On simulation-based metrics that characterize the behavior of RTL errors

Recent advances in automated debugging offer significant reductions in the manual effort required to localize RTL errors. These tools return relatively compact sets of RTL locations that can be potential error sources. However, once these locations are returned, the engineer still has to perform detailed analysis to discard irrelevant locations and identify the culprit. This process happens without any further guidance from the debugger. In this work, we perform a statistical analysis that exposes a significant discrepancy between RTL errors and other unrelated locations returned by these tools. The analysis is conducted on industrial designs and is based on metrics extracted from simulation. Our methodology determines that specific continuous distributions can effectively characterize the behavior of RTL errors. Using these welldefined metrics one can automate the process of further pruning the RTL locations returned by debuggers, effectively accelerating the localization of error sources.