Divide-and-conquer is essential to address state explosion in model checking. Verifying each individual component in a system, in isolation, efficiently requires an appropriate context, which traditionally is obtained by hand. This paper presents an efficient modular model checking approach for asynchronous design verification. It is equipped with a novel abstraction refinement method that can refine a component abstraction to be accurate enough for successful verification. It is fully automated, and eliminates the need of finding an accurate context when verifying each individual component, although such a context is still highly desirable. This method is also enhanced with additional state space reduction techniques. The experiments on several nontrivial asynchronous designs show that this method efficiently removes impossible behaviors from each component including ones violating correctness requirements.

Compositional verification is essential for verifying large systems. However, approximate environments are needed when verifying the constituent modules in a system. Effective compositional verification requires finding a simple but accurate overapproximate environment for each module. Otherwise, many verification failures may be produced, therefore incurring high computational penalty for distinguishing the false failures from the real ones. This paper presents an automated method to refine the state space of each module within an overapproximate environment. This method is sound as long as an overapproximate environment is found for each module at the beginning of the verification process, and it has less restrictions on system partitioning. It is also coupled with several state-space reduction techniques for better results. Experiments of this method on several large asynchronous designs show promising results.