Analysis of polymerization and nanoparticle formation in silane plasma by unsupervised learning method and statistics in complex chemical network

Chemical reactions in low-temperature plasma are so functional for industrial processes, but are too complicated to understand what are going on in a specific process, and we frequently detect their outputs without their internal cause-effect relationships. Some previous studies [1] proposed visualization and centrality analysis based on complex network science to unveil such underlying processes. Following them, in this study, we investigate silane plasma chemistry, in which polymerization and nanoparticle formation are active, by two analytical methods. Using community detection, which is one of the unsupervised learning methods, we detect automatically one species group that contributes to this chemical processes. Furthermore, when we trace polymer molecules in the corresponding degree-distribution spectrum, we observe significant descents of degrees as the count of Si atoms involved in a molecule increases [2]. This fact indicates that, starting from a silane molecule which is the mother gas species, polymer growth makes heavier species less important, with possible removal of nanoparticles as process outputs without impacts on this chemical system. [1] T. Murakami and O. Sakai, Plasma Sources Sci. Technol. 29, 115018 (2020). [2] O. Sakai, S. Kawaguchi and T. Murakami, Jpn. J. Appl. Phys. 61 (2022) (in Press).