Compilation Strategies and Design Space Exploration for Parallel Embedded Systems

The focus in embedded system design is moving away from single processor implementations towards (heterogeneous) multiprocessor system-on-chip (MPSoC) architectures. While offering high scale integration, high computing power and low power consumption, MpSoC designs face enormous challenges in terms of software design: choice of a suitable specification model, automatic or semi-automatic parallelization, scalability, efficiency in terms of run-time and energy consumption, predictable behavior in terms of functional and non-functional properties. Our research concentrates on the challenge of providing an environment for programming parallel applications and mapping them to heterogeneous distributed architectures. We try to tackle this challenge by leveraging an appropriate model of computation, namely Kahn process networks, using a dedicated application programming interface, and applying appropriate methods for mapping optimization. Much of our results are integrated into the high-level compilation framework DOL (Distributed Operation Layer).

Furthermore, embedded systems are usually domain-specific and try to use the characteristics of the particular application domain in order to arrive at competitive implementations. The combination of a huge design space on the one hand and the complexity in interactions on the other hand necessitate automatic or semi-automatic (interactive) methods for exploring different designs. Here, we use our experience in multiobjective optimization to arrive at viable system implementations.

More information can be found on the home pages of the specific projects we are involved in:

Some relevant publications on our work since 2000: Publications on MPSOC, Publications on Embedded System Design, Publications on Reconfigurable Computing, Publications on Design Space Exploration.