An efficient algorithm to fixed-outline floorplanning based on instance augmentation

In this paper, we present an efficient and effective algorithm addressed to fixed-outline floorplanning. Unlike simulated annealing based algorithms, the proposed algorithm starts with sub-instances of the given instance. When feasible solution of a sub-instance is found, it augments this instance and tries to find a feasible solution of the augmented instance. This process continues until feasible solution of the given instance is found, or some stopping criteria met. Local search without hill-climbing is adopted to search for feasible solution of each instance. To optimize wirelength, once feasible solution of current instance is found, instead of augments it immediately, the proposed algorithm performs another local search for this instance. Experiments on MCNC and GSRC benchmarks show that in most cases the proposed algorithm can obtain feasible floorplans in quite short time, even with tight outlines or outlines with large aspect ratios imposed, and the results of wirelength optimization are also quite promising.