초록/요약

In mobile devices, using multicore and high operational frequency has recently caused severe power consumption and heat generation. Meanwhile, since the way to reduce the heat is limited, thermal mitigation is being conducted by software. This dissertation proposes a thermal mitigation algorithm (TMA) for a central processing unit (CPU), applying a power curve of a CPU to increase average operational frequency. A conventional TMA of a CPU conducts thermal mitigation using only current temperature information of a CPU. However, the heat generation of a CPU shows various characteristics depending on the CPU. Therefore, a study of a TMA for a target system is proposed in this dissertation, applying the power curve that is characteristic of the specific semiconductor chip that causes heat. A TMA of a CPU is software that dynamically adjusts the CPU’s maximum allowed frequency to mitigate the heat when it occurs. To apply the CPU power curve to conventional TMAs, power consumption and Dhrystone million instructions per second (DMIPS) of a target CPU were measured. The result of the measurement showed that power consumption compared to the DMIPS increases rapidly in the high operational frequency range. It means that using the high operational frequency range for high performance causes rapid heat generation, and a number of maximum allowed frequency levels have to be lowered by TMA. To analyze the total influence of these effects on the average operational frequency, the analysis of the thermal and performance variation was conducted by designing a TMA that limits the usage of the maximum operating frequency, in which power consumption compared to performance is most inefficient. A Qualcomm Snapdragon 801 was used as the target AP for this experiment. An AnTuTu Benchmark, version 4.5.1, was run for performance measurement, and a temperature sensor of a crystal oscillator (XO) was used for temperature measurement, which is embedded in AP. When comparing two cases of applying the conventional TMA and proposed TMA for these experiment environments, average temperature was reduced by 1.5̊C and average performance was enhanced by 2.8%. In addition, the average value of operational frequencies was enhanced by 52MHz. This results shows that the average operational frequency is rather decreased by severe heat generation even when the maximum operational frequency is used. As a result, the proposed TMA reduces the thermal effect and increase the average frequency and system performance by reducing the most inefficient power consumption, which occurs when the maximum frequency is used. This study is significant in presenting a new direction of study to improve the conventional TMA that uses only the current temperature information of a CPU.