On improving genetic optimization based test generation

Test generation procedures based on genetic optimization were shown to be effective in achieving high fault coverage for benchmark circuits. In a genetic optimization procedure, the crossover operator accepts two test patterns t(1) and t(2), and randomly copies parts of t(1) and parts of t(2) into one or more new test patterns. Such a procedure does not take advantage of circuit properties that may aid in generating more effective test patterns. In this work, we propose a representation of test patterns where subsets of inputs are considered as indivisible entities. Using this representation, crossover copies all the values of each subset either from t(1) or from t(2). By keeping input subsets undivided, activation and propagation capabilities of t(1) and t(2) are captured and carried over to the new test patterns. The effectiveness of this scheme is demonstrated by experimental results.