Techniques for Reducing the Number of Decisions and Backtracks in Combinational Test Generation

Combinational test pattern generation (TPG) is basically a search in a finite state space. In general, the search is performed in a branch-and-bound fashion. The branch-and-bound search builds a decision tree using two basic operations: decision making and backtracking. The size of the decision tree, and hence the efficiency of the branch-and-bound search, is directly dependent on the number of decisions made. This paper proposes a set of novel techniques for reducing the number of decisions and the size of the decision space. These techniques work directly on the maximum number of potential ways of justifying a given logical assignment. This maximum number is reduced by exploiting the properties of prime-and-irredundant covers. These same properties are also used to reduce the number of backtracks by implying a maximum number of necessary assignments. The reduction of the number of decisions and the identification of a maximum of necessary assignments make the proposed TPG method highly efficient as demonstrated by experimental results. This paper also proposes a novel combination of prime-and-irredundant cover extraction and transitive closure computation for a more efficient TPG process.