In advanced device manufacturing processes for semiconductors and displays, thin-film structures are essential, and real-time, high-precision measurement and monitoring of thin-film thickness are critically required to ensure product quality and improve production efficiency. Recently, studies have applied deep learning algorithms to optical spectroscopic reflectance measurement methods, attracting significant attention due to advantages such as increased analysis speed and the ability to overcome local minimum problems.
However, conventional deep learning models have typically been trained only on theoretical datasets generated based on the Fresnel equations. As a result, random noise and fluctuations arising in real measurement environments are not sufficiently reflected in the training process, which may degrade the reliability of thin-film thickness measurements performed using deep learning algorithms. To overcome these limitations, this study aims to improve the reliability of real-time thin-film thickness measurement by constructing datasets that systematically incorporate reflectance variations caused by actual measurement environments, such as temperature changes, vibrations, and optical intensity fluctuations, and using these datasets for deep learning training.
To verify the effectiveness of the proposed approach, a theoretical dataset and a measurement-environment–aware dataset were each used to train artificial neural networks with identical architectures, and their performances were compared. The uncertainty evaluation results of thin-film thickness measurements showed that the neural network trained with the dataset reflecting measurement environment effects exhibited lower measurement uncertainty than the neural network trained with the purely theoretical dataset. These results demonstrate that incorporating real measurement environments into deep learning training is effective in improving measurement reliability.
Based on these findings, it is expected that deep learning–based real-time thin-film thickness measurement technologies can be successfully applied to actual device manufacturing processes.
