Uncovering atomistic mechanisms of crystallization using Machine Learning.

The process of solidification has been used through centuries to create materials with specific shapes and compositions, since much before microscopy was able to reveal the microstructure of materials. With the maturation of computational and microscopy techniques, a mechanism-based understanding of the crystallization process was developed. Yet, current methods fall short of delivering full atomistic understanding of solidification kinetics due to difficulties in capturing and classifying the complex atomic-scale processes at play. Here, we present an approach to uncover the fundamental atomistic events leading to silicon crystallization. Such events are classified according to the microstructure surrounding the crystallizing atoms using a Machine Learning algorithm to optimally categorize the local atomic structure. We discover that this approach reveals a local-structure dependent kinetic model for crystallization with predictive capabilities. From the model we also draw new insights on the role of interface-induced ordering of the liquid on the kinetics of crystallization.