Illuminating the mechanical responses of amorphous boron nitride through deep learning: A molecular dynamics study

Abstract

Amorphous boron nitride (a-BN) has recently emerged as a promising ultralow-dielectric-constant material for interconnect isolation in integrated circuits. However, its mechanical properties remain poorly understood. This study establishes a deep learning potential (DLP) for hexagonal boron nitride (h-BN), cubic boron nitride (c-BN), and a-BN based on density functional theory data. The DLP accurately captures energies, forces, and virial stresses compared to direct DFT calculations. Molecular dynamics simulations employing the DLP are conducted to investigate the mechanical behaviors of a-BNs with varying densities under uniaxial tension. The results reveal that higher density a-BN exhibits increased Young's modulus and yield stress but reduced ductility, while lower density a-BN displays superior ductility and higher ultimate strain. The formation of local distortion and evolution of cracks are examined using atomic local shear strain. This work elucidates structure–property relationships in a-BN through simulations, providing useful insights for integrating a-BN in device applications.