Machine Learning and Molecular Dynamics for the thermal conductivity of doped semiconductors

Semiconductors are fundamental components of modern industry, as they serve as the backbone of electronics and energy technologies. Among their many characteristics, thermal properties play a crucial role in ensuring both performance and long-term reliability. Doped semiconductors, in particular, display unique and useful electronic properties; however, the introduction of impurities generally alters their thermal behavior, often leading to a reduction in thermal conductivity. To study this effect, equilibrium molecular dynamics (EMD) combined with the Green–Kubo formalism can be employed to calculate the thermal conductivity of doped semiconductor systems. In addition, the use of interatomic potentials, such as the Tersoff model, provides a framework to capture the underlying atomic interactions. By integrating machine learning techniques with molecular dynamics, it becomes possible to predict thermal properties across different doping levels and defect concentrations. Machine learning models, trained on simulation data, can reduce the computational cost of traditional simulations, which are typically both time- and resource-intensive. The results highlight how doping and defects modify thermal conductivity and help establish practical limits for impurity levels that still allow semiconductors to remain attractive for technological applications. The results are also of interest to determine the figure of merit of doped semiconductors. Understanding this relationship is essential for designing advanced materials that balance performance, efficiency, and reliability.