Machine learning-driven insights into biaxial strain-induced anomalous thermal conductivity enhancement of boron arsenide

Boron arsenide (BAs) is a promising integrated circuit thermal management material with ultra-high thermal conductivity (κ) and exceptional semiconductor properties. In practical devices, mechanical stress and strain caused by thermal expansion inevitably affect the thermodynamic properties of heat sink materials. Consequently, investigation of the effects of stress and strain on theκof BAs is essential. Accurate assessment and regulation of theκand phonon heat transport of BAs requires consideration of both three-phonon (3ph) and four-phonon (4ph) interactions. However, anharmonicity computation using density functional theory is computationally expensive, particularly for the fourth-order interatomic force constants. This work uses a machine learning-driven moment tensor potential method to evaluate higher-order anharmonicity and extends it to include fourth-order forces, accelerating the calculations significantly. Results show that theκof BAs calculated using the moment tensor potential method when considering both 3ph and 4ph processes agrees well with experimental data. In contrast, considering the 3ph interaction alone causesκto be overestimated. The machine learning approach reduces computational costs by approximately 94% while maintaining comparable accuracy to density functional theory. By solving the phonon Boltzmann transport equation, the thermal transport properties of BAs under biaxial tensile strains ranging up to 7% are evaluated. BAs shows an anomalousκenhancement when including both 3ph and 4ph interactions, with a maximum enhancement of 15% at a small strain of 1%, primarily because of the weakening of the 3ph scattering. Beyond 2% strain, both the 3ph and 4ph scattering rates increase significantly, causing shorter phonon lifetimes and a monotonic reduction inκ. Additionally, theκtemperature dependence under strain is explored, highlighting temperature’s role in modulating phonon scattering. This study provides machine learning-driven insights into the BAs thermal response under biaxial tensile strain and offers a new perspective for understanding similar anomalous strain-induced phenomena in other materials.