Low energy consumption has emerged as one of the most important design objectives for many modern embedded systems, particularly the battery-operated PDAs. For some soft real-time applications such as multimedia applications, occasional deadline misses can be tolerated. How to leverage this feature to save more energy while still meeting the user required quality of service (QoS) is the research topic this thesis focuses on. We have proposed a new probabilistic design methodology, a set of energy reduction techniques for single and multiple processor systems by using dynamic voltage scaling (DVS), the practical solutions to voltage set-up problem for multiple voltage DVS system, and a new QoS metric.

Most present design space exploration techniques, which are based on application's worst case execution time, often lead to over-designing systems. We have proposed the probabilistic design methodology for soft real-time embedded systems by using detailed execution time information in order to reduce the system resources while delivering the user required QoS probabilistically.

One important phase in the probabilistic design methodology is the offline/online resource management. As an example, we have proposed a set of energy reduction techniques by employing DVS techniques to exploit the slacks arising from the tolerance to deadline misses for single and multiple processor systems while meeting the user required completion ratio statistically.

Multiple-voltage DVS system is predicted as the future low-power system by International Technology Roadmap for Semiconductors (ITRS). In order to find the best way to employ DVS, we have formulated the voltage set-up problem and provided its practical solutions that seek the most energy efficient voltage setting for the design of multiple-voltage DVS systems. We have also presented a case study in designing energy-efficient dual voltage soft real-time system with (m, k)-firm deadline guarantee.

Although completion ratio is widely used as a QoS metric, it can only be applied to the applications with independent tasks. We have proposed a new QoS metric that differentiates firm and soft deadlines and considers the task dependency as well. Based on this new metric, we have developed a set of online scheduling algorithms that enhance quality of presentation (QoP) significantly, particularly for overloaded systems.