Quality and Quantity in Robustness-Checking Using Formal Techniques

Fault tolerance is one of the main challenges for future technology scaling to tolerate transient faults. Various techniques at design level are available to catch and handle transient faults, e.g., Triple Modular Redundancy. An important but missing step is to verify the implementation of those techniques since the implementation might be buggy itself. The thesis is focusing on formally verifying digital circuits with respect to fault-tolerant aspects. It considers transient faults and basically checks whether these faults can influence the output behavior of sequential circuits for any kind of scenarios. As a result the designer is pin-pointed directly to critical parts of the design and gets a prove about the absence of faulty behavior for non-critical parts. The focus of the verification is completeness with respect to the analysis. Three issues need to be adequately addressed: 1) cover all input stimuli, 2) all possible transient faults, and, 3) all possibly exponential long (wrt. to number of state bits) propagation paths. All three issues are addressed in different engines. A tool called RobuCheck has been implemented and evaluated on different academic benchmarks from ITC'99 and industrial benchmarks from IBM.