Résumé
Power efficiency of embedded systems is nowadays a challenge to reach the increasing performance requirements while limiting the power consumption. Adaptive architecture allows to adapt the operating point of each power domain of a MultiProcessor System-on-Chip using independent Dynamic Voltage and Frequency Scaling techniques. Furthermore the optimal operating point of each power domain is dependent on the current variability state, thus it appears essential to monitor locally and on-line the environmental variability. To limit the power consumption and hardware overhead the monitoring system must have the lowest cost and the best possible efficiency. This paper presents a monitoring system based on two complementary methods to estimate dynamically the local conditions using low-cost digital sensors. A fast voltage drop monitoring method is performed every 0.66μs and a low-cost calibration method is also proposed based on the Voltage-Temperature state of the platform at startup.