ISBN 9783843921923

978-3-8439-2192-3, Reihe Informatik

Joachim Falk
A Clustering-Based MPSoC Design Flow for Data Flow-Oriented Applications

267 Seiten, Dissertation Universität Erlangen-Nürnberg (2014), Hardcover, B5

Zusammenfassung / Abstract

This book presents advances for state-of-the-art design flows tackling the challenging problem of system design for embedded systems to program multiple connected and heterogeneous resources by: (1) introducing a modeling language suited to these systems; (2) contributing a methodology called clustering that fits the concurrency of an application to the degree of concurrency provided by the architecture; and (3) discussing the synthesis of the modeled applications via a synthesis back-end that supports multiple targets such as single core software implementations, virtual prototypes of distributed Multi-Processor System-on-Chip (MPSoC) systems, and dedicated hardware implementations.

Some embedded systems like automotive controller networks are naturally heterogeneous distributed systems, while heterogeneous distributed architectures have been chosen for other embedded systems in portable devices. This choice is due to the dichotomy between the power requirements induced by the rising hardware capabilities and the comparatively slow improvements in battery technology. Here, the power-efficient implementations of functionality enabled by heterogeneous implementations using power-efficient dedicated accelerators and multicore processors is exploited to resolve this dichotomy.

Problems arise when these embedded systems are implemented via traditional sequential programming languages since these languages are not designed to handle concurrency well. Concurrency, however, is inherently present in distributed systems and must be handled efficiently to exploit their advantages. In contrast, the data flow paradigm is well-suited to express concurrency and, hence, is a natural fit for programming heterogeneous distributed systems. Hence, this book is concerned with a design flow starting from applications specified via this paradigm and targeting multiple architectures such as single- and multicore software implementations, virtual prototypes of distributed MPSoC systems, and dedicated hardware implementations. However, to create an efficient implementation of a data flow-oriented application for a given target architecture, the concurrency of the application must be fitted to the concurrency provided by the architecture. To resolve this problem, a novel clustering methodology is presented in this book.