High-Quality, Low-Cost Quasi-Bulk Hexagonal Boron Nitride Crystals by LP-HVPE: Machine Learning-Assisted Process Optimization via XRD and Raman Characterization

Abstract Highly ordered, quasi-bulk hexagonal boron nitride (hBN) films were grown on c-plane sapphire (α-Al2O3) substrates via low-pressure hydride vapor-phase epitaxy (LP-HVPE), using industrial-grade boron trichloride (BCl3) and ammonia (NH3) as low-cost boron and nitrogen sources, respectively. The effects of four key growth parameters (growth temperature, III/V source flow ratio, growth ambient, and growth pressure) on the crystalline quality of hBN films were systematically investigated via the single-variable method. A gradient boosting regression machine learning (ML) model was constructed based on 24 groups of experimental datasets to achieve high-precision prediction of hBN crystalline quality, with determination coefficients (R2) of 0.96 and 0.94 for the prediction of X-ray diffraction and Raman characteristic peak full width at half maximum, respectively. The optimized growth condition was determined as follows: a growth temperature of 1350 °C, a III/V ratio of 1:3, N2 ambient, and a growth pressure of 100 Pa, which is consistent with the global optimal parameters obtained by Bayesian optimization combined with the ML model. This work provides a low-cost, high-efficiency route for the controllable preparation of high-quality quasi-bulk hBN crystals via LP-HVPE with ML-assisted optimization.