Classification of Clock Domain Crossing (CDC) Data using Machine Learning
DescriptionThis paper proposes a new methodology for classifying clock domain crossing results with machine learning. As SOC design complexity increases, the time spent on CDC analysis impacts project schedule and quality. Typically, most of the reported CDC paths are false alarms, and many CDC waivers are one-time used & abandoned. Also, the analysis of the CDC path is based on the designer's experience. We extract machine-learning information (CDC paths, CDC attributes, labels) from previous projects to imitate designer reviews on CDC paths. The trained model uses this learning information and provides classification/grouping results. It can be used for users to get insight into CDC paths and reduce CDC signoff time.
TimeTuesday, July 11th3:30pm - 3:45pm PDT
Location2010, 2nd Floor