Managing the Complexity in Embedded and Cyber-Physical System Design : System Modeling and Design-Space Exploration

Sammanfattning: To cope with the increasing complexity of embedded and cyber-physical system design, different system-level design approaches are proposed which start from abstract models and implement them using design flows with high degrees of automation. However, creating models of such systems and also formulating the mathematical problems arising in these design flows are themselves challenging tasks. A promising approach is the composable construction of these models and problems from more basic entities. Unfortunately, it is non-trivial to propose such compositional formulations today because the current practice in the electronic design automation domain tends to be on using imperative languages and frameworks due to legacy and performance-oriented reasons.This thesis addresses the system design complexity by first promoting proper formalisms and frameworks for capturing models and formulating design-space exploration problems for electronic system-level design in a declarative style; and second, propose realizations based on the industrially accepted languages and frameworks which hold the interesting properties such as composability and parallelism.For modeling, ForSyDe, a denotational system-level modeling formalism for heterogeneous embedded systems is chosen, extended with timed domains to make it more appropriate for capturing cyber-physical systems, and mapped on top of the IEEE standard system design language SystemC. The realized modeling framework, called ForSyDe-SystemC, can be used for modeling systems of heterogeneous nature and their composition to form more sophisticated systems and also conducting parallel and distributed simulation for boosting the simulation speed. Another extension to ForSyDe, named wrapper processes, introduces the ability to compose formal ForSyDe models with legacy IP blocks running in external execution environments to perform a heterogeneous co-simulation.In platform-based design flows, the correct and optimal mapping of an application model onto a flexible platform involves solving a hard problem, named design space exploration. This work proposes Tahmuras, a constraint- based framework to construct generic design space exploration problems as the composition of three individual sub-problems: the application, the platform, and the mapping and scheduling problems. In this way, the model of the design space exploration problem in Tahmuras is automatically generated for each combination of application semantics, target platform, and mapping and scheduling policy simply by composing their respective problems. Using constraint programming, problems can be modeled in a declarative style, while they can be solved in a variety of different styles, including imperative solving heuristics commonly used to solve difficult problems. Efficient parallel solvers exists for constraint programming.