RTC Magazine

Getting the maximum performance out of embedded PCs while keeping their size, weight and cost within reasonable limits requires careful thermal design. Excessive internal temperatures can wreak havoc on embedded PCs and their electronic components. Exceeding thermal limits results in loss in performance first, and ultimately, component failures and downtime.

Thermal management has always been important. As processor operating frequencies go up and IC processes continue to shrink, the issue has turned from important to critical. Hot spots occur easily and are quite a challenge in today’s embedded PCs. An executive from a well-known processor vendor at a recent technology forum remarked, “Thermal issues are the number one problem we face today.”

Embedded systems cover a wide range of mechanical variations, from rackmount VME systems to small portable designs. The computer architecture in embedded systems also varies tremendously, from low power, special-function RISC processors to general-purpose processors common in desktop or laptop PCs. PC processors such as Intel’s Pentium and Motorola’s PowerPC are very common in embedded systems because their use in consumer systems drives the cost down.

That said, these processors and their support circuitry lack the power and thermal management capabilities suited to the rigors of industrial embedded systems. For example, the narrow slot spacing in a VME chassis forces a designer to use alternatives to the large heatsinks and fans found in desktop motherboards. Kontron’s ASM3-VME chassis shown in Figure 1 illustrates the relatively small inter-board space common to VME.

System designs using embedded PCs require heat dissipation from a few watts for a 386-based system to over 100 W for a dual Pentium 4 system. Compounding the thermal management task are the special environmental, high-availability and longevity requirements common in embedded systems. Dissipating even 10 W can be difficult in a system that can’t use fans, must be completely sealed, and must operate in an environment where the ambient air temperature is 50°C and the maximum operating temperature is 85°C. Such restrictions are common in many industrial and military applications. Obtaining the proper thermal design requires both thermal modeling and verification by actual measurements.

Rings of Thermal Management

It’s helpful to model thermal management in an embedded PC-based system as a set of concentric rings. Each ring represents a thermal interface that transports heat away from the ring inside of it. Within the innermost ring is the silicon of the high-performance ICs used in the system that are the primary sources of heat. The first thermal interface layer is the packaging provided by the IC vendor. The outermost ring is almost always thermally stable ambient air, but this is determined or controlled by the user.

The system designer must implement application-appropriate thermal interfaces to transfer the heat of the system, starting at the ICs to this outer, stable heat reservoir of air. Table 1 shows the layers and simple description of their function. The number and type of layers in the thermal solution varies from system to system according to the environmental and application-specific requirements. In any given system, not all layers may be present, or allowed, and the layers might vary in order from system to system.

The packaging and features of the components, the source of system heat, represent the first layer of thermal management. Packaging material is commonly plastic for economic reasons, but in high-power ICs a metal heat slug or “flip chip” style packaging allows closer thermal coupling to the silicon die for more efficient heat transport. Many ICs, particularly processors, offer programmable power reduction features such as dynamic clock speed control. Low-power processors may be required in situations where active cooling isn’t allowed and space limitations exist.

Board Selection Key

For most system designers, the use of embedded PCs starts by purchasing a board-level product. Board-level embedded PCs typically provide only layers 0 and 1 of the thermal solution. To implement a complete system based on this type of product typically requires a heat sink and/or fan. Following the thermal model, in that case layer 2 is absent, layer 3 represents the processor to heatsink interface, and layers 4 and 5 are implemented by the heatsink and fan. That scenario so far only transfers the heat into the air inside the enclosure, so there must be some airflow through the enclosure or some other means to transfer the heat to the environment—layer 6.

Board-level PCs can be difficult to manage in a demanding thermal environment. The system designer may have to analyze the cooling of a number of individual heat-producing devices on the board and then provide thermal layers 2 through 6 for each of these devices as well as for the processor. Often large ICs such as PC Northbridge chips and video controllers require either passive or active heatsinks. This can lead to a complex thermal solution that’s closely tied to a specific board-level PC design.

Some board-level products, such as component SBCs or System-On-Module type embedded PCs, offer a thermal interface at layer 2 in order to simplify system design. The cover plate of those modules is a heat spreader assembly that dissipates the heat of the major components across its surface. Essentially conduction is used as the next transport layer. As long as the heat spreader temperature limit is met, the system designer need not manage the cooling of each specific heat-producing component on the module. For thermal purposes the module can be treated as a single component.

In some component SBC families, the heat spreader interface is standardized for all products across a wide range of processor and performance choices. This allows the system designer to reuse a single thermal solution for successive generations of systems. Although the lowest layers of the thermal solution are specific for each PC module design, these layers become the responsibility of the SBC vendor and the system designer sees only the standardized heat spreader interface. Typically the heat spreader will be coupled to a system enclosure that serves as a heatsink. Alternatively, a conventional heatsink can be attached to the heat spreader to implement natural or forced convection cooling. An example of Kontron’s ETX-P3 SBC with and without heat spreader is shown in Figures 2a and 2b.

There are several ways to implement thermal layer 6 of a system. In some cases the enclosure itself can be used to transfer heat to the environment—sometimes the exterior of the enclosure is finned to facilitate this. Where airflow through the enclosure is permitted, fans or blowers aid the process. Obtaining acceptable audible noise levels for fan cooling involves the optimum choice of fan size, blade pitch, number of blades, airflow volume and rotational speed. A smart fan controller IC and a temperature sensor can be used to control the fan according to the enclosure temperature, minimizing fan noise and power consumption. The use of smart fan control can also help extend a fan's mean time between failure (MTBF)—a common concern in fan-cooled systems.

Another solution is the use of heatpipes if a large amount of heat must be transferred out of a sealed enclosure. A heatpipe transfers heat by evaporating a liquid into a gas at one end and condensing the gas back to a liquid at the other. Typically a fan and heatsink are provided at the condenser end of the heatpipe, outside of the sealed enclosure, in order to transfer the heat to ambient air.

Chassis Build or Buy Trade-off

System designers have the choice to buy a board-level product and do their own mechanical integration or buy a chassis/system-level product. If the user puts the system together they have to deal with thermal management problems themselves. Part of the make or buy decision involves thermal knowledge, overall system cost and standard versus custom design.

An off-the-shelf embedded PC can address a variety of applications when the system supplier has addressed the thermal layers and specified the limits. Applying an embedded PC in an “easy” environment, such as systems for use in an air-conditioned office, is straightforward. However, a factory floor or mobile application may require operating temperatures well above 110°F (44°C). In such cases, an off-the-shelf system is probably not acceptable and a custom solution should be considered.

To illustrate such a custom approach, it’s useful to examine the details of an example where Kontron worked closely with a particular system designer and addressed his application’s thermal problem with a custom solution. In this example the constraints were defined up front, so the system designer had a number of options to deal with the thermal problem. The full extent of the thermal layers could be determined and addressed at the system level design.

One of the tools used in the analysis was Flotherm airflow and heat transfer analysis software from Flometrics. Flotherm was used to perform a computational fluid dynamic evaluation. This tool allows users to simultaneously look at the airflow distribution and temperature distribution within the air in a 3-dimensional finite element model to avoid costly over design. Using this design tool, Kontron designers provided the initial enclosure design shape and dimensions, and the Flotherm software divided the unit into cells to form a computational grid. By applying boundary conditions including ambient temperature, known mass flow rates and heat sources, designers determined hot spots within the enclosure.

Thermal Software Guides Choices

Using the system designer’s input for a custom embedded PC, a feasibility study was performed on a sealed enclosure with no fans using the Flotherm software. The enclosure was modeled as a simple aluminum box, without fins, with a wall thickness of 0.125 inches (3.18 mm) (Figure 3a). The CPU was modeled as a 533A Celeron FC-PGA processor.

The preliminary model used a heatsink that served as a coupling from the CPU die to the case of the chassis. The initial Flotherm analysis for the high temperature range indicated the CPU would run approximately 40°C above ambient. Other critical components including the core chipset were calculated to run at approximately 50°C above ambient. This initial analysis indicated that the design would not meet the customer’s requirements.

Subsequent revisions were made to the thermal model. By varying the wall thickness of the enclosure and the thickness of the heatsink, it was possible to optimize the tradeoff between enhancing manufacturability and lowering the CPU and hard disk drive temperatures. Next, more mesh was added to the heatsink model to obtain a more accurate result. Meanwhile the contact area between the heatsink and the enclosure was increased.

Using data from the initial Flotherm analysis, an enclosure was fabricated and tested to validate the thermal model. Thermal test data showed the CPU to run at approximately 72°C at 25°C ambient, while sitting on a bench oriented vertically as it would be mounted in the end application. According to the updated thermal model, the unit should have measured 53.6°C at 25°C ambient.

A number of revisions were evaluated based on the combination of the computer-aided analysis, testing on the physical model, and the designers’ previous experience and insight into thermal management issues. Key revisions included welding the chassis sides together and increasing the surface area of the chassis. Meanwhile, a different heatsink concept was evaluated. Perforating the back of the enclosure enabled improved airflow. Another revision idea was to evaluate a 1000 cubic-ft/min. fan.

The chosen enclosure produced the temperature results shown in Figure 3b. The final selection was made based on manufacturability as well as meeting the end customer’s performance requirements. Table 2 shows the temperature rise above ambient that occurred for three final configurations.

Think Thermal Early On

There’s no doubt that temperature rises occurring in today’s embedded PCs must be addressed early in the system design. A thermal management model discussed above uses thermal layers from the heat sources to the heat reservoir. Understanding the thermal design issues should be kept in mind as system designers choose among board-level, module and chassis system products available in the marketplace.

Special environments may call for a custom solution, along with a detailed specification of the environment. Making good use of simulation tools and thermal measurements on early prototypes of the hardware helps system designers achieve their performance and cost goals, while eliminating the need for expensive field testing or retrofitting.

Kontron America
San Diego, CA.
(858) 677-0877.
[www.uskontron.com]