Congestion-Driven Floorplanning with Module Reshaping

In this paper, we implement a two-stage congestion-driven floorplanner. In first stage, an area-wire length-oriented method using simulated-annealing (SA) with the conception of ant algorithm (SANTA) was introduced and a congestion-driven module reshaping was introduced in second stage. The floorplanner which Simultaneously minimizes area, the wire congestion and the total wire-length generates the feasible solution. The objective of the paper is simultaneously area, the total wire length and the wire congestion. With the sequential-pair (SP) presentation, we first apply the SANTA improves the neighbor searching of traditional simulated-annealing (SA), to increase the solution quality and decrease the runtime. For the intermediate floorplan, the two adjacent soft modules located at the congested regions are selected and divided into a set of connected sub-rectangles to enlarge the common boundary between the adjacent modules. The longer common boundary actually reduces the total wire-length between the pins of two modules and minimizes the wire congestion. A nonlinear programming (NLP) method is applied for module reshaping mentioned above to further improve the congestion without the area penalty. Compared to traditional SA, SANTA can achieve an average improvement on area, total wire length and runtime by 3% and 8.1% and 23.0%, respectively. For the randomly selected 3)0 floorplan samples from Microelectronics Center of North Carolina (MCNC) benchmarks, the result also shows experimentally that the reshaping method obtains the improvement on the wire congestion and total wire length by 22% and 1.54%, respectively.