12th World Congress on Industrial Process Tomography
Encoder-Decoder Deep Learning Model for Electrical Capacitance Tomography Image Reconstruction in Semiconductor Post-Fabrication Process Monitoring
M. Jeon1, A. K. Khambampati1, S. Ko1, K. Y. Kim1,2*
1Electronic Engineering Department, Jeju National University, South Korea, 63243
2 Kyoto College of Graduate Sudies for Informatics, Kyoto, Japan, 606-8225
*Email: kyungyk@jejunu.ac.kr
ABSTRACT
In semiconductor post-fabrication, deposit accumulation in pipes hinders efficiency and poses safety risks. Electrical Capacitance Tomography (ECT) offers a non-invasive solution for monitoring these deposits to optimize maintenance. This paper proposes a novel encoder-decoder deep learning model for high-resolution permittivity distribution reconstruction from ECT capacitance measurements in these crucial industrial environments.
The proposed model was trained using capacitance values and corresponding permittivity distributions generated via numerical simulations. Its performance was then evaluated against the conventional Newton-Raphson (NR) method and a Fully Connected Neural Network (FCNN). In quantitative evaluation, the proposed encoder-decoder model significantly outperformed the FCNN in RMSE, indicating superior accuracy in estimating actual permittivity values, especially at complex gas-deposit interfaces.
Keywords: Decoder-encoder, Artificial neural network, semiconductor manufacturing process, image reconstruction
Industrial Application: AI for tomography
Sign-in to access the full text
Copyright © International Society for Industrial Process Tomography, 2025. All rights reserved.