Multi-scale modeling of gas-phase reactions in metal-organic chemical vapor deposition growth of WSe₂
Summary¶
The paper builds a multi-scale framework for gas-phase chemistry in tungsten diselenide MOCVD from W(CO)\(_6\) and H\(_2\)Se precursors. DFT is used to propose reaction classes and to supply a training set to extend ReaxFF to W/H/C/O/Se. ReaxFF-based reactive MD in periodic gas-phase cells enumerates major pathways and barriers that are converted to Arrhenius form. Those kinetics, together with transport and thermodynamic property data, feed a detailed gas-phase mechanism in a reacting-flow CFD model of the cold-wall horizontal reactor, so predicted tungsten chalcogenide species fluxes to the growth zone can be compared to measured film-thickness trends across the wafer (surface growth itself is not resolved atomistically).
Methods¶
The abstract stresses that experimental MOCVD chambers rarely provide spatially resolved gas-phase composition maps, motivating simulation. QM/ReaxFF: DFT calculations suggest reaction types and provide data to fit ReaxFF; ReaxFF simulations map pathways and barriers. Kinetics: compiled Arrhenius parameters and species thermophysical data enter the CFD model. CFD: solves reacting flow with thermal and mass transport for the MOCVD geometry used experimentally. Validation compares gas-phase concentrations of tungsten chalcogenides near the substrate to experimental measurements of average film thickness across the wafer. Surface growth steps are not resolved atomistically—the focus is gas-phase precursors feeding deposition.
ReaxFF gas-phase MD (reaction enumeration). The DFT training set and ReaxFF molecular dynamics are implemented in LAMMPS-class workflows (as stated in the article) using periodic simulation boxes for gas-phase species; NVT- or NVE-like stages (with a Nose–Hoover-type thermostat when NVT is used) run for sufficient duration to sample reaction pathways and barrier-like events feeding Arrhenius fits—timestep and ns-scale trajectory lengths are in the Methods (not all repeated here). Barostat for gas cells: N/A or NPT-like only if the article uses variable volume (confirm in PDF). Electric field: N/A. Metadynamics / umbrella / replica exchange: N/A for the standard gas-phase ReaxFF leg as summarized in the abstract.
ReaxFF parameterization (FF training). Parent ReaxFF is extended for W/H/C/O/Se; DFT PBE-class (or as stated) reaction and cluster data form the training set; CMA-ES/genetic-style optimization of ReaxFF parameters is described in the paper; validation uses both QM and reactor thickness trends as reference data.
Findings¶
The coupled model reports that predicted tungsten chalcogenide fluxes near the growth surface correlate with measured thickness trends as a function of position, supporting the hypothesis that film growth is linked to deposition of these gas-born species. The authors position the workflow as transferable to other CVD chemistries and useful for reactor optimization and reproducibility across chambers. Sensitivity is to inlet flow, heater layout, and chemistry completeness in the compiled mechanism; corpus honesty: use the J. Cryst. Growth PDF for figure-level validation and any uncorrected proof sibling at 2019xuan-venue-paper only for ingest notes.
Limitations¶
Reactor geometry, inlet mixing, and substrate temperature boundary conditions feed directly into CFD concentration fields; any multi-scale coupling to surface CVD models must import those gas-phase fluxes as boundary data rather than assuming uniform precursor delivery. Atomistic surface reactions and nucleation on WSe₂ are outside the paper’s explicit modeling scope. CFD predictions depend on mechanism completeness and boundary conditions; experimental diagnostics inside the chamber remain sparse. Adri C. T. van Duin co-authorship connects the work to corpus ReaxFF practice but does not guarantee parameter reuse without retraining checks for new chemistries.
Confidence rationale: high—primary journal article with detailed abstract.