Abstract
Technological advances of recent years laid the foundation consolidation of informatisationof society, impacting on economic, political, cultural and socialdimensions. At the peak of this realization, today, more and more everydaydevices are connected to the web, giving the term ”Internet of Things”. The futureholds the full connection and interaction of IT and communications systemsto the natural world, delimiting the transition to natural cyber systems and offeringmeta-services in the physical world, such as personalized medical care, autonomoustransportation, smart energy cities etc. . Outlining the necessities of this dynamicallyevolving market, computer engineers are required to implement computingplatforms that incorporate both increased systemic complexity and also cover awide range of meta-characteristics, such as the cost and design time, reliabilityand reuse, which are prescribed by a conflicting set of functional, technical andconstruction constraints. This thesis aims to ad ...
Technological advances of recent years laid the foundation consolidation of informatisationof society, impacting on economic, political, cultural and socialdimensions. At the peak of this realization, today, more and more everydaydevices are connected to the web, giving the term ”Internet of Things”. The futureholds the full connection and interaction of IT and communications systemsto the natural world, delimiting the transition to natural cyber systems and offeringmeta-services in the physical world, such as personalized medical care, autonomoustransportation, smart energy cities etc. . Outlining the necessities of this dynamicallyevolving market, computer engineers are required to implement computingplatforms that incorporate both increased systemic complexity and also cover awide range of meta-characteristics, such as the cost and design time, reliabilityand reuse, which are prescribed by a conflicting set of functional, technical andconstruction constraints. This thesis aims to address these design challenges bydeveloping methodologies and hardware/software co-design tools that enable therapid implementation and efficient synthesis of architectural solutions, which specifyoperating meta-features required by the modern market. Specifically, this thesispresents a) methodologies to accelerate the design flow for both reconfigurableand application-specific architectures, b) coarse-grain heterogeneous architecturaltemplates for processing and communication acceleration and c) efficient multiobjectivesynthesis techniques both at high abstraction level of programming andphysical silicon level.Regarding to the acceleration of the design flow, the proposed methodologyemploys virtual platforms in order to hide architectural details and drastically reducesimulation time. An extension of this framework introduces the systemicco-simulation using reconfigurable acceleration platforms as co-emulation intermediateplatforms. Thus, the development cycle of a hardware/software productis accelerated by moving from a vertical serial flow to a circular interactive loop.Moreover the simulation capabilities are enriched with efficient detection and correctiontechniques of design errors, as well as control methods of performancemetrics of the system according to the desired specifications, during all phasesof the system development. In orthogonal correlation with the aforementionedmethodological framework, a new architectural template is proposed, aiming atbridging the gap between design complexity and technological productivity usingspecialized hardware accelerators in heterogeneous systems-on-chip and networkon-chip platforms. It is presented a novel co-design methodology for the hardwareaccelerators and their respective programming software, including the tasks allocationto the available resources of the system/network. The introduced frameworkprovides implementation techniques for the accelerators, using either conventionalprogramming flows with hardware description language or abstract programmingmodel flows, using techniques from high-level synthesis. In any case, it is providedthe option of systemic measures optimization, such as the processing speed,the throughput, the reliability, the power consumption and the design silicon area.Finally, on addressing the increased complexity in design tools of reconfigurablesystems, there are proposed novel multi-objective optimization evolutionary algo-rithms which exploit the modern multicore processors and the coarse-grain natureof multithreaded programming environments (e.g. OpenMP) in order to reduce theplacement time, while by simultaneously grouping the applications based on theirintrinsic characteristics, the effectively explore the design space effectively.The efficiency of the proposed architectural templates, design tools and methodologyflows is evaluated in relation to the existing edge solutions with applicationsfrom typical computing domains, such as digital signal processing, multimedia andarithmetic complexity, as well as from systemic heterogeneous environments, suchas a computer vision system for autonomous robotic space navigation and manyacceleratorsystems for HPC and workstations/datacenters. The results strengthenthe belief of the author, that this thesis provides competitive expertise to addresscomplex modern - and projected future - design challenges.
show more
