Worst-Case Temperature Analysis for Different Resource Availabilities: A Case Study

With three-dimensional chip integration, the heat dissipation per unit area increases rapidly and may result in high on-chip temperatures. Real-time constraints cannot be guaranteed anymore as exceeding a certain threshold temperature can immediately reduce the systems reliability and performance. Dynamic thermal management methods are promising methods to prevent the system from overheating. However, when designing modern real-time systems that make use of such thermal management techniques, the designer has to be aware of their effect on the maximum possible temperature of the system. This paper proposes an analytic framework to calculate the worst-case temperature of a system with general resource availabilities. The event and resource model is based on real-time and network calculus so that the method is able to handle a broad range of uncertainties in terms of task arrivals and available computational power. In various case studies, the proposed framework is applied to an advanced multimedia system to analyze the impact of dynamic frequency scaling and thermal-aware scheduling techniques on the worst-case temperature of an embedded real-time system.