Fault Dictionary Compaction Using Structural and Tree-Based Techniques

摘要

For repeated diagnosis of different copies of the same circuit, the faultdictionary is typically better than dynamic diagnosis, because it requires substantially less time. In this work, fault dictionary compaction has been addressed from the point of view of two relatively (but not entirely) orthogonal tasks. The first of these tasks has been identified as identifying diagnostically useful information. Previous techniques attempting to solve this problem had several limitations, thus limiting their applicability to large practical circuits. Efficient techniques were proposed to identify diagnostically useful information, thus resulting in dictionaries with satisfactory resolution and are feasible to be generated on large practical circuits. The second task is one of representing identified information efficiently. This task is especially significant in applications where the first task cannot be performed. This is true in many faulty chips where the fault modeling process is not accurate, i.e., to say that the presence of unmodeled faults could have caused the observed errors. A novel approach to storing all the information in the full fault dictionary based on the use of unlabeled tree encoding has been proposed as an alternative to reducing the size of storage considerably. The proposed storage structure uniquely captures the indistinguishability class information at various stages of the diagnosis process and, hence, is shown to exhibit better behavior than currently known techniques for the purpose of matching the faulty symptoms with the stored data.