High-level synthesis using a genetic algorithm.

This paper proposes high-level synthesis using a genetic algorithm that can realize register transfer level circuits from behavioral descriptions. Until now this type of synthesis problem has been separated into scheduling and allocation problems, and the two problems have been solved separately. This procedure has a difficulty that the process may easily fall into a local solution.This paper proposes a simultaneous solution of those problems, so that a circuit better than in the conventional algorithm can be synthesized. The proposed approach, on the other hand, has the disadvantage that the computation time tends to be longer, since the length of the chromosome is increased. To avoid this difficulty, several measures are used such as modification of a faulty gene and discrimination among circuits requiring the same number of resources. A result better than in the conventional approach is obtained for the benchmark problem. © 1999 Scripta Technica, Electron Comm Jpn Pt 3, 83(4): 24–32, 2000 [ABSTRACT FROM AUTHOR]