Transferring component and circuit heat from its various sources to that mystical, magical, cooler place called “away” in order to keep a board or assembly cool is an ongoing battle. Just when you think there’s no more thermal headroom, demands increase and somehow new cooling needs must be met with innovative approaches.
How much so? Numbers vary, but a study by Uptime Institute, LLC (an “IT service management company”) maintains that per-rack density in 2020 was about 8.4 kilowatts (kW), up from just 2.4 kW in 2011 (Figure 1), with 46% of the racks dissipating between 5 and 9 kW (Figure 2); those numbers are reasonably in line with others I have seen. I was curious and did some informal historical research, and it appears that the typical rack’s thermal load in the bad old days of those power-hungry vacuum tubes was between one to three kilowatts, for comparison.
Figure 1 Average rack thermal dissipation has more than tripled in the past ten years. Source: Uptime Institute
Figure 2 Just under half the rack installations showed dissipation of between five and nine kilowatts. Source: Uptime Institute
There’s some irony in these numbers. When vacuum tubes were displaced by solid-state devices (first transistors, then ICs) it seemed like power problems and dissipation issues would be greatly reduced—and they were, as long as you were replacing an existing function, such as a radio receiver. However, as ICs got smaller and lower power, the sophisticated systems they went into were expected to dissipate power at an even higher rate.
In effect, the demands placed on systems and the components which implement them grew faster than the rate of power decrease. You might call it yet another manifestation of the law of unintended or unforeseen consequences. As a result of this increased dissipation, basic assembles such as embedded-system card cages are now being severely stressed thermally.
There are two aspects to removing the heat; of course, they are closely related, and one can’t be accomplished without the other:
- micro or localized cooling, which means keeping individual excessively hot components—usually processors or power devices—cool enough via heats sinks, heat pipes, and more.
- the broader macrocooling issue of getting that aggregate dissipated heat away from the board and enclosure.
How much heat can a system handle? There are many guidelines out there, as the answer depends on multiple factors. One metric which popped up several times during my research as a starting reference point was 1 W/in2 gold about 0.15 W/cm2 for maximum board dissipation, which is far higher than it was just a few years ago.
While every design has its own issues, many systems are based on multiple PC-board cards slipped into a card cage or chassis. Early card cages and chassis were just that: a simple metal frame with nylon rails (guides) into which the PC boards could slide, Figure 3. In many cases, the boards had nylon ejectors at the two visible corners to facilitate pulling the board out from its backplane connector and from the cage (Figure 4). Cooling would be via forced or unforced convection, with little or no conduction cooling through the cage structure itself.
Figure 3 The early PC-board card cage was just that: a holder for a parallel set of cards, usually with a common backplane for card-to-card interconnection. Source: Andy’s Arcade
Figure 4 Cards often had nylon ejectors at their front corners to provide a handle and add leverage when extracting the card from the backplane connector and cage. Source: Vector Electronics & Technology, Inc.
However, that basic cage or chassis adds nothing to the cooling solution but just hold boards in place, and has largely been supplanted by cages and chassis where the enclosure and boards form an intimate cooling relationship. To facilitate this, the PC board has copper thermal paths along its edges (Figure 5), and these edges, in turn, are fitted with wedgelocks which contact the cage guide rails (Figure 6).
Figure 5 Copper strips along the two side edges of the PC board are an important first step in providing a low-resistance thermal path off board (here, on a prototyping board). Source: PJRC Electronic Projects
Figure 6 The edges of the boards are then fitted with wedgelocks which contact the cage guide rails. Source: Advanced Cooling Technologies
These wedgelocks along the edges of the boards are engaged with a twist of an internal screw, tightening them against the guide rails so there is solid physical contact between the card edge and the metal rail. This provides a low-impedance thermal path from the card to the cage or chassis and then to “away.”
As entire cage or chassis is a part of the heat-sinking thermal path, that physical contact between the board-edge wedge and the guide rail is critical. You might think that this wedge action is about as good as it can get for low thermal impedance, but it’s not. Cooling experts are always looking for ways to eke out modest yet meaningful gains in the thermal situation.
Recently, thermal-management device vendor Advanced Cooling Technologies introduced an enhanced card-edge wedgelock which they call “Ice-Lok” (Picture 7). These wedgelocks target embedded computing applications, are VITA 48.2 compliant, and compatible with standard VITA 3U, 6U and 9U cards.
Picture 7 Like standard wedgelocks, the “Ice-Lok” wedgelocks are VITA 48.2 compliant. Source: Advanced Cooling Technologies
Instead of just providing up-and-down thermal paths from the card edge, through the wedgelock, and to the cage/chassis guide rails, these proprietary wedgelocks also offer an outward path. The wedges are cut at a bias, so that the wedge can expand in two directions to allow the ICE-Lok to contact all four sides of the thermal interface, thus increasing the surface area for heat transfer and maximizing contact pressure (Picture 8).
Picture 8 The “Ice-Lok” wedgelocks offer additional low-resistance thermal paths from board edge to rail guides. Source: Advanced Cooling Technologies
You might not think that this modest amount of additional contact area would make a big difference in the overall thermal situation, but ACT tests show that it reduces thermal resistance by more than one-third compared to a standard wedgelock (Figure 9), so component temperatures are lowered by 10°C at 100 W power input.
Figure 9 By adding the additional thermal path, thermal resistance is reduced by about one-third and temperature at the board is reduced, or higher dissipation is possible at the same temperature limit. Source: Advanced Cooling Technologies
Once again, as with most advances, there are big-step improvements and smaller-step ones. In some cases, the accumulation of those smaller ones adds up to a significant overall improvement. That’s been the situation with thermal considerations as modest gains due to better IC lead frames, integral thermal pads under the IC, enhanced device packaging, additional heat-flow paths on the PC board, and lowering of thermal resistance to the “outside” add up . Even without active cooling via fans or liquid-based approaches are used, there passive enhancements which enable significantly higher thermal loads than was possible even a few years ago.
- Uptime Institute, “Rack Density is Rising”
- Advanced Cooling Technologies, “ICE-Lok® Performance Data”
- Advanced Cooling Technologies, “Ice-Lok – Thermally Enhanced Wedgelocks”
- Embedded Computing Design, “Characterizing Thermal Performance Through Card Retainers”
- Curtiss-Wright, “The VPX Ecosystem”
- Curtiss-Wright, “VITA 48.8 Air Flow Through Cooling Standard Lowers SWaP-C on Deployed VPX Systems”
- VITA Technologies, “VITA 48 and 50 cooling standards”
- LinkedIn, “A New trend in military electronics cooling-Air Flow Through and a case study”
- Mobility Engineering, “Maximizing Thermal Cooling Efficiencies in High-Performance Processors”
- LCR Embedded Systems, “Stretching the Limits of VITA 48.2 Chassis Design”
Bill Schweber is an EE who has written three textbooks, hundreds of technical articles, opinion columns, and product features.