Image acquisition and image processing enhancement of 3D X-ray CT images of electronic devices

• Main Author: Ong, Maverique Wen-Kai
• Other Authors: Gan Chee Lip
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/176083
• Institution: Nanyang Technological University

In the realm of hardware assurance, 3D X-ray Computed Tomography (CT) has been beneficial in enabling users to view the features within an electronic device without having to break it open. However, with devices continuously shrinking in size and increasing in complexity, such a technique may encounter difficulties in producing high-resolution images with low noise, which can cause submicron features to be distorted by noise and consequently not be properly identified. Therefore, this makes analysis of very-large-scale integration (VLSI) digital integrated circuits, to understand their functionality or to detect any malicious modifications, extremely difficult. This project explores three methods that can be applied post-scan to determine if they are capable of improving the quality of images derived from 3D X-ray Microscopy (XRM), a subset of 3D X-ray CT. These methods are ZEISS’s DeepRecon reconstruction algorithm, filtering on Object Research Systems’ Dragonfly software, and Python. Dataset from a scan on an Apple A13 Bionic system on a chip (SoC) motherboard underwent the DeepRecon and filtering process, whereas dataset from a scan on two other printed circuit boards (PCBs) underwent the Python process. Overall, all three methods demonstrated improvements in image quality with DeepRecon and filtering creating images with higher resolution and a better signal-to-noise ratio, while Python created images that has reduced shadows and artifacts within them. However, the DeepRecon process resulted in an image that has a low field-of-view (FOV) while the filtering and Python processes were mostly conducted manually. Therefore, future work can explore the potential of automating the filtering and Python processes to increase the efficiency of analysis, as well as the potential of overcoming the trade-off between resolution and FOV for DeepRecon datasets.