Competitive Hydrogen Migration in Silicon Nitride Nanoclusters: Reaction Kinetics Generalized from Supervised Machine Learning

• Published in: The journal of physical chemistry. A, Molecules, spectroscopy, kinetics, environment, & general theory Vol. 126; no. 17; pp. 2677 - 2689
• Main Authors: Choi, Yeseul, Adamczyk, Andrew J
• Format: Journal Article
• Language: English
• Published: United States American Chemical Society 05.05.2022
• Subjects: A: Structure, Spectroscopy, and Reactivity of Molecules and Clusters
• ISSN: 1089-5639; 1520-5215
• DOI: 10.1021/acs.jpca.2c01050

The rate coefficients for 52 hydrogen shift reactions for silicon nitrides containing up to 6 atoms of silicon and nitrogen have been calculated using the G3//B3LYP composite method and statistical thermodynamics. The overall reaction of substituted acyclic and cyclic silylenes to their respective silene and imine species by a 1,2-hydrogen shift reaction was sorted by three different types of H shift reactions using overall reaction thermodynamics: (1) endothermic H shift between N and Si:, (2) endothermic H shift between Si and Si:, and (3) exothermic H shift between Si and Si:. Endothermic H shift reactions between Si atoms have one dominant activation barrier where the exothermic H shift reaction between Si atoms has two barriers and a stable intermediate. The rate-determining step was determined to be from the intermediate to the substituted silene, and then kinetic parameters for the overall reaction were calculated for the two-step pathway. The single event pre-exponential factors, Ã, and activation energies, E a, for the three different classes of hydrogen shift reactions of silicon nitrides were computed. The hydrogen shift reaction was explored for acyclic and cyclic monofunctional silicon nitrides, and the type of hydrogen shift reaction gives the most significant influence on the kinetic parameters. Using a supervised machine learning approach, the models for predicting the energy barrier of three different hydrogen shift reactions were generalized and suggested based on selected descriptors.