TECHNOLOGY IN CONTEXT
Developments in VME
Air and Conduction Cooling for 3U COTS Cards: An Overview
Thermal knowledge and innovation continue to improve cooling limits for air-cooled and conduction-cooled cards, and this benefits 3U cards greatly due to their shorter width and lower power.
BY IVAN STRAZNICKY, CURTISS-WRIGHT CONTROLS EMBEDDED COMPUTING
Page 1 of 1
There are many trends at the die level and circuit card level that can drive the decision of which cooling method to use for 3U COTS cards in rugged applications. One of these trends is the increased use of multicore processors on 3U cards. Placing more processor cores on a die increases power dissipation. However, as more cores are placed on a die, the size of the die increases, which actually decreases the power density in terms of W/cm2 (Figure 1). This is a good thing from a thermal standpoint because it reduces the spreading resistance down the heat removal path. Unfortunately, increasing device power dissipation, combined with the long-term trend of decreasing junction temperatures (i.e., from 125°C to 105°C, or 100° or less), tends to override the small benefit obtained from the decrease in power density.
Rising power dissipation and peaking heat density on successive processor generations.
One of the main causes of increased power dissipation for processors is the trend toward increasingly smaller transistor geometries, which results in large increases in static (or leakage) power. Some modern processors use new types of transistor materials that reduce the amount of static power. For example, Intel uses hafnium dioxide on its 45nm processors. These new transistor materials reduce the amount of gate oxide tunneling and sub-threshold leakage current, two of the dominant forms of leakage current.
At the board level, component miniaturization and the use of more highly integrated devices are increasing the functional density on 3U cards, which is directly related to heat density. As the amount of processing power per square inch per 3U card has increased, it has driven up heat density. Another factor driving up allowable power dissipations on 3U cards is the support for higher voltages in the new VPX (VITA 46) standards. VITA 46 defines support for 12V and even 48V, compared to the standard, traditional 3.3 and 5V supported by VME. VPX also supplements the traditional 0.8” pitch of VME with 0.85” and 1.0” pitches. This increase in pitch enables the use of more, hotter devices on the rear side of the circuit card, increasing the power dissipation per unit area and volume. The outcome is an almost exponential increase of power at the circuit card level. Susceptibility to this trend depends on the card’s functionality. The higher power cards are typically DSP cards that have multiple multicore processors on board for number crunching. In comparison to DSP cards, general-purpose processor and I/O cards typically follow more of a flat curve in terms of power.
While direct air and conduction cooling have been able to keep up with these power increases to date, it has been a challenge. The amount of thermal design, analysis and testing that is required on a rugged military COTS 3U card is many times what it was five years ago. This increased work is basically the result of the increase in power dissipation.
Direct Forced Air Cooling
Direct forced air cooling is typically the starting point in terms of cooling approaches for military COTS cards simply because most software and system development begins in a laboratory environment with air-cooled cards in a benchtop rack. Consequently, these cards are usually commercial temperature rated and not rugged.
Cooling begins at the device die and on most modern, high-power devices, the die is exposed. This is because most commercial cooling approaches employ an air-cooled heatsink on top of the die, which provides the shortest and lowest resistance heat removal path to the air. While forced air cooling takes advantage of this arrangement, it may not always be desirable to have a large piece of metal on the die. For this reason a heat spreader is sometimes placed between the heatsink and the die to spread the heat and to provide some protection for the die. Unfortunately, for today’s higher power devices, standard off-the-shelf aluminum heatsinks, with a few fins per inch, may be insufficient. To address these hotter devices the heatsink will likely need to be optimized. System designers can use computation fluid dynamics (CFD) tools to design the heatsink they require for air cooling. There are subroutines available within CFD tools to design the heatsink’s optimum number of fins per inch, thickness of fins, optimum gap and height, etc. Another trend in air cooling today is the availability of higher flow and pressure fans and blowers, which provide increased airflow for such high pressure drop heat sinks.
These new approaches coupled with advances in air cooling have resulted in significantly improved cooling numbers. For example, today, a 150W 6U card can be sufficiently cooled with a 71°C air inlet. Just five years ago, many designers would have been hard pressed to believe this was achievable. For 3U cards, air cooling is even more attractive because there are far fewer hot components on the card that need to be cooled by the same air stream (by virtue of the 3U geometry, i.e. 100 mm in the airflow direction vs. 233 mm).
There are, however, drawbacks for air cooling. Air-cooled cards are typically not very stiff compared to conduction-cooled cards. Unlike conduction-cooled cards, they lack a metal stiffener plate on top to increase the stiffness. When air-cooled cards are subjected to significant vibration, they can displace quite a bit and end up experiencing fatigue problems more quickly than on stiffened cards. With that said, this is less of a problem for air-cooled 3U cards than for 6U cards, because they have a much shorter span between guide rails (Figure 2).
Photo of 3U air-cooled product
Another drawback for air cooling stems from the use of non-sealed chassis to enable air to be blown directly over the electronics. Because the chassis isn’t sealed, the cards and electronics may be exposed to contaminants, such as salt fog/sand/dust, etc., in the ambient environment. Filtering the air will remove some of these contaminants, but will come at the cost of an increase in pressure drop due to the filter.
One more drawback associated with air cooling that is worth noting is that because air is compressible and has low heat capacity and low thermal conductivity, it is not a very good coolant. Because of these properties, an increase in airflow rate over a card can result in an asymptotic curve with respect to cooling vs. airflow. At the same time, the pressure drop is increasing steadily so the cooling limit is reached relatively quickly with airflow cooling.
Conduction cooling is a popular choice and used quite widely for deployed 3U systems because it is inherently more rugged (Figure 3). The “backbone” of conduction cooling is the stiffening and thermal frame, typically made of aluminum. Today’s increased power dissipation levels are leading to increased use of copper as well.
Photo of 3U conduction-cooled product
Because of copper’s high density it is used only where needed, for heat spreading for example. Composite materials are also showing some promise, not necessarily as a replacement for the thermal frames, but localized in areas of the thermal frames. However, most composites are orthotropic with regards to thermal conductivity, with the in-plane thermal conductivity being quite high. For example, the claimed thermal conductivity for pyrolytic graphite is very good—about 1500 watt per meter per degree Kelvin (W/m°K) in either in-plane direction. Compared to aluminum, which is 180 W/m°K, pyrolytic graphite should provide performance almost an order of magnitude better. However, the through thickness thermal conductivity for pyrolytic graphite is only around 20 W/m°K, and needs to be taken into account. If large amounts of heat are being moved across large planes or long distances, a composite can be a good solution. But trying to move heat through the thickness is more difficult. There are several research groups working on developing materials that have isotropic conductivity better than copper (which is 400 W/m°K), have the density of aluminum or lighter, and can be produced at reasonable cost. However, because they are highly engineered, cost often ends up being an issue.
Another option for increasing the thermal conductivity of thermal frames on 3U cards is to use phase change devices such as heat pipes or vapor chambers. Curtiss-Wright has undertaken substantial research and development with heat pipes and has successfully implemented them in rugged products. Heat pipes are very effective at moving heat with a very low temperature drop. A drawback of heat pipes though, is that that they are orientation dependent, which makes it critical to understand how they behave under the effects of body forces such as acceleration, vibration and gravity. The various performance limitations of heat pipes, such as capillary, entrainment and condenser limits, also must be understood to implement them properly. When properly implemented, heat pipes can handle fairly high power densities and power dissipations. A heat pipe implemented in the axial direction provides effective conductivity in the thousands of W/m°K, which is an order of magnitude higher than the metals used for cooling.
Advances in conduction-cooled cards now make it possible to cool in excess of 170W at an 85°C card edge temperature, the upper limit of what is seen with military COTS circuit cards. A standard chassis with air-cooled sidewalls, though, is not capable of cooling even one of these cards. Air cooling through the side of a chassis wall, with conduction cards slotted between rails, is just not capable of cooling the level of power density at the circuit card edge. In such a case, other approaches, such as a liquid-cooled chassis, in which liquid is flowed through the sidewalls, must be employed.
Thermal resistances in the heat path of a conduction card have improved greatly, for example thermal interface materials, over the last decade. Ten years ago the best thermal interface material you could find was in the order of 1-3 W/m°K. Now there are several in the order of 10-15 W/m°K, which is basically an order of magnitude improvement. However, you do have to be careful with some of these materials. When tested, some that have been advertised as having a certain thermal conductivity actually exhibited only 1/10th of that value. Had that material been designed into products without testing, thermal failures may have resulted. Because these materials have been designed for use in commercial applications, they also need to be tested for rugged properties such as long-term thermal cycling, exposure to humidity, etc., to ensure that they can be used in rugged applications.
The perceived “Achilles’ heel” of conduction cooling is the thermal contact resistance between the conduction card edge and the chassis rail that results from metal-to-metal contact. These metal surfaces are not truly flat and smooth at the microscopic level. Metal peaks, called asperities, contact each other and create low resistance paths through which much of the heat flows. The heat flowing across this junction is a surprisingly complex phenomenon. The thermal contact resistance for a given pair of surfaces depends on a complicated combination of variables including contact area, contact pressure, surface flatness, surface roughness and hardness.
Typical values of thermal contact resistance used for VME or VPX cards are in the range of 0.3-0.5°C per watt. So for a 100W card, with 50W going to either card edge, the 50W is multiplied by the .5°C per watt resulting in a 25°C temperature difference across that interface. This is huge in terms of the temperature budget between the coolant and the electronics. For example, the temperature budget for cooling a part that has 100°C maximum junction temperature with 55°C ambient air is only 45°C. By investigating and optimizing the contributing factors to thermal contact resistance, much lower values in the range of 0.1-0.2°C/W can be achieved. Note that these values are at sea level and they will change (increase) at altitude. With these numbers and our example of a 100W card, a temperature difference of only 5 to 10 degrees can be obtained. This is still significant and must be accounted for in system level thermal analyses, but 5-10°C is certainly more appealing to system integrators than 25°C. These improvements should be sufficient for 3U cards for the foreseeable future, making the 3U form factor a good choice for small form factor (SFF) electronics.
Curtiss-Wright Controls Embedded Computing