Few-shot AI segmentation of semiconductor device FIB-SEM tomography data
Few-shot AI segmentation of semiconductor device FIB-SEM tomography data
Thursday, October 31, 2024: 8:40 AM
202 (Hilton San Diego Bayfront)
Summary:
SEM images of semiconductor device structures are usually analyzed manually by experts using specialist knowledge for correct identification of functional structures and defects. This approach can hardly be scaled to analyze a larger number of images and features than a human user can process. This is especially true for FIB-SEM tomography image series. To take full advantage of such data during inspection, it is necessary to segment the various structural components before visualizing them. Conventional image segmentation methods are not suitable for this in most cases, while AI based segmentation produces much superior results. Recent progress in this field made it possible to significantly reduce the amount of training data to be manually annotated by an expert user to a few images only ("few-shot" segmentation). We present examples with workflows and applications.
SEM images of semiconductor device structures are usually analyzed manually by experts using specialist knowledge for correct identification of functional structures and defects. This approach can hardly be scaled to analyze a larger number of images and features than a human user can process. This is especially true for FIB-SEM tomography image series. To take full advantage of such data during inspection, it is necessary to segment the various structural components before visualizing them. Conventional image segmentation methods are not suitable for this in most cases, while AI based segmentation produces much superior results. Recent progress in this field made it possible to significantly reduce the amount of training data to be manually annotated by an expert user to a few images only ("few-shot" segmentation). We present examples with workflows and applications.