Design Space ExplorationThe area of research I am interested in can be described as design space exploration. I have been (and still am) working on several aspects of design space exploration for embedded systems: performance analysis, modelling of applications and architectures, and multiobjective optimization. In the domain of embedded systems, I and my colleagues focus mainly on the design space exploration for network processors. Performance AnalysisReal-time calculus is an analytical method used to assess the performance of embedded systems. We use this method to evaluate the performance of design points during the exploration. In general, the methods used for performance analysis in the context of design space exploration have to be rather fast because many designs have to be evaluated during an exploration run. Currently, we are investigating how we can speed up the analytical method and how we can combine real-time calculus with other performance analysis methods. OptimizationAnother important aspect of design space exploration is optimization. We use black-box multiobjective optimizers and de-couple problem-specific parts (e.g. the evaluation of a design point, or finding its neighborhood) and the problem-independent (generic) parts of the DSE framework using the PISA interface (Platform- and programming language independent Interface for Search Algorithms). More information about PISA and many different optimizers can be found here. We have tested the performance of several different optimizers for design space exploration of network processors using the PISA-compliant EXPO tool. Based on the insights from the performance comparisons, we are currently developing new evolutionary algorithms for multiobjective optimization. ToolsI am involved in the development of a few tools (parts of tools) for design space exploration and performance analysis.
Projects:
|