Discrete Floorplanning by Multidimensional Pattern Matching

In this paper, we show a new approach for solving discrete floorplanning problems. Modules to be placed and the target architecture are modelled by periodic graphs. The objective is, to find a valid assignment of module nodes to slot nodes of the target architecture, such that a cost function on the placement will be minimized. We use an algorithm which is abstracted and derived from a traditional pattern matching technique. Our method has time complexity O(m d nd a), where m is the number of modules to place, d is the dimension of the periodic graph, n is the maximum node count in the target architecture along a dimension and a is the complexity to decide the validity of an assignment.