Design for Thermal Issues

Unit 16: Practical aspects of enclosures

Our focus now turns from the board and components to the enclosure and the whole of the system. Building on the previous information on conduction and convection, we will be considering how all the elements in the cooling environment are brought together to create a complete piece of equipment.

This unit and the next are deliberately slightly shorter than their immediate predecessor, realising that you will be at the stage where you will want to concentrate on Assignment 2. In consequence, the material is supplemented by a number of items of background information, which may be treated as resources to dip into if you come across the issues in the real world.

Unit contents

What is an enclosure?

Manufacturing companies frequently speak of the process of “box build”, to distinguish the manufacture of the board assembly (colloquially known as “board stuffing”) from the many different processes that are used to provide both protection and an interface with the outside world around assemblies, whether a single board or a complete rack-full of boards. From the thermal perspective, we prefer the term “enclosure”, as this emphasises the way in which the “box” also provides a thermal boundary for the system, as well as introducing a number of constraints on the way in which heat flows.

The system view

Before looking at the way in which an enclosure is made, we should perhaps think generically of the elements that compose the complete thermal system. As shown in Figure 1, an electronic enclosure is made up of a number of elements:

Figure 1: Schematic view of an enclosure as a system

Schematic view of an enclosure as a system

 

Depending on the application, the fourth and fifth levels may be integrated, with the circuit assembly being mounted directly onto the enclosure, and there may be a “mezzanine” between the second and third levels made up of Multi-Chip Modules, hybrid microcircuits or other daughter boards.

Not only are these parts of the system joined together, but they are also coupled thermally, which means that the thermal performance of any one part is directly dependent on other parts of the system. The heat transport mechanisms can become very complex, but we need to build our system from the ground up, examine the main heat transport mechanisms at work, and simplify the view as far as possible.

[ back to top ]


Elements of the system

In Unit 13, we discussed the role of the circuit board in spreading heat, but perhaps did not stress sufficiently the consequence that the spread of heat between chips may result in the critical chip not necessarily being the one that is the largest, or has the highest power dissipation. So a system thermal analysis needs to look at the junction temperature of each device, and ensure that it is within its design specification.

In more complex systems, boards with assemblies will be fitted in multiples within a card-holder, usually referred to as a ‘shelf’ or ‘cage’. This acts as a mechanical housing, into which boards are normally inserted by using card guides. Although other wiring methods are available, the most usual configuration is where the shelf supports a backplane, fitted with connectors, into which the board fits. In many cases, the board-to-connector mechanical attachment is an important part of the overall rigidity of the system, although some boards will also have front plates that are used to secure the otherwise loose end to the top and/or bottom edges of the cage.

Thermally, each of the board assemblies is in free air, and there are few paths for heat transfer from board to shelf by conduction. However, this is not necessarily the case: in some specialist equipment, particularly in the avionics area, substantial conduction cooling is taken through to the individual boards within the cage.

Within each card-holder there will be opportunity for the circulation of air, but with significant resistance to flow in cases where boards are closely packed together, or heavily populated with tall components. In this case, as we saw in Unit 15, the fan will need to be chosen to work satisfactorily against the back-pressure.

The enclosure itself, particularly in the bigger systems, is a frame that houses a number of shelves, each of which will be supported by the enclosure, but not totally surrounded. The heat generated within the card frame/shelf will be transferred to the surrounding air, and the overall enclosure airflow will exit through the vent holes. Unfortunately, in many configurations, an easier path may exist for the air from entrance to outlet, bypassing card-holders and cards. Most of the air will take this “path of least resistance”, with the result that the cooling efforts are less effective than might have been expected.

Comment

One of the real benefits of simulation is that it makes it much easier to model the effect of “baffles”, modifications to the flow path that are deliberately introduced in order to balance the flow of air between different sections, and ensure that air is directed at those parts of the system requiring the most cooling. Whilst introducing any kind of baffle inevitably increases the system impedance, the equalisation of cooling that results more than outweighs the reduced overall airflow.

 

There is usually reasonable thermal contact between shelf and system through the mechanical connection, but both convection and radiation also play a part in transferring heat from the card tray to the system enclosure, which is typically at a temperature similar to that of the surrounding environment. Surprisingly, the radiation component can be significant, especially in a system cooled by natural convection.

Much of the energy dissipated within the system will be transferred to air passing through the system, especially in the case of forced convection. However, any analysis or simulation should not neglect the coupling of the enclosure to the surroundings through radiation and natural convection. The amount of heat transferred can vary significantly, depending on the system surroundings – contrast a system in a climatically-controlled building, where there is significant external airflow to cool the enclosure, and a free-standing system exposed to sunlight!

[ back to top ]


The enclosure specification

Exercise

Before reading further, make a list of the elements that should be considered when defining the requirement for an enclosure for an electronics system. Then check off your own list against the items that we mention – and they won’t all be in this section!

 

Elements of the specification

Typically there will be information on the mechanical constraints for the enclosure, including a fully dimensioned drawing and Bill of Materials. In many cases an enclosure will have several stages of build, and many elements. The development of international standards for enclosures is of great assistance to the designer, as the specification task becomes the responsibility of the supplier, assisted by the Standards Authority.

There will be more about standard enclosures in the section Detailing the specification, but we would offer a word of warning, that buying a standard configuration, such as a 19-inch rack, merely buys you one of a wide range of products that may be significantly different from the thermal management point of view.

A key element of the specification is the environment in which the system is to operate, under both storage and functional conditions. As we point out in our paper The “real world”, there are considerable variations in the natural environment, as well as those imposed from elsewhere in the system. Some products will operate in relatively benign environments, whereas automotive and avionics applications are considerably more challenging. And the challenge is not just temperature and humidity, but a combination of the two, together with mechanical challenges (vibration and shock) and a range of external influences ranging from salt spray, sand and dust, through to chemical and biological agents. In our paper Simulating the real world in the test house, we say something about these requirements and the way that products are tested.

An interesting concept within that paper is that of a “climatogram” an example of which is repeated as Figure 2. This is one of a whole range of similar illustrations, based on BS EN 60721, that indicate the range of conditions that an equipment is expected to survive.

Figure 2: Climatogram for Class 7K3

Climatogram for Class 7K3

Class 7K3: Use at totally or partially weather protected locations in areas with Warm Temperate, Warm Dry, Mild Warm Dry, Extremely Warm Dry, Warm Damp types of climates and at non-weather protected locations covered by the Restricted group of open air climates.

For many purposes, however, this envelope will be substituted by a statement of the operating temperature range. This is dangerous, firstly because it ignores the influence of relative humidity. Although water vapour in the air only marginally improves its ability to transmit heat, condensed moisture is a hazard for any electronic system.

Secondly, the temptation is to focus on the maximum temperature, and on keeping the maximum junction temperature that corresponds to that high ambient within a safe operating area. Unfortunately, forgetting performance at the bottom of the temperature range can be highly dangerous. As Tony Kordyban puts it:

Quote

It is easy to think that the “worst case” for any electronic assembly is at the maximum ambient. (I have been guilty of this myself.) But maybe your product has to operate outdoors or in an unheated underground vault.

Many commercial components don’t work below 0°C. Things like batteries, capacitors and crystal oscillators behave very strangely.

You can’t rely on your high ambient instincts. At high ambient, you test the fully-equipped system at maximum power. The worst case at the low end may be with a partially-equipped system at minimum power. And slapping on a heat sink to fix a problem at the high ambient may make the cold end worse.

If, as I did, you think this challenge is trivial, just talk to an electronics engineer in the auto industry.

 

Comment

For those who enjoy a light-hearted look at serious subjects, Tony Kordyban is always a good read; his regular kordyban’s korner in Electronics Cooling is our first port of call in the magazine, and you will enjoy the article Ten stupid things engineers do to mess up their cooling, of which the extract above is Stupid Thing No.7. If you have time, we really would encourage you to buy a copy of his book Hot air rises and heat sinks: Everything you know about cooling electronics is wrong and to read its sequel More Hot Air. As Electronics Cooling put it “They tell stories . . . including the thermal challenges in developing a telepathy-based telecom system. All of them are educational. Some are even true.”

 

Another key element of the specification, and one that is frequently not well defined by the end user, is the nature of the external environment; whether any cooling is available, or conversely, whether thermal gain is to be expected from the surroundings.

The specification should also indicate how the enclosure is to be tested to verify that the specification is met. In the second half of Simulating the real world and the test house, we make reference to a series of tests now enshrined in BS EN 600068. However, one has to be careful to apply appropriate levels of stress, particularly mechanical stress, bearing in mind that this range of specifications covers the whole of the electronics spectrum from components to systems.

Of course, thermal considerations are only one of the elements to be balanced by the enclosure designer, and whilst most specifications will indicate the need for testing over a full range of temperature, they may be less clear as to other requirements. Depending on the application, the enclosure may have to provide protection against mechanical damage and chemical and biological adversaries in the environment, as well as the more common challenges caused by humidity. Yet these are the main factors that will determine whether or not the enclosure is sealed, and the materials from which it is made.

[ back to top ]


Closed or open box?

Another aspect of the enclosure that will be the subject of a specification is its resistance to water in the environment. Typically this is specified in terms of an “IP rating”, as described at this link. The pair of numbers used to indicate the protection provided by the enclosure relates both to solid foreign objects and water. The higher the rating, the more likely is the enclosure to be sealed.

Sealed enclosures present considerable problems of thermal management, especially when dissipation is high. However, it is still possible to have a sealed system with forced convection, provided that the system includes a heat exchanger or equivalent that is capable of transferring heat by conduction from the inside of the enclosure to the outside. How this might be carried out is indicated in Figure 3.

Figure 3: Closed box air-cooling configuration with double-sided heat sink air-to-air heat exchanger

Closed box air-cooling configuration with double-sided heat sink air-to-air heat exchanger

 

The source of this diagram is the article by Robert Simons, Estimating temperatures in an air-cooled closed box electronics enclosure, which adopts an analytical approach to the solution of the problem, although this can also be modelled. Note that, in the implementation shown in Figure 3, a fan is used to cool the external heat sink, although having a fan in this position might give rise to other reliability problems in a location with a high level of particulates.

Sealed enclosures are frequently used in outdoor applications, and a wide range of solutions may be applied to the task of cooling them. As well as the air-to-air heat exchanger considered above, standard air conditioners, thermoelectric coolers, and heat pipes may be employed. An alternative, which would apply in the common situation where an enclosure heats up during the day, but is cooled naturally at night, is to use a phase change material to absorb peak energy loads during the day and reject that heat load later. PCM materials typically have high heats of fusion, allowing small volumes of material to store large amounts of energy simply by changing phase. More information in Marongiu and Clarksean, Thermal management of outdoor enclosures using phase change materials.

A final application of sealed systems is where the enclosure is to be used in a flammable atmosphere, and the requirement is for intrinsic safety. Here the regulatory requirements are considerable, so specialist advice needs to be taken.

Comment

It is highly tempting to assume that sealed enclosures are permanently sealed. In practice, as Jim Wilson discusses in Moisture permeation in electronics, most polymeric materials are permeable to some extent, so maintaining a dry condition inside an enclosure is a considerable challenge.

 

[ back to top ]


Detailing the specification

Equipment practice

The form of the enclosure will be affected not only by environmental requirements, but also by:

The last two of these are inter-related, many industry sectors having preferences for particular configurations, and corresponding standards for what is referred to as “equipment practice”.

The task of creating a detailed enclosure specification involves a fine compromise between a wide range of requirements, electronic, thermal and mechanical. Fortunately, the need to make quite a few decisions will be removed by a suitable choice of standard enclosure. Whilst there are many suppliers and variants, there are broadly compatible international standards both for small cases and especially for the larger rack-mounted systems. The existence of standard measurements and fixings has enabled a very useful degree of harmonisation and reduction in piece-part count.

Although both larger and smaller versions are available, probably the most commonly-encountered standard is that of the “19-inch rack”, a design which originated before World War II as a mounting system for telephone switching equipment. Although metric equivalents have been devised, many of the dimensions for standard equipment practice are still inch-based.

The 19-inch rack is used widely for commercial and professional applications, whether computing, telecommunications or process control. As a well-established standard, it is particularly useful for system integrators who are purchasing pieces of standard equipment. Many designers of items such as standalone test equipment will choose the 19-inch rack as a basis, and supply an instrument either in a rack case (that is, a local enclosure for a single unit), or as a bare unit in a sub-rack for integration into a larger rack. Such modules will typically have some local screening, for physical protection and EMC compliance, and this will have considerable impact on the airflow, even where the internal covers are perforated or louvered.

The mounting fixture on a rack consists of two parallel metal strips, also referred to as “rails”. Of standard dimensions, they are separated by a 17¾" gap, giving an overall rack width of 19". The rails have holes at regular intervals and a standard pitch: these are usually square holes that allow clip mounting, or can be adapted for use with bolts by inserting a “cage nut”, which consists of a captive nut within a spring steel cage.

The pattern of holes repeats every 1¾", and racks are divided into regions 1¾" in height, within which there are three pairs of holes in a vertically symmetric pattern. This region is commonly referred to as a “U”, standing for unit, and heights within racks are measured in multiples of this unit1. All rack-mountable equipment is designed to occupy an integral number of U – measuring equipment might be 4U high, whereas a rack-mounted computer might be only 1U or 2U, with the trend towards the smaller dimension. This means that a “tray” of fans at the bottom of the unit is often designed to be just 1U high, which has implications for the type of fan used.

1 Is it just coincidence that a standard floppy disk is 2U square, and its predecessor had a 3U dimension? Scope for a “History of Science” investigation . . .

 

The rails are usually not flat strips, but folded strips arranged around the corner of the rack, or girder sections. This helps stiffen the rail, but additional support is needed for heavy equipment. This can be provided by a second pair of mounting rails at the back of the equipment, or by using a pair of light rails screwed to the rack, with the equipment sliding into the rack along the rails. Such rail-mounted systems are particularly useful if they are able to support a unit that has been slid clear of the rack, as this allows easy inspection or maintenance.

The basic outline of a 19-inch sub-rack is shown in Figure 4. Boards may be fitted to guide rails, and plugged into connectors on the backplane, or may be fitted with front panels secured to the top and bottom front rail. Alternatives available, depending on the manufacturer, include box type plug-in units and complete sub-racks.

Figure 4: Overall view of a typical 19-inch sub-rack

Overall view of a typical 19-inch sub-rack

 

Whilst board connectors with spills may be secured to inner rails, and then wire-wrapped, a more usual method of interconnecting boards is to use a backplane. This is a circuit board that in its earliest form merely made parallel connections between all the connectors, providing a computer bus structure, although of course a number of these connections needed to be dedicated to power and ground.

Using a board with connectors is a more reliable solution than using cables, particularly as cables that are hard-wired to the card will be flexed every time that a card is removed for inspection, and this flexure eventually causes mechanical failure. A backplane, however, has a service life that is limited only by the number of mating cycles (insertion, and subsequent removal) that the connector can withstand.

Since this original simple backplane was developed, the concept has grown enormously in complexity, and most backplanes are now custom items. Not only is the connectivity specific to the application, but the backplane will feature low-impedance connections for power/ground, controlled impedance connections for high-speed signals, and access for test points. Backplanes will often themselves be assemblies, with active components providing facilities such as bus driving. “Active” backplanes are those that include circuitry to buffer the signals.

The differences between a backplane and a typical rack card are both electronic (typically there is no on-board processing power) and mechanical – a backplane needs to be relatively rigid, so is often made of substantial laminate (say 3.2mm thick, instead of 1.6mm for a standard board).

But the backplane is more than this – from the thermal point of view, it usually presents a barrier that is substantially non-conductive, and that closes off the channel between the cards. Fortunately, where both cards and case are vertical, this has little effect on airflow.

The detailed design of a 19-inch rack will vary from supplier to supplier, although the basic dimensions are constant, and there is a pleasing degree of interchangeability between components from different suppliers. There is however more variety in the thermal management aspects of these systems, and a designer will take some time to become conversant with the wide variety of accessories and options.

It is not possible to consider a rack-mounted system in isolation from its interconnections. The standards for the racks are thus closely aligned with standards for the matching connectors, and the accessories available with the system will include methods of cable management, both for connections between units and externally.

Where the 19-inch rack is not appropriate, there are less comprehensive ranges for stand-alone applications, the “small boxes”. Here there is also more variety in appearance, both in terms of the surface finish and of the enclosure material itself.

[ back to top ]


Component selection

Our preliminary discussion has been on equipment practice, because this affects every aspect of the overall design. But we still need to consider all levels, from component through to enclosure.

Starting with active components, individual parts need to be compatible with the application. Thermally, the maximum acceptable temperatures should be higher than those predicted from the thermal model. How much margin is required will depend on the accuracy of the model, allowing for some variability between nominally identical components.

This requirement applies equally to passive components. Most surface mount parts are not problematic, but a number of larger components, such as relays, are made of materials that have a definite maximum temperature rating which, for optimum reliability, should not be exceeded. Suitability for application will depend on the local temperature rise, on the environmental specification, on the choice of materials for the part, and on any heat generated. For example, relay coils generate heat, but so also do conductors and connections used at moderate/high current.

When assessing the thermal properties of a part, it is prudent to make some allowance for expected changes with life – all connections, whether permanent or remakeable, become more resistive with use and consequently run “hotter”.

Specifications

In the same way that the electronic component specification may be influenced by its expected performance during life, we may wish to make provision for possible causes of failure. What happens if a motor is overloaded, or a filter becomes clogged? In this case, the motor may overheat and the insulation will deteriorate, and may break down. This is not a problem where the motor has been oversized and underrated, as used always to be the case, but the continuing move towards motors that are physically smaller for the same rated output means that less margin is available. In consequence, fast methods of protection are used, such as embedding a PTC thermistor to provide protection against burnout and automatic re-setting once the cause of the fault condition has been removed.

Earlier Units have dealt with the specification of board, heat sinks and fans, all of which can be modelled as part of the overall enclosure. As always, this work should be done early in the design process, so that thermal performance can be included and problems avoided, rather than just measured and fixed after prototyping. This is particularly important where significant changes are being made at the board level. In her paper Solve thermal issues earlier in updated board designs, Alexandra Francois-Saint-Cyr promotes early thermal evaluation to highlight potential problem areas, and allow engineers more flexibility in providing a cost-effective resolution to any problems.

Noise

Every time you fit a fan, you are fitting a noise-generating component. This may be a “whisper” fan, but experience indicates that initially quiet fans may become noisier with time and wear. In Unit 15 we emphasised the need to choose a fan running at relatively low speed and feeding a low-impedance network as a way of reducing noise.

System noise will therefore be an (increasingly important) aspect of the enclosure specification that is agreed with the customer. Unfortunately, most enclosures incorporate flat panels that are supported at a small number of points, and are otherwise in free air. Such panels can form resonating surfaces, and care should be taken that these are not “excited” by the fan.

The solution to a noise problem may involve compliant mountings for the fan, or some degree of damping for enclosure panels. But damping by using means such as compliant gaskets may have the unwanted effects of reducing heat loss by conduction and modifying the heat flow.

Additional information

With projected power dissipation levels continuing to increase, there is likely to be a corresponding increase in the emission of acoustic noise. However, particularly for office environments, even quite low levels of acoustic noise have negative effects. In his paper Acoustic noise emission and communication systems in the next century, Quinlan identifies the need for radical changes to cabinet design and a concerted effort to understand the basis of noise generating mechanisms in small air moving devices.

 

Comment

In this section we have concentrated on just a few aspects of the specification, but the designer has many more requirements to consider when devising a complete system. For example, telecoms equipment used in the North American market has to comply with NEBS requirements that relate to a wide range of environmental issues. In Thermal design and NEBS compliance, Majid Safavi of Lucent Technologies considers the design issues, of which thermal issues are just one aspect.

 

[ back to top ]


Where to put the box

Enclosure best practice

Units 14 and 15 have already discussed some suggested elements of best practice as regards the enclosure. These involve the positioning of the components and the selection of fan and heat sinks. As well as the points considered there, any design has to consider the way in which cables are routed – it is not a good ideal to allow cables to “flap in the breeze”, as this generates noise, reduces life expectancy, and impedes airflow. Wiring should be carefully routed and secured.

The layout and cabling of a system will also impact on its EMC performance, and there is often some conflict between the EMC requirement for a “leak-tight” enclosure to prevent EM emissions and the thermal management requirement to have at least some perforations in the outer skin to facilitate airflow. The Flomerics paper, EMC and Thermal Design Conflicts in a PC, which used for its study a Pentium PC, showed the effect of non-continuous grounding, and how the overall assembly could be managed to best effect. It is perhaps not surprising that Flomerics have developed EMC management software that co-exists with Flotherm.

Cooling the whole room

So far we have concentrated on individual enclosures, and only mentioned some of the specification issues associated with multiple rack-mounted systems closely packed within a computer room.

Because of the high thermal dissipation within a computer room, the room is normally air conditioned as a separate operation by a specific CRAC (Computer Room Air Conditioned Unit). The room will normally have a false floor, with the space beneath the cabinets managing the airflow to perforated tiles in the floor, and also providing a routing space and protection for the extensive cabling needed by such equipment. Typically, a CRAC returns air to the top of the room and provides negative pressure, so as to extract warm air from beneath the floor.

The layout of the enclosures within the computer room will be determined by their interconnectivity, by other signal and power cabling, by the ability of the floor to withstand the cabinet loading, and by the requirement for maintenance. Whilst the most obvious way of arranging the units is in parallel rows, and this is convenient, there is a problem associated with the fact that most equipment has air intake grills at the front, which allow air to be drawn in across the internal components, before being then exhausted to the rear. But if the racks are configured all facing the same way, the exhaust of the first row will be drawn into the air intake of the second row, and so on. To avoid cumulative heat build up, IXE recommend that the racks be arranged to face in alternate directions, so that the aisles are successively “cold” and “hot” corridors.

Other practicalities indicated in this reference are a suggestion that doors should not be fitted, or at least should be fully ventilated. This is because there is little gap between rack and doors, and in consequence, considerable turbulence near them.

[ back to top ]


Choice of cooling regime

Almost all the material so far, whether concerning the specification, the modelling, or the equipment practice, is independent of the selection of the particular cooling regime. Indeed, many 19-inch rack systems manage perfectly well when cooled by natural convection, provided that there are proper conduction heat transfer paths to disperse the heat from the critical areas, and that the circulation of air is facilitated by appropriate placement of vents. Adding a fan, and creating a forced convection regime, allows the amount of heat generated to be increased substantially. However, the trend is towards ever-increasing power densities, and the question inevitably arises as to the limitations of air cooling. This is something we touched on in Unit 14, when we looked at enhancing cooling for components and modules, where an air-cooled heat sink did not have the required performance.

Limits of convection cooling

As you will expect from your experience in other technologies, continual improvements to the basic processes and materials mean that their capability continues to improve, so the point at which alternative methods need to be adopted moves to yet higher dissipation.

Here we need to take two opposite views of the problem. The first is a global view of the whole enclosure, as suggested by Tony Kordyban in The weakest link in air cooling. This is that the limitation on total power dissipation in a system depends ultimately on the effect that this has on the human environment, by heating the ambient to an extend that cannot be tolerated by the equipment operators. This is a very valid approach, which is obviously room-dependent, but does explain why extreme thermal challenges are typically dealt with in ways that involve cooling the whole environment.

The second is the more critical from the point of view of component reliability. This is to look at the power dissipation of individual components, and determine whether the junction temperature can be kept within safe limits. This is an issue of spreading the heat, operating on a local scale, rather than removing it totally. Where the local temperature is the main concern, then strategies such as heat pipes become an effective way of extending the scope of air cooling.

Figure 5: Temperature differences available as a function of heat fluxes for various heat transfer modes and coolant fluids

Temperature differences available as a function of heat fluxes for various heat transfer modes and coolant fluids

 

So what are the limits? Kraus and Bar-Cohen suggest in Figure 5 a range of applicability for natural and forced convection that relate the local heat flux to temperature difference. An alternative way of presenting similar information is given by Simons in Figure 6.

Figure 6: Range of heat transfer coefficients for various coolants

Range of heat transfer coefficients for various coolants

 

Note the dependence on materials, and the more evident overlap between forced and natural convection in air, and the relative place of water cooling. Whilst this was pioneering work, and now some 20 years old, the basic physics of cooling changes little, although there would be some improvement (a move to the right in Figure 6) were these figures to be re-evaluated for today’s situation.

And there are of course some alternative methods that these comparisons do not show. One example is the use of a heat pipe; another is to use impingement cooling. But there are cost implications to both techniques, and, in the second case, jet impingement may be unacceptably noisy. But these heat sink improvements are substantial, as witness Figure 7, which was taken from Alphonso Ortega’s What are the heat flux limits of air cooling?.

Figure 7: Achievable power consumption for various cooling modes

Achievable power consumption for various cooling modes

 

Supplementary information

For further information on enhanced air cooling, and a practical illustration of state-of-the-art cooling applied to a computer application, see Banton and Blanchet, Utilizing advanced thermal management for the optimization of system compute and bandwidth density.

 

[ back to top ]


Water cooling

Water cooling was used in computers at least as early as 1964, and mainframe systems continued to use water cooling until the early 1990s, when the market shifted from bipolar technologies to CMOS, and took advantage of the lower power dissipation by moving to air cooling. More recently, liquid cooling has seen something of a resurgence, to meet increasingly challenging applications. In Liquid cooling is back, Roger Schmidt of IBM discusses the rationale behind the reintroduction of water cooling, stressing the need to take into account the interests and preferences of their customers.

If water cooling is necessary, then how do we do it? Two options are to cool from the edge, allowing board conduction to do the rest of the work, and to pass coolant through a cavity in the centre of the card. How this is used in an avionics application can be seen in Part 1 of Kaveh Azar’s articles Cooling Technology Options.

More typically we will use a cold plate, which we need to make as effective as possible. As well as the style of cold plate previously described in Unit 14, a number of special-purpose designs have been developed, aiming to improve the conductance of the cold plate. Given that there is a temperature rise from inlet to outlet, it is not surprising that several of these involve alternating inlet and outlet channels in order to improve the distribution of coolant.

Supplementary information

The idea of “intercoolers”, water-cooled heat exchangers used to reduce exhaust temperatures for high-dissipation applications such as computer racks, is addressed in Robert Simon’s Estimating the effect of intercoolers for computer rack cooling. The reduction in exit air temperature is very dramatic!

 

In most designs, fluid is confined to specific tubes, holes or channels formed in the surface by machining or etching. An alternative is to create a ‘heat transfer matrix’, where the void inside a cold plate is filled by sintered metal powder, in order to enhance the area of contact.

Comment

David Copeland makes the point that, because the volumetric thermal capacity of water is close to 4,000 times that of air, the volume of water required per hour will be slightly lower than the volume of air required per second to provide the same thermal capacity. His article Review of low profile cold plate technology for high density servers is a reminder that overall system costs may be reduced by using water to cool the processes, because it is then possible to transfer heat to the external environment through the water supply, bypassing the CRAC (Computer Room Air Conditioning) system. This significantly reduces the cost, energy consumption and space required by the CRAC units.

 

Supplementary information

Further information on this aspect is available in the paper Liquid cooling of a high-density computer cluster by LaPlante et al, where a duplex system, with both an indoor chiller and outdoor condenser, is used very effectively to meet a 10kW+ challenge, reducing the temperature rise from over 40°C and increasing to a stable 11°C.

 

A liquid system is fundamentally more effective, but gives a number of challenges. One of these is the conceptual problem that we are merely using water to remove heat from where it is generated to the outside, and need to manage the discharge of heat to the environment. In some applications it may be acceptable to discharge warm water to a drain, but this is costly and a typical cooling system will actually be a water-to-air hybrid, as described in Figure 1 of Robert Simons’ Estimating temperatures in a water-to-air hybrid cooling system.

Comment

However large the external heat exchanger, and however effective its ventilation, the temperature of the fluid circulating in the system can never be as low as the ambient air temperature. In conditions where the ambient temperature is high, it may be necessary to look at refrigeration or similar techniques. More ideas on this in Part 2 of Kaveh Azar’s article.

 

As Chris Soule of Aavid points out in The benefits of liquid cooling over air cooling for power electronics: “Liquid cooling counters almost every drawback of air cooling. It can dissipate more heat with considerably less flow volume, maintain better temperature consistency, and do it with less local acoustic noise”. But there are drawbacks, and the proposed solutions need to have a reduced risk of leakage and less obtrusive and cheaper support equipment.

In Threatened with a pipe, a contribution to “kordyban’s korner” in Electronics Cooling, Tony Kordyban puts an amusing spin on the moves towards using liquid cooling as a means of extending the scope of convection cooling. He points out four problem areas that, whilst fairly obvious, still need to be kept in mind:

  1. Some training will be required. “Liquid cooling will force mechanical engineers to learn whole new disciplines – plumbing and industrial controls. Liquid is not just an air that is a lot denser. The end user of the product will also need lots of additional training and/or customer support to learn how to operate and maintain the complex liquid cooling system.”
  2. Liquid carries solids. Any plumbing system is subject to becoming clogged by dirt, and the channels may corrode. Where filters are used to control this problem, they too will clog up, and the margin for failure is smaller with liquid than with a filter on a convection system.
  3. Balancing the flow is not a trivial problem. Unfortunately, modifying the flow in any particular channel by changing its resistance will have a knock-on effect in other branches.
  4. Leaks. Whilst an air-cooled system can leak air with no detrimental effect, the same cannot be said of the water in a fluid-cooled system! As Kordyban puts it, there is “nothing like a chassis full of corrosion and mildew to put the ‘ick’ back into electronics”!

The points that Kordyban makes are an important reminder, were one needed, that introducing water-cooled systems in a new application should be approached with some caution. Even where the end-user is accustomed to dealing with water-cooled systems, such as in the military, there is a significant overhead involved in training and maintenance terms.

Comment

Whilst water has much better thermal properties than air, it is still far inferior to metals as a heat transport medium. And Simons showed early on that many other fluids were not as effective as water. In the article by Nguyen and his colleagues, Heat transfer enhancement by using nanofluids for cooling of high heat output microprocessor, there is an interesting insight into how a dispersion of extremely fine particles can improve the thermal properties of a conventional fluid. Watch “nanofluids” for developments in this area.

 

In his editorial in the May 2005 edition of Electronics Cooling, Kaveh Azar suggests that the issues relating to liquid cooling will not be resolved by discussion, but only by “market acceptance”. In other words, does the market segment for the final product accept liquid cooling as a possible solution, with all its associated costs?

In the commercial field, high-performance “over-clocked” PCs frequently use liquid cooling on a small scale, but the majority of users are in specialist applications that are already equipped to handle a liquid-cooling requirement. Examples of market segments where liquid cooling is available, and constraints on thermal dissipation and reliability allow its extra cost to be justified, are high-power lasers, medical equipment power supplies, power electronics, military electronics and high-capacity computing. Azar makes the comment that offering systems with liquid cooling to other customers, such as those in telecommunications and data communications is likely to be unattractive.

Quote

“The reason a more complex liquid-cooled system may have been chosen by an engineer is that, we hope, all simpler systems involving air cooling were exhausted. In other words, to bring the product to market, liquid cooling was the only alternative if the system structure (packaging) could not be changed.”

Kaveh Azar, Electronics Cooling, May 2005

 

Cooled systems

The cooling methods discussed so far, with the exception of the brief mention of refrigerated systems, are only able to achieve temperatures that are above the ambient, but there are some specific cases where performance of semiconductors can be improved by operating at sub-ambient temperatures. Here refrigeration methods, involving the compression, cooling and evaporation of volatile materials offer the only solutions. But this is not an area of operation for the faint hearted, because there are reliability issues to consider at low temperatures, and a major potential for the build-up of damaging moisture.

[ back to top ]


Resources for this Unit

This list is in the order in which the material is referenced in the Unit text.

Optional links and information

[ back to top ]