System-level synthesis of adaptive computing systems
Sandeep K. Neema, János Sztipanovits · 2001
With the rapid proliferation of embedded computer technology over the years, functional and performance demands from embedded computer systems have increased severely. In addition to the resource and size constraints, these systems may need to function in rapidly changing environments where performance and functional goals change over time. Adaptive Computing has been considered to meet the contradicting demands of high-performance with minimal resources and changing functional and performance goals over time. The main theme of the adaptive approach is to develop system architectures that can be modified or reconfigured dynamically upon mode changes, to match the algorithm and maximize performance. The primary challenge of the Adaptive Computing approach is in system design. The complexity of the design process multiplies, as the system designer needs to design and maintain a large number of different system architectures that exist at different times in the lifetime of the operational system. A methodology and extensible software toolset for design and synthesis of adaptive computing systems is presented. The methodology is based on Model Integrated Computing (MIC), an approach for design and synthesis of computer-based systems. A graphical modeling environment has been developed for design capture and representation of adaptive systems. The environment facilitates creation of large flexible design spaces by explicit representation of design alternatives. Large design spaces improve trade-off and optimization opportunities; however, exploration in such large design spaces is a major challenge. A constraint-guided design space exploration method has been developed, that uses constraints to prune the design space. By selectively applying different constraints to the design space, an engineer can explore and trade-off by quickly “zooming-in” to different regions of the design space. A constraint language based on Object Constraint Language (OCL) is presented for expression of a wide variety of user-defined constraints. A symbolic constraint satisfaction method based on Ordered Binary Decision Diagrams (OBDD) has been developed for application and satisfaction of constraints in the design exploration method.