@User: Massimo Sivilotti wrote it. . . He's was CTO at Tanner Research and I don't know what his position is now since they were bought by Mentor. That's about all I know. I can describe the hashing and partitioning algorithm (it's quite clever) but it's difficult to conceptualize clearly.

It's based on finding equivalence classes IIRC. I think I want to add this to my class this quarter and wanted a reference. I think it's similar to state minimization of FSMs where each state has a similar signature. In that, you end up with states being the same are redundant. In the case of netgen two items in a class are matched.