Accurate image classification method for Computational Fluid Dynamics image

Abstract

Computational Fluid Dynamics is employed to optimize the structural design of processing equipment used for treating semiconductor wafers with chemical solutions. Chemical flows affect the processing conditions of wafers as the flow pattern differs based on the position inside the equipment, causing undesired variations in the condition of each wafer. To address this challenge, a structure that stabilizes the flow pattern is required. Accurate analysis of the flow patterns is therefore critical and advanced image clustering techniques are crucial in this analysis. However, conventional methods based on Autoencoder fall short in terms of clustering accuracy and fail to consider the positional information within the equipment. This paper proposes an image synthesis method that preserves the positional information in the equipment and a flow pattern image clustering method using IDFD. Our experiment demonstrated a significant improvement in clustering accuracy, as evidenced by a silhouette coefficient increase from 0.11 to 0.57.