Breakthrough cooling scheme crashes 30-watt-per-slot barrier

Feb. 1, 2004
Designers of embedded systems for military and severe-environment applications face a painful dilemma today.

By Doug Patterson, director of marketing, Vista Controls

Designers of embedded systems for military and severe-environment applications face a painful dilemma today. While the latest and greatest silicon offers extraordinary advances in speed and integration in line with Moore's law, these designers frequently are unable to make full use of that silicon because of thermal-management issues.

Microprocessors running at or near 1 GHz, to begin with, can put a strain on the cooling capabilities of even commercial systems operating in benign environments. Yet in more environmentally rigorous environments, functionally dense processor boards threaten to exceed the power and heat limitations of standard backplane architectures. While today's leading-edge single-board computers allow designers to host applications on ever-smaller systems with fewer and fewer boards, the result is ever-higher power density. This presents designers with a painful choice: either settle for less than state-of-the-art capabilities or implement some exotic, expensive, and doubtlessly nonstandard cooling technique.

Vista Controls, a Curtiss-Wright company in Santa Clarita, Calif., has developed a thermal-management scheme, Enhanced Cold-wall Thermal Interface (ECTI), that handles the heat dissipated by the latest and greatest chips and boards, and it does so without violating Eurocard mechanical standards. The physics of thermal dynamics limit today's conduction-cooled IEEE 1101.2 Eurocard boards to a maximum of approximately 30 watts in one slot. ECTI technology, however, can handle boards dissipating upwards of 60 watts.

The genesis of ECTI

ECTI came from Vista Controls' efforts to design conduction-cooled 6U VMEbus boards called the PPC G4D family. One design called for a pair of Motorola G4 Series PowerPC 7455/7457 RISC microprocessors with integrated AltiVec accelerators, a certain complement of memory and I/O, and a pair of PMC expansion sites.

The power dissipation of the two 12.8-watt 750-MHz PowerPC 7455 processors selected and the 14.7 watts of the other on-board circuitry added up to 40.3 watts — well beyond the 30-watt barrier without even taking into account the power budget of PMC boards, each dissipating as much as 7.5 watts. The processor board could have been packaged as a dual-slot solution, but such a waste of real estate is an undesirable approach.

Conduction-cooled boards for rugged and military environments conventionally carry heat away from hot components and out to the sides of the boards; the thermal path consists of a thermal-transfer medium or "gap pad," a metal heat cap, a board stiffener/thermal shunt, and copper layers or thermal planes buried within the printed wiring board (PWB). The gap pad is most commonly a small elastomeric pad containing a thermally conductive material that, when compressed, spreads out uniformly to create a consistent thermal interface between two surfaces.

When a designer inserts a board into a chassis, Allen screws under the extraction handles activate wedgelocks along the sides of the board; the wedgelocks expand to push the board tightly against the sidewalls of the chassis, securing the board firmly in place. The wedgelocks also establish the primary thermal interface between the rear of a board and one chassis sidewall.

This conductive thermal path carries heat from the on-board circuitry out to the edge of the board immediately under the wedgelock, through the back of the board to the chassis, and then to the ambient air. This primary board-to-chassis thermal interface, however, provides only about 3.4 square inches of physical area (1.7 square inches on each side of a board) for heat conduction on the rear of the board — hence the 30-watt-per-slot barrier.

Some rugged-board vendors have added secondary thermal-management techniques in an attempt not only to improve the thermal conduction capabilities of this scheme, but also to help deal with the high-power microprocessors of today. These secondary thermal-management techniques create an additional thermal path for processor heat that avoids injecting that heat back down into the board. These secondary heat paths typically consist of a compressed gap pad over the microprocessor die topped off with a heat sink and heat shunt over the top of the wedgelock.

The problem with this common approach is that the cantilevered beam of the structure causes the shunt to rise at an angle when the wedgelock expands to engage the chassis sidewall. This phenomenon loses much of the positive contact between the gap pad and wedgelock/sidewall, which badly undermines the efficiency of the thermal path from the processor die. Moreover, the necessary high degree of gap-pad compression stresses the processor die.

ECTI technology uses an extended heat shunt like other thermal-management techniques, yet the shunt maintains its thermal interfaces even when the wedgelock activates; an ECTI thermal path does not degrade across numerous wedgelock expansion/relaxation cycles.

The heart of the matter

At the heart of ECTI is an extended C-shaped heat sink/shunt that conducts more than 90 percent of the heat of one or more microprocessors directly to the "cold-wall" interface of a chassis sidewall. The lip of the heat sink extends to lie atop the wedgelock and, upon wedgelock activation, presses against the opposite sidewall of the chassis to make a secondary thermal path. This keeps latent heating effects to a minimum and halves the thermal impedances on a board without breaking any Eurocard physical or electrical standards or specifications. The PPC G4D in particular can dissipate 60 watts of total on-board power to a chassis sidewall of 75 degrees Celsius while maintaining processor temperatures at less than 95 degrees C.

The ECTI C-shaped heat shunt essentially is a plated, heat-treated copper alloy with unique properties. Structurally, the bottom of the shunt attaches firmly to the processor substrate with a trio of precision set screws, while a set of factory-set precision spacers ensures the mechanical integrity and physical dimensions of the processor/gap-pad/heat-sink interface.

The activation of the wedgelock on an ECTI board does not pull the heat-sink metal up off the gap pad and microprocessor, unlike a conventional heat shunt, and does not undermine the thermal interface to a chassis sidewall.

A proprietary manufacturing technique called FlexJoint gives the bend of the ECTI heat sink and a section at the top of the shunt a special heat treatment that allows them to flex. This provides the ability to endure several flexures without suffering mechanical structure microfractures, which over time would decrease thermal conductivity.

When an activated wedgelock and raised heat shunt rise to contact a sidewall, the shunt's bend — which operates essentially as a hinge — widens to eliminate the force that raises the heat shunt off the gap pad in conventional designs.

The ECTI scheme effectively doubles the size of the thermal interfaces between the board and chassis sidewalls and handles the vast majority of microprocessor-dissipated heat independently. In fact, a battery of thermal tests that Vista officials recently conducted demonstrated that 95 percent of microprocessor heat transfers directly through the new ECTI shunt to the chassis sidewall, bypassing the board entirely.

Voice your opinion!

To join the conversation, and become an exclusive member of Military Aerospace, create an account today!