Mechanistic insights into the thermal decomposition of hexamethyldisilane: a reactive molecular dynamics study

Hexamethyldisilane (HMDS) serves as a critical single-source precursor for the chemical vapor deposition (CVD) of silicon carbide (SiC), yet its atomic-level pyrolysis mechanism and the kinetics of radical generation remain unclear. This study investigates the thermal decomposition behavior of HMDS to provide theoretical guidance for optimizing SiC deposition processes. The results demonstrate that HMDS pyrolysis follows first-order kinetics with an apparent activation energy of 44.47 kcal/mol, a value significantly lower than the theoretical dissociation energy of the Si-Si bond. By combining this kinetic data with reaction pathway analysis, it is concluded that the decomposition is governed by a multistep cooperative mechanism rather than simple homolytic bond cleavage. The reaction proceeds through three distinct stages: initial precursor decomposition dominated by C-Si bond dissociation, secondary reactions of intermediates involving cascading demethylation, and small-molecule formation accompanied by radical recombination. Methyl radicals (CH₃) are identified as the primary chain carriers, which are ultimately converted into thermodynamically stable methane (CH₄) via hydrogen abstraction. Furthermore, temperature is found to critically regulate the generation and accumulation behavior of CH₃ radicals.

Density functional theory (DFT) calculations were carried out with Gaussian 16 at the unrestricted ωB97XD/6-311G(d,p) level to optimize geometries and train the force field. A broken-symmetry strategy with guess = (mix,always) and nosymm was adopted to reliably describe bond dissociation and radical behaviors. Using the high-quality DFT data, the ReaxFF force field was further optimized. Reactive molecular dynamics simulations were then performed in LAMMPS with the optimized potential under the NVT ensemble at 2500-4000 K with a 0.1-fs time step. A cubic box with 100 HMDS molecules and periodic boundary conditions was adopted, and each condition was run three times for statistical reliability.