Thermal modeling and control in mobile and server systems

Mohammad Javad Dousti · University of Southern California Digital Library · 2015

This dissertation deals with thermal modeling and control issues in two types of systems: servers and mobile devices. For server systems, thermoelectric coolers (TECs) are considered as the cooling solution. Despite their unique benefits, TECs generate heat during their operation due to the Joule heating effect. This reduces their cooling efficiency and necessitates careful design and control in order to enable their effective utilization. In this dissertation, three key issues are identified and addressed. First, it is noted that the traditional definition of TEC coefficient of performance (COP) is not useful for electronic cooling packages due to strong dependence of the circuit leakage power on the die temperature. Hence, the COP is redefined to consider the effect of leakage power. Second, it is found that in a cooling package comprised of a fan and TECs, the TECs driving current and fan rotation speed should be properly set to avoid power losses. Accordingly, two optimization problems are set up and solved: One that tries to minimize the maximum die temperature and another that aims to minimize the cooling power consumption subject to the temperature constraint. Last, it is observed that hot spots are spatially and temporally distributed on the surface of a chip. As a result, conventional control of all TECs as a single unit is too coarse-grained and tends to result in power inefficiencies. To address this issue, TECs are divided into a set of clusters, where each cluster is instrumented with a bypass switch. The presence of these switches enables a controller to selectively turn on and off each TEC cluster, thereby, significantly enhancing the power efficiency of the TEC-based cooling solution. For mobile devices, a tool called ThermTap is introduced to enable system and software developers find power and thermal bugs in the design. ThermTap comprises of a power analyzer called PowerTap and an online thermal simulator called Therminator 2. Equipped with accurate power macro-models and utilizing operating system kernel device drivers, PowerTap collects the activity profiles of major components of a portable device in an event-driven manner, which are in turn analyzed to produce power dissipation profiles (i.e., power traces) for these components. Therminator 2 subsequently reads these traces and, using a compact thermal model of the device, generates various temperature maps including those for the device skin and the aforesaid components. Accurate per-process and per-application temperature maps produced by ThermTap enable software and system developers to find thermal bugs in their software.

Read the paper · More papers on PaperTik