Real Time Predictions of VGF-InP Growth Dynamics by Recurrent Neural Networks

In this study, the Recurrent Neural Network (RNN) was combined with Computational Fluid Dynamics (CFD) to predict the temperature and crystal growth height during the growth of indium phosphide single crystal by Vertical Gradient Freeze (VGF) method. The effect of using Long Short-Term Memory (LSTM) model and Gated Recurrent Unit (GRU) to predict this task was compared. It was found that the prediction results of using Gated Recurrent Unit were more accurate and took less time. The problem of invisible growth of indium phosphide single crystal by vertical gradient freezing method is solved, which provides a solution for the automation and process control of single crystal growth.