Models and formal verification of multiprocessor system-on-chips

In this article we develop a model for applications running on multiprocessor platforms. An application is modelled by task graphs and a multiprocessor system is modelled by a number of processing elements, each capable of executing tasks according to a given scheduling discipline. We present a discrete model of computation for such systems and characterize the size of the computation tree it suffices to consider when checking for schedulability.

Analysis of multiprocessor system on chips is a major challenge due to the freedom of interrelated choices concerning the application level, the configuration of the execution platform and the mapping of the application onto this platform. The computational model provides a basis for formal analysis of systems.

The model is translated to timed automata and a tool for system verification and simulation has been developed using Uppaal as backend. We present experimental results on rather small systems with high complexity, primarily due to differences between best-case and worst-case execution times. Considering worst-case execution times only, the system becomes deterministic and using a special version of {Uppaal}, where the no history is saved, we could verify a smart-phone application consisting of 103 tasks executing on 4 processing elements.