Synthesis and Optimization of Digital Hardware/Software Systems

Abstract

In this introductory paper to the field, it is our goal to provide a new unified look at synthesis problems that is independent from the level of abstraction like system, RTL, and logic (for refinements targeted to hardware), or process- and basic block level (for refinements targeted to software). For each level of our model called ”double roof”, synthesis requires the solution of three basic problems, namely allocation (of resources), binding, and scheduling Based on the ”double roof” model, we present a graph-based formulation of the tasks of system-synthesis: Contrary to former approaches that consider system-synthesis as a bi-partition problem (e.g., earlier hardware/software partitioning algorithms), we consider also as well the allocation of components like micro- and hardware coprocessors as part of the optimization problem as scheduling of tasks including communication scheduling. The approach is flexible enough to be applied to different other abstraction levels. Finally, we introduce the problem of design space exploration as a new challenge in synthesis. For the typically multi-objective nature of synthesis problems, not only one optimum is wanted, but an exploration of a complete front of optimal solutions called Pareto points.