Cooling Methodologies—The Tail Wagging the Dog



Discovering how to maximize cooling at the lowest cost and highest efficiency.

Complex military and industrial applications require complex computing solutions. The end user, or committee of end users, writes a requirements list, pretty much shooting for the moon. The specification is released to the programming group, and they define the processing platform required to execute the new code. This platform might include two, or more, 8, 10 or 12-core processors, a ton of RAM, one, or more, high performance GPUs, video processing, multiple high-performance hard drives and user I/O boards plus the power supply. The environment and physical envelope will have been defined. However, the last thing the people in this process think about is how to cool this power-hungry beast. How is the heat pulled out of the system, and where is it going to be sent? The code may be developed not on the end-use processing hardware, but instead in a large desktop enclosure sitting in a cool office setting. When deployed to the field, heat issues ultimately limit performance and reliability.

Figure 1: Inside a Heat Pipe

Figure 1: Inside a Heat Pipe

Heat is the enemy of modern electronic systems. Integrated circuits, especially processors, can generate significant heat that must be removed to keep the system operational.

At the atomic level, heat is expressed as rapidly vibrating atoms (and molecules). The higher the temperature, the faster the atoms vibrate. Heat conduction occurs when hot, or rapidly vibrating, atoms interact with neighboring atoms and transfer some of their vibration to these neighboring atoms, which transfers the heat energy. This energy transfer continues to happen until the solid is at equilibrium with the same temperature across the entire material. The goal of cooling is to take the heat off the solid as fast as it is being introduced by the electronics.

Nature provides five mechanisms to move heat: conduction, convection, advection, radiation and phase change of a material. Ultimately the heat needs to be dumped into a larger sink such as the atmosphere or the ocean in the case of a submarine. Objects in space dump waste heat to space through radiation. The goal is to move the heat from the source to the sink in the most efficient and cost-effective manner.

Heat transfer is the exchange of thermal energy from an area of higher energy to an area of lower energy, or hot to cold. The First Law of Thermodynamics says the total energy in a sealed system is constant. Therefore, energy into a system in the form of electricity has to be removed from that system as heat, or the total energy, as expressed by system temperature, will continue to increase. The Second Law of Thermodynamics governs the direction, hot to cold, to maximize system entropy.

Radiation is not an effective mechanism to cool electronics and will not be discussed here. The opposite is usually the case where electronic systems are exposed to sunlight or are located adjacent to a high temperature component that is radiating infrared, and the systems become heated as a result. Testing for the effects of solar radiation is covered in MIL-STD-810G, Method 505.5.

Conduction is the transfer of heat from a hot area to a colder area through a solid. Cooling electronics always starts with conduction. The circuits inside an integrated circuit transfer their heat to the surrounding case matrix via conduction. Conduction can include transfer of heat from one material to another, not necessarily through the same body. For example, a processor conducts heat through thermal compound to an attached heat sink made of aluminum or copper, all different materials.

Convection is the transfer of heat from a solid surface to a fluid, which includes air or other gas. Convection includes natural convection and forced convection. Natural convection uses gravity to move the fluid. As the fluid heats, it expands, becomes less dense and rises, carrying the heat from the hot object. Cooler fluid flows in to continue the circulation. Forced convection uses forced movement of the fluid to increase the flow over the item being cooled. A typical heat sink on a processor in a computer includes a fan to force air through the fins, in effect blowing the heat off the heat sink.

Advection is the use of a moving fluid to carry heat from one location to another. The cooling system in a car is a classic example of advection, using coolant to carry waste heat from the engine to the radiator where it conducts the heat to the cooler radiator fins and thence to the air before recycling back to the engine. Technically, convection includes the sum of transport by both diffusion and advection.

All systems use a combination of conduction, convection and advection. The trick is to find the best combination of all the available physical mechanisms to maximize cooling at the lowest cost and highest efficiency.

Phase Change
One of the effects of changing a material from one phase to another is it requires heat to effect that change. This can be solid to liquid or, more common to electronic cooling, liquid to gas. When the gas is again condensed into a liquid, it gives up that heat. That is why you feel cool when sweating and a breeze evaporates the sweat. The liquid sweat is evaporated into a gas, water vapor, which chills your skin.

Heat pipes are an example of a phase change cooling system. A liquid is converted to the gas phase inside the sealed pipe at the heat source, the gas travels to a cooler heat exchanger or condenser where it gives up the heat and transitions back to a liquid. The liquid flows back to the heat source, usually through capillary action, and the cycle continues. Water is commonly used for heat pipes for room temperature applications such as computer processor cooling. Heat pipes are much more effective at moving heat than a solid block of copper. The energy needed to evaporate water is 540 times the energy required to raise the same weight of water 1°C and that heat can be rapidly moved to the heat exchanger.

The effective thermal conductivity of a well-designed heat pipe solution can be from 10 to 10,000 times the thermal conductivity of copper. Heat pipes have an extremely effective high thermal conductivity. While solid conductors such as aluminum, copper, graphite and diamond have thermal conductivities ranging from 250 W/m•K to 1,500 W/m•K, heat pipes have effective thermal conductivities that range from 5,000 W/m•K to 200,000 W/m•K.

Thermal Resistance
Cooling system performance can be modelled similar to an electrical circuit where different materials and interfaces between materials in the circuit offer resistance to the heat flow. Thermal resistance is expressed in degrees Centigrade per watt (°C/W). A heat sink rated at 1°C/W means the item connected to the heat sink will increase 1°C per watt of heat.

It is important to remember that the total energy in a system should be constant. Electricity in, measured in watts, is equal to the heat out, also measured in watts. There will be inefficiency in the power supply, which generates some waste heat the power supply has to dissipate, but the rest of the power goes into the electronics, which comes back as heat. The lower the thermal resistance in the circuit, the lower the IC temperature, all other considerations being equal.

Different materials offer different thermal conductivity properties (Table 1). This can be referenced as BTU/(hr ft°F) or W/m•K.

Table 1

Table 1

A solid diamond heat sink would be the most effective method to move heat off an IC, but the cost would probably not fit the budget. Copper, on the other hand, has half the resistance of aluminum, the other common heat sink material, though copper is more expensive and heavier than aluminum. Higher performance heat sinks will use a copper slug in contact with the IC to distribute the heat into a larger aluminum finned part to realize the benefits of copper and the lower cost and weight of aluminum.

Initial System Design
During initial system design, the total system requirements need to be examined and specified. There is always a tradeoff between required (desired) processing horsepower and available cooling capacity. Tradeoffs include:

• Budget
• Available rack space or allocated space for the electronic system
• Required processors, memory, GPU cards, etc.
• Required video performance
• Allowable noise
• Environmental factors including ambient temperature, shock and vibration, etc.
• Available sink for waste heat

These factors very often contradict, forcing compromise in system features. The usual design process is the user defines the system application, writes the software and finds a desktop computer to do initial system debug. Everything runs fine in the lab. They then go to industry to create a rugged solution with a defined physical spec and environmental requirements. Unfortunately, the thermal requirements of the system components won’t fit within the allowable enclosure dimensions and environment.

Thus it is often a better approach to determine the allowable system heat dissipation characteristics and work from that to find an appropriate motherboard/processor solution. Otherwise, the system may overheat in the field if the programmer’s wish list is realized.

A general rule of thumb is a 10°C increase in system temperature halves the MTBF of the system. It is therefore critical to keep system and component temperature as low as possible.

Available rack space is a key factor in cooling capacity, especially for air-cooled system. A 1U system is only 1.75-inches tall allowing the use of 40mm fans. It does not matter how many fans you can install, there will be a limited amount of cooling air flow available. A 4U system, on the other hand, allows the use of 120mm fans, which can move significantly more air at a higher pressure for more flow.

More processing power generally requires more watts to operate. Newer chip architectures can provide higher performance for lower power requirements but that overhead is soon used up with new system development. Processors now offer multiple cores and motherboards can support 2 and even 4 processors. Each processor will support ever-larger RAM sticks, which use more power. It is a never-ending cycle.

Noise can be a huge issue. Moving significant amounts of air with a fan will generate noise, especially for smaller 1U systems using 40mm fans. A Sunon 40x40x56mm fan providing up to 31.7CFM will generate 66dB of noise. That is for one fan. In addition, the fan is running at 18,000 to 21,500RPM so the generated noise is very high pitched. A chassis will have 5 or 6 of these fans, which multiplies the noise. It becomes a health hazard to be in close proximity to these chassis for extended periods of time, especially if there are multiple systems mounted nearby. These little fans also can draw 16W or 100W for 6 of them.

On the other hand, a 4U enclosure will accommodate 2 or 3 120mm fans. A Sunon 120x120x38mm fan can provide 190CFM. That is six times the air flow per fan in free air versus the 40mm fan. Noise is 54dB each. Still pretty loud for the maximum performance fan. Noise can drop to 37dB for 93CFM.

Noise is measured in dB where an increase in 10dB doubles the perceived noise. A single 66dB fan will be more than twice as loud as a single 54dB fan. Six fans will be much louder than two fans.

A 1U chassis will be challenged regarding air inlets and exhausts. Any drives mounted in the front panel reduces that area that can be used for air inlets. If the motherboard is mounted with the I/O panel in the rear panel and should a plug-in card be installed, there will be very little space left in the rear panel for exhaust ports.

A 4U chassis will offer much more open slots for cooling requiring a much lower pressure to move more air through the system plus much larger fans can be installed.

Another way to lower fan noise is to use an intelligent fan controller such as Chassis Plans’ Syscool, which will measure system temperature and adjust fan speed accordingly. Fan noise drops significantly with lower fan speed. Other benefits include longer fan life and lower filter maintenance.

Environmental factors are an obvious concern. Higher ambient temperatures require more, maybe significantly more, cooling. Higher altitude lowers the cooling system effectiveness. Shock and vibration can impact large, heavy heat sinks.

And where the waste heat is eventually exhausted can be an issue. The F35 has been reported as having electronic cooling issues because the aircraft uses the fuel supply as a heat dump. The problem is the fuel in the fuel trucks heats up while the trucks are parked in the sun. The hot fuel is loaded into the F35 so there is nowhere for the waste heat from the copious electronics which make up the F35 to be sent.

Dense electronics bays can overheat if heat disposal is not considered early in the design.

Heat Management Devices
There is a multitude of methods to remove heat from an electronic enclosure. All of these methods are interdependent. A heat pipe initially uses conduction to take the heat off the IC and also to conduct to the condenser. Ultimately the heat is typically dumped to the atmosphere, which requires convection.

Passive Heat Sink
A passive heat sink is the simplest and quietest cooling device. It is generally a block of copper or aluminum with fins or pins to increase the surface area for heat transfer to the air. With the cost of copper so high, there exist passive heat sinks with a core of copper pressed into an aluminum body. The aluminum spreads the heat while the copper, with its higher heat transfer capability, is in contact with the processor. Passive heat sinks are common in 1U enclosures because of the limited overhead space. A well-designed chassis will provide fans in front of the heat sink blowing air through the fins for forced convection. Ducts inside the chassis can be used to direct the air flow to hotter components.

Pros—Lowest cost. Highest reliability with no moving parts.

Cons—Limited heat dissipation capability. Requires some external air flow to move the heat off. The system enclosure may not have sufficient room for installation. Component heat is dumped into the system enclosure, which may impact down-stream components.

Passive Heat Sink with Fan
As the description implies, a passive heatsink with fan is simply a passive heat sink with a fan screwed or clipped to the top or side. The fan blows into the heat sink to force convection and flow through the fins. These are simple, low-cost devices and are sufficient for most processors if sufficient flow through the chassis is available to move the heat out.

Pros—Low cost and reasonable to good heat dissipation.

Cons—High performance, high-power processors will require a large heatsink which may be a weak point in high vibration, high shock environments. May not provide sufficient performance in high-powered systems. Reliability is dependent on the fan reliability. May plug up with dirt if the system air is not filtered. The system enclosure may not have sufficient room for installation. Component heat is dumped into the system enclosure, which may impact down-stream components.

Figure 2: Heat Sink with Heat Pipes and Forced Convection (Courtesy of PC Cooler)

Figure 2: Heat Sink with Heat Pipes and Forced Convection (Courtesy of PC Cooler)

Heat Sink with Heat Pipes
Modern high-performance heatsinks typically include heat pipes. As discussed in the section on Phase Change above, heat pipes can have 10 to 10,000 times the cooling performance of solid copper solutions. A block of copper or aluminum will be in contact with the IC to spread the heat. Heat pipes will be embedded in that block and terminated in the finned part of the heat sink. A fan will be used to provide forced convection. These are relatively inexpensive and can have very high cooling performance. The downside is they can be fairly large and subject to vibration and shock stress.

Pros—Low to medium cost and good to great heat dissipation. Able to cool highest power processors.

Cons—High performance, high-power processors will require a large heatsink which may be a weak point in high vibration, high shock environments. May not provide sufficient performance in very high-powered systems. Reliability is dependent on the fan reliability. May plug up with dirt if the system air is not filtered. The system enclosure may not have sufficient room for installation. Component heat is dumped into the system enclosure, which may impact down-stream components.

Liquid Cooled
Liquid cooling uses hollow heat sinks attached to the high-power ICs in a computer. Fluid is piped through these heat sinks taking the heat off the ICs and out of the chassis to an external heat exchanger. Plain water is typically used as the cooling fluid though more sophisticated fluids could be used.

Pros—Very quiet. High power density. Ability to use off-the-shelf PC motherboards and plug-in cards. Their small size allows the use in smaller space-constrained rack mount 1U and 2U enclosures.

Cons—More expensive and more complex than air cooled systems. Possibility of fluid contamination inside the chassis. Requires an external heat exchanger which increases system complexity and space requirements. Higher maintenance costs to service the cooling system.

Heat Pipe to Case
Virtually all the sealed, fanless systems use heat pipes to couple the internal heat sources directly to the case. This allows the internal components to be completely isolated from the outside environment and not subject to dirt and moisture contamination. The case enclosure will typically be manufactured from aluminum for reasonable thermal conduction properties, lightweight and low cost. These enclosures can be extruded, cast, machined or a combination these technologies. They will have built-in exterior fins for increasing the surface area for increased convection. A properly designed system using low-power components in a defined temperature environment can be successfully implemented. PC/104 boards are commonly used in this installation because of the small size, wide availability and varied feature set.

Pros —The principle advantage of these systems is their rugged construction and complete immunity to environmental factors other than high temperature. They are also silent.

Cons—The downside is system heat dissipation is through the case surface via convection. This can be enhanced by placing the device in a moving air stream. They are limited in how much power can be dissipated which precludes use of power-hungry high performance CPUs, graphics, and so forth.

Heat Spreader to Case
High-performance purpose-built systems principally targeted at the military markets may incorporate custom board sets that include a heat dissipating plate (heat spreader) in the PCB. The ICs are thermally coupled to the heat spreader layer. The system enclosure will provide card locks that clamp the PCBs to the case by their edges providing a thermal path for the heat spreaders in the PCBs. The case will typically be double-walled with a sealed inner chamber surrounded by a fluid passage through which air or fluid is pumped to take the heat off the system.

Pros—Very high card density in that all the cooling is via the card heat spreaders so that no space is required in the system for air flow. Potentially the highest heat dissipation. High performance at high elevations such as in aircraft and UAVs.

Cons—Very expensive with a long lead time for new system development. The enclosures are very complex and require machined and brazed components. Typically a sole-source vendor relationship limiting cost reduction through competition. Does not support off-the-shelf PC cards.

Spray Cool
Spray Cool is a division of Parker Aerospace and offers enclosures, both custom and standard, which cools the electronics by deploying a fine mist of non-corrosive, non-toxic, non-conductive liquid (Fluorinert™ by 3M) sprayed in a thin layer, which evaporates and cools electronics. The enclosures are sealed. They offer enclosures that accommodate 3U and 6U VPX, VXS, cPCI and VME boards allowing off-the-shelf electronic solutions. An external heat exchanger cools the liquid.

Pros—Very high card density and power budget. Open system architecture. Environmentally sealed. Limited effects from altitude. The heat exchanger can be mounted remotely from the electronics.

Cons—High acquisition cost. Requires an external heat exchanger. Semi-custom or fully custom solutions with attendant high cost and long lead times. Single source limits competitive pricing. Does not accommodate lower-cost PC cards.

Submersion and Direct Flow
A variety of vendors offer solutions in which the boards are submerged in a non-conductive fluid similar to mineral oil. The cooling fluid is circulated through an external heat exchanger. An alternative solution is encapsulating off-the-shelf board in sealed hot-swappable canisters filled with a non-conductive dielectric fluid. A secondary fluid flow surrounds the canisters and carries the heat away.

Pros—High power density. System heat can be piped (literally) any distance to the ultimate end-point heat exchanger. Cooling flow can be cooled below ambient for additional reliability and power capability. Removes the fans from the equation for increased system reliability. Suitable for high ambient temperatures.

Cons—Installed boards are contaminated by the cooling fluid making maintenance and repair difficult. The systems tend to be heavy. High acquisition costs.

The Right Path
It is apparent that heat mediation should be considered very early in the system design process. Hot components may not function and will be less reliable. Before settling on the highest-performance processor, memory, etc., it is important to determine how much heat those components will generate, how that heat will be removed, and where that heat will be dispersed. Assuming a high-performance system is required, heat mediation should be the next determining factor to specify the cooling methodology; is an air cooled system sufficient or is a custom liquid-cooled method mandated?

Engage the selected vendor early in the process. The vendor will be able to analyze system cooling requirements prior to irreversible decisions being made that impact the end product delivery or reliability. It might turn out that system performance has to be lowered to meet the allowable heat budget.

Chassis Plans has a long history of working closely with customers creating sophisticated systems operating in extreme environments. Being engaged early in the specification process always leads to a successfully deployed solution.

______________________________________________________________________________________________________

David-Head-ShotDavid Lippincott is Chief Technology Officer, Chassis Plans. He founded Chassis Plans to provide custom industrial and military computer designs allowing customers to have these computers manufactured locally. The company morphed from an engineering design firm to a full-service manufacturer designing and manufacturing highly regarded rugged computer and LCD display systems to all branches of the military as well as all the prime contractors and leading industrial companies

Share and Enjoy:
  • Digg
  • Sphinn
  • del.icio.us
  • Facebook
  • Mixx
  • Google
  • TwitThis