BROWSE ARTICLES BY TECHNOLOGY

DIGITAL EDITION

RTC Magazine Digital Edition

INDUSTRY NEWS

QUICK DOWNLOADS

RTEC10 is an index made up of 10 public companies which have revenue that is derived primarily from sales in the embedded sector. The companies are made up of both software and hardware companies being traded on public exchanges.

COMPANY PRICE
(USD)
CHANGE
 
Adlink
1.22
-1.781%
Advantech
3.02
-0.889%
Concurrent Comp
3.58
-3.241%
Elma
474.00
0.173%
Enea
5.31
-1.918%
-   Interphase5.130.000%
-   Kontron0.00
Mercury Comp
14.04
1.299%
Performance Tech
1.83
-2.032%
PLX
3.22
-0.617%
Radisys
7.39
0.271%
52 WK HIGH 52 WK LOW MKT CAP (Million USD)
1.24
1.15
167.08
3.06
3.02
1,668.57
3.66
3.51
32.95
474.00
474.00
108.30
5.34
5.00
93.75
5.155.1235.37
0.000.000.00
14.05
13.69
429.77
1.83
1.72
20.36
3.25
3.20
143.40
7.52
7.23
204.97
RTEC10 Index: 603.86 (-4.75%)
RTEC10 is sponsored by VDC research

SOLUTIONS ENGINEERING

Cooling Large Systems

Beating the Heat

Traditional thermal solutions fall short in handling the power requirements and heavy loading of new system designs. Solutions to mounting and planar alignment issues include bearing-cooled fan-sinks, reliable sealing technology and flexible heat pipes.

MARTIN MAYER, ADVANCED DIGITAL LOGIC

  • Page 1 of 1
    Bookmark and Share

The three most important aspects in designing successful high-performance embedded systems are efficient PCB design, meeting power requirements and an effective cooling solution. Since heat reduces system life, long-term reliability is a direct result of effective cooling solutions to heat dissipation issues. OEMs that partner with their solution providers early in the design and integration process will find that a variety of cooling techniques are available. Often, innovative solutions in heat dissipation issues create an improved operational envelope and lower total cost of ownership for those systems.

A successful cooling solution meets or exceeds the performance goals of the OEM’s design. High-reliability designs must thrive in high-temperature environments while maintaining sufficient processor throughput to meet the computational demands of the embedded application. Market forces are driving designs to optimize both price and performance, and to continue the trend toward full CPU utilization for a given deployment. Thus, it is important for thermal management solutions to be capable of handling worst-case loads.

Designers of high-performance embedded systems have

also been challenged by increasing power consumption requirements, another driver of thermal management solutions. The incorporation of Intel’s Pentium M CPU into small form-factors such as PC/104 has pushed PC/104 solution power envelopes to new levels. In addition, full utilization of Intel’s Core Duo processor will more than double the 15W maximum power consumed by the previous Pentium III generation.

Traditional convective and conductive cooling solutions may not be adequate to meet the new demands and will end up costing more over the life of the system. Ineffective mounting and planar alignment problems degrade the performance of most thermal management solutions. By bringing aerospace technologies to address this issue, system life can be extended and both time-to-market and the overall cost of ownership can be reduced.

Traditional Thermal Management Solutions

One popular method for traditional conductive applications is the deployment of a flat-surface heat-spreader. Such a spreader is designed to make thermal contact with the CPU and other support chips, providing a highly heat-conductive path from the contacted chips directly to the chassis wall. Unfortunately, these one-piece, rigid solutions have the undesirable result of transferring the shock from an impact on the chassis directly to the CPU die. Additionally, in order to accommodate interior flat chassis wall mounting, heat-conductive solutions must adapt the board topography to a flat planar mounting surface.

For heat-convective applications that allow active cooling, the traditional extruded heat sink can be improved with the addition of an appropriately sized fan. Cooling solution attachment points are usually required on each chip to accomplish this. In the case of a Pentium M system, individual heat sinks for the CPU and the highly integrated 82855GME chipset are required. However, the ownership costs of maintenance—e.g., renewal of worn fans and cleaning of contaminated heat sinks—may only be acceptable to OEMs in certain market segments, or during early design, proof-of-concept creation and prototype phases.

In cases where traditional extrusion offers insufficient performance, pin-fin heat sinks can be considered. Integrated combinations of pin-fin heat sink and fan, or fan-sinks, are available to obviate the problem of bolting fans to a pin-fin heat sink. For example, Advanced Digital Logic’s fan-sink combines a flat motor design with a customized pin-fin heat sink base (Figure 1). The fan-sink solution can dissipate more than 30W when the full base area is in contact with the heat source, in a volume of space 37% smaller then equivalent traditional extruded heat sink solutions.

This fan-sink also protects the bearings, the most common point of failure in mechanical devices. By directing air flow through the central opening in the impeller, and therefore through the motor, the precision double ball bearings of the fan-sink are placed directly in the path of the inlet air flow. This keeps the bearings cooler than conventional brushless-DC motor designs, in which air is drawn around the motor, leaving the bearings shielded in a zone where heat can concentrate.

Enclosure Considerations

Once conductive thermal solutions have transported heat from the PCB to the chassis wall, convective solutions are often required for drawing heat away from the chassis wall. Chassis exteriors are often ribbed to induce natural heat convection over very large surface areas. The chassis can also be outfitted with a secondary thermal interface to carry heat into the structure on which it is mounted. OEM designers are free to leverage the moderate complexity and larger surface area of the chassis exterior.

To keep chassis design simple and cost-effective, a flat interior interface surface is desirable. With traditional thermal solutions, the chassis surface must be planar with the PCB or die face for optimum heat transfer. In reality, this planarity is practically impossible to achieve. Extra work, design cost and installation cost are often required just to meet minimum requirements. For example, accommodating PC/104’s mechanical stacking and interior thermal attachment is accomplished with ten loose-fitting through-holes, drilled through the flat interior area of the chassis. For sealed chassis, precision drilling, gasket washers and counter-sinking of exterior holes may be deployed to produce a gas-tight system.

A flat mounting surface is also commonly found at the bottom of the chassis system. In such a case, a PC/104 stack is then constructed off the pin-side of the PC/104 interconnect system. In many cases, this type of mounting has the added benefit of locking connector headers in place when the CPU is mounted.

In all cases, a conductive solution that adapts the board topography to a flat chassis planar mounting surface is required for optimal performance and reliable system longevity.

Planarity Control Problems

Since the inside wall of the OEM chassis is a flat plane, all attachment points provided by the CPU must match this plane. To meet this need, it is necessary to resolve the uncontrolled nature of the CPU and chipset planes that source the heat supply of the thermal conductive interconnect.

In Flip-Chip Ball-Grid Array (FCBGA) chip packages, the thermal interface plane height varies with respect to the plane of the ball contact to the PCB (Figure 2). Solder ball plane-to-top surface height ranges are published for FCBGA-packaged chips delivered by Intel and other suppliers. Along with the published ranges, a friendly disclaimer that reflow processing can affect both installed height and plane alignment is included for design considerations.

Considering that both the Pentium M CPU and the popular 82855GME chipset are housed in FCBGA packages, it can be concluded that there is also an uncontrolled planar relationship between the thermal interface points of each of the die.

In one-piece heat-spreader solutions, the spreader must be mounted so that contact with the CPU die is maximized. Use of a high-quality thermal compound is a must, because this mounting often results in a small gap of about 0.005 in. between the chipset’s die and the bottom of the heat-spreader. Several re-workable two-part gap filler compounds are available to provide the heat-conductive path to the support chip in a manner that does not increase mechanical stress on the board.

Once mounted, a single-piece heat-spreader defines the mounting plane that will attach to the inside wall of the enclosure. For example, the standoffs used for PC/104 structural mounting must share the same plane as that of the enclosure. Achieving this may involve machining overly long standoffs to fit a particular unit, or adding small precision shims to raise short standoffs to the plane of the heat-spreader.

Achieving maximum performance with a one-piece heat-spreader requires attention to detail and patience. By performing these operations before shipment and providing fitted standoffs, the OEM receives an out-of-the-box solution that can be bolted on to the end application with minimal effort. By designing this into the solution, the OEM does not need to perform the alignment process in the field.

When designing one-piece heat-spreaders, it is also critical to choose the nominal height of the spreader in relation to commonly available standoffs recognized by the PC/104 standard. Research shows that 15 mm is the predominant inter-board spacing in European integrations, while 0.600 in. has become the Society of Automotive Engineers (SAE) equivalent. The nominal difference between the two is roughly 10 mils, or 0.010 in. This is significant because original board planarity should be maintained to one-tenth of this difference, or 0.001 in. Adapters may be required to satisfy both SAE and European customers.

Gaining Control

The total cost of ownership is thus impacted by the detailed manufacturing required to achieve reliable performance with one-piece heat-spreader solutions. Since each spreader is custom-fitted to each CPU, there is little opportunity to benefit from economies of scale. This results in a significant per-unit cost.

A new approach to multi-chip conductive thermal management can improve and standardize European and SAE dimensional differences, reduce installation cost, lower the total cost of ownership, stabilize the Gaussian performance characteristics across all installations and improve reliability. To achieve these goals, Advanced Digital Logic is partnering directly with its customers’ PCB layout teams to implement a superior attachment method for the thermal solution.

After considerable review and parallel testing, the final solution (Figure 3) is a highly reliable, RoHS-compliant, soldered interconnect heat pipe solution. Capable of handling the maximum heat loads of Intel’s Core Duo processor and advanced chipset offerings, this flexible solution meets all of the design goals. It is also backward compatible for OEMs with existing designs based on one-piece conductive interfaces.

Each chip to be cooled is attached to one of the adapter plates. The adapter plate makes contact with the CPU or chipset die and each is allowed to independently match the final plane of the assembled chip. As is the case with Intel’s desktop options, the motherboard’s fiberglass is utilized as the tension mechanism. Full planar adjustment for each device is possible through tri-point mounting.

Differences in planarity are compensated for by the adiabatic section of the individual heat pipe, which loops above the chip to connect with the main thermal interface plate. Sufficient flex of the heat pipe easily accommodates differences in board-to-chassis planarity, so that any standoffs from 15 mm (0.5905 in.) to 15.875 mm (0.625 in.) may be utilized.

NASA has studied the reliability of heat pipe assemblies for deployment aboard spacecraft, in particular to regulate the thermal environment of DSP devices. NASA’s results conclude not only that heat pipes accomplish the design goal, but that their reliability is directly correlated to the sealing process utilized.

Highly reliable sealing technology and the deployment of a sintered powder-metal wick allow freeze-thaw stability and zero-gravity operation. The CPU unit can be mounted in any orientation and in any environment, from a static factory floor to an acrobatic aircraft.

For the supplier, flexible heat pipe solutions reduce assembly time via their self-adjusting and simple fitting procedures. The OEM can choose structural standoffs, simplifying the supply chain. The ability to field-swap a CPU unit onto an existing heat pipe unit also contributes to lower cost of ownership compared to one-piece conductive interfaces.

OEMs that partner early in the integration process with their suppliers to achieve effective thermal solutions will find that they can save considerable cost over the life of their systems. Traditional thermal solutions may be inadequate at handling the power requirements and the heavy loading of new system designs. Mounting and planar alignment are real issues that need real solutions. These solutions are available in the form of bearing-cooled fan-sinks, reliable sealing technology and flexible heat pipes.

Advanced Digital Logic
San Diego, CA.
(858) 490-0597.
[www.adlogic-pc104.com].