Ants Can Solve Fundamental System Synthesis Problems

Numerous design challenges in system synthesis can be formulated as combinatorial optimization problems. Unfortunately, most of these problems are NP-complete or NP-hard. As fabrication technology improves and transistors become smaller and more numerous, we will continue to rely on design automation techniques to create and use future computing systems. Therefore, there is a substantial need for scalable algorithms to solve these increasingly complex design problems. In this paper, we present our work of using the recently developed ant colony optimization for tackling fundamental system synthesis problems. Using the ant system meta-heuristic, we give an effective method to solve multi-way system partitioning and instruction scheduling. The results of applying this approach are very promising; they indicate that our novel heuristics are quantitatively close to optimal and scale well to large problem instances.