Self-validating diagnosis of hypercube systems

Abstract

A novel approach to the diagnosis of hypercubes, called self-validating diagnosis (SVD), is introduced. An algorithm bared on this approach, called the SVD algorithm, is presented and evaluated. Given any fault set and the resulting syndrome, the algorithm returns a diagnosis and a syndrome-dependent bound, T/sub /spl sigma//, with the property that the diagnosis is correct (although possibly incomplete) if the actual number of faulty units is less than T/sub /spl sigma//. The average of T/sub /spl sigma// is very large and the diagnosis is almost complete even when the percentage of faulty units in the system approaches 50%. Moreover, the diagnosis correctness can be validated deterministically by individually probing a very small number of units. These results suggest that the SVD algorithm is suitable for applications requiring a large degree of diagnosability, as is the case for wafer-scale testing of VLSI chips, where the percentage of faulty units may be as large us 50%.