Design of experiments in BDD variable ordering
Justin E. Harlow, F. Brglez · 1998
Applying the Dmign of Experiments methodology to the evaluation of BDD van'able orden.ngalgon.thmshas yielded a number of conclusive Tesults.The methodology reiies on the rtient~y i~tTOdUCd equivalence classes of functionally petiurbed circuits that maintain logic invariance, , ...)-mintems of the original refeTence OT aTe urithin {1, 2 circuit function, abo maintaining entTopy-invan.ante.FOT some of the current vam.ableOTde~ng algom.thmsand tools, the negative raults include: (1) statistically significant sensitivity to naming of vatiables, (2) confirmation that a numbeT of variable orden.ngalgorithms aTe statistically equivalent to a random variable order assignment, and (3) obsemation of a statistically anomalous variable ordefl-ng behavioT of a wellknown benchmaTk circuit isomorphic class when analyzed un-deT a single and multiple outputs.On the positive side, the methodology supports a statistically significant mem.t evaluation of any newly introduced vatiable oTdem.ngalgon.thm,including the one bm.e~yintroduced in this papeT.