Investigation into energy-efficient co-synthesis of distributed embedded systems

Abstract

The first part of this thesis addresses dynamic power minimisation for data and control dominated embedded systems.  A novel conditional behaviour-aware DVS (CBADVS) targeting embedded systems expressed as conditional task graphs has been proposed.  The CBADVS exploits the slack time taking into account the conditional behaviour of the application, so that energy dissipation is reduced and, at the same time, timing feasibility is guaranteed for all possible condition values.  Furthermore, a genetic algorithm based mapping is introduced to optimise the system implementation towards effective exploitation of the proposed CBADVS, hence, leading to further energy reduction.  The technique has been validated extensively including a real-life example of vehicle cruise controller, and it has been shown that up to 42% energy saving is achieved with 5 second computational time, and with no penalty in meeting real-time constraints.

The second part of the thesis addresses the impact of communications on dynamic power minimisation in data and control dominated systems design.  It is shown how the concept of enhanced system model, which captures the time and power costs of communications, allows the design of energy-efficient embedded systems by integrating communications with the above CBADVS based cy-synthesis technique.  The computational time of the communication-integrated co-synthesis technique has been reduced by analysing the deficiency of concurrent task and communication mapping, and decoupling communication from task mapping.  A large number of experiments have been used to investigate the effect of alternative communication architectures on system quality in terms of energy efficiency.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata