Section: Overall Objectives
The problems tackled by the team R2D2 relate to the design of specialized systems on reconfigurable platforms. A hardware platform is a structure of Integrated Circuits (IC) containing a set of programmable components –general purpose or specific processor cores–, memories and generally specialized components. Such a platform can be seen as an integrated architecture scheme, common to numerous algorithms belonging to a given application domain. This notion is the answer given by the designers of embedded systems to the increasing difficulty they have to implement their applications  . One can consequently imagine that in the future, most of the ICs necessary to the design of a complex system will be derived from a given existing platform. This design approach is an alternative to the IP-based (Intellectual Property ) design approach, in which the system is built by assembling separately designed components. A reconfigurable platform includes a set of reconfigurable components (blocks of reconfigurable logic, reconfigurable data-path, flexible communication networks). In terms of area and power consumption, the reconfigurable resources enable a far more efficient use of the silicon than in programmable processors or in specialized components.
Future platforms will be highly parallel, heterogeneous, programmable and reconfigurable. Parallelism is the only way of reaching the performance level required by future applications. Heterogeneity results from the report that an efficient design is often composed of several subsystems, characterized by well-differentiated computation requirements. Programmability avoids freezing the functionalities. Finally, reconfigurability combines the speed of specialized solutions and the flexibility of traditional programmable components.
Our scientific objectives seek to profit from various methods (very high-level synthesis, behavioral synthesis, flexible compilation, floating-point to fixed-point conversion, etc.), contributing each one with its specificities, to the design of a part of a specialized system. The models and the underlying techniques allow the use of estimators, thus contributing to the choices of implementation, with a precise knowledge of the performance of the system, of its complexity and its power consumption.