Vacancy-induced phonon localization in boron arsenide using a unified neural network interatomic potential
Junjie Zhang, Hao Zhang, Jing Wu, Xin Qian, Bai Song, Cheng‐Te Lin, Te‐Huan Liu, Ronggui Yang
Abstract
Boron arsenide, considered an ideal semiconductor, inevitably introduces arsenic defects during crystal growth. Here, we develop a unified neural network interatomic potential with quantum-mechanical precision that accurately describes phonon transport properties in both perfect and defective boron arsenides. Through molecular dynamics simulations, we quantitatively explore the degree of phonon localization in boron arsenide caused by arsenic vacancies. We confirm that this localization primarily affects vibration modes within the frequency range of 2.0–4.0 THz, which is a challenge for conventional first-principles approaches. In addition, we examine the fluctuation of the heat flux autocorrelation function, which reveals the extent of phonon phase disruption resulting from arsenic voids and lattice anharmonicity from a more fundamental perspective. Our study highlights the applicability of molecular dynamics simulations in conjunction with neural network interatomic potential for defective systems, laying the theoretical groundwork for phonon engineering in real semiconductor crystals.