Breadth-first manipulation of very large binary-decision diagrams

P This paper presents an e cient method for manipulating very large Shared inary-Deciston Diagrams (SBDD’S) which am too large to be stored within main memory. In contrast that the conventional depth-first aigorithm causes random access of memory,, the proposed method is intended to cause sequential access of memory. The main idea o our method is leveli by-level manipulation of Share Quasi-reduced BDD’s (SQBDD’S) upon a breadth-first algorithm. A garbage collection al orithm based on sliding type compaction i ts also intro ,uced m order to reduce page faults in succeeding manipulation. We implemented and evaluated the proposed method on the workstation Sun SPARC Station 10 with 6 i megabyte main memory and a one gigabyte har~ dis drive. As a result it took onl 5.6 hours to obtain an “? SQBDD of more than 12 mt hon nodes, which re re/ f sents all the primary outputs o a 15-bii mult~p ~er, from its circuit description. I we use the conventional SBDD manipulator instead, it is estimated that it would take about 1,900 hours.