Date of Award

Spring 1-1-2014

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical, Computer & Energy Engineering

First Advisor

Fabio Somenzi

Second Advisor

Aaron R. Bradley

Third Advisor

Pavol Cerny

Fourth Advisor

Sriram Sankaranarayanan

Fifth Advisor

Niklas Sorensson

Abstract

Model checking has become a widely adopted approach for the verification of hardware designs. The ever increasing complexity of these designs creates a continuous need for faster model checkers that are capable of verifying designs within reasonable time frames to reduce time to market. IC3, the recently developed, very successful algorithm for model checking safety properties, introduced a new approach to model checking: incremental, inductive verification (IIV). The IIV approach possesses several attractive traits, such as stability and not relying on high-effort reasoning, that make its usage in model checking very appealing, which motivated the development of another algorithm that follows the IIV approach for model checking ω-regular languages. The algorithm, Fair, has been shown to be capable of dealing with designs beyond the reach of its predecessors.

This thesis explores IIV as a promising approach to model checking. After identifying IIV's main elements, the thesis presents an IIV-based model checking algorithm for CTL: the first practical SAT-based algorithm for branching time properties. The algorithm, IICTL, is shown to complement state-of-the-art BDD-based CTL algorithms on a large set of benchmarks. In addition to fulfilling the need for a SAT-based CTL algorithm, IICTL highlights ways in which IIV algorithms can be improved; one of these ways is addressing counterexamples to generalization, which is explored in the context of IC3 and is shown to improve the algorithm's performance considerably. The thesis then addresses an important question: for properties that fall into the scope of more than one IIV algorithm, do these algorithms behave identically? The question is answered negatively, pointing out that the IIV framework admits multiple strategies and that there is a wide spectrum of possible algorithms that all follow the IIV approach. For example, all properties in the common fragment of LTL and CTL—an important class of properties—can be checked with Fair and IICTL. However, empirical evidence presented in the thesis suggests that neither algorithm is always superior to the other, which points out the importance of being flexible in deciding the strategy to apply to a given problem.

Share

COinS