Feld, Dustin (2017). FieldPlacer - A flexible, fast and unconstrained force-directed placement method for heterogeneous reconfigurable logic architectures. PhD thesis, Universität zu Köln.
|
PDF (PDF-X3)
DustinFeld_PhDThesis.pdf - Published Version Download (32MB) |
Abstract
The field of placement methods for components of integrated circuits, especially in the domain of reconfigurable chip architectures, is mainly dominated by a handful of concepts. While some of these are easy to apply but difficult to adapt to new situations, others are more flexible but rather complex to realize. This work presents the FieldPlacer framework, a flexible, fast and unconstrained force-directed placement method for heterogeneous reconfigurable logic architectures, in particular for the ever important heterogeneous FPGAs. In contrast to many other force-directed placers, this approach is called ‘unconstrained’ as it does not require a priori fixed logic elements in order to calculate a force equilibrium as the solution to a system of equations. Instead, it is based on a free spring embedder simulation of a graph representation which includes all logic block types of a design simultaneously. The FieldPlacer framework offers a huge amount of flexibility in applying different distance norms (e. g., the Manhattan distance) for the force-directed layout and aims at creating adapted layouts for various objective functions, e. g., highest performance or improved routability. Depending on the individual situation, a runtime-quality trade-off can be considered to either produce a decent placement in a very short time or to generate an exceptionally good placement, which takes longer. An extensive comparison with the latest simulated annealing placement method from the well-known Versatile Place and Route (VPR) framework shows that the FieldPlacer approach can create placements of comparable quality much faster than VPR or, alternatively, generate better placements in the same time. The flexibility in defining arbitrary objective functions and the intuitive adaptability of the method, which, among others, includes different concepts from the field of graph drawing, should facilitate further developments with this framework, e. g., for new upcoming optimization targets like the energy consumption of an implemented design.
Item Type: | Thesis (PhD thesis) | ||||||||
Translated abstract: |
|
||||||||
Creators: |
|
||||||||
URN: | urn:nbn:de:hbz:38-75751 | ||||||||
Date: | 2017 | ||||||||
Publisher: | Fraunhofer Verlag | ||||||||
ISBN: | 978-3-8396-1155-5 | ||||||||
Language: | English | ||||||||
Faculty: | Faculty of Mathematics and Natural Sciences | ||||||||
Divisions: | Faculty of Mathematics and Natural Sciences > Department of Mathematics and Computer Science > Institute of Computer Science | ||||||||
Subjects: | Data processing Computer science Mathematics |
||||||||
Uncontrolled Keywords: |
|
||||||||
Date of oral exam: | 24 October 2016 | ||||||||
Referee: |
|
||||||||
Refereed: | Yes | ||||||||
URI: | http://kups.ub.uni-koeln.de/id/eprint/7575 |
Downloads
Downloads per month over past year
Export
Actions (login required)
View Item |