It is predicted by the Boltzmann transport equation (BTE) that the thermal conductivity of boron arsenide (BAs) at room temperature is as high as 1441 W/(m K). However, the experimental measurements exhibit striking variations, ranging from 53 to 2100 W/(m K). To reveal the divergence, high-precision neuroevolution potential (NEP) is first trained using density-functional theory (DFT) data and machine learning methods. Subsequently, NEP-based molecular dynamics (MD) simulations are used to investigate the growth mechanism of BAs, the formation and recombination of native defects, and lattice thermal conductivity. At 900 K, the formation enthalpy and cohesive energy for the cubic phase BAs are −0.017 and −5.38 eV, indicating that vapor B and As atoms can form the BAs phase instead of precipitating elemental phases B and As. At high temperatures, ${\mathrm{V}}_{\mathrm{B}}$, ${\mathrm{V}}_{\mathrm{As}}$, ${\mathrm{A}}_{\mathrm{B}}$, and ${\mathrm{A}}_{\mathrm{As}}$ exhibit low formation energy, suggesting that they may exist during crystal growth. Furthermore, we observe that ${\mathrm{A}}_{\mathrm{B}}\text{−}{\mathrm{A}}_{\mathrm{As}}$ pairs undergo recombination. Our investigation suggests that the ${\mathrm{A}}_{\mathrm{B}}\text{−}{\mathrm{A}}_{\mathrm{As}}$ pairs can be eliminated by reducing the cooling rate of the crystal, which provides guidance for crystal growth. Applying the homogeneous nonequilibrium molecular dynamics (HNEMD) approach, the thermal conductivity for defect-free BAs is predicted to be 1333 W/(m K). Due to defects scattering phonons, the thermal conductivity is suppressed to 44–208 W/(m K) for a system with a defect concentration of 1.40 $\times {10}^{20}{\mathrm{cm}}^{-3}$, which explains the underestimation of thermal conductivity in experimental measurements. This work establishes a pathway for identifying and understanding the functionality of defect suppression materials.

Close

PrevNext

View figure in article

• Defects
• Thermal conductivity
• Machine learning
• Molecular dynamics