Transformational partitioning for the co-design of mixed hardware/software systems

This thesis develops a new methodology for hardware/software co-design. It introduces a transformational partitioning approach capable of handling distributed systems and multiprocessor target architectures. This semi-automatic partitioning approach allows to link a system-level specification to a hardware/software architecture in a short design time with fast exploration of the design space. The main application domain of this work is the design of distributed architectures composed of software processors (e.g. Instruction-level programs) and hardware processors (e.g. ASICs). In contrast to the classical co-design approach implemented by most of existing co-design systems, this approach deals with flexible architectures composed of processors communicating through a complex network. The main contribution of this work is the application of the transformational approach to hardware/software co-design of multi-processor architecture. The user starts the design process with a system-level specification and an architectural solution in mind and realizes refinements to transform the initial specification into the architectural solution. Each incremental refinement step fixes some implementation detail. In this model, the initial specification is given in the SDL language. This specification is converted to an intermediate model adapted to the synthesis algorithms. The intermediate model is refined and implemented on a distributed C/VHDL model. The estimation methods developed allows: a fast exploration of the design space; to reduce the implementation costs; to verify the consequence of different refinement strategies; and study the different optimization criteria of architectures. The methodology presented here is implemented in a co-design environment called Cosmos.
Ce travail de thèse développe une nouvelle méthodologie pour la conception conjointe du logiciel et du matériel. Cette méthodologie met en pratique une approche transformationnelle de découpage capable de manipuler des systèmes distribués et des architectures multiprocesseurs. Cette approche semi-automatique de découpage réalise le lien entre une spécification au niveau système et une architecture logicielle/matérielle de manière rapide permettant une exploration rapide de l'espace des solutions. Le principal domaine d'application de ce travail est la conception des architectures distribuées, composées, de processeurs logiciels (e.g. programmes) et processeurs matériels (e.g. ASICs). Par rapport à l'approche classique de conception, utilisée dans la plus grande partie des systèmes de conception conjointe existants, cette approche supporte des architectures flexibles composées de processeurs qui communiquent en utilisant un réseau de communication complexe. La principale contribution de ce travail est l'application de l'approche transformationnelle à la conception conjointe logicielle/matérielle d'architectures multiprocesseurs.