The design of a complex real-time embedded system requires the specification of its functionality, the design of the hardware and software architectures, the implementation of hardware and software components and finally the system validation. The designer, starting from the specification, refines the solution trying to minimize the system cost while satisfying functional and non functional requirements. The automatic code generation from models and the introduction of the platform-based design methodology can drastically improve the design efficiency of the software partition, while maintaining acceptable the cost overhead of the final system. In this approach, both top-down and bottom-up aspects are considered and solutions are found by a meet-in-the-middle approach that couples model refinement and platform modeling. In more details, given a model of the implementation platform, which describes the available services and data types, the algorithms captured by models are refined and then automatically translated to software components. These components are integrated with handwritten (e.g. legacy) software modules together with the software platform. A final validation phase on the real target is performed to finally validate the functionality and to guarantee that the performance constraints are met.
The methodology described in this paper has proven in the years of deployment its validity and maturity level. The effective results are the improvement of the time-to-market and the capability to cope with the complexity of modern embedded controllers for power-train. The selected automatic code generation environment (the model compiler) has been instrumental in implementing our model based design methodology.
In the future, the platform based design methodology will allow an easy accommodation of the new automotive software architecture standard promoted by the AUTOSAR consortium.