Data science of the in silico crystallization

A data-driven approach is used to describe the growth of crystals and thin films based on high-throughput numerical experiments. This way the Machine Learning (ML) helps to reveal hidden patterns in the results from our atomistic-scale model which combines Cellular Automaton and Monte-Carlo moduli shaped as a Kossel crystal, through a carefully devised monitoring scheme, i.e. our computational nanoscope. Thus, from the quantities monitored in real-time, like growth rate and surface roughness, evolves the powerful concept of a crystal growth kernel — an ML-based comprehensive digital library that describes the growth of a crystal adequately across all the chosen ranges of system parameters, sizes, from atomistic to macroscopic, and growth modes — step-flow growth and instabilities, nucleation, rough growth, etc. Here we start with the analysis of how the kernel`s prediction accuracy and time depend on the number of parameters.