TY - GEN
T1 - Lazy abstraction and SAT-based reachability in hardware model checking
AU - Vizel, Yakir
AU - Grumberg, Orna
AU - Shoham, Sharon
PY - 2012
Y1 - 2012
N2 - In this work we present a novel lazy abstraction-refinement technique for hardware model checking, integrated with the SAT-based algorithm IC3. In contrast to most SAT-based model checking algorithms, IC3 avoids unrolling of the transition relation. Instead, it applies local checks, while computing over-approximated sets of reachable states. We find IC3 most suitable for lazy abstraction, since each one of its local checks requires different information from the checked model. Similarly to IC3, our algorithm obtains a series of over-approximated sets of states. However, when constructing the series, different abstractions are used for different sets. If an abstract counterexample is obtained, we either find a corresponding concrete one, or apply refinement to eliminate all counterexamples of the same length. Refinement makes the abstractions more precise as needed, and where needed. After refinement, the computation resumes from the same step where it was interrupted. The result is an incremental abstraction-refinement algorithm where the abstraction is lazy. We implemented our algorithm, called L-IC3, and compared it with the original IC3 on large industrial hardware designs. We obtained significant speedups of up to two orders of magnitude.
AB - In this work we present a novel lazy abstraction-refinement technique for hardware model checking, integrated with the SAT-based algorithm IC3. In contrast to most SAT-based model checking algorithms, IC3 avoids unrolling of the transition relation. Instead, it applies local checks, while computing over-approximated sets of reachable states. We find IC3 most suitable for lazy abstraction, since each one of its local checks requires different information from the checked model. Similarly to IC3, our algorithm obtains a series of over-approximated sets of states. However, when constructing the series, different abstractions are used for different sets. If an abstract counterexample is obtained, we either find a corresponding concrete one, or apply refinement to eliminate all counterexamples of the same length. Refinement makes the abstractions more precise as needed, and where needed. After refinement, the computation resumes from the same step where it was interrupted. The result is an incremental abstraction-refinement algorithm where the abstraction is lazy. We implemented our algorithm, called L-IC3, and compared it with the original IC3 on large industrial hardware designs. We obtained significant speedups of up to two orders of magnitude.
UR - http://www.scopus.com/inward/record.url?scp=84874602520&partnerID=8YFLogxK
M3 - منشور من مؤتمر
SN - 9781467348324
T3 - 2012 Formal Methods in Computer-Aided Design, FMCAD 2012
SP - 173
EP - 181
BT - 2012 Formal Methods in Computer-Aided Design, FMCAD 2012
T2 - 12th Conference on Formal Methods in Computer-Aided Design, FMCAD 2012
Y2 - 22 October 2012 through 25 October 2012
ER -