Multiscale modeling: from Electromagnetism to the GRID – MEG

1- contexte scientifique et objectifs du projet In the field of Sciences and Technologies for Information and Communication, there are key scientific challenges to develop miniaturized systems embarking more and more intelligence both at the physical and the software layer, with architectures increasingly complex. At the same time, it is strongly required to have robust design methodologies in order to accelerate the design cycle and minimize the prototyping phases. In this context, one issue gaining more and more importance is dealing with the design and optimization of the physical layer of the communication medium. Today, the complexity turns out to make very difficult the modeling and design of such systems and impossible their optimization due to a dramatic increase of the number of unknown parameters. 2- description du projet, méthodologie In this project the use of electromagnetic methods of computational electromagnetic (CEM) in grid computing (GC) environments is at stake. The scope of the present work is to define an environment and tools for the electromagnetic (EM) simulation of future radio frequency (RF) systems. This objective will be pursued by formulating a multi-scale and adaptive use of increasingly available distributed resources (computing nodes network architecture memory storage) and the specific EM methodologies at disposal. 3- résultats attendus The challenge in the present project will be, in a first phase, to optimize the grid use with respect to each method, and later, with the combination (hybridization) of the two in a second phase. A two-fold approach can be envisioned for the enhanced use of GC by exploiting the intrinsic characteristics of the TLM and SCT methods. In the case of the TLM arbitrary scalability of the grid resources is allowed by the possibility to have arbitrary decomposition of the simulation domain While for the SCT adaptive scalability of the grid resources is enabled by the intrinsic scalability of the method. The computation of each domain can be mapped to an equivalently large set of computing nodes. The hybridization of the two TLM and SCT techniques could be later on exploited in order to address their use on different portions of the same EM problem. This operation represents an important objective as it would significantly increase the efficiency of the modeling in terms of both speed and memory requirements.