The document discusses several approaches to floorplan optimization including geometric programming, convex programming, and linear programming. Specifically, it summarizes works that formulated floorplanning with soft modules as a non-linear geometric programming problem, a new convex programming formulation that reduced the number of variables and constraints, and a linear programming based formulation that handles different module types and solves iteratively to obtain the global minimum solution faster than previous mixed integer approaches.
The document discusses several approaches to floorplan optimization including geometric programming, convex programming, and linear programming. Specifically, it summarizes works that formulated floorplanning with soft modules as a non-linear geometric programming problem, a new convex programming formulation that reduced the number of variables and constraints, and a linear programming based formulation that handles different module types and solves iteratively to obtain the global minimum solution faster than previous mixed integer approaches.
The document discusses several approaches to floorplan optimization including geometric programming, convex programming, and linear programming. Specifically, it summarizes works that formulated floorplanning with soft modules as a non-linear geometric programming problem, a new convex programming formulation that reduced the number of variables and constraints, and a linear programming based formulation that handles different module types and solves iteratively to obtain the global minimum solution faster than previous mixed integer approaches.
recently been devoted to various new representations of non-slicing floorplans to improve space utilization. The authors of [Lai and Wong, 2001] showed that a simple compaction procedure extends the capability of normalized polish expression to represent non-slicing floorplans. They conclude that slicing tree is a complete floorplan representation for all non-slicing floorplans as well. ILP-based Floorplanning Algorithm In [Moh et al., 1996], floorplanning with soft modules is formulated as a non-linear geometric programming. It is well-known that the geometric pro- gramming problem, after a simple transformation, can be converted into a convex optimization problem, which in turn can be solved efficiently by a non- linear solver. The advantage of this approach over the mixed integer linear programming by [Sutanthavibul et al., 1991] is that it does not suffer from the solution degradation caused by various schemes to linearize non-linear objectives and constraints. The number of variables and constraints used in [Moh et al., 1996] is signif- icantly reduced in a new convex programming formulation presented in [Chen and Fan, 1998]. Since the complexity of solving a convex programming prob- lem typically increases dramatically with the numbers of variables and con- straints, this new formulation leads to a significant reduction of computational effort in solving the floorplan area minimization problem. The works by [Moh et al., 1996] and [Chen and Fan, 1998] are not scalable because they suffer from the high complexity to solve a convex program- ming problem. The authors of [Chen and Kuh, 2000] presented a new linear programming (LP) based formulation that handles soft, hard, and pre-placed modules. They solve a set of LP problems in an iterative fashion to obtain global minimum solution. They do not use integer variables and constraints, thereby further speeding up the runtime compared to [Sutanthavibul et al., 1991]. The authors of [Ekpanyapong et al., 2004] adopted the mixed integer linear programming formulation of [Sutanthavibul et al., 1991] to perform floorplan- ning with micro-architectural modules. A major difference between floorplan- ning with circuit modules and micro-architectural modules is that