E-QED: Electrical Bug Localization During Post-silicon Validation Enabled by Quick Error Detection and Formal Methods

During post-silicon validation, manufactured integrated circuits are extensively tested in actual system environments to detect design bugs. Bug localization involves identification of a bug trace (a sequence of inputs that activates and detects the bug) and a hardware design block where the bug is located. Existing bug localization practices during post-silicon validation are mostly manual and ad hoc, and, hence, extremely expensive and time consuming. This is particularly true for subtle electrical bugs caused by unexpected interactions between a design and its electrical state. We present E-QED, a new approach that automatically localizes electrical bugs during post-silicon validation. Our results on the OpenSPARC T2, an open-source 500-million-transistor multicore chip design, demonstrate the effectiveness and practicality of E-QED: starting with a failed post-silicon test, in a few hours (9 hours on average) we can automatically narrow the location of the bug to (the fan-in logic cone of) a handful of candidate flip-flops (18 flip-flops on average for a design with ~ 1 Million flip-flops) and also obtain the corresponding bug trace. The area impact of E-QED is ~2.5%. In contrast, determining this same information might take weeks (or even months) of mostly manual work using traditional approaches.