Fast and Accurate Machine Learning Potential of hBN to Simulate Synthesis and Predict Thermal Conductivity

Hexagonal Boron Nitride (hBN) is a 2D material with high in-plane thermal conductivity and low out-of-plane thermal conductivity. Due to its anisotropic and dielectric properties, hBN films have a high potential for the thermal management of heterogeneous-integrated (HI) electronics. The construction of Process-Structure-Property (PSP) relations of hBN film can help in accelerating the synthesis of hBN films with desired thermal properties. Multi-scale numerical simulations are an essential component for the successful construction of PSP relations. The biggest difference between hBN compared to 3D crystals is that the interatomic interactions in cross-plane are very weak compared to the in-plane. Covalent bonds connect atoms in a layer while Van der Waals forces are dominant for the interactions of atoms on different layers. In-plane force fields are well developed but the out-of-plane force fields considering Van der Waals forces lead to the prediction of in-accurate thermal properties. There are mainly two interlayer potentials for hBN: one is Lennard-Jones (LJ) potential [1] and the other is Interlayer Potential (ILP) [2]. LJ potential is fast but inaccurate, and ILP potential is accurate but slow. For efficient data generation of hBN deposition and property prediction, an accurate and fast interlayer force field is needed.<br/>To achieve high accuracy and speed in predicting interlayer forces for hBN, a machine learning potential (MLP) will be developed. Data sets will be generated using density functional theory from various geometries of hBN considering different types of defects, which will be used to train and develop efficient MLP. This potential will be verified by calculating thermal conductivity and comparing it with other accurate force fields or measured values. After determining the potential is valid, the deposition of hBN will be simulated based on various process conditions used in Pulsed Layer Deposition such as temperature, pressure, etc. Various structures will be simulated, and the corresponding thermal conductivity will be calculated. Using this fast and high-fidelity MLP, PSP relations for hBN will be constructed, which will guide the experimentalists to accelerate the optimization of hBN synthesis to achieve high thermal conductivity values.<br/>[1] Neek-Amal, M., and F. M. Peeters, Applied Physics Letters, 104.4 (2014).<br/>[2] W. Ouyang, D. Mandelli, M. Urbakh and O. Hod, Nano Lett. 18, 6009-6016 (2018).