Uncovering atomistic mechanisms of crystallization using Machine Learning

< back to research overview

Crystal growth is a challenging process to model quantitatively because of the notoriously complex atomic environments near solid-liquid interfaces. In order to make sense of such atomic environments we have introduced a novel data-driven approach to systematically detect, encode, and classify all atomic-scale mechanisms of crystallization. Machine learning and atomistic simulations were employed together to uncover the role of the liquid structure on the process of crystallization and derive a predictive kinetic model of crystal growth.

Figure 1: Snapshots of a crystal growth simulation of silicon using molecular dynamics. All atomic crystallization events are systematically detected, encoded, and classified according to the local structure using machine learning.

Crystallization from the melt (fig.1) is a pervasive process in industry, from metal casting for structural applications to the Czochralski process for semiconductor wafer growth for electronics. It is important to control and understand the crystal growth process because it is at this stage that the material’s microstructure morphology is created, which in turn defines the material’s properties. Consequently, a great deal of effort has been put into understanding the complex interplay between structure, thermodynamics, and kinetics that governs the process of crystal growth. This has led to a mechanism-based understanding of crystallization in terms of the microstructural elements of the crystallite being formed. Considerably less attention has been put in understanding the effects that the liquid adjacent to the solid–liquid interface has on the process of crystal growth.

In this work we have introduced a novel data-driven approach to systematically detect, encode, and classify all atomic-scale mechanisms of crystallization according to the local structure using machine learning. Our approach leads to the construction of a structural quantity (namely softness) that captures the propensity for atomic rearrangements leading to crystallization to occur (Fig.2a). We show that our strategy naturally leads to a predictive kinetic model of crystallization that takes into account the entire zoo of microstructural elements present at solid-liquid interfaces. Our results were only made possible by employing atomistic simulations and machine learning together. The strength of this combined approach is that one can perform complex simulations and yet glean physical insight from notoriously complex atomic environments.

Figure 2: (a) Cross section of a snapshot of the initial stages of silicon growth. Liquid atoms are colored according to their softness value, while atoms in the crystalline phase are colored in gray. (b) and (c) Illustration of how interface-induced ordering of the liquid alters the local structure around crystallizing atoms and affects the activation energy for solidification.

Our new approach also leads to the development of novel conceptual knowledge about the process of crystallization. We have observed that the solid-liquid interface induces partial ordering of the nearby liquid during crystal growth. Using the new insights obtained from the machine learning results we discovered that the ordering of liquids strongly affects the process of crystal growth in metals and semiconductors. It is found that the modified structure of the liquid nearby solid–liquid interfaces reduces the mobility of liquid atoms and thus slows down the crystal growth kinetics. The physical mechanism behind the interface-induced ordering of the liquid is explained in fig.2b and 2c. The crystallizing atom (green) has its local structure (illustrated here only by its first neighbors) affected by the nearby solid–liquid interface. High-index or rough interfaces (fig.2b) interact strongly with the liquid and cause significant ordering of the liquid, which becomes rigid, resulting in large activation energies (ΔEa) when compared to the barrier for diffusion in the liquid (ΔEd). Low-index interfaces (fig.2c) interact weakly with the liquid and cause very small ordering, resulting in low ΔEa.

Our work elevates the liquid structure to the same level of importance as the crystal surface morphology in understanding crystallization, a knowledge that can enable material advances through the incorporation of liquid-structure engineering as a novel pathway for synthesis. This innovative application of ML in materials science blends conventional scientific methods with data science tools to produce physically consistent predictive models and novel conceptual knowledge.

Contact: Rodrigo Freitas, freitas_at_stanford.edu


Freitas, R. and Reed, E. J., Uncovering the effects of interface-induced ordering of liquid on crystal growth using machine learning. Nature Communications, 11 3260 (2020).