Melting of thin silicon films: a molecular dynamics study with two machine learning potentials

TL;DR

This study uses molecular dynamics with machine learning potentials to investigate the thermal stability and melting behavior of thin silicon films and silicene, revealing how film thickness influences decomposition and melting processes.

Contribution

It compares two machine learning potentials, SNAP and GAP, for modeling silicene and thin silicon films, highlighting their strengths and limitations in predicting melting behavior.

Findings

Silicene melts at 500 K, losing structure.

Melting temperature increases with film thickness, saturating at 28 layers.

GAP potential fails to accurately model gas phase decomposition.

Abstract

Thermal stability of silicene and thin silicon films is studied by molecular dynamics using two machine-learning potentials, SNAP and GAP. For SNAP potential, systems ranging from a single silicene layer to films of 36 layers are considered. Silicene is found to lose its structure at 500 K. The decomposition temperature increases with film thikness and reaches saturation at about 28 layers, corresponding to the bulk melting point of the SNAP model (1380 K). Thin films up to 8 layers exibit two-phase coexistence upon decomposition, while thicker films undergo surface melting followed by complete collapse into the liquid state. The GAP potential, although more accurate for bulk silicon, fails to describe the gas phase: silicene modelled with GAP decomposes into a set of small clusters. The results are compared with earlier data for the Stillinger-Weber potential.