The Dagstuhl Seminar on "Dynamically Reconfigurable Architectures" brought together 42 participants from 11 different countries. As with its predecessor seminar held in 1998, the participants came from three distinct communities, concerned with:
Some of the specific questions addressed by the seminar were:
There were 34 talks given by the participants, each shedding some light on one or more of the above issues. A lively discussion session on the Wednesday evening provided an opportunity to synthesise the differing viewpoints of the participants. Interestingly, one topic that provoked much discussion was how to define the term `dynamically reconfigurable architecture', with differing positions emerging on how dynamic the reconfiguration should be expected to be, and on what types of architecture might be included.
The seminar emphasised that technological advances have opened up new ways of implementing complex systems, ways that blur the traditional barriers between hardware and software components. Because of this, existing design tools do not seem to be adequate for the necessary new design styles - in fact, at one extreme, it is possible to let hardware evolve by itself, learning the required functions. There are many applications, ranging from computing on spacecraft to the control of motors, that should be able to derive benefits from enabling novel types of configurable computing systems.
A recurrent theme was whether the traditional Von Neumann model of computation has now run out of steam, as sequential processors become just one type of component in parallel and distributed systems. Irrespective of opinion on this, the need for robust computational models for run-time reconfigurable systems and evolvable hardware was highlighted, possibly following the general philosophy of the RMESH model for reconfigurable meshes. A thorough understanding of the computational possibilities is of great importance in assessing the benefits of dynamic reconfiguration for real-life applications.
Based on such an understanding, it will then be possible to make progress on finding apt high-level languages and notations for expressing computations based upon dynamically reconfigurable architectures. While mature concepts from both hardware and software engineering will influence this exercise, it seems that these must be augmented by natural high-level ways of expressing features such as concurrency, topology and reconfiguration.
Bridges then have to be built between the high level descriptions of function and the physical underlying architectures, which may be based on a variety of technologies, including optoelectronics. Robust computational models will also enable the development of apt virtual machines, that mask the particular technical details of specific reconfigurable computing machines.
The seminar pointed the way to how progress might be made in these directions, and new collaborations were initiated. The pleasant atmosphere of Schloß Dagstuhl was an important incentive for the lively interaction between the participants. We thank all who contributed to the success of the seminar. Furthermore, we gratefully acknowledge financial support for the seminar from Xilinx Inc.
Gordon Brebner, Karl-Heinz Brenner, Hossam El Gindy and Hartmut Schmeck