Enhancing the efficiency of Bayesian network based coverage directed test generation

Abstract

Coverage directed test generation (CDG) is a technique for providing feedback from the coverage domain back to a generator, which produces new stimuli to the tested design. Recent work showed that CDG, implemented using Bayesian networks, can improve the efficiency and reduce the human interaction in the verification process over directed random stimuli. This paper discusses two methods that improve the efficiency of the CDG process. In the first method, additional data collected during simulation is used to "fine tune" the parameters of the Bayesian network model, leading to better directives for the test generator. Clustering techniques enhance the efficiency of the CDG process by focusing on sets of non-covered events, instead of one event at a time. The second method improves upon previous results by providing a technique to find the number of clusters to be used by the clustering algorithm. Applying these methods to a real-world design shows improvement in performance over previously published data.