Modèles, Méthodes et Outils pour Améliorer la Productivité des Systèmes Embarqués de Traitement du Signal

The complexity of systems is currently growing faster than the productivity of system designers and programmers. This phenomenon is called Design Productivity Gap and results in inflating design costs. This Habilitation à Diriger des Recherches (HDR) report present models, methods and tool for improving the Design Productivity (DP) of embedded Digital Signal Processing systems and reducing this Design Productivity Gap. The notion of Design Productivity is commonly used without defining it. This report starts by precisely defining DP before presenting different methods to improve and evaluate it.In a first part, methods based on Models of Computation (MoCs) are explored. They constitute our past and present work on the subject. Dataflow MoCs are used to study application properties (liveness, schedulability, parallelism, etc.) at a high level of abstraction, often before implementation details are known. We have over the last seven years explored how dataflow MoCs can influence the performance and design efforts of modern multicore Digital Signal Processing systems.In the second part of this report, the focus is shifted on the notion of Model of Architecture (MoA) we recently introduced. Parallel and heterogeneous platforms are becoming ever more complex. MoAs have the potential to counteract the DP reduction caused by this rising complexity and constitute the main subject I intend to continue studying in the next years. MoAs, and in general Model-Based Design, can have a great impact on future design methods. Together with the concepts of Cyber-Physical Systems (CPS), Approximate Computing and Rapid Prototyping, they represent new opportunities for reducing the Design Productivity Gap of future DSP systems