Machine learning-based SEM images contour extraction data augmentation

Scanning Electron Microscopy (SEM) images contour extraction is crucial due to its ability to provide precise surface morphology analysis. Accurate contour extraction is helpful for quality control, failure analysis, defect detection and the development of advanced devices. However, existing machine learning-based contour extraction methods face the problem of difficulty in obtaining enough training datasets, particularly the lack of effective SEM image/contour pairs, resulting in defects in the extracted contours, such as open line ends and discontinuous lines. In this work, we propose a cGAN-based data augmentation pipeline to solve these issues. We generate defective contours by artificially introducing defects into ideal contours, which then serve as input to a cGAN-based model to generate synthetic SEM images with low contrast or weak signal in the corresponding parts. These generated SEM images, paired with original ideal contours, are then used to train a new contour extraction model, which significantly improve the model’s ability to handle previous failed cases. The experimental results demonstrate that our approach greatly enhances the contour accuracy, resolving previously challenging cases.