Variable ordering optimization of ROBDD using genetic algorithm

Abstract

The reduced ordered binary decision diagram (ROBDD) has been known as an effective data structure for representing and manipulating a Boolean function. The size of ROBDD is proportional to computational complexity in the Boolean function manipulation and critically depends on the order of the Boolean function variables. Hence, in most applications where Boolean function is central, finding the order leading to the smallest size of ROBDD is very important. Recently, several genetic algorithms have been developed to tackle ROBDD ordering optimization which have shown better results than other conventional approaches such as sifting, window permutation, and simulated annealing, etc. However, genetic algorithms have been limited in applying for large-sized Boolean functions because of the hugh computation time and lack of computer memory. The above mentioned weaknesses can be overcome by simplifying the procedure of genetic algorithm, improving construction methods of ROBDD, and developing new genetic operation methods. In this thesis, we investigate previous researches and propose new methodologies to access better results in these ways. Additionally, interesting experimental results for large-sized Boolean functions are introduced.