Dictionary - Less Defect Diagnosis as Real or Surrogate Single Stuck - At Faults by Chidambaram Alagappan

With Moore’s law prediction already reaching its final stages, the number of transistors present in a chip has already touched more than a billion. At this stage, if a chip fails during manufacturing testing, it is becoming highly difficult to diagnose the root cause of the failure with precision. The other major contributing factor to this problem is that the actual defects are always not the same as the modeled defects considered during diagnosis. A diagnostic procedure should be able to give the possible locations where the defect could be present, given the failing occurrences of a chip. In this thesis, we propose a diagnostic algorithm that identifies classic single stuck-at faults as surrogate fault suspects for non-classical faults like multiple faults by analyzing failing circuits. Several diagnostic procedures have been proposed previously, but not many of them could directly associate the diagnosis results to the actual fault present in the circuit, the reason being, the diagnosis procedure is aimed at diagnosing pre-modeled defects. Few diagnostic procedures that somewhat try to associate with the actual defect are extremely complex. Hence, in this thesis no specific type of defects are assumed. Complexity is kept under control by considering only single stuck-at faults, which are highly analyzable. The single stuck-at faults are identified not as real suspects but as surrogates for actual defects. The diagnostic algorithm, therefore, works with the possibility that the observed behavior of the defective circuit may not exactly match with that produced by the surrogate single stuck-at faults. The algorithm is based on effect-cause analysis and proves to be less complex than existing methods. This diagnostic procedure involves adding or removing faults from a set of suspected faults based on the observed circuit outputs and minimal fault simulation to finally obtain a small set of candidate faults. Since the proposed algorithm does not require