Data Mining Diagnostics and Bug MRIs for HW Bug Localization
Monica Farkash1,a, Bryan Hickerson2 and Balavinayagam Samynathan1,b
1University of Texas, Austin, TX, USA.
2IBM, Austin, TX, USA.
This paper addresses the challenge of minimizing the time and resources required to localize bugs in HW dynamic functional verification. Our diagnostics solution eliminates the need to back trace from point of failure to its origin, decreasing the overall debugging time. The proposed solution dynamically analyses data extracted from sets of passing and failing tests to identify behavior discrepancies, which it expresses as source code lines, coverage events and timing during simulation. It also provides a visual diagnostic support, an image of the behavior discrepancies in time which we call a Machine Reasoning Image (MRI). This paper describes in detail our data mining solution based on coverage data, HDL hierarchies and time analysis of coverage events.
Our approach brings a data mining solution to the problem of HW bug localization. It defines new concepts, provides in-depth analysis, presents supporting algorithms, and shows actual results on archetypical problems from PowerPC core verification as an industrial application.
Keywords: Diagnostics, Bug localization, Debugging, Verification, EDA tools.
Full Text (PDF)