High-speed flow field prediction and process parameters optimization in a vertical MOCVD reactor based on a hybrid RSM-KNN model

Abstract Metal-Organic Chemical Vapor Deposition (MOCVD) is an important method of epitaxial growth used to obtain high-quality film and mass production. Computational Fluid Dynamics (CFD) method is widely used to simulate and study the growth process of MOCVD, but the complex process parameters and high computational cost have limited the real-time application. A hybrid Response Surface Methodology–K-Nearest Neighbor (RSM-KNN) model is developed to reach the goal of the high speed of flow field prediction and process parameters optimization. In this paper, a 3-dimensional ZnO-MOCVD model is built and then validated using the experimental and simulation results. The Design of Experiment (DOE) method is used to improve data collection performance. Results of deposition rate and uniformity are combined with the RSM model to analyze the influence of process parameters and to explore the relationship between deposition rate, uniformity and flow state. The results of the flow field are combined with the machine learning method KNN to study and predict the flow field and flow states in the reactor. The hybrid model is also presented and optimized to realize the growth of high deposition rate, and the uniformity is increased by 30% at a steady flow state of plug flow.