Democritus University of Thrace (DUTH)
Σχεδιασμός υλικού και λογισμικού επαναδιαμορφούμενων συστημάτων VLSI με χαμηλή κατανάλωση ισχύος
Abstract
dc:descriptionThis PhD deals with the development of novel methodologies for hardware/software co-design, targeting to develop more efficient reconfigurable architectures. Even though the proposed methodologies are applicable to any device, throughout this thesis they are applied in order to improve the most common reconfigurable architecture, the FPGA (Field-Programmable Gate Array). More specifically, during this thesis a novel reconfigurable architecture was designed in full-custom approach. This device exhibits remarkable power/energy savings compared to existing approaches, without affecting other crucial design parameters, such as delay and/or silicon area. An equivalent important problem is the development of supporting algorithms and CAD tools that fully employ the features introduced by the target reconfigurable architecture. Due to this, a design framework targeting to FPGAs was also developed. Based on existing studies among the alternative platforms for realizing digital applications onto hardware, it is evident that the reconfigurable architectures exhibit remarkably better results compared to general purpose (CPU) and special purpose (DSP) processors. However, they are not so efficient as the application specific hardware (ASIC). This occurs due to the limitations of interconnection network, which occupies almost the 80% of the device area, while it consumes about 85% of the total energy dissipation. Moreover, if we take into consideration that the interconnection resources are not fully utilized over the reconfigurable architecture, we might conclude that a more careful design is required. In order to resolve this problem, throughout this thesis a methodology for designing general-purpose heterogeneous interconnection networks is proposed. This methodology meets the requirements of the employed algorithms (i.e. placement and routing), as well as the application domains (Le. multimedia, DSP, etc.) that are realized onto the FPGA. The increased power/energy consumption of reconfigurable architectures, among others affects the on-chip temperature during their operation. This issue was recently described from the industry development of reconfigurable architectures, as one of the most important problem, which needs to be resolved immediately. Moreover, this problem is even more crucial than the corresponding one from ASIC devices, due to the increased power density of FPGAs. Among other problems that might occur due to the high temperature values, are the reliability issues, as well as the life-time reduction. In order to face this problem, during this thesis, a novel methodology that manages the on-chip temperature is proposed. This methodology guarantees to reduce the maximal temperature values and to distribute their power/temperature sources over the whole architecture. Apart from the disadvantages that reconfigurable architecture exhibits, compared to ASIC devices, it has a competitive advantage, as it permits the implementation of every application. More specifically, the opportunity for altering the functionality of the reconfigurable hardware let FPGAs realize applications in short time and with reduced cost. We have to mention that, in order this opportunity to be fully employed, there is a demand for CAD tool that manage efficiently all the parameters introduced by this procedure. Due to this, I propose a methodology for calculating the configuration data regarding the reconfigurable architectures, which is independent of the target FPGA device, while it supports a number of features (Le. partial, dynamic, run-time configuration, etc.) which leads to more efficient application realization. The modem digital applications demand even more powerful devices that operate in higher operation frequencies, while they consume less power. In order the current architectures to satisfy this demand; they integrate more logic blocks onto the same chip. As a result, the new devices exhibit additional problems that are related to the increased wire-length. Such a design constraint has turned the industry to employ alternative fabrication technologies targeting to the design of architectures. These approaches are apart from the conventional CMOS technology, as the device scaling has almost reached the fabrication limits of silicon. One of the emerging technologies that shows promising results for this problem is the three dimensional integration, which is thought to be used for the designing of upcoming devices. However, apart from the advantages that this technology provides, up to now there are no available CAD tools that quantify the potential gains of employing such a new design approach. Due to this, the last issue that is discussed in this PhD thesis affects the development of a novel design methodology, as well as the supporting CAD tools, targeting to 3D reconfigurable architectures.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2008
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Siozios, Konstantinos
- Σιώζιος, Κωνσταντίνος
Subjects
dc:subject × 12- Νόμος Moore
- Ανάπτυξη ψηφιακών σχεδιασμών
- Επαναδιαμορφούμενες αρχιτεκτονικές
- Εργαλεία CAD
- Αρχιτεκτονική δικτύου διασύνδεσης
- Δίκτυα διασύνδεσης υψηλών επιδόσεων
- Μεθοδολογία απεικόνισης εφαρμογών καθοδηγούμενη απο τη θερμοκρασία
- Μεθοδολογία δημιουργίας αρχείων προγραμματισμού
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/20928