Cut-Based Inductive Invariant Computation

This paper presents a new way of computing inductive invariants in sequential designs. The invariants are useful for strengthening inductive proofs in difficult unbounded model checking instances. Candidate invariants are derived from a set of m-feasible cuts in the logic network and proved by induction. Thus, the proposed computation is very scalable, and it is possible to flexibly trade computational effort for the expressiveness of the proved invariants. Experimental results on several benchmark families show that the proposed strengthening proves many hard properties that are unsolved by other model checkers. The implementation is publicly available in the synthesis and verification system ABC.