Multi-objective design space exploration of embedded systems

In this paper, we address the problem of the efficient exploration of the architectural design space for parameterized embedded systems. The exploration problem is multi-objective (e.g., energy and delay), so the main goal of this work is to find a good approximation of the Pareto-optimal configurations representing the best energy/delay trade-offs by varying the architectural parameters of the target system. In particular, the paper presents a Design Space Exploration (DSE) framework to simulate the target system and to dynamically profile the target applications. In the proposed DSE framework, a set of heuristic algorithms have been analyzed to reduce the overall exploration time by computing an approximated Pareto set of configurations with respect to the selected figures of merit. Once the approximated Pareto set has been built, the designer can quickly select the best system configuration satisfying the constraints. Experimental results, derived from the application of the proposed DSE framework to a superscalar architecture, show that the exploration time can be reduced by three orders of magnitude with respect to the full search approach, while maintaining a good level of accuracy.