi: 10.1063/1.98366

    work page doi:10.1063/1.98366 1987

  3. [3]

    Molecular beam epitaxy,

    A. Y . Cho and J. R. Arthur, “Molecular beam epitaxy,” Progress in Solid State Chemistry, vol. 10, pp. 157–191, Jan. 1975, doi: 10.1016/0079-6786(75)90005-9

    doi:10.1016/0079-6786(75)90005-9 1975

  4. [4]

    Chemical V apor Deposition Growth and Applications of Two-Dimensional Materials and Their Heterostructures,

    Z. Cai, B. Liu, X. Zou, and H.-M. Cheng, “Chemical V apor Deposition Growth and Applications of Two-Dimensional Materials and Their Heterostructures,” Chem Rev, vol. 118, no. 13, pp. 6091– 6133, Jul. 2018, doi: 10.1021/acs.chemrev.7b00536

    work page doi:10.1021/acs.chemrev.7b00536 2018

  5. [5]

    van der Waals epitaxy of transition metal dichalcogenides via molecular beam epitaxy: looking back and moving forward,

    D. K. Singh and G. Gupta, “van der Waals epitaxy of transition metal dichalcogenides via molecular beam epitaxy: looking back and moving forward,” Mater Adv, vol. 3, no. 15, pp. 6142– 6156, 2022, doi: 10.1039/D2MA00352J

    work page doi:10.1039/d2ma00352j 2022

  6. [6]

    Bayesian Optimization in Bioprocess Engineering—Where Do We Stand Today?,

    F. Gisperg, R. Klausser, M. Elshazly, J. Kopp, E. P. Brichtová, and O. Spadiut, “Bayesian Optimization in Bioprocess Engineering—Where Do We Stand Today?,” Biotechnol Bioeng, vol. 122, no. 6, pp. 1313–1325, Jun. 2025, doi: 10.1002/bit.28960

    doi:10.1002/bit.28960 2025

  7. [7]

    Hyperparameter Optimization for Machine Learning Models Based on Bayesian Optimization,

    J. Wu, X. Y . Chen, H. Zhang, L. D. Xiong, H. Lei, and S. H. Deng, “Hyperparameter Optimization for Machine Learning Models Based on Bayesian Optimization,” Journal of Electronic Science and Technology, vol. 17, no. 1, pp. 26–40, Mar. 2019, doi: 10.11989/JEST.1674-862X.80904120

    work page doi:10.11989/jest.1674-862x.80904120 2019

  8. [8]

    Bayesian Optimization for Adaptive Experimental Design: A Review,

    S. Greenhill, S. Rana, S. Gupta, P. Vellanki, and S. Venkatesh, “Bayesian Optimization for Adaptive Experimental Design: A Review,” IEEE Access, vol. 8, pp. 13937–13948, 2020, doi: 10.1109/ACCESS.2020.2966228

    arXiv 2020

Show all 70 references
  1. [9]

    One-Factor-at-a-Time versus Designed Experiments,

    V . Czitrom, “One-Factor-at-a-Time versus Designed Experiments,” Am Stat, vol. 53, no. 2, pp. 126–131, May 1999, doi: 10.1080/00031305.1999.10474445

    1999 arXiv

  2. [10]

    Bellman, Dynamic programming

    Richard. Bellman, Dynamic programming. Dover Publications, 2003

    2003

  3. [11]

    Bayesian optimization for stable properties amid processing fluctuations in sputter deposition,

    A. Shrivastava, M. Kalaswad, J. O. Custer, D. P. Adams, and H. N. Najm, “Bayesian optimization for stable properties amid processing fluctuations in sputter deposition,” Journal of Vacuum Science & Technology A, vol. 42, no. 3, May 2024, doi: 10.1116/6.0003418

    2024 doi

  4. [12]

    Prospective on methods of design of experiments for limited data scenarios in materials design and engineering,

    E. Ryan, A. Roshandelpoor, Z. Pollard, J. L. Goldfarb, and P. Vakili, “Prospective on methods of design of experiments for limited data scenarios in materials design and engineering,” MRS Commun, vol. 13, no. 6, pp. 1087–1101, Sep. 2023, doi: 10.1557/s43579-023-00478-4

    2023 doi

  5. [13]

    Traditional or adaptive design of experiments? A pilot-scale comparison on wood delignification,

    H. Rummukainen, H. Hörhammer, P . Kuusela, J. Kilpi, J. Sirviö, and M. Mäkelä, “Traditional or adaptive design of experiments? A pilot-scale comparison on wood delignification,” Heliyon, vol. 10, no. 2, p. e24484, Jan. 2024, doi: 10.1016/j.heliyon.2024.e24484

    2024 doi

  6. [14]

    Google Vizier,

    D. Golovin, B. Solnik, S. Moitra, G. Kochanski, J. Karro, and D. Sculley, “Google Vizier,” in Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, New York, NY , USA: ACM, Aug. 2017, pp. 1487–1495. doi: 10.1145/3097983.3098043

    2017

  7. [15]

    Packwood, Bayesian optimization for materials science

    Daniel. Packwood, Bayesian optimization for materials science. Springer, 2017

    2017

  8. [16]

    Garnett, Bayesian optimization

    Roman. Garnett, Bayesian optimization. Cambridge University Press, 2023

    2023

  9. [17]

    Active learning in materials science with emphasis on adaptive sampling using uncertainties for targeted design,

    T. Lookman, P. V . Balachandran, D. Xue, and R. Yuan, “Active learning in materials science with emphasis on adaptive sampling using uncertainties for targeted design,” NPJ Comput Mater, vol. 5, no. 1, p. 21, Feb. 2019, doi: 10.1038/s41524-019-0153-8

    2019 doi

  10. [18]

    Benchmarking the performance of Bayesian optimization across multiple experimental materials science domains,

    Q. Liang et al., “Benchmarking the performance of Bayesian optimization across multiple experimental materials science domains,” NPJ Comput Mater, vol. 7, no. 1, p. 188, Nov. 2021, doi: 10.1038/s41524-021-00656-9

    2021 doi

  11. [19]

    Adaptive Bayesian optimization for epitaxial growth of Si thin films under various constraints,

    K. Osada et al., “Adaptive Bayesian optimization for epitaxial growth of Si thin films under various constraints,” Mater Today Commun, vol. 25, p. 101538, Dec. 2020, doi: 10.1016/j.mtcomm.2020.101538

    2020

  12. [20]

    Bayesian optimization with experimental failure for high-throughput materials growth,

    Y . K. Wakabayashi, T. Otsuka, Y . Krockenberger, H. Sawada, Y . Taniyasu, and H. Yamamoto, “Bayesian optimization with experimental failure for high-throughput materials growth,” NPJ Comput Mater, vol. 8, no. 1, p. 180, Aug. 2022, doi: 10.1038/s41524-022-00859-8

    2022 doi

  13. [21]

    Machine-learning-assisted thin-film growth: Bayesian optimization in molecular beam epitaxy of SrRuO3 thin films,

    Y . K. Wakabayashi, T. Otsuka, Y . Krockenberger, H. Sawada, Y . Taniyasu, and H. Yamamoto, “Machine-learning-assisted thin-film growth: Bayesian optimization in molecular beam epitaxy of SrRuO3 thin films,” APL Mater, vol. 7, no. 10, Oct. 2019, doi: 10.1063/1.5123019

    2019 doi

  14. [22]

    Bayesian optimization-driven enhancement of the thermoelectric properties of polycrystalline III-V semiconductor thin films,

    T. Ishiyama, K. Nozawa, T. Nishida, T. Suemasu, and K. Toko, “Bayesian optimization-driven enhancement of the thermoelectric properties of polycrystalline III-V semiconductor thin films,” NPG Asia Mater, vol. 16, no. 1, p. 17, Mar. 2024, doi: 10.1038/s41427-024-00536-w

    2024 doi

  15. [23]

    Bayesian Optimization for Controlled Chemical Vapor Deposition Growth of WS 2,

    F. Zhang, R. Tamura, F. Zeng, D. Kozawa, and R. Kitaura, “Bayesian Optimization for Controlled Chemical Vapor Deposition Growth of WS 2,” ACS Appl Mater Interfaces, vol. 16, no. 43, pp. 59109–59115, Oct. 2024, doi: 10.1021/acsami.4c15275

    2024 doi

  16. [24]

    Cross-Modal Characterization of Thin- Film MoS2 Using Generative Models,

    I. A. Moses, C. Chen, J. M. Redwing, and W. F. Reinhart, “Cross-Modal Characterization of Thin- Film MoS2 Using Generative Models,” Advanced Intelligent Systems, p. 2500613, 2025, doi: 10.1002/AISY .202500613;PAGEGROUP:STRING:PUBLICATION

    2025 doi

  17. [25]

    Quantum and classical machine learning investigation of synthesis–structure relationships in epitaxially grown wide band gap semiconductors,

    A. S. Messecar, S. M. Durbin, and R. A. Makin, “Quantum and classical machine learning investigation of synthesis–structure relationships in epitaxially grown wide band gap semiconductors,” MRS Communications 2024 14:4, vol. 14, no. 4, pp. 660–666, Jul. 2024, doi: 10.1557/S435...

    2024 doi

  18. [26]

    Simultaneous prediction of structural properties in epitaxially–grown GaN with quantum and conventional multi–output learning algorithms,

    A. S. Messecar, E. F. Tamayo, S. M. Durbin, and R. A. Makin, “Simultaneous prediction of structural properties in epitaxially–grown GaN with quantum and conventional multi–output learning algorithms,” MRS Advances 2025, pp. 1–7, Oct. 2025, doi: 10.1557/S43580-025-01420-7

    2025 doi

  19. [27]

    Multimodal Machine Learning Analysis of GaSe Molecular Beam Epitaxy Growth Conditions,

    M. Yu, I. A. Moses, W. F. Reinhart, and S. Law, “Multimodal Machine Learning Analysis of GaSe Molecular Beam Epitaxy Growth Conditions,” ACS Appl Mater Interfaces, vol. 17, no. 23, pp. 34707–34716, Jun. 2025, doi: 10.1021/ACSAMI.5C02891

    2025 doi

  20. [28]

    Machine-learning-enabled on-the-fly analysis of RHEED patterns during thin film deposition by molecular beam epitaxy,

    T. C. Kaspar et al., “Machine-learning-enabled on-the-fly analysis of RHEED patterns during thin film deposition by molecular beam epitaxy,” Journal of Vacuum Science & Technology A, vol. 43, no. 3, p. 32702, May 2025, doi: 10.1116/6.0004493/3341018

    2025 doi

  21. [29]

    Machine-learning-assisted and real-time-feedback-controlled growth of InAs/GaAs quantum dots,

    C. Shen et al., “Machine-learning-assisted and real-time-feedback-controlled growth of InAs/GaAs quantum dots,” Nature Communications 2024 15:1, vol. 15, no. 1, pp. 1–11, Mar. 2024, doi: 10.1038/s41467-024-47087-w

    2024 doi

  22. [30]

    Analyzing the impact of Se concentration during the molecular beam epitaxy deposition of 2D SnSe with atomistic-scale simulations and explainable machine learning,

    J. R. Chin et al., “Analyzing the impact of Se concentration during the molecular beam epitaxy deposition of 2D SnSe with atomistic-scale simulations and explainable machine learning,” Mater Today Adv, vol. 28, p. 100640, Dec. 2025, doi: 10.1016/J.MTADV .2025.100640

    2025

  23. [31]

    Crystalline–Crystalline Phase Transformation in Two-Dimensional In2Se3 Thin Layers,

    X. Tao and Y . Gu, “Crystalline–Crystalline Phase Transformation in Two-Dimensional In2Se3 Thin Layers,” Nano Lett, vol. 13, no. 8, pp. 3501–3505, Aug. 2013, doi: 10.1021/NL400888P

    2013 doi

  24. [32]

    Polymorphism and Ferroelectricity in Indium(III) Selenide,

    C. K. Y . Tan, W. Fu, and K. P. Loh, “Polymorphism and Ferroelectricity in Indium(III) Selenide,” Chem Rev, vol. 123, no. 13, pp. 8701–8717, Jul. 2023, doi: 10.1021/ACS.CHEMREV .3C00129

    2023 doi

  25. [33]

    Gate Controlled Photocurrent Generation Mechanisms in High-Gain In2Se3 Phototransistors,

    J. O. Island, S. I. Blanter, M. Buscema, H. S. J. Van Der Zant, and A. Castellanos-Gomez, “Gate Controlled Photocurrent Generation Mechanisms in High-Gain In2Se3 Phototransistors,” Nano Lett, vol. 15, no. 12, pp. 7853–7858, Dec. 2015, doi: 10.1021/ACS.NANOLETT.5B02523

    2015 doi

  26. [34]

    Phase-Engineering-Driven Enhanced Electronic and Optoelectronic Performance of Multilayer In2Se3 Nanosheets,

    W. Feng et al., “Phase-Engineering-Driven Enhanced Electronic and Optoelectronic Performance of Multilayer In2Se3 Nanosheets,” ACS Appl Mater Interfaces, vol. 10, no. 33, pp. 27584–27588, Aug. 2018, doi: 10.1021/ACSAMI.8B10194

    2018 doi

  27. [35]

    Self-Assembly High-Performance UV– vis–NIR Broadband β-In2Se3/Si Photodetector Array for Weak Signal Detection,

    Z. Zheng, J. Yao, B. Wang, Y . Yang, G. Yang, and J. Li, “Self-Assembly High-Performance UV– vis–NIR Broadband β-In2Se3/Si Photodetector Array for Weak Signal Detection,” ACS Appl Mater Interfaces, vol. 9, no. 50, pp. 43830–43837, Dec. 2017, doi: 10.1021/ACSAMI.7B16329

    2017 doi

  28. [36]

    High-performance and flexible photodetectors based on chemical vapor deposition grown two-dimensional In2Se3 nanosheets,

    Y . Wang et al., “High-performance and flexible photodetectors based on chemical vapor deposition grown two-dimensional In2Se3 nanosheets,” Nanotechnology, vol. 29, no. 44, p. 445205, Sep. 2018, doi: 10.1088/1361-6528/AADC73

    2018 doi

  29. [37]

    Extraordinary Photoresponse in Two-Dimensional In2Se3 Nanosheets,

    R. B. Jacobs-Gedrim et al., “Extraordinary Photoresponse in Two-Dimensional In2Se3 Nanosheets,” ACS Nano, vol. 8, no. 1, pp. 514–521, Jan. 2013, doi: 10.1021/NN405037S

    2013 doi

  30. [38]

    Dynamic observation of phase transformation behaviors in indium(iii) selenide nanowire based phase change memory,

    Y . T. Huang et al., “Dynamic observation of phase transformation behaviors in indium(iii) selenide nanowire based phase change memory,” ACS Nano, vol. 8, no. 9, pp. 9457–9462, Sep. 2014, doi: 10.1021/NN503576X/SUPPL_FILE/NN503576X_SI_005.A VI

    2014 doi

  31. [39]

    Indium selenide nanowire phase-change memory,

    B. Yu et al., “Indium selenide nanowire phase-change memory,” Appl Phys Lett, vol. 91, no. 13, Sep. 2007, doi: 10.1063/1.2793505/334327

    2007 doi

  32. [40]

    Indium selenide (In2Se3) thin film for phase-change memory,

    H. Lee, D. H. Kang, and L. Tran, “Indium selenide (In2Se3) thin film for phase-change memory,” Materials Science and Engineering: B, vol. 119, no. 2, pp. 196–201, May 2005, doi: 10.1016/J.MSEB.2005.02.060

    2005 doi

  33. [41]

    Pressure-induced phase transformation of In2Se3,

    A. M. Rasmussen, S. T. Teklemichael, E. Mafi, Y . Gu, and M. D. McCluskey, “Pressure-induced phase transformation of In2Se3,” Appl Phys Lett, vol. 102, no. 6, Feb. 2013, doi: 10.1063/1.4792313/26004

    2013 doi

  34. [42]

    Electrically Driven Reversible Phase Changes in Layered In2Se3 Crystalline Film,

    M. S. Choi et al., “Electrically Driven Reversible Phase Changes in Layered In2Se3 Crystalline Film,” Advanced Materials, vol. 29, no. 42, p. 1703568, Nov. 2017, doi: 10.1002/ADMA.201703568

    2017 doi

  35. [43]

    Out-of-Plane Piezoelectricity and Ferroelectricity in Layered α-In2Se3 Nanoflakes,

    Y . Zhou et al., “Out-of-Plane Piezoelectricity and Ferroelectricity in Layered α-In2Se3 Nanoflakes,” Nano Lett, vol. 17, no. 9, pp. 5508–5513, Sep. 2017, doi: 10.1021/ACS.NANOLETT.7B02198

    2017 doi

  36. [44]

    Intercorrelated In-Plane and Out-of-Plane Ferroelectricity in Ultrathin Two- Dimensional Layered Semiconductor In2Se3,

    C. Cui et al., “Intercorrelated In-Plane and Out-of-Plane Ferroelectricity in Ultrathin Two- Dimensional Layered Semiconductor In2Se3,” Nano Lett, vol. 18, no. 2, pp. 1253–1258, Feb. 2018, doi: 10.1021/ACS.NANOLETT.7B04852

    2018 doi

  37. [45]

    Prediction of intrinsic two-dimensional ferroelectrics in In2Se3 and other III2-VI3 van der Waals materials,

    W. Ding et al., “Prediction of intrinsic two-dimensional ferroelectrics in In2Se3 and other III2-VI3 van der Waals materials,” Nature Communications 2017 8:1, vol. 8, no. 1, pp. 1–8, Apr. 2017, doi: 10.1038/ncomms14956

    2017 doi

  38. [46]

    In-plane anisotropic electronic properties in layered α′-In2Se3,

    Z. Liu, J. Wu, and J. Li, “In-plane anisotropic electronic properties in layered α′-In2Se3,” J Appl Phys, vol. 130, no. 1, Jul. 2021, doi: 10.1063/5.0050979/158520

    2021 doi

  39. [47]

    In-plane ferroelectric tunnel junctions based on 2D α-In2Se3/semiconductor heterostructures,

    Z. Liu, P. Hou, L. Sun, E. Y . Tsymbal, J. Jiang, and Q. Yang, “In-plane ferroelectric tunnel junctions based on 2D α-In2Se3/semiconductor heterostructures,” npj Computational Materials 2023 9:1, vol. 9, no. 1, pp. 1–7, Jan. 2023, doi: 10.1038/s41524-022-00953-x

    2023 doi

  40. [48]

    In-Se (indium-selenium),

    H. Okamoto, “In-Se (indium-selenium),” Journal of Phase Equilibria and Diffusion 2004 25:2, vol. 25, no. 2, pp. 201–201, Apr. 2004, doi: 10.1007/S11669-004-0031-Y

    2004 doi

  41. [49]

    Evidence for In-Plane Electrical Polarization in 3R-β’-In2Se3 Thin Films Grown by Molecular Beam Epitaxy,

    D. S. H. Liu et al., “Evidence for In-Plane Electrical Polarization in 3R-β’-In2Se3 Thin Films Grown by Molecular Beam Epitaxy,” ACS Appl Mater Interfaces, vol. 17, no. 24, pp. 35661– 35672, Jun. 2025, doi: 10.1021/ACSAMI.5C01185

    2025 doi

  42. [50]

    Epitaxial growth of γ-InSe and α, β, and γ-In2Se3 on ε-GaSe,

    N. Balakrishnan et al., “Epitaxial growth of γ-InSe and α, β, and γ-In2Se3 on ε-GaSe,” 2d Mater, vol. 5, no. 3, p. 035026, Jun. 2018, doi: 10.1088/2053-1583/AAC479

    2018 doi

  43. [51]

    Atomically Resolving Polymorphs and Crystal Structures of In2Se3,

    L. Liu et al., “Atomically Resolving Polymorphs and Crystal Structures of In2Se3,” Chemistry of Materials, vol. 31, no. 24, pp. 10143–10149, Dec. 2019, doi: 10.1021/ACS.CHEMMATER.9B03499

    2019 doi

  44. [52]

    Atomic Imaging of Electrically Switchable Striped Domains in β′-In2Se3,

    Z. Chen et al., “Atomic Imaging of Electrically Switchable Striped Domains in β′-In2Se3,” Advanced Science, vol. 8, no. 17, Sep. 2021, doi: 10.1002/ADVS.202100713

    2021 doi

  45. [53]

    Molecular beam epitaxy synthesis of In2Se3 films,

    C. A. V oigt, M. Reingold, A. Dube, L. S. Early, B. K. Wagner, and E. M. V ogel, “Molecular beam epitaxy synthesis of In2Se3 films,” Journal of Vacuum Science & Technology A, vol. 42, no. 3, May 2024, doi: 10.1116/6.0003508/3283074

    2024 doi

  46. [54]

    High Mobility γ-Phase Indium Selenide on Si(100) Grown by Molecular Beam Epitaxy,

    A. Salhi, A. Abutaha, A. Zekri, M. Pasha, A. Samara, and S. Mansour, “High Mobility γ-Phase Indium Selenide on Si(100) Grown by Molecular Beam Epitaxy,” ACS Appl Electron Mater, vol. 7, no. 4, pp. 1398–1407, Feb. 2025, doi: 10.1021/ACSAELM.4C01800

    2025 doi

  47. [55]

    Molecular-Beam Epitaxy of Two-Dimensional In2Se3 and Its Giant Electroresistance Switching in Ferroresistive Memory Junction,

    S. M. Poh et al., “Molecular-Beam Epitaxy of Two-Dimensional In2Se3 and Its Giant Electroresistance Switching in Ferroresistive Memory Junction,” Nano Lett, vol. 18, no. 10, pp. 6340–6346, Oct. 2018, doi: 10.1021/ACS.NANOLETT.8B02688

    2018 doi

  48. [56]

    Recent advances of phase transition and ferroelectric device in two-dimensional In2Se3,

    W. Han et al., “Recent advances of phase transition and ferroelectric device in two-dimensional In2Se3,” Appl Phys Rev, vol. 11, no. 2, Jun. 2024, doi: 10.1063/5.0190609/3289204

    2024 doi

  49. [57]

    Quantitative Raman and x-ray photoelectron spectroscopy of mixed-phase indium selenide films,

    C. A. V oigt, M. Tian, R. Peacock, B. K. Wagner, and E. M. V ogel, “Quantitative Raman and x-ray photoelectron spectroscopy of mixed-phase indium selenide films,” J Appl Phys, vol. 135, no. 21, p. 215301, Jun. 2024, doi: 10.1063/5.0202596/3295804

    2024 doi

  50. [58]

    Sticking coefficients of selenium and tellurium,

    D. S. H. Liu, M. Hilse, and R. Engel-Herbert, “Sticking coefficients of selenium and tellurium,” Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, vol. 39, no. 2, Mar. 2021, doi: 10.1116/6.0000736

    2021 doi

  51. [59]

    Molecular-beam epitaxy growth of ZnSe using a cracked selenium source,

    H. Cheng, J. M. DePuydt, M. A. Haase, and J. E. Potts, “Molecular-beam epitaxy growth of ZnSe using a cracked selenium source,” Journal of Vacuum Science & Technology B: Microelectronics Processing and Phenomena, vol. 8, no. 2, pp. 181–186, Mar. 1990, doi: 10.1116/1.584850

    1990 doi

  52. [60]

    Equilibrium Composition of Selenium Vapor; the Thermodynamics of the Vaporization of HgSe, CdSe, and SrSe,

    J. Berkowitz and W. A. Chupka, “Equilibrium Composition of Selenium Vapor; the Thermodynamics of the Vaporization of HgSe, CdSe, and SrSe,” J Chem Phys, vol. 45, no. 11, pp. 4289–4302, Dec. 1966, doi: 10.1063/1.1727488

    1966 doi

  53. [61]

    Comment on the Composition of Selenium Vapor,

    J. Berkowitz and W. A. Chupka, “Comment on the Composition of Selenium Vapor,” J Chem Phys, vol. 48, no. 12, pp. 5743–5744, Jun. 1968, doi: 10.1063/1.1668677

    1968 doi

  54. [62]

    Growth of Bi2Se3 topological insulator films using a selenium cracker source,

    T. P. Ginley and S. Law, “Growth of Bi2Se3 topological insulator films using a selenium cracker source,” Journal of Vacuum Science & Technology B, Nanotechnology and Microelectronics: Materials, Processing, Measurement, and Phenomena, vol. 34, no. 2, Mar. 2016, doi: 10.1116/1....

    2016 doi

  55. [63]

    On the distribution of points in a cube and the approximate evaluation of integrals,

    I. M. Sobol’, “On the distribution of points in a cube and the approximate evaluation of integrals,” USSR Computational Mathematics and Mathematical Physics, vol. 7, no. 4, pp. 86–112, Jan. 1967, doi: 10.1016/0041-5553(67)90144-9

    1967 doi

  56. [64]

    On the Scrambled Soboĺ Sequence,

    H. Chi, P. Beerli, D. W. Evans, and M. Mascagni, “On the Scrambled Soboĺ Sequence,” 2005, pp. 775–782. doi: 10.1007/11428862_105

    2005 doi

  57. [65]

    Quasi-Van der Waals Epitaxial Growth of γ′-GaSe Nanometer-Thick Films on GaAs(111)B Substrates,

    M. Yu et al., “Quasi-Van der Waals Epitaxial Growth of γ′-GaSe Nanometer-Thick Films on GaAs(111)B Substrates,” ACS Nano, vol. 18, no. 26, pp. 17185–17196, Jul. 2024, doi: 10.1021/ACSNANO.4C04194

    2024 doi

  58. [66]

    Tuning the phase stability of Mo-based TMD monolayers through coupled vacancy defects and lattice strain,

    Q. Tang, “Tuning the phase stability of Mo-based TMD monolayers through coupled vacancy defects and lattice strain,” J Mater Chem C Mater, vol. 6, no. 35, pp. 9561–9568, Sep. 2018, doi: 10.1039/C8TC03430C

    2018 doi

  59. [67]

    Atomic and structural modifications of two-dimensional transition metal dichalcogenides for various advanced applications,

    B. Kirubasankar, Y . S. Won, L. A. Adofo, S. H. Choi, S. M. Kim, and K. K. Kim, “Atomic and structural modifications of two-dimensional transition metal dichalcogenides for various advanced applications,” Chem Sci, vol. 13, no. 26, p. 7707, May 2022, doi: 10.1039/D2SC01398C

    2022 doi

  60. [68]

    Defects and Defect Engineering of Two- Dimensional Transition Metal Dichalcogenide (2D TMDC) Materials,

    M. F. Hossen, S. Shendokar, and S. Aravamudhan, “Defects and Defect Engineering of Two- Dimensional Transition Metal Dichalcogenide (2D TMDC) Materials,” Nanomaterials, vol. 14, no. 5, p. 410, Mar. 2024, doi: 10.3390/NANO14050410

    2024 doi

  61. [69]

    {Bayesian Optimization}: Open source constrained global optimization tool for {Python},

    Fernando Nogueira, “{Bayesian Optimization}: Open source constrained global optimization tool for {Python},” 2014, https://github.com/bayesian-optimization/BayesianOptimization

    2014

  62. [70]

    A Unified Approach to Interpreting Model Predictions,

    S. M. Lundberg, P . G. Allen, and S.-I. Lee, “A Unified Approach to Interpreting Model Predictions,” Adv Neural Inf Process Syst, vol. 30, 2017, Accessed: Nov. 03, 2025. [Online]. Available: https://github.com/slundberg/shap Supplementary Information Machine-Learning-Guided Po...

    2017

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.