Dependency driven partitioning objects generation for hardware/software partitioning

Hardware/software partitioning is a key issue in the design of embedded systems where performance, chip area and/or power dissipation constraints have to be met. The granularity for automatic partitioning has great impact on the run-time of the partitioning and the quality of the final implementation. In this paper we present a new approach that generates partitioning objects using the dependency of the operations in the system. By clustering the dependent operations into one object, the number of the objects and the coupling between them are lowered. Compared to the fine-grain partitioning, our approach, which is called dependency driven partitioning, prunes a great portion of invalid solution space during partitioning objects formation, so as to expedite the partitioning without loss of the quality. Experiments with simulated annealing optimization show a faster convergence than that of fine-granularity with comparable partitioning quality