Recent advancements in 3D deep learning have led to significant progress in improving accuracy and reducing processing time, with applications spanning various domains such as medical imaging, robotics, and autonomous vehicle navigation for identifying and segmenting different structures. In this study, we employ the latest developments in 3D semi-supervised learning to create cutting-edge models for the 3D object detection and segmentation of buried structures in high-resolution X-ray semiconductors scans. We illustrate our approach to locating the region of interest of the structures, their individual components, and their void defects. We showcase how semi-supervised learning is utilized to capitalize on the vast amounts of available unlabeled data to enhance both detection and segmentation performance. Additionally, we explore the benefit of contrastive learning in the data pre-selection step for our detection model and multi-scale Mean Teacher training paradigm in 3D semantic segmentation to achieve better performance compared with the state of the art. Our extensive experiments have shown that our method achieves competitive performance and is able to outperform by up to 16% on object detection and 7.8% on semantic segmentation. Additionally, our automated metrology package shows a mean error of less than 2 m for key features such as Bond Line Thickness and pad misalignment.
Deep Learning is being widely used to identify and segment various structures in 2D and 3D scans in fields such as robotics and medical imaging. We leverage this exciting technology to train state-of-the-art models for 3D object detection and segmentation for various buried structures such as Through Silicon Vias (TSVs), memory bumps, and logic bumps. We show in detail how we fabricate our wafers and generate 3D scans. Thereafter, we explain our approach in locating these different structures in 3D scans and how we further segment these structures into solders, voids, Cu-Pillars, and Cu-Pads for 3D metrology and defect identification. We compare our approach with state-of-the-art techniques and perform a thorough analysis to discuss the advantages and disadvantages of various approaches in each step.
The 2D-3D metrology is a critical step for in-line inspection and off-line failure analysis. Due to lack of relevant data and complexity of embedded components, identifying and segmenting defects such as voids, pad misalignments in 2D and 3D voxel data has been a challenge in the semiconductor industry. Addressing this problem has the potential to further improve fault detection in this field significantly. This work follows our previously published works in EPTC 2020, ECTC 2021, introducing a cost-effective and non-destructive approach using deep learning and 3D x-ray microscopy. In particular, we apply our 3D object detection and Semi-Supervised Learning (SSL) image segmentation on High Bandwidth memory and logic bumps (HBMs). This paper introduces new detection and segmentation methods that overcomes issues in the current data such as data imbalance or defective bumps. We applied better 2D-3D detection strategy and loss and activation functions for 3D semicon data. We describe the data features, our new approach on 2D-3D scanned data, methods developed to perform better object detection and segmentation to classify each pixel into individual categories such as solders, voids, Cu-Pillars, and Cu-Pad. We analyze in-depth observations from our new models and discuss the benefits and improvements of our revised approach.