Temperature-Power Consumption Relationship and Hot-Spot Migration for FPGA-Based Systems

Introduction And Background

Dealing with power consumption efficiency in modern telecommunication became a major topic leading to a new field referred to as Green Communication. Green Communication (GC) is a set of concepts and technologies that aims at mitigating the impact of communication devices on the environment. As a matter of fact, currently, 3% of the word-wide energy is consumed by the ICT (Information and Communications Technologies) infrastructure that causes about 2% of the word-wide CO2 emissions (which is comparable to the world-wide CO2 emissions from airplanes or one quarter of the word-wide CO2 emissions from cars)(IN-EN.COM, 2010).

Recently, Cognitive Radio (CR), supported by software defined platforms, has been proposed as a GC technology enabler (Jacques, 2009; Honggang, 2008). In a nutshell, CR equipment is a communication system aware of its environment as well as of its operational abilities. Thus, it is a device that has the ability to collect information through it sensors and that can use the past observations on its surrounding environment to improve its behaviour consequently. Thermal and power consumption properties are counted, as crucial information CR equipment has to deal with.

To limit the impact of ICT infrastructures on the environment, one of the main issues GC is dealing with concerns power consumption in CR equipment. Related to this matter, heat emission and temperature control on an electronic device are known to be highly correlated to power consumption as well as to the equipment’s reliability. Thus, tackling such matter is essential during designing process since power losses can occur in the form of heat due to manufacturing imperfections. Moreover, equipment overheating often leads to errors in the functionality of the chip or unexpected results because of the increase in temperature. Indeed, leakage current in CMOS based FPGA increases with temperature since a positive feedback loop exists between leakage power and temperature (Heo, 2003; Velusamy, et al., 2005).

When dealing with equipment heat limitation, one can think, mainly, of three kinds of solutions: Firstly, the most common solution is to increase the power of cooling systems, which reduces, efficiently, circuit temperature leading, unfortunately, to added power requirements. A second solution would probably be to reduce computing power by changing the functional frequency or supported voltage and thus control the heat production. In that case, the result of heat reduction is obtained by the victimization of performance. Finally, overheated point migration is a third technique by which the functional module can be migrated to a lower temperature area. Such overheated point in a chip is known as “hot-spots.” Power losses can occur in the form of heat because of manufacturing imperfections and more than often lead to errors in the functionality of the chip or unexpected results due to the increase in temperature. Overheating is becoming a serious concern as more densely packed transistors within the chips are generating more heat per unit area. In multi-processor systems, this last solution is widely applied to limit heat increase by migrating software tasks from one processor to another. This solution can be similarly, immigrated on FPGA-based System-on-Chip (SoC) platforms.