Revisiting temperature induced metallicity of the Si(001) Surface: insights from molecular dynamics simulations with machine learned interatomic potentials

Despite the importance of Si in modern technology, some microscopic phenomena of Si remain mysterious. One example is the temperature driven metallization of the Si(001) surface, on which surface Si atoms were found to dimerize. To understand this phenomenon, we investigate temperature dependent structural dynamics of Si dimers by performing molecular dynamics simulations with Behler-Parrinello Neural Network (BPNN), Gaussian approximation potential (GAP), and MACE machine learned interatomic potentials trained on a database derived from density functional theory-based calculations. Despite the differences in the architecture of these models, they are in great agreement with each other for simulation of the Si(001) surface. We find that the dimers can occupy the higher energy symmetric configuration (associated with metallic behavior) even at temperatures as low as 300 K with a finite probability, which increases with temperature as the dimers flip rapidly between their ground state configurations, i.e., the asymmetric (buckled) dimers, with the flipping rates of 10⁷-10⁹ s^-1. We also find that starting around 700 K dimer breaking does occur with a low probability. These results are in accord with experimental observations of metallicity on Si(001) with an onset around 400 K which increases with increasing temperature followed by an abrupt rise around 700 K. By capturing the dynamics of dimer flipping, these simulations provide a rationale for the origin of the observed metallicity.