Analysis and optimization of MOCVD flow ratio based on machine learning and PSO algorithm

Abstract

Metal-Organic Chemical Vapor Deposition (MOCVD) is a widely used technology for epitaxial growth. The metal-organic (MO) source enters the reaction chamber through several pairs of gas inlets, and the flow ratio among these inlets has a significant influence on epitaxial results. Based on Computational Fluid Dynamics (CFD) software, a numerical model of MOCVD was established to study the influence of the flow ratio of MO source inlets on the flow states in reaction chamber. It was found that the flow ratio has no effect on the flow states in general, though there are some differences in the flow field. Back Propagation (BP) neural network was used to approximate the relationship between MO flow ratio and deposition results, which showed high accuracy. Swarm Intelligence algorithm was used to optimize the flow ratio of MO source. The coefficient of variation was reduced by 38.1%, which indicated that it was an efficient way to adjust the flow ratio and improve the film quality.