Skip to content

Reactive Force Field Molecular Dynamics Studies of the Initial Growth of Boron Nitride Using BCl3 and NH3 by Atomic Layer Deposition

Evidence and attribution

Authority of statements

Prose sections below (Summary, Methods, Findings, etc.) are curated summaries of the publication identified by doi, title, and pdf_path in the front matter above. They are not new primary claims by this wiki.

For definitive numerical values, reaction schemes, and interpretations, use the peer-reviewed article (and optional records under normalized/papers/ when present)—not this page alone.

Summary

Develops a ReaxFF description for BN ALD from BCl\(_3\) + NH\(_3\), trains bonded/geometry-sensitive terms against DFT, and runs cycle-resolved ReaxFF MD mimicking pulse–purge ALD steps. The growth story is decomposed into surface diffusion, BN cluster nucleation/growth, HCl formation/diffusion/desorption, and temperature-sensitive competition between 3D cluster growth vs 2D film growth across five simulated cycles. Substrate temperature modulates initial growth mode and film thickness—too-high T accelerates desorption of gases/clusters, suppressing film thickening in the regimes explored. Industrial ALD of h-BN is motivated by dielectric and diffusion-barrier applications; BCl\(_3\)/NH\(_3\) chemistry is highly reactive, so atomistic models must treat halogen byproducts explicitly.

Methods

  • QM reference / ReaxFF training: Parameters are trained against density functional theory data describing BCl\(_3\) geometries and BCl\(_3\)/NH\(_3\) surface reactions that produce BN films and HCl (abstract and §2 opening in normalized/extracts/2024naoya-uene-j-phys-chem-reactive-force_p1-2.txt). The peer-reviewed article specifies the DFT program, functional, basis/cutoffs, and training sets; those details are not fully reproduced in the short checked-in extract (truncated mid-§2.1).
  • ALD cycle in ReaxFF MD: The process is modeled as four repeated steps: (1) BCl\(_3\) pulse, (2) first purge, (3) NH\(_3\) pulse, (4) second purge (abstract). The abstract reports five simulated ALD cycles.
  • Growth decomposition (abstract): (i) BCl\(_3\)/NH\(_3\) surface diffusion, (ii) BN cluster formation/growth, (iii) HCl formation, (iv) HCl surface diffusion, (v) HCl desorption.
  • Reactive MD implementation: ReaxFF bond-order dynamics as described in §2.1 of the article; integration timestep, thermostat, substrate supercell size, and temperature setpoints for each pulse are given in the full PDF at pdf_path—consult it for values not stated in the extract on disk.

1 — MD application (ReaxFF production). Engine / code: ReaxFF reactive MD (§2.1; LAMMPS-class usage is typical for this corpus—confirm in the VOR PDF). System and boundaries: BCl\(_3\) and NH\(_3\) on a surface slab with PBC as defined in Computational Methods; atom counts, full cell vectors, and layer counts are not in the short p1–2 extract (truncated mid-§2.1). Cycle protocol (abstract): (1) BCl\(_3\) pulse, (2) first purge, (3) NH\(_3\) pulse, (4) second purge, repeated for five ALD cycles. Stages (abstract): (i) surface diffusion, (ii) BN cluster nucleation/growth, (iii) HCl formation, (iv) HCl diffusion, (v) HCl desorption. Ensemble, timestep, thermostat, total trajectory length, barostat, stress, E-field, enhanced sampling: N/A in this page summary — the indexed excerpt does not list NVT thermostat coupling, fs timestep, or ps/ns duration of each pulse/purge (total production time per cycle is in the full article at pdf_path). N/A — no NPT/barostat, external electric field, or umbrella/metadynamics protocol is stated in the abstract-level text summarized here (confirm negation in the full article if needed).

2 — Force-field training. Parent: ReaxFF parameterization for B/N/Cl/H and BN from BCl\(_3\)/NH\(_3\) surface reactions. QM reference: DFT (program, functional, basis, cutoffs) in §2 for BCl\(_3\) and surface pathways to BN and HCl—full tables not in the p1–2 extract on disk. Training set and optimization: QM bond/angle/energy targets and iterative ReaxFF refinement (see article §2–3). Reference / validation data: DFT (primary); experimental comparisons as cited by the authors.

3 — Static QM — DFT is the reference for fitting, not a separate static-only results paper; detailed settings are in the article text/SI.

Findings

Outcomes and growth modes. Across five ALD cycles, simulations report coexisting 3D cluster growth and 2D film growth. HCl formation, surface diffusion, and desorption couple to BN coalescence.

Comparisons and levers (temperature). Substrate temperature shifts the balance: moderate T supports BN cluster formation/growth, while excess T increases desorption of gas-phase species and BN clusters, reducing net film thickening in the regimes highlighted in the abstract—so precursor flux alone does not control smoothness if byproduct removal is too fast.

Authored limitations. Chamber-scale fluid/reactor physics and long-time industrial ALD are outside the periodic slab model (see ## Limitations).

Limitations

Simulation duration/cycles are still far below industrial wafer-scale times; surface models omit full reactor fluid dynamics—ALD macroscale coupling is named as future work in the abstract.

Chamber wall reactions, precursor parasitic CVD, and wafer temperature nonuniformity during pump/purge cycles are not represented in the periodic slab ALD loop; use the ReaxFF trends as relative comparisons between temperature setpoints rather than absolute Å/cycle numbers for production tools.

Relevance to group

Industrial ALD collaboration (Japan Advanced Chemicals / Tohoku) with van Duin co-authorship; complements gas-phase BN synthesis papers (20220000-0002-1558-1560-x-reaxff-force) with a surface-process viewpoint.

Citations and evidence anchors

https://doi.org/10.1021/acs.jpcc.3c06704 — Abstract (~pp. 1–2) lists ALD loop stages and growth steps (i–v).