BROWSE BY TECHNOLOGY



RTC SUPPLEMENTS


SOLUTIONS ENGINEERING

Cooling Large Systems

VITA 48/REDI: Keeping the Future Cool

As high-end embedded systems continue to push the heat envelope with their thirst for more performance, VITA 48/REDI is poised to provide the infrastructure support needed for keeping them cool.

ALAN STORROW, RADSTONE EMBEDDED COMPUTING

  • Page 1 of 1
    Bookmark and Share

The advent of VPX, formerly VITA 46, in the VMEbus world brings with it some significant strides forward that will deliver state-of-the-art performance to the high-end embedded computing applications of today and tomorrow. Specifically, through its adoption of the 6.25 Gbit/s Tyco Multigig RT2 connector, it brings support for serial switched fabrics together with other high-speed serial I/O technologies, such as digital video, serial ATA and FPGA interconnects. At the same time, VPX offers complete electrical and dimensional compatibility with existing 6U and 3U VME products. VPX boards are, therefore, VME boards as we know them in all important respects except one, the connector.

However, as it has always done, this increased processing power and throughput capability comes at a price: the additional heat that must be dissipated. Serial switched fabrics are, in themselves, a technology response to the need to provide ever higher bandwidth interconnects between ever more capable processors. Not only does the current generation of processors dissipate more heat than their predecessors, but the very high speeds at which the fabrics are driven also adds to the total thermal load.

Enter VITA 48/REDI

The arrival of VPX has, then, refocused attention on power dissipation and the requirement for cooling paradigms and technologies that enable systems designers to leverage the inherent capability of today’s embedded commercial components. In parallel to the development of the VPX standard is the work that has been undertaken on VITA 48, now known as Ruggedized Enhanced Design Implementation (REDI).

There are three important things to understand about REDI. First, while it describes a standard for creating rugged systems wherever they may be deployed, this standard is primarily driven by unusually harsh environments such as those found in military and aerospace computing. It is also driven by the fact that no commercial standards have yet addressed this requirement, just as they have not addressed requirements to deal with shock, vibration, humidity and ambient temperature extremes.

Second, REDI is a mechanical standard that is entirely independent of the VPX electrical standard. Third, although REDI is unquestionably highly complementary to VPX and the two are often discussed in the same breath, REDI should be viewed as a discrete standard that is highly applicable outside the VPX world.

REDI is thus a standard for ruggedization, which means that it deals with all aspects of creating high-performance embedded systems for environments that are far from benign. It describes board area and module volume, as well as structural ruggedization. It specifically addresses two-level maintenance, an approach that delivers greater logistical simplicity and the resulting substantial cost saving. Finally, it deals with thermal management.

How REDI Addresses Cooling

How, then, does REDI address the cooling issue? As with all VITA standards, VITA 48 comprises a number of disparate but related elements (Figure 1). It is expected that all will have been approved by the end of 2006. In its comprehensiveness, REDI goes beyond the work previously done by the VITA 34 working group. While VITA 34 was designed to accommodate liquid cooling, VITA 48 addresses it specifically. Because REDI deals with all forms of cooling within a single standard, it embraces common mechanical design approaches. This makes it simpler for manufacturers to create single board designs that can be cooled by convection, by conduction or by liquid, an important step forward.

One of the first things that REDI does is increase the maximum pitch between modules from the IEEE 1101 standard of 0.8 in. to 1.0 in. This increase is designed to do three things. First, it allows for the greater PCB thickness demanded by today’s sophisticated multi-layer boards, which can reach up to 24 layers. Second, the greater pitch makes it possible for more thermally demanding components to be placed on the secondary side of the board. Third, it provides a number of cooling options to board manufacturers and systems designers. Perhaps the most important of these options is the ability to fit side covers that can be used for either conduction or liquid cooling (Figure 2).

These can be used for cooling components immediately adjacent to them. Alternatively, the secondary side plate can be thermally linked to a primary side plate for increased heat transfer. An additional benefit of the ability to apply two plates to a board is that provision can be made for protection against electrostatic discharge (ESD), complementing the EDS protection that is inherent in the design of the Tyco connector. This allows in-field handling of the board and thus enables the implementation of two-level maintenance practices.

The Need for Liquid Cooling

Conduction cooling has historically been the preferred cooling approach for applications such as those found in military and aerospace environments. However, the increasing demand for more processing power and its concomitant effect on cooling is driving an increase in the adoption of liquid cooling methods, which REDI addresses.

To some extent, liquid cooling has thus far been a solution looking for a problem, since the applications and environments that generate the level of heat where liquid cooling becomes mandatory have been few and far between. Although such methods have been around in military/aerospace environments for several decades, the exceptions of yesterday are becoming the commonplaces of today. While VMEbus slots were limited to a maximum of 90W at 5V, REDI envisages a future in which power per module will be on the order of 200W, rising to 500W for liquid cooling.

Challenges of Liquid Cooling Methods

The lack of a mainstream requirement for liquid cooling, including spray cooling, has meant that the liquid cooling industry has been somewhat fragmented, with incompatibilities among different vendors’ offerings. Previously, boards that would be liquid-cooled were often designed specifically for that environment, making these boards more expensive than their more conventional counterparts. A key benefit of REDI, via VITA 48.3, is that it will introduce a commercial standard, leading to greater degrees of interoperability, increased competition and improved price/performance.

Although REDI includes a provision for spray cooling, it is anticipated that liquid flow-through (LFT) cooling will likely be the more common approach, not least because of problems with material compatibility and the maintenance of “wet” electronics. However, in the medium to long term, spray cooling has significant potential. For LFT, REDI

describes a paradigm in which the primary side cover has liquid flowing through it. There is no reason why the secondary side cover could not also be so configured, although coupling it to the front side cover is a more likely scenario. Manufacturers are evaluating different schemas for the liquid to pass through the plate in order to maximize the plate’s efficiency.

The lack of interoperability in liquid cooling solutions often forces users to employ one or another proprietary solution to the exclusion of all others.

Perhaps the greatest contributor to this lack of interoperability has been the lack of commonality in the design of quick disconnects (QDs), which provide the critical physical interface between the liquid distribution rail and the liquid flow-through plate. REDI defines a standard in which the QDs from various vendors will be capable of interconnecting.

Liquid-Cooled and Hybrid Chassis

Radstone has provided liquid-cooled ATR chassis for a number of years (Figure 3). These chassis are designed to accommodate the high pressures normally associated with liquid-cooled infrastructures. The materials, coatings and construction techniques are compatible with the standard range of coolants, which are polyalphaolefin, water/glycol or fluorinated liquids. The coolant is coupled into the chassis by means of no-loss quick disconnect connectors.

These liquid-cooled chassis are designed to handle high-power systems of 1 to 2 kW, populated with conduction-cooled circuit cards. However, they can easily be modified to accommodate liquid flow-through-cooled VITA 48.3 modules by extending the fluid path to interface to the blind-mating quick-disconnect liquid connectors of the modules.

It is anticipated that many hybrid chassis will be developed that allow populations of legacy conduction-cooled VME and VPX boards to be mixed with new VITA 48.3 modules. These modules will be designed to meet the significant increase in power dissipation of the next generation of DSP applications, especially radar processing. These systems will be expected to handle thermal power dissipations between 2 and 4 kW.

A 1ATR Short VITA 48.3 concept chassis being developed for this purpose has a liquid path through the side walls and entry points into the main chassis for coupling to VITA 48.3 modules (Figure 4). Further work and study are required before the concept chassis can be characterized for backpressure, thermal performance and balance of liquid flow through the side walls and modules.

Of course, in liquid cooling as with any other form of cooling, heat still has to be removed from the enclosure. An approach that will likely come into its own is the use of an external heat rejection unit. It combines a pump, heat exchanger and associated control systems in a single enclosure capable of dissipating 1 kW. This enables the deployment of a complete, self-contained liquid-cooled subsystem that includes an ATR enclosure and heat rejection unit.

There are a number of attractions to the use of an external heat exchange unit. First, it allows liquid cooling technology to be inserted where no liquid cooling infrastructure had previously been provided. Second, it allows the system designer greater flexibility if two smaller boxes need to be accommodated within the platform rather than one larger one. Finally, it also allows the rejection unit to be installed in the best place for transferring the heat away. However, the ongoing emphasis on smaller, lighter platforms in high-performance systems such as military/aerospace will continue to put pressure on cooling solutions to reduce size and weight, since those must be factored in as part of the total solution.

The REDI standard can be seen as a vital contribution by VITA to the continued success of commercial solutions in high-performance embedded systems. Power and heat are becoming increasingly important considerations in semiconductor and processor design, with even Intel focusing more on the overall balance between raw processing power and heat dissipation. Nevertheless, high-end embedded systems continue to push the envelope with their unquenchable thirst for more performance. That thirst would remain unsatisfied were it not for the parallel availability of pragmatic, commercial cooling solutions. The REDI standard will be instrumental in providing the infrastructure support necessary for the intensive computing requirements of the future.


Radstone Embedded Computing
Billerica, MA.
(800) 368-2738.
[www.radstone.com].