Thermal conductivity of h-BN monolayers using machine learning interatomic potential.

Area of Science:

• Materials Science
• Computational Physics
• Nanotechnology

Background:

• Miniaturized electronic devices require efficient thermal management materials.
• Evaluating thermal conductivity through theoretical design is computationally intensive.
• Hexagonal boron nitride (h-BN) is a promising material for thermal management.

Purpose of the Study:

• To apply machine learning interatomic potentials (MLIP) for evaluating thermal conductivity in hexagonal boron nitride (h-BN) monolayers.
• To assess the accuracy and efficiency of MLIP compared to traditional methods.
• To establish reliable benchmarks for MLIP quality in predicting lattice dynamical properties.

Main Methods:

• Developed a machine learning interatomic potential (MLIP) using the Gaussian approximation potential (GAP) method.
• Calculated lattice dynamical properties and thermal conductivity of h-BN monolayers using MLIP.
• Compared MLIP results with explicit frozen phonon calculations.

Main Results:

• Accurate thermal conductivity predictions were achieved using MLIP trained on approximately 30% of representative configurations.
• High-order force constants proved to be a more reliable indicator of MLIP quality than the harmonic approximation.
• MLIP significantly reduces the computational cost associated with thermal conductivity evaluation.

Conclusions:

• MLIP offers an efficient and accurate method for predicting the thermal conductivity of h-BN monolayers.
• The developed MLIP approach can accelerate the design and discovery of novel thermal management materials.
• High-order force constants are crucial for validating MLIP performance in lattice dynamics.