Performance oriented partitioning for time-multiplexed FPGA's

Abstract

Time-multiplexing is a promising method to reduce the cost of FPGA based systems. It means execution of logic in consecutive steps with reconfiguration taking place between these steps. The use of time-multiplexing makes it possible to reduce the size of FPGAs but requires a new step in the design flow. The circuit has to be divided into sequential steps, partitions. In this paper we present an algorithm which partitions sequential circuits for time-multiplexing. The algorithm is based on list scheduling. Our experiments show that the algorithm is fast. It is able to partition a design with 4000 nodes in less than 4 seconds. The generated partitions have small size overhead, up to 3.2%, while no time overhead is allowed, besides the necessary reconfiguration time.