An approach to automated hardware/software partitioning using aflexible granularity that is driven by high-level estimation techniques

摘要

Hardware/software partitioning is a key issue in the design ofnembedded systems when performance constraints have to be met and chipnarea and/or power dissipation are critical. For that reason, diversenapproaches to automatic hardware/software partitioning have beennproposed since the early 1990s. In all approaches so far, thengranularity during partitioning is fixed, i.e., either small systemnparts (e.g., base blocks) or large system parts (e.g., wholenfunctions/processes) can be swapped at once during partitioning in ordernto find the best hardware/software tradeoff. Since the deployment of anfixed granularity is likely to result in suboptimum solutions, wenpresent the first approach that features a flexible granularity duringnhardware/software partitioning. Our approach is comprehensive in so farnthat the estimation techniques, our multigranularity performancenestimation technique described here in detail, that controlnpartitioning, are adapted to the flexible partitioning granularity. Innaddition, our multilevel objective function is described. It allows usnto tradeoff various design constraints/goals (performance/hardware area)nagainst each other. As a result, our approach is applicable to a widernrange of applications than approaches with a fixed granularity. We alsonshow that our approach is fast and that the obtained hardware/softwarenpartitions are much more efficient (in terms of hardware effort, fornexample) than in cases where a fixed granularity is deployed