A Methodology for efficient architectural exploration of energy-delay trade-offs for embedded systems
Sami M. G.,
Contributo in atti di conferenza
The main goal of this paper is to identify the best architecture of an embedded system by considering at the same time energy and delay, avoiding the comprehensive analysis of the architectural design space. We adopt the Energy-Delay Product (EDP) as the evaluation metric to compare the alternative architectures of the target system. The paper analyzes an extended adaptive random search algorithm (ADGREED) to efficiently explore the architectural design space. The ADGREED algorithm is a pseudo-random optimization algorithm that combines the best potentialities of the adaptive random search (ADRAS) and the Greedy deterministic algorithm. The analysis has been carried out through the architectural optimization of the memory subsystem of a real-word embedded system executing the set of Mediabench benchmarks for multimedia applications. The reported experimental results have shown a reduction up to one order of magnitude of the number of design alternatives analyzed during the exploration phase, while maintaining very high accuracy.