Trash or Treasure? Machine-learning based PCB layout anomaly detection with AnoPCB

Designing PCBs requires experience and knowledge about bad designs, e.g. sensitive signals being influenced by aggressively switching ones in close proximity. Those interactions can hardly be detected by formal checks and are usually treated in review processes or by lengthy design iterations. An automated way for identifying potentially harmful regions is still missing. This contribution introduces AnoPCB, a tool for automated detection of such potentially harmful regions. Due to the lack of pre-classified data and tremendous potential for bad design choices, AnoPCB is designed as unsupervised method and detects deviations from well-known design practices. In addition, we made AnoPCB freely-available as plugin for the open-source KiCad PCB design environment. Instead of using top-down imagery, AnoPCB processes geometrical relationships and signal properties in terms of layout slices containing category-based signal annotations. After training the anomaly detection on wellfunctional PCB layouts, our system is able to identify novel and potentially anomalous design patterns in new PCB layouts. We demonstrate our approach using freely available PCB layouts from the HackRF projects and showcase how novel design patterns are detected by AnoPCB.