The Research On Dynamic Low Power Design Policies In Embedded Sytem

Embedded system designers usually focus on the stability and real-time during previous designing procedure. Nowadays, minimization of power consumption is one of the most challenging problems during embedded system designing. Power reduction in embedded system can be realized by static and dynamic techniques, and the former has been the dominated mothod. However, static techniques can not satisfy the system requirements on power consumption absolutely with the improvement on system function and integration density. Recent research has concentrated on dynamic power reduction techniques, which include DPM (dynamic power management) and DVS (dynamic voltage scaling) . Based on the exploring of DPM and DVS, various dynamic power-reduction policies, applicable to different system environment, are presented in this paper.DPM aims at minimization of power consumption of electronic systems by dynamically switching the power state of power manageable components according to the variations of workloads. An adaptive timeout policy named CVAAS is introduced for exploiting the nonidealities of real-life battery cells. By importing battery effect, CVAAS can further reduce the system power, and extend the adaptive object of DPM simultaneously. PBALT is provided against the defect of ALT model. By characterizing the device activity in probability, PBALT has higher hit ratio and can optimize the balance between performance and power. In addition, it is shown that traditional DPM policies observe requests only at the target device and lost the extra information from the OS, so an on-line device scheduling policy called EODSA is presented in view of task level. EODSA provides more information to power manager about future requests by classifying requests according to their sources, and can reduce more system power accordingly.The above DPM methods are not applicable to embedded real-time systems for unsatisfying the real-time requirement. Fortunately, DVS can solve the associative requirement of performance and power consumption. An optimal dynamic voltage scaling model expressed as EOVSM and a sub-optimal model named ESOVSM for sporadic task sets are discussed on the basis of theorems proving. The experimental results show that both models can satisfy the tasks' real time and reduce power significantly.For the sake of combining DPM whith DVS efficiently, an online voltage scaling policy ECVSP for mixed task sets consisting of periodic and sporadic tasks is presented on the foundation of EOVSM/ESOVSM model and task regrouping algorithm. The experimental results show that the ECVSP policy is superior in power saving to general DPM policies while satisfying the tasks' deadline.Lastly, a summary of thesis contributions is given and other directions for work on the power-reduction techniques are discussed. Simultaneously, we show prospect for the future exploration.