Semi-Supervised Automated Layer Identification of X-ray Tomography Imaged PCBs

Abstract

Reverse engineering (RE) for printed circuit boards (PCB) can be achieved through X-ray Computed Tomography (CT) in a non-destructively yet predominantly manual process. For practical RE to be possible in applications such as obsolescence replacement and hardware assurance, it is important that this process be as automated and fast as possible. This paper introduces a framework to identify which slices of an X-ray CT 3-D PCB stack belong to what layer on a physical PCB in an automated and generalizable fashion for any X-rayed PCB. To the best of our knowledge, this is the first method for identifying the correspondence between an X-ray CT slice and layers of a PCB using reference design information, demonstrated on a 6 layered PCB. First, a spatial pyramid bag-of-visual-words technique is leveraged to enable a semi-supervised approach where the user has either digital or X-ray layout information available for heterogeneous image comparison between X-ray CT image slices. Second, a weighting scheme based on entropy of an image region is proposed. Our results show near perfect quantitative and qualitative layer identification. The completion of this step facilitates improved performance in later RE stages, such as feature extraction or analysis of an X-rayed PCB.