Utilizing Synthesis Methods in Accurate System-level Exploration of Heterogeneous Embedded Systems

In this paper, we present a new mapping strategy for efficient design space exploration of heterogeneous embedded systems. The new mapping approach is developed within the context of the Sesame framework [1], [2] to improve its simulation accuracy and capabilities. In Sesame, separate application and architecture models together with an explicit mapping step are recognized within a system simulation. Originally, Sesame maps application models onto architecture models using trace-driven co-simulation of both models. However, this approach is limited by the fact that information is lost in the linear application traces that drive the architecture model. This may hamper accurate architecture simulation. The new mapping approach, which is based on Integer-controlled Dataflow actors, removes these limitations at the cost of a slight increase in complexity regarding the mapping. We compare our new approach with the current state-of-the-art mapping strategies under multiple criteria. Also with examples and a simple experiment, we illustrate how we integrate our new mapping strategy into Sesame’s mapping layer. To appear in Proc. of the IEEE Workshop on Signal Processing Systems, Seoul, Korea, Aug. 2003.