Improving 3D-Floorplanning using smart selection operations in meta-heuristic optimization

In 3D-Floorplanning even more than in 2D-Floorplanning new objectives, e.g. temperature, TSV-Planning or IR-Drop are considered. This increases the complexity of the problem formulation and, therefore, of the optimization algorithm, dramatically. Apart from some analytical approaches, simulated annealing based algorithms (SA) are widely used for 3D-Floorplanning. To increase the solution quality of classical SA, a common approach is to adapt the selection operations, improving local search. While previous work proposes selection operations which consider mostly one single design issue (e.g. temperature or fixed-outline), we propose a comprehensive multiobjective floorplan optimization methodology (smart SA) which is capable of efficiently considering several objectives and constraints (area, wirelength, fixed-outline, maximum number of TSVs and maximum temperature) at the same time. For the objectives and constraints we present simplified analysis models. Experimental results show that our extended SA algorithm outperforms the classical one and finds valid solutions where classical SA fails.