Dependability Driven System Level Co-Design and Optimization of Embedded Systems

Embedded systems are becoming pervasive in diverse application domains such as automotive, avionic, medical, control and their functionality is increasingly defined by software (SW). Such systems especially in safety-critical (SC) applications, with implications on system dependability and real-time must be designed to be dependable (fault tolerant) enough and have to meet timing requirements in order to avoid any potential catastrophic consequences. More and more new and innovative functionality is being integrated into such systems, invariably leading to a heterogeneous environment consisting of applications of mixed-criticality (SC and non-SC), each with associated extra-functional requirements such as dependability, timing, resources and power consumption. Efficient system design methods and techniques are needed to be developed to integrate these diverse applications across limited hardware (HW) resources. This thesis develops a novel dependability-driven system level SW-HW co-design methodology which systematically guides the design and optimization of such embedded systems from requirements analysis phase through integration to the prototyping. We first develop the concept of a consolidated mapping of SC and non-SC applications onto a common distributed computing architecture such that their operational delineation is maintained over the integration. We then devise an optimization based co-design approach through quantifying the various design objectives/variables. Our aim is to develop the design methodology for an integrated embedded architecture. A heuristic based systematic mapping process is elaborated for integrating varied criticality applications. A set of functional and extra-functional requirements and constraints are satisfied during the mapping. At an early design stage, the mapping considers rigorous design strategies such as fault tolerance, fault/error containment, robust partitioning, timeliness, resource and power consumption. Dependability is ensured through replication of application jobs with high criticality and a schedulability analysis is presented for guaranteeing the timeliness properties. The developed mapping algorithm generates an initial feasible solution and guides the optimization in a unified and efficient way. We develop a comprehensive multi variable optimization (MVO) framework which quantifies and optimizes a set of competing variables from dependability, real-time and resource perspectives. During the optimization process the satisfaction of constraints is maintained. The key aspect of the approach is to enhance dependability by using fault containment mechanisms including the quantification and estimation of the considered design variables. The framework is extended by quantifying and modeling the reliability and system level power consumption as design variables. In order to evaluate and validate the developed methods and techniques presented in the thesis, we have performed extensive experiments. Throughout the thesis we illustrate our ideas and concept using real-life automotive examples (where these techniques were actually validated). The concept is applied to a supporting tool set where we develop a prototype of the system level co-design approach. The prototype is created adhering to a transformation based design process