The research activities described in the following address different aspects of research in embedded systems architectures, which are listed in the following:
- In the DFG project Self-Adaptive Coarse-Grained Reconfigurable Architectures as Reliability Enhancers in Embedded Systems we devise methods to develop and implement a multi-functional, self-adaptive coarse-grained reconfigurable core as a reliability enhancer. This core will be used to lower the reliability cost of embedded systems. CGRAs already contain many structures that allow high reliability and high yield embedded systems: reconfigurable, interchangeable and redundant components, a communication network, parallelism, and the tools to use this new kind of architecture. However, what is missing are new methods that exploit these features for better reliability, as current work has only focused on the flexibility aspects of CGRAs. This work will close this gap with methods that produce high yield and highly reliable embedded systems.
- A second research activity is done in close cooperation with the Federal University of Pernambuco (UFPE) in Brazil in the area of communication synthesis techniques and tools for platform design of embedded systems. In this research cooperation we also explore FPGAs to accelerate data intensive applications. Especially seismic computations offer the opportunity to exploit both data-parallel processing and more utilization through pipelined approaches, thus arrays of FPGAs become even more attractive.
- A third research activity addresses the analysis and optimization of hybrid high-performance processing architectures. The approach works out decision criteria and techniques for an integration of accelerator components into a high-performance processing architecture, in terms of optimizing the characteristics and reducing the total costs of a given complex system application running on that hybrid scenario. IBM Deutschland Research & Development GmbH sponsored and co-operated this research in which mainframe systems are extended with blade units of different architectures. We show that a many applications can be adopted to such a hybrid system. A very successful example is to execute insurance calculation on network-attached accelerator blades. This results in improved interactive usage and batch processing.