Pith. sign in

REVIEW 4 major objections 4 minor 9 references

This paper uses real-time tracking of 474 individual MoS2 crystals to show that CVD growth is edge-attachment-limited, that crystals grow independently without competing for precursors, and that each flake is fed by its own central MoOxSy n

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Real-time tracking of 474 MoS2 crystals shows growth is edge-attachment-limited and neighbor-independent until impingement.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection A genuinely new dataset and tracking pipeline for MoS2 CVD growth, but the headline mechanistic claim rests on a correlation that needs more statistical and spatial controls before I'd take it as settled. the 4 major comments →

arxiv 2607.13893 v1 pith:77VCVA5I submitted 2026-07-15 cond-mat.mtrl-sci

Revealing the MoS2 Growth Mechanism in Chemical Vapor Deposition: Real-Time Imaging and Statistical Analysis

Faizal Arifurrahman , Tianshu Zhai , Yuguo Wang , Jing Zhang , Andrew L. Hitt , Jun Lou , Ming Tang This is my paper
classification cond-mat.mtrl-sci PACS 81.15.Gh
keywords MoS2chemical vapor depositiongrowth mechanismedge-attachment-limited kineticsin-situ optical microscopycrystal trackingMoOxSy seedingtwo-dimensional materials
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 sets out to settle whether chemical vapor deposition of monolayer MoS2 is controlled by precursor diffusion to the crystal or by the reaction at the crystal edge. By filming growth in a custom reactor and automatically tracking 474 crystals, it finds that instantaneous growth rate scales with crystal size in a way that matches edge-attachment-limited kinetics. It also finds that neighboring crystals grow at the same rate regardless of how crowded they are, until they physically touch, implying they do not compete for the same precursor. The reason proposed is a central MoOxSy nanoparticle in each flake that acts as a private molybdenum source, isolated by a step-edge barrier. If true, seed density—not just temperature and precursor flow—becomes the main lever for controlling crystal size.

Core claim

On the paper's own terms, the central discovery is that MoS2 monolayer growth at temperatures above ~850 °C is edge-attachment-limited: the rate at which a crystal's area increases is proportional to its edge length (dA/dt ∝ √A), not constant as diffusion-limited growth would predict. A corollary, shown by comparing growth rates against the number and coverage of neighbor crystals and by tracking centroid distances of approaching pairs, is that MoS2 crystals grow independently until impingement, without forming depletion zones. The authors attribute this to self-seeding: each crystal nucleates on a MoOxSy nanoparticle that supplies Mo locally, while a Schwoebel-Ehrlich barrier keeps that mol

What carries the argument

The load-bearing element is the semi-automated image-processing pipeline that turns an in-situ optical video into complete growth trajectories for 474 crystals—segmentation by a random forest classifier, frame-to-frame centroid tracking, single-crystal versus polycrystal classification, and shape-feature extraction. The decisive statistical test is a comparison of instantaneous growth rates dA/dt at a fixed time (t = 50 s) against crystal area: a positive correlation with size picks out edge-attachment-limited kinetics, while diffusion-limited growth would show size-independent rates. The explanatory mechanism is the MoOxSy nanoparticle seen at the center of nearly every flake, which acts as

Load-bearing premise

The inference that growth is edge-attachment-limited rests on the assumption that every crystal in the field of view experienced the same local precursor supersaturation at the moment growth rates were compared; if supersaturation varied from place to place, the correlation between size and growth rate could arise from local supply differences rather than edge kinetics.

What would settle it

Measure, in the same reactor, the instantaneous growth rates of crystals of equal area that sit in regions with different local seed densities or different distances from the precursor inlet; if equal-area crystals grow at different rates when local precursor supply differs, the edge-attachment-limited interpretation is confounded. Alternatively, vary the density of MoOxSy seeds on the substrate and check whether final flake size tracks seed density as the paper's mechanism predicts.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If edge attachment limits growth, then enlarging flakes means extending growth time or raising edge incorporation rate, not just increasing precursor flux.
  • Seed density, rather than vapor supersaturation alone, determines how many crystals form; fewer seeds should yield larger flakes.
  • Neighboring crystals do not steal precursors from each other, so local crowding is not a barrier to uniform growth until crystals physically merge.
  • Avrami/JMAK analysis of ensemble data alone can misclassify the growth mechanism; individual-crystal trajectories are required to distinguish interface control from transport control.
  • The tracking workflow transfers to other 2D materials grown by CVD, making mechanism identification statistical rather than anecdotal.

Where Pith is reading between the lines

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

  • A direct test of the seed hypothesis would be to deliberately vary MoO3 source temperature or substrate pretreatments to change MoOxSy particle density and check whether final flake size scales inversely with seed density.
  • The non-competitive conclusion implies that in this regime, dense arrays of MoS2 crystals can be grown without sacrificing per-crystal growth rate, which is favorable for wafer-scale uniformity—though the authors do not state this explicitly.
  • The size-rate correlation at fixed time assumes uniform local supersaturation; if a future experiment measures gas-phase or surface precursor concentration near crystals and finds spatial gradients, the edge-attachment assignment would need revisiting.
  • One might extend the analysis to bilayer or multilayer growth: the same tracking could reveal whether second-layer nucleation obeys the same edge-attachment kinetics or becomes diffusion-limited as the terrace widens.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper reports a semi-automated image-processing pipeline for tracking the nucleation and growth of more than 400 MoS2 crystals in a single in-situ optical microscopy CVD experiment. From the extracted growth trajectories, the authors conclude that MoS2 growth is edge-attachment-limited rather than diffusion-limited, based on a positive correlation between instantaneous growth rate and crystal area at a fixed time (Figure 4). They further report that neighboring crystals grow independently until physical impingement, based on the absence of a correlation between growth rate and local neighbor density (Figure 6) and on centroid-distance measurements for selected crystal pairs. The authors propose that the non-competitive growth arises from seeded growth: MoOxSy nanoparticles at crystal centers act as local Mo reservoirs. They also use a JMAK analysis of total coverage to illustrate that ensemble-level fits can be misleading.

Significance. If the central mechanistic conclusion is correct, it would provide direct in-situ evidence for edge-attachment-limited growth of MoS2 under high-temperature CVD conditions and would identify seed density as a practical control parameter for crystal size. The high-throughput tracking of several hundred individual crystals is a genuine methodological advance, and the cautionary JMAK discussion is a useful contribution to the interpretation of ensemble growth data. However, the load-bearing evidence for the headline claim is currently qualitative and does not exclude plausible confounding effects. The paper would be strengthened substantially by a quantitative statistical analysis that controls for nucleation time and spatial position.

major comments (4)
  1. [Figure 4 / 'Diffusion- vs. Edge-attachment-limited growth'] The central claim that growth is edge-attachment-limited rests on a qualitative positive correlation between dA/dt and crystal area at t=50 s. The manuscript does not report a fit of dA/dt = C*sqrt(A), the corresponding R^2 or confidence intervals, or a statistical comparison against the alternative null model dA/dt = constant. More importantly, the comparison does not control for nucleation time or spatial position, despite the authors' own Fig. 3c,d showing a strong dependence of growth rate on tnucl. Since earlier-nucleated crystals are systematically larger at t=50 s and grew under higher effective supersaturation, the observed size-rate correlation is equally consistent with a nucleation-time effect under diffusion-limited kinetics. The assertion that 'all crystals should experience similar supersaturation' is an assumption that needs to be tested. Please add spatial maps of growth
  2. [Figure 6 / 'Growth competition between MoS2 crystals'] The non-competition conclusion is based on the statement that 'the average crystal growth rate does not exhibit a noticeable correlation' with Nsurr or fsurr. No statistical test, effect size, or confidence interval is given, and the analysis does not control for nucleation time or location. If early-nucleated crystals are spatially clustered or if neighbor density correlates with local precursor flux, the average correlation could be masked. The centroid-distance analysis is a useful complement, but it appears to be limited to a small number of selected pairs, and the selection criteria and uncertainties are not reported. Please provide a regression of growth rate on Nsurr/fsurr with tnucl and coordinates as covariates, or an equivalent matched analysis, and report the number of pairs used for centroid-distance measurements.
  3. [Section 'Seeded MoS2 crystal growth'] The proposal that MoOxSy nanoparticles at crystal centers supply Mo and that Schwoebel-Ehrlich barriers confine Mo feedstocks is an interesting hypothesis but is not directly tested. The SEM observation shows nanoparticles, yet no quantitative link is established between seed size or density and growth rate, no controlled variation of seed density is reported, and no diffusion measurements are presented. Consequently, the conclusions that 'the number of MoS2 crystals formed on the substrate is controlled by the seed density' and that growing larger flakes requires fewer MoOxSy deposits go beyond the evidence. Please clearly present this as a hypothesis and discuss alternative explanations, such as a dominant gas-phase Mo supply with a large diffusion length.
  4. [Methods / 'Crystal tracking algorithm'] The Methods state that the minimum object size of 15 pixels (~27 um^2) means the 'nominal nucleation time differs from the real nucleation time due to the resolution limitation,' but the consequent delay is not quantified. If the detection delay depends on growth rate, the size-rate correlation in Figure 4 could be biased. A sensitivity analysis (e.g., varying the minimum object size or back-extrapolating the nucleation time from early linear growth) would help confirm that the central result is robust to this segmentation choice.
minor comments (4)
  1. [Results text] There are un-rendered cross-reference placeholders ('Error! Reference source not found.') in the Results section; these should be fixed before publication.
  2. [Abstract / Introduction] Minor typographical issues: 'transitional metal dichalcogenides' should be 'transition metal dichalcogenides'; 'Kolmogolov' should be 'Kolmogorov'; 'retrain' should be 'retain'; 'flaks' should be 'flakes'; and the Conclusions use both 'MoSxOy' and 'MoOxSy' for the same species.
  3. [Figure 5b] The claim that growth rates of different orientation groups are 'not statistically distinguishable' would benefit from reporting the actual statistical test, p-values, and error bars, especially since only a thresholded orientation assignment is possible with optical microscopy.
  4. [Eq. (3) / JMAK discussion] The JMAK fit yields n = 1.71, but no uncertainty or goodness-of-fit value is reported. The decomposition into (a, b, c) is acknowledged to be non-unique; stating this explicitly in the main text would prevent readers from over-interpreting the illustrative assignment.

Circularity Check

0 steps flagged

No significant circularity: central claims rest on direct observation and hypothesis testing, not on assumptions that presuppose the conclusions.

full rationale

The paper's conclusions are drawn from direct in-situ measurements rather than from a derivation whose inputs already contain the results. The edge-attachment-limited claim is based on the t = 50 s size–rate correlation in Figure 4, which is compared against the two competing predictions: diffusion-limited growth gives dA/dt constant (Eq. 1), while edge-attachment-limited growth gives dA/dt ∝ √A (Eq. 2). This is a hypothesis test, not a fit: Eq. (2) is not fitted to the data and then recycled as a prediction. The same-time comparison is intended to control for global supersaturation; the assumption that all crystals experience similar local supersaturation is a testable confound (spatial gradients could in principle mimic the correlation), but that is a validity concern, not circularity. The JMAK fit (Avrami exponent 1.71) is used only to show that JMAK alone is ambiguous and would incorrectly suggest diffusion-limited growth; it is not used to support the edge-attachment conclusion. The non-competition conclusion is based on neighbor-density correlations and centroid-distance trajectories, both observational. The seed mechanism is an ex-post explanation supported by SEM imaging and prior literature (e.g., Cain et al., Zhou et al.); it does not define the conclusion into existence. The only self-citations (Refs. 8 and 32) are contextual descriptions of the miniaturized CVD apparatus and a related review; neither carries the load of the central mechanistic claim. No equation is equal by construction to an input, and no fitted parameter is renamed as a prediction. Therefore no circular step is present.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The central claims rest on several hand-chosen analysis thresholds and domain assumptions about uniform supersaturation, validity of standard growth-law relations, and correctness of the tracking pipeline. The seed-based mechanism is an additional explanatory hypothesis that is not independently verified. No new physical entities are introduced.

free parameters (6)
  • JMAK exponent n and rate constant k = n = 1.71, k not reported
    Fitted to the coverage-vs-time curve (Eq. 3) to illustrate the ambiguity of JMAK analysis; not used in the central mechanistic claim.
  • Area-growth exponent = ~2.9 (text garbled)
    Fitted to the ensemble area coverage growth; descriptive only and not load-bearing.
  • Neighbor radius threshold = 3R
    Chosen by hand to define neighbors in the competition analysis (Fig. 6). The absence of correlation could be sensitive to this choice.
  • Orientation tolerance = ±15°
    Chosen by hand to assign crystals to 0° or 60° orientation groups; affects orientation distribution statistics.
  • Minimum object size = 15 pixels (~27 µm²)
    Set to suppress noise in detection; biases measured nucleation times and may exclude small crystals from analysis.
  • Centroid matching distance = 15 pixels
    Threshold for linking crystal objects across frames; tracking errors could propagate into growth trajectories and affect all downstream statistics.
axioms (5)
  • domain assumption All crystals in the field of view experience the same precursor supersaturation at a given time.
    Stated before Fig. 4: 'since all crystals should experience similar supersaturation.' Required to interpret the size-dependence of dA/dt as edge-attachment kinetics.
  • standard math Diffusion-limited growth implies dA/dt is size-independent; edge-attachment-limited growth implies dA/dt ∝ √A.
    Eqs. 1 and 2, taken from Balluffi et al. (ref 49). Used as the benchmark to classify the observed mechanism.
  • domain assumption The image tracking pipeline correctly identifies and tracks individual crystals across frames.
    The growth trajectories are the primary data. The RF segmentation achieves 96% accuracy on a validation set, but the tracking module includes 'redraw' and polycrystal classification steps that may introduce errors.
  • domain assumption MoOxSy nanoparticles at flake centers supply Mo feedstock locally, and a Schwoebel-Ehrlich barrier prevents Mo from escaping the crystal.
    Proposed in the 'Seeded MoS2 crystal growth' section, based on SEM images and refs 37,63–66. This is the explanation for non-competitive growth but is not directly tested in this paper.
  • domain assumption The substrate and gas flow are uniform across the field of view.
    The competition analysis assumes that local crystal density is the only spatial variable affecting growth, ignoring possible inhomogeneities in temperature, gas flow, or substrate condition.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Revealing the MoS2 Growth Mechanism in Chemical Vapor Deposition: Real-Time Imaging and Statistical Analysis." pith.science (2026). https://pith.science/paper/77VCVA5I

@misc{pith2026260713893,
  author       = {Pith},
  title        = {Pith review of: Revealing the MoS2 Growth Mechanism in Chemical Vapor Deposition: Real-Time Imaging and Statistical Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/77VCVA5I}},
  note         = {Machine review of arXiv:2607.13893}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Chemical Vapor Deposition (CVD) is a promising method for scalable synthesis of two-dimensional transitional metal dichalcogenides (TMDs) such as MoS2, but challenges in reproducibility and controllability persist due to an incomplete understanding of their dynamic growth mechanisms. While in-situ characterization methods could provide valuable insights, it remains challenging to track a large ensemble of crystals to enable quantitative, statistical analysis. Here, we address this gap by developing and applying a semi-automated image processing pipeline to analyze in-situ optical microscopy footage of MoS2 growth. This framework enables the high-throughput reconstruction of complete growth trajectories for over 400 individual crystals from a single experiment. Our statistical analysis demonstrates that MoS2 crystallization is governed by an edge-attachment-limited mechanism rather than by precursor diffusion. Furthermore, MoS2 crystals exhibit non-competitive growth, indicating that precursor supply does not limit the growth of neighboring flakes until physical impingement occurs. These findings provide direct, quantitative evidence that advances the fundamental understanding of TMD growth, establishing a powerful methodology for rational optimization of the CVD growth of two-dimensional materials.

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

9 extracted references · 4 canonical work pages

  1. [3]

    These features include centroid location, orientation, size and boundary contour

    Crystal feature extraction After each unique MoS2 crystal is identified and labeled throughout the image sequence, the workflow extracts time-dependent shape features of all the crystals from the labeled frames. These features include centroid location, orientation, size and boundary contour. Because segmented crystals are often not perfectly polygonal, w...

    2014

  2. [5]

    (29) Dumcenco, D.; Ovchinnikov, D.; Marinov, K.; Lazić, P.; Gibertini, M.; Marzari, N.; Sanchez, O

    https://doi.org/10.1038/srep10257. (29) Dumcenco, D.; Ovchinnikov, D.; Marinov, K.; Lazić, P.; Gibertini, M.; Marzari, N.; Sanchez, O. L.; Kung, Y. C.; Krasnozhon, D.; Chen, M. W.; Bertolazzi, S.; Gillet, P.; Fontcuberta I Morral, A.; Radenovic, A.; Kis, A. Large-Area Epitaxial Monolayer MoS2. ACS Nano 2015, 9 (4), 4611–4620. https://doi.org/10.1021/acsna...

    work page doi:10.1038/srep10257 2015

  3. [7]

    (19) Zhan, K.; Chen, Y.; Xiong, Z.; Zhang, Y.; Ding, S.; Zhen, F.; Liu, Z.; Wei, Q.; Liu, M.; Sun, B.; Cheng, H

    https://doi.org/10.1038/srep40091. (19) Zhan, K.; Chen, Y.; Xiong, Z.; Zhang, Y.; Ding, S.; Zhen, F.; Liu, Z.; Wei, Q.; Liu, M.; Sun, B.; Cheng, H. M.; Qiu, L. Low Thermal Contact Resistance Boron Nitride Nanosheets Composites Enabled by Interfacial Arc-like Phonon Bridge. Nat. Commun. 2024, 15 (1). https://doi.org/10.1038/s41467-024-47147-1. (20) Choi, S...

    work page doi:10.1038/srep40091 2024

  4. [255]

    The image frames are then subject to a series of operations as described below

    After cropping the frames, the region of interest (ROI) typically contains 600 – 2000 pixels in both dimensions, with a pixel size of 1.35 × 1.35 µm2. The image frames are then subject to a series of operations as described below. Histogram equalization is applied to images to enhance the contrast between MoS2 and the surrounding substrate. As the field o...

    2000

  5. [602]

    (24) Zhang, J.; Zhai, T.; Yan, Y.; Fang, Q.; Liu, Y.; Zhu, Y.; Tu, W.; Lin, C.-Y.; Wang, Y.; Lou, J

    https://doi.org/10.1038/s41467-026-68320-8. (24) Zhang, J.; Zhai, T.; Yan, Y.; Fang, Q.; Liu, Y.; Zhu, Y.; Tu, W.; Lin, C.-Y.; Wang, Y.; Lou, J. Scalable Mechanical Exfoliation of Two-Dimensional Nanosheets by Polymer-Assisted Dry Ball-Mill of Layered Materials and Insights from Machine Learning. Mater. 28 Today Nano 2025, 30, 100604. https://doi.org/http...

    arXiv 2025

  6. [1939]

    G.; Lin, Y.-C.; Vincent, T.; Maruyama, M.; Aji, A

    (58) Suenaga, K.; Ji, H. G.; Lin, Y.-C.; Vincent, T.; Maruyama, M.; Aji, A. S.; Shiratsuchi, Y.; Ding, D.; Kawahara, K.; Okada, S.; Panchal, V.; Kazakova, O.; Hibino, H.; Suenaga, K.; Ago, H. Surface-Mediated Aligned Growth of Monolayer MoS2 and In-Plane Heterostructures with Graphene on Sapphire. ACS Nano 2018, 12 (10), 10032–10044. https://doi.org/10.10...

    arXiv 2018

  7. [2019]

    (3) Wang, H

    https://doi.org/10.1002/adma.201801583. (3) Wang, H. S.; Chen, L.; Elibol, K.; He, L.; Wang, H.; Chen, C.; Jiang, C.; Li, C.; Wu, T.; Cong, C. X.; Pennycook, T. J.; Argentero, G.; Zhang, D.; Watanabe, K.; Taniguchi, T.; Wei, W.; Yuan, Q.; Meyer, J. C.; Xie, X. Towards Chirality Control of Graphene Nanoribbons Embedded in Hexagonal Boron Nitride. Nat. Mate...

    arXiv 2021

  8. [2020]

    Prehistory

    (45) Mao, Y.; Dong, N.; Wang, L.; Chen, X.; Wang, H.; Wang, Z.; Kislyakov, I. M.; Wang, J. Machine Learning Analysis of Raman Spectra of MoS2. Nanomaterials 2020, 10 (11), 1–13. https://doi.org/10.3390/nano10112223. (46) Tang, B.; Lu, Y.; Zhou, J.; Chouhan, T.; Wang, H.; Golani, P.; Xu, M.; Xu, Q.; Guan, C.; Liu, Z. Machine Learning-Guided Synthesis of Ad...

    work page doi:10.3390/nano10112223 2020

  9. [2022]

    (21) Zhai, W.; Li, Z.; Zhai, L.; Yao, Y.; Li, S.; Wang, Y.; Ren, Y.; Chi, B.; Ge, Y.; Zhang, H

    https://doi.org/10.1038/s41467-022-29182-y. (21) Zhai, W.; Li, Z.; Zhai, L.; Yao, Y.; Li, S.; Wang, Y.; Ren, Y.; Chi, B.; Ge, Y.; Zhang, H. Epitaxial Growth of Wafer-Scale Single-Crystal Transition Metal Dichalcogenide Monolayers for Future Electronics. Matter. Cell Press August 3, 2022, pp 2405–2408. https://doi.org/10.1016/j.matt.2022.05.008. (22) Fu, J...

    work page doi:10.1038/s41467-022-29182-y 2022

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.