Guiding Sensitization Searches Using Problem Reduction Graphs

In an earlier paper [10] the generation of problem reduction graphs from a search goal definition and a design language description of circuit behavior was presented. The objective of forming such a graph was to provide heuristic guidance for state space searches. Most state space searches use both heuristic cost weighting on the nodes and heuristic input vector generation. In this paper an approach is provided for computing the heuristic cost function based on its contribution to the search objective and the frequency with which that node has been reached in past searches. This is done by weighting the nodes of the problem reduction graph associated with the state search problem based on the frequency with which those problem reduction graph nodes have been reached in past searches. From this weighting follows an heuristic cost function which can be used to quickly compute node weights during the actual state space search. Input vectors are also suggested by the problem reduction graph. Because digital circuits so often partition naturally into control and data sections, the input vectors which can be applied at a given control state are taken from the problem reduction graph and so grouped. Four case studies demonstrate the usefulness of the method, while also showing that neither heuristic weighting nor input selection alone is adequate to provide general purpose state space search guidance.