For an electronic device, heat removal is critical to assuring proper operation and long-term reliability. Elevated operating temperatures can introduce circuit stability problems. They also can take semiconductor junctions to the point where they break down electrochemically and fail. Over the long term, high temperatures can degrade low-power electrical components, insulation, adhesives, and other structural components. To further complicate reliability and cooling, more and more power is being forced through semiconductors. At the same time, the semiconductors are becoming more functionally dense. This trend is causing greater heat localization in smaller and smaller physical areas.

As a result of these changes, a process that was once as simple as adding a fan to move air through a system chassis has become a complex discipline. It requires expertise in design, materials science, manufacturing, and assembly. The first step in achieving effective cooling is to quantify the process. That way, subsequent steps can be taken in a deliberate fashion with some confidence in the outcome. Unfortunately, the flow of cooling air inside a cabinet is not always easy to predict. It can change considerably in response to fairly minor changes in the placement of equipment, cables, and other components.

A simplified model of enclosure cooling makes it clear that the engineer must account for all of the electrical power entering a system (FIG. 1). Any power that is not re-routed to external devices will ultimately be converted to heat within the enclosure. The dissipation of heat from the points where it is created must overcome the path's "thermal resistance." Furthermore, the final temperature of the equipment will always be some number of degrees above ambient temperature—unless pre-cooled air is forced through the system.

Thermal resistance is a function of the distance and materials through which the heat must flow. It is commonly signified with the lower-case Greek letter theta (θ). It is expressed in units of °C/Watt. A simplified equation that relates temperature to power dissipation is:

Temperature differential = system power dissipation × thermal resistance, or

T_SYSTEM − T_AMBIENT = P × θ

For example, take a system in which the net power to be dissipated is 100 W, the ambient temperature is 30°C, and θ = 0.25°C/W. Obviously, a system temperature of 55°C can be expected.

T_SYSTEM − T_AMBIENT = P × θ

T_SYSTEM = P × θ + T_AMBIENT

T_SYSTEM = 100 W × 0.25°C/W + 30°C

T_SYSTEM = 25°C + 30°C = 55°C

Note that the simplified formula treats heat as a point source. Real-world systems, in contrast, typically represent a much more complex, distributed heat source. Lower figures for θ are more desirable, as they indicate a smaller temperature rise for a given amount of energy introduced into a system. Thermal resistance can be reduced through means like increasing airflow, pre-cooling incoming air, providing adequate heatsinking, and using high-performance thermal interface materials between heat-generating devices and heatsinks.

Simple power devices, such as regulators and power transistors, begin to require external heatsinks at dissipations of around a few hundred milliwatts. But today's microprocessors and switching components are electronically denser and faster. The latest generation of microprocessors operates at a supply range of approximately 1.4 to 1.9 V and current levels of 40 to 50 amperes. They can dissipate upwards of 100 W of heat.

Though the maximum recommended microprocessor temperatures vary by model, they all range from approximately 65° to 85°C (149° to 185°F). For equipment installed in elevated temperature environments, this range does not provide much margin for cooling.

Because microprocessors are not always the most power-hungry components in a system, these figures do not represent worst-case cooling scenarios. Take the packet-switched-backplane (PSB) architecture for CompactPCI, which is described in the PICMG 2.16 specification. A CompactPCI switch-fabric system can consist of more than 20 cards. Of those cards, one or two will be switching cards. Power consumption in each of these switching slots can exceed 70 W. This high consumption will be added to the heat load produced by microprocessors and components on other circuit cards.

Finally, the cooling systems themselves usually impose an additional heat burden. Fan arrays can be placed at the intake or the exhaust side of the system. The choice is usually a tradeoff between fan reliability and additional heat stress on the system. Fans placed at the intake of a system draw cool air on themselves. But their exhaust heat is added to the heat load present in the system. In contrast, fans placed at the exhaust side of an enclosure draw hot air from the system across themselves. This may contribute to shorter bearing life.