Coarse-grain reconfigurable architectures have been a matter of intense research in the last few years. They promise to be more adequate for computational tasks due to their better efficiency and bigger speed. As the coarse granularity implies also a reduction of flexibility, a universal architecture seems to be hardly feasible. Based on the KressArray architecture family, a design-space exploration system is being implemented, which supports the designer in finding an appropriate architecture for a given application domain. By analysing the results of a number of different experimental mappings, the system derives suggestions how the architecture can be enhanced. This paper presents an analyser tool, which allows the generation of such suggestions using approximate reasoning based on fuzzy logic. The tool is flexible enough to support different data gathering methods and an extensible rule set.
Reiner W. Hartenstein, Michael Herz, Thomas Hoffma