Testing ROBDD construction with a massive parallel processing genetic algorithm

A Reduced Ordered Binary Decision Diagram (ROBDD) is a data structure that has benn widely used in CAD tools for VLSI logic synthesis design. ROBDDs have some good properties and some drawbacks. The most important drawback is that the size of a ROBDD depends on input variable ordering. Depending on this ordering, one can find ROBDDs for the same logic function with widely different numbers of nodes. The problem of finding the best input variable ordering can be modeled as a Travelling Salesman Problem (TSP). In this communication, in order to find such optimum ordering, Genetic Algorithms (GAs) are used. Due to the high computational cost, a parallel aprroach to GAson a Massive Parallel Processing (MPP) system is followed. Our results show that GAs fit well for finding good variable ordering for ROBDDs on a MPP environtment. The efficiency of such algorithms was 73\% using 128 Processing Elements (PEs)