Aristotle University Of Thessaloniki (AUTH)
Ανάπτυξη μεθοδολογίας σχεδιασμού επεξεργαστών ειδικού σκοπού
Abstract
dc:descriptionThis Ph.D. dissertation unfolds a methodology for the development and design of application-specific processors, suitable for incorporation in embedded systems, supporting application sets that feature common characteristics. Representative examples of the considered application domains belong to the digital signal processing field, without fundamental reasons for this methodology not being applicable to other realistic/industrial application environments as well. Often, it is required that these applications have to be serviced under constraints of portability, sufficient processing performance, adaptability/flexibility to potential changes to the standards that they ought comply, and the time limitations from the starting point of developing the associated product (to which they are embedded to) to its introduction to the market (time-to-market). In most cases, these requirements cannot be satisfied by deploying the traditional solutions of either Application-Specific Integrated Circuits (ASICs) due to their lack of flexibility, and the associated high development costs and financial risks, or General-Purpose Processors (GPPs) which present increased power/energy consumption and relatively low performance per area unit. Application-Specific Processors and especially their subset of Application-Specific Instruction-set Processors (ASIPs) present a solution to the aforementioned problem. ASIPs provide sufficient flexibility through their programmability by the use of an instruction repertoire, while sustaining the high performance of a dedicated ASIC solution. An application-specific processor is designed for either an application or an application set that present similarities. The success of an ASIP design is determined by the optimal utilization of the respective characteristics of the applications, in order to satisfy requirements on performance, circuit area occupation and energy dissipation, while maintaining the desired level of flexibility. For this purpose, the effort is focused in optimizing the computationally-intensive tasks that are critical for the overall performance of the processor under design. The aims of this Ph.D. dissertation that have been achieved are: the formulation of a methodology for the development of an ASIP design flow, the development of novel techniques at the level of architectural and microarchitectural design for accomplishing the aforementioned requirements regarding processor metrics, and the extension and adaptation of algorithms such as for the generation of custom instructions, for underlying tasks involved in the realization of individual parts of the methodology. For establishing the applicability of the methodology in practice, strong design automation elements have been introduced in both the overall design flow and parts of the methodology in the form of prototype software tools for application analysis, custom instruction generation and selection under constraints, and performance estimation. The merit of the methodology has been evaluated with the end-to-end development of processor prototypes, the design of the corresponding simulators and programming tools, and their implementation in FPGA devices for extensive testing of indicative application fields.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2008
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Καββαδίας, Νικόλαος
- Kavvadias, Nikolaos
Subjects
dc:subject × 18- Ενσωματωμένα συστήματα
- Μικροεπεξεργαστές
- Συστήματα πραγματικού χρόνου
- Μεταγλωττιστές
- Γλώσσα προγραμματισμού C
- Εργαλεία αυτοματοποίησης σχεδιασμού
- Σχεδιασμός υλικών
- Embedded systems
- Microprocessors
- Real time systems
- Compilers
- C programming language
- Design automation tools
- Hardware design
- Φυσικές Επιστήμες
- Φυσική
- Natural Sciences
- Physical Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/18967