Nanotwinning induced decreased lattice thermal conductivity of high temperature thermoelectric boron subphosphide (B12P2) from deep learning potential simulations

Abstract

Boron subphosphide (B12P2) is a promising high temperature thermoelectric material due to its good thermal stability, and chemical inertness. However, the thermal properties of B12P2 have not been well revealed so far. Here, we first develop a deep learning potential for B12P2 based on quantum mechanical calculations. Then the isotropic lattice thermal conductivity (LTC) of crystalline B12P2 is predicted to be 39.70 ± 4.38 W/m⋅K from molecular dynamics simulations using this deep learning potential. The LTC exhibits the relationship of κL∼1/T in the temperature range of 300 ∼ 1500 K. More important, a twin boundary strategy is proposed to reduce the LTC of B12P2. In nanotwinned B12P2, the phonon transport in all directions is significantly suppressed by twin boundaries (TBs) with the isotropic LTC of 15.85 ± 2.70 W/m⋅K, especially in the direction normal to the TB plane. The decrease of vibrational density of states and phonon participation ratio due to TBs' phonon scattering is the main reason of the low LTC in nanotwinned B12P2. In addition, the elastic moduli (B and G) of B12P2 crystal decrease by less than 7% after inducing TBs, which suggests that the mechanical properties are not significantly affected by TBs. Overall, this work enriches our understanding of the thermal properties of B12P2 and offers a promising approach, i.e., introducing TBs, to design high-performance thermoelectric materials.