Why is combinational ATPG efficiently solvable for practical VLSI circuits?

Empirical observation shows that practically encountered instances of combinational ATPG are efficiently solvable. However, it has been known for more than two decades that ATPG is an NP-complete problem (Ibarra and Sahni, IEEE Transactions on Computers, Vol. C-24, No. 3, pp. 242-249, March 1975). This work is one of the first attempts to reconcile these seemingly disparate results. We introduce the concept of cut-width of a circuit and characterize the complexity of ATPG in terms of this property. We introduce the class of log-bounded width circuits and prove that combinational ATPG is efficiently solvable on members of this class. The class of of log-bounded width circuits is shown to strictly subsume the class of k-bounded circuits introduced by Fujiwara (International Symposium on Fault-Tolerant Computing, June 1988, pp. 64-69). We provide empirical evidence which indicates that an interestingly large class of practical circuits is expected to have log-bounded width, which ensures efficient solution of ATPG on them.