Low Cost Dynamic Voltage and Frequency Policy based upon Robust Control Techniques

Sylvain Durand, Suzanne Lesecq, Edith Beigné and Christian Fabre.
Software Technologies Concertation on Formal Methods for Components and Objects. October 3rd-5th, 2011, Torino, Italy.
Today mobile computing platforms need ever-increasing computational performances while their energy consumption is drastically limited by battery lifespan. An optimal operating point is obtained thanks to a compromise between performance and power consumption. For distributed architectures (e.g. MultiProcessor System On Chip), the supply voltage and the operating frequency of each processing element are usually tuned dynamically to reach efficient performance/power consumption trade-offs. These control strategies (also called Dynamic Voltage and Frequency Scaling (DVFS)) can be implemented as closed-loop systems using control theory. This leads to the “sense and react” paradigm where information on the system state
(especially actual temperature, occurrence of timing faults, supply voltage changes must first be
gathered, possibly using data fusion techniques. Then from this information, and with robust and/or adaptive control approaches, the power consumption of the system can be finely controlled