A Multiobjective Tabu Search Algorithm for the Design Space Exploration of Embedded Systems

An important step during the design of embedded systems is to allocate suitable architectural components and to optimally bind functions (tasks) to these components. This design step is called system synthesis. The automation of system synthesis is limited in recent research by developing models only for standard optimization algorithms. This paper describes the first approach to improve a standard optimization technique itself for the use in embedded system design. Our solution extends the heuristic optimization algorithm tabu search by multiobjective optimization. Using the multiobjective approach, domain specific heuristics could easily be included into the algorithm. By performing experiments with the new algorithm, a new effect was discovered: In contrast to known results from literature, the quality of optimization was depending on the size of the neighborhood if the moves in the neighborhood were sorted by domain specific estimation.