A Survey of Co-Design Ideas and Methodologies (draft)

Designing embedded systems becomes more and more complex due to the increasing size of integrated circuits, increasing software complexity and decreasing time-to-market requirements and product costs. A possible way to deal with this complexity is the use of high-level system design languages. Such approaches combine modelling of software and hardware, and are called Co-Design approaches. A fundamental difference between hardware components and software components is in the way commands are executed: parallel vs. sequential. A modelling approach needs to combine these, to allow efficient synthesis and to prevent a (very expensive) paradigm shift. The differences in execution between software-and hardware components can be analysed using computational models , which are mathematical descriptions of the behavior of a component. Most Co-Design approaches use a modelling language which is based on one or more computational models. The amount of computational models used in the modelling language (Internal Design Representation or IDR) of a Co-Design method can be used to discriminate between 2 types of approaches. Some approaches combine multiple computational models in the IDR; these are the most ambitious approaches. A single IDR to model the entire system has the advantage that the decision can be postponed what to implement in hardware and what in software. Many researchers are doubting whether such an approach will work, because it is very difficult to combine various models of computation in a single IDR. Therefore there are other approaches that explicitly use more than one IDR. Here the choice between hardware and software components is made before the modelling phase starts, and each component is modelled in an IDR specially targetted towards that type of computation. The integration offered by this approach is less deep, but simulation is used to validate the design and integration of the various IDR (this is called co-simulation). Future research in Co-Design will be concentrated in the following directions: • Special complete design flows based on single IDR models • Research into richer and better IDRs • Approaches that use more than one IDR (co-simulation based approaches). These have design flows that are partly manual (and in which the hardware-software split is made earlier) • Approaches based on a (much) lower level of abstraction, that are focussed on component reuse (platform-based design). ii ABSTRACT iii For Chess the most valuable investment would be research into the first and third type of approaches: special complete design flows and co-simulation based approaches. …