ALaRI Hang Glider

Search form

Education and Innovation in Embedded Systems Design

USI Università della Svizzera italiana, USI Faculty of Informatics, Advanced Learning and Research Institute USI Università della Svizzera italiana USI Faculty of Informatics USI Advanced Learning and Research Institute
TitleMultiprocessor System-on-Chip Design Space Exploration based on Multi-level Modeling Techniques
Publication TypeConference Paper
Year of Publication2009
AuthorsMariani, G., G. Palermo, C. Silvano, and V. Zaccaria
Conference NameProceedings of IEEE IC-SAMOS'09 - International Conference on Embedded Computer Systems: Architectures, MOdeling, and Simulation
Date PublishedJuly
Conference LocationSamos, Greece
Keywordsartificial neural network, design space exploration, genetic algorithm, multi level modelling, multiprocessor system-on-chip (MPSoC), platform-based design
Abstract

Multi-processor Systems-on-chip are currently designed by using platform-based synthesis techniques. In this approach, a wide range of platform parameters are tuned to find the best trade-offs in terms of the selected system figures of erit (such as energy, delay and area). This optimization phase is called Design Space Exploration (DSE) and it generally consists of a Multi-Objective Optimization (MOO) problem. The design space of a Multi-processor architecture is too large to be evaluated comprehensively. So far, several heuristic techniques have been proposed to address the MOO problem, but they are haracterized by low efficiency to identify the Pareto set. In this paper we propose a methodology for heuristic platform based design based on evolutionary algorithms and multi-level simulation techniques. In particular, we extend the NSGA-II with an approximate neural network meta-model for multi-processor architectures in order to replace expensive platform simulations with fast meta-model evaluation. The model accuracy and efficiency is improved by exploiting high-level platform simulation techniques. High-level simulation allows us to reduce the overall complexity of the neural network and improving its prediction power. Experimental results show that the proposed techniques is able to reduce the number of simulations needed for the optimization without decreasing the quality of the obtained Pareto set. Results are compared with state of the art echniques to demonstrate that optimization time due to simulation can be sped up by adopting multi-level simulation techniques.

DOI10.1109/ICSAMOS.2009.5289222