Skip to content

ReaxFF studies of surface fluorination of alumina and etching of alumina/aluminum metal heterostructures under gas-phase hydrogen fluoride exposure

Reactive molecular dynamics with a merged Al/O/H/F ReaxFF parameterization is used to study HF-driven fluorination and etching of α-Al₂O₃(0001) and alumina / Al / alumina heterostructures. The work emphasizes how surface termination (Al-rich versus O-rich), HF chemical potential (replenishment and gas-phase loading), and temperature (1250 K for bare alumina; 750 K for heterostructures to remain below bulk Al melting) control conversion, HF dissociation kinetics (Arrhenius analysis), and volatile products (including AlHₓ in metal-containing stacks).

Summary

The study investigates self-limiting surface conversion of alumina and oxide–metal–oxide stacks under gas-phase HF using ReaxFF. Simulations compare 100% Al–, 50% Al–, and 100% O-terminated α-Al₂O₃(0001) models and several sandwiched Al heterostructures (thick crystalline oxide, thin amorphous oxide, thin crystalline oxide). DFT (VASP for periodic HF adsorption on alumina; Jaguar M06-2X for small gas-phase HF/H₂O and Al-oxide fragments) supports training and spot checks of HF dissociation pathways relative to ReaxFF.

Methods

  • Force field: ReaxFF formulation with EEM charges, bond-order bonded terms, Coulomb + van der Waals nonbonded terms, and Taper truncation; the manuscript merges prior Al/O/H and Al/F/H training and extends the set for HF on α-Al₂O₃(0001) (adsorption structures, barriers) as described in Sec. II.C and Fig. 1.
  • DFT reference: VASP, GGA-PBE, PAW, 500 eV cutoff, (2×5×1) k-mesh for a 12-layer (2×1) α-Al₂O₃(0001) slab; Jaguar M06-2X / 6-311++G(d,p) for constrained HF/H₂O and cluster fragments.
  • MD engine and protocol: Simulations in LAMMPS; slabs and heterostructures are relaxed (NPT, Berendsen thermostat and barostat, 300 K, pressure damping 5000 fs, temperature damping 100 fs), then heated (NVT, Berendsen, 0.02 K/fs) to 1250 K (alumina-only) or 750 K (heterostructures). Production runs use NVT at the target T; time step 0.25 fs; HF replenishment every 0.5 ns (non-rarefied, ≥120 HF) or 0.25 ns (rarefied, 10 HF); H₂O product molecules removed every 5000 steps; equilibration judged when HF is no longer consumed.
  • Kinetics: HF → H⁺ + F⁻ rate constants from Arrhenius fits of ln k vs 1/T after initial HF loading; additional analysis references Figs. S1–S3 and Table I for slab thinning and AlFₓ coordination.

1 — MD application (atomistic dynamics). Engine: LAMMPS reactive MD with the merged Al/O/H/F ReaxFF; slab α-Al₂O₃(0001) and oxide/Al/oxide heterostructures; 3D PBC; NPT relaxation then NVT production (see above); time step 0.25 fs; multistage heating and HF replenishment/H₂O removal; T = 1250 K (bare alumina) or 750 K (stacks, below Al melting). N/A — no replica/metadynamics; N/A — no static electric field; N/A — no NPT barostat during the quoted NVT etch legs.

2 — Force-field training — The manuscript merges prior Al/O/H and Al/F/H training; N/A as a net-new “from scratch” ReaxFF paper in the narrow sense, but new DFT-guided data for HF on α-Al₂O₃(0001) are added for reactive MD (Sec. II.C, Fig. 1).

3 — Static QM / DFT (spot QM and barrier checks). VASP GGA-PBE PAW (2×5×1) k-mesh 12-layer (2×1) slab 500 eV; Jaguar M06-2X/6-311++G(d,p) cluster work on gas-phase fragment and barrier checks.

Findings

  • On α-Al₂O₃(0001) at 1250 K with replenished HF, 100% Al-terminated surfaces show the largest reacted-Al fraction and most complete fluorination versus 50% Al and especially 100% O-terminated surfaces (the O-rich case forms dense hydroxyl coverage and little Al–F despite facile HF dissociation). Al–F coordination and alumina thinning track the degree of fluorination; –OH:–F ratios distinguish residual hydroxyls versus fluoride.
  • HF concentration on 100% Al-terminated surfaces is non-monotonic: higher HF loading can raise the barrier but also the pre-exponential factor, with an optimal loading (e.g., 280 HF in the reported series) giving the deepest conversion before very high loading (360 HF) suppresses dissociation via hydrogen-bonding and H transport through AlFₓ.

  • For oxide/metal/oxide stacks at 750 K, Al diffusion from the metal layer couples to AlHₓ and AlFₓ products; thinner oxides etch more strongly, and amorphous vs crystalline thin oxide kinetics differ in barrier vs pre-factor (collision frequency). Rarefied vs non-rarefied HF exposes kinetic vs transport limits on ns trajectory scales (see Figs. S1–S3/Table I).

  • Limitation (authored view): Idealized defect-free terminations isolate stoichiometry; outlook toward rough real films is in ## Limitations and the main text.

Limitations

The repository PDF is an author proof / galley (AIP query boilerplate appears in extracts); page and issue numbers in the file may be placeholders until the version of record is confirmed. Reactive dynamics are classical (ReaxFF) and omit explicit excited-state chemistry. Idealized defect-free terminations isolate stoichiometry effects but omit real polycrystalline / rough microstructures.

Relevance to group

Adri C. T. van Duin (Penn State) co-authors; the paper extends group ReaxFF development toward halogen plasma / ALE-relevant alumina and metal–oxide interfaces.

Citations and evidence anchors