Products for thermal management are rarely the subject of spectacular innovations—nevertheless, manufacturers continue to make incremental progress to keep pace with system-design demands. New products in thermal management formed a significant proportion of the exhibits at this year's PCIM Exhibition, held in Nuremberg in May.

The designer's first line of defence is invariably the conventional heatsink, spreading the heat from a component or board over an increased area of metal that is exposed to a convected or forced air flow. When you exceed the limits of air as a working fluid for heat removal, you can consider using a liquid cooling system. Liquid cooling has been an option for cooling electronics systems for a long time. Designers have most commonly turned to this technique in very high power systems such as traction, other motor drives and inverters. Typically, power devices such as IGBT modules. Typically, they mount to a cold plate: a coolant liquid circulates, either through pipework embedded in the cold plate, or through channels or chambers formed directly in the metal (usually—but not exclusively—aluminium) of the plate itself. Pumped circulation transfers the heated liquid to a heat exchanger of some form—a radiator if the system is losing its heat to ambient air, or possibly a liquid/liquid exchanger if the system is part of a more complex system and returns it to the cold plate in a closed cycle. There is in principle nothing about such a system that a comparison with a water-cooled internal combustion engine—or with a domestic heating system—cannot illustrate. In fact, allowing for scale, the components often look remarkably similar. Systems can be sealed and pressurised, or unpressurised with a gravity-fed filling loop (for use, obviously, with products of fixed orientation).

Designers have been less willing to turn to such methods at the circuit-board level, and with good reason: as well as adding to design complexity and bill-of-materials costs, adding liquid cooling imposes a maintenance burden on the customer—often in ways that are familiar from the use of liquid cooling in those other engineering disciplines. The most common liquid you can use is water, which may have additives such as glycol to provide protection against freezing: pure water has a higher heat capacity but is liable to corrode metals over time. Just as with those other liquid heat-transfer systems, you may require chemical inhibitors to protect against corrosion, especially if the cooling loop is made of more than one metal. Other system elements that will not be maintenance-free might include filters to protect against build-up of contamination.

Where thermal density forces water-cooling onto designers at the PC-board level, the latter most often employ it to remove heat from certain specific components that have very high dissipation in a small area—invariably, microprocessors or other large chips running at very high clock rates. Some of the more exotic heatsink/fan combinations have served the enthusiast, performance-PC builder market—and a range of water-cooling products, sometimes retailed as a means of eliminating the fan noise inevitable in an enthusiast-built (and possibly overclocked) PC, also serves that same market.

As a sidelight on that design space, in the case of the PC and the dissipation of its CPU, the worst may be over. Over recent years, the power that the successive generations of Pentium processors and their competitors dissipate has climbed inexorably towards the 90- or 100-W level. In a recent presentation outlining some of the parameters of forthcoming chips, Intel has charted a path that, to some extent, reverses that trend. Increasing the GHz clock rate is no longer seen as the way forward: the company is set on the multicore path and now cites power levels of around 65W per package as a likely dissipation. It may be worth noting, however, that if you have an application that fully utilises the capabilities of these multicore chips, Intel and its partners will be expending great efforts in the operating system to keep the two (or more) cores fully and symmetrically loaded: you may find little difference between peak and average power.

Aavid Thermalloy has produced a kit-based approach to fitting a liquid-cooling loop to a product design, under the HydroSink name. This kit contains a pump to circulate the coolant, a heat exchanger, pipework and fittings, and a reservoir to maintain the level of coolant in the system. Aavid matches the kit with a cold plate that the user will specify according to his task. This plate may vary from a spot-cooling plate that replaces the conventional heatsink on a large IC package to a larger rectangular plate designed to carry, for instance, several IGBT modules. The company makes plates in a variety of formats, including its Blister technology—the technique bonds an aluminium plate to a stamped aluminium sheet that carries the cooling channels—and its HiContact technology, which embeds a copper plate within the aluminium. The base kit includes a heat exchanger (fan and radiator) measuring 510×390×200 mm: this supports heat dissipation up to 3.5 kW to a 30°C ambient.

An alternative means of removing heat from a specific source point and transporting it to ambient is the heat pipe, again a relatively mature technology. A heatpipe is a sealed tube containing a working fluid that at ambient is close to its boiling point. Heat applied at one end (by contact with the hot component) will vaporise the fluid, the vapour will travel to the cooler region of the pipe (attached to a cold wall or other means of disposing of heat) and give up the transported heat as it condenses. The bi-phase system, using the latent heat of vaporisation and condensation, carries much more heat than is possible with a single-phase (liquid) medium. Heat pipes can contain a capillary medium that acts as a return path for the liquid phase of the working fluid, or they may rely on gravity for that effect, requiring the "cold" end of the system always to be above the heat source. An increasing trend is for manufacturers to embed heatpipes within the base plates of otherwise conventional heatsinks, once again to cater for components that generate a great deal of heat over a very small area. These embedded heatpipes lower the transverse thermal resistance across the base plate, making it very nearly isothermal. With uniform heatsink temperature the sink is more effective, giving a lower overall thermal resistance from the relevant component to ambient. Vendors including—once again—Aavid Thermalloy presented examples of new designs employing this construction; design-consulting company CRS Engineering can provide custom thermal solutions involving heatpipe technology.

Extruded aluminium heatsinks are a well-established technology, but here too manufacturers continue to make incremental improvements. At PCIM 2006, Pinbloc demonstrated a heatsink pressed from a single aluminium block, and taking the form of an array of pins or posts of circular cross-section (Figure 1). One side of the aluminium ingot that is the starting point for the device forms the base of the sink, and the pins are formed in a single pressing operation—the company carefully distinguishes its manufacturing process from extrusion. The material is pure aluminium, with a high heat conductivity of 220 W/mK. Pinbloc explains that the key to the performance of the design is the precise form, dimensions and spacing of the cylindrical pins. When an airflow blows through the forest of pins, each one gives rise to or "sheds" (as any object placed in a fluid stream will tend to do) vortices that propagate downstream in the flow. The proportions of its design, a spokesman for Pinbloc explains, results in the vortex shedding giving rise to almost completely turbulent flow around the pins, with no flow shadows or stagnant regions. This yields optimal heat transfer from the pins to the airflow, and the crystalline structure of the metal that forms the pins in the pressing process, the company says, improves thermal conductivity. Pinbloc says its design is also very effective in convective cooling, as well as in the forced-air configuration demonstrated at PCIM.

The company claims that independent tests have measured thermal performance 30% better than extruded heatsinks of comparable dimension, and 40% better than cast structures. Pinbloc can use its process to make heatsinks with base plates of differing shapes, and can also supply a composite copper/aluminium base layer to improve heat distribution.

Alutronic is employing manufacturing technology that explicitly is extrusion: the company was demonstrating heatsinks with tall (high aspect ratio, up to 15:1 height:width) fins (Figure 2). Alutronic says that using new designs of manufacturing tools it can produce extruded finned heatsinks in formats that manufacturers have previously achieved by assembling fins to a separate base plate. Without that thermal discontinuity between base plate and fin, the design achieves a lower thermal resistance.

If you are designing a PC board that does not host individual components with very high dissipations, you may have been able to rely on heat absorption into the board itself—either via thermal contact pads on the underside of surface-mount devices, or simply by heat conduction through contact leads or pads. With ever-increasing component densities, designers are finding more and more often that this option is insufficient. While it may be feasible to conduct the heat away in this fashion, the steady-state average board temperature is unacceptably high.

Companies with a materials focus are constantly introducing products that conduct heat away from the top surfaces of the components, either via thermal pads provided for heat-sink contact, or simply through the component's encapsulation, to reduce the amount of heat that reaches the board's core. One such company is Bergquist: at PCIM 2006 it introduced a new gap-filling material, Gap Pad 5000S35, which it says is the softest material available that offers a thermal conductivity of 5W/mK. The material can therefore conform to a range of component heights, transferring those components' heat to a heatsink or a case wall, while accommodating assembly tolerances (Figure 3). From the same vendor comes a solution for designers who have no choice but to absorb heat in the plane of the printed board. Intended for use in designs such as dc/dc converters, IsoEdge heat plates are coated with a dielectric material that yields 2,250V dc isolation, which allows them to be placed in close proximity to heat-dissipating components without separate insulation spacers (Figure 4). Their thermal resistance is 0.3 to 0.4 W/mK and, the company says, designers can use both new products together to achieve optimum heat spreading on small assemblies.