Multi-objective optimization in fixed-outline floorplanning with reinforcement learning

Abstract

Floorplanning is a crucial step in integrated circuit design. To address the fixed-outline floorplanning problem more effectively, we formulate it as a multi-objective optimization issue and employ multi-objective simulated annealing to simultaneously optimize both area and wirelength. Additionally, we apply deep reinforcement learning to learn from optimization experiences. This enables the exploration of more balanced multi-objective heuristics, thereby improving the results of multi-objective optimization. Test results on public benchmarks demonstrate the robust generalization capabilities of the proposed model. Compared to other advanced methods, our approach not only ensures a 100% success rate but also delivers superior performance in terms of wirelength. The deep reinforcement learning-assisted multi-objective simulated annealing method proposed in this paper can effectively address the fixed-outline floorplanning problem.