Deep learning for defect detection

• Main Author: Low, Edwin Xuan Hao
• Other Authors: Qian Kemao
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/171974
• Institution: Nanyang Technological University

The increase in demand for technology because of mankind’s reliance on it has led to an increase in manufacturing output for Integrated Chips. Quality assurance of Integrated Chips cannot be understated as it directly affects the operability of the many devices we rely on. Although measures have been put in place by manufacturers, most are largely reliant on manual inputs. The main objective of this project is to develop a deep learning-based method to detect defects in semiconductor applications. A Convolutional Autoencoder model was proposed in this project and tested on an industry-partner-provided dataset. Many current defect detection methods in semiconductor applications are image-based and the Convolutional Autoencoder model has shown its successes in image-based defect detection in other fields. Other image-based tasks of Convolutional Autoencoders include image denoising and compression. In this project, the benefits, and potential problems of Convolutional Autoencoders were discussed. Comparisons to other models were also discussed and quantitative metric comparisons to other state-of-the-art models were done. Suggestions for improving the proposed model were also provided at the end of the report.