By Randy Banton
The trend of escalating power consumption of processors and communications infrastructure chips presents an increasing challenge for deploying high-performance processing systems in military environments. This challenge exists regardless of the cooling method: air-cooled, conduction-cooled, spray-cooled, or liquid-flow-through systems.
The existing 6U printed circuit board and enclosures infrastructure was designed more than 20 years ago with significant headroom for power distribution and heat dissipation. Yet, relatively recent escalation of power consumption of many components has stretched that infrastructure to the breaking point in high-performance systems.
There is a new 6U infrastructure for air cooling and conduction cooling, however, which is extensible to 3U systems, and spray-cooling and liquid-flow-through cooling systems. This infrastructure will enable broader use of commercial off-the-shelf (COTS) systems in deployed environments that are today still implemented using proprietary methods, since COTS solutions that provide two-level maintenance and/or liquid-cooling are not available.
For forced-air systems, Mercury Computer Systems in Chelmsford, Mass., developed a patented technique called ManagedAir that applies to a 6U or 3U form factor. A board cover, which creates a defined and controlled plenum, mates with the chassis infrastructure to increase the efficiency of air-cooling methods. This method also provides significant enhancement to the structural integrity of the module, as well as the primary means of physical protection for two-level maintenance.
Conduction-cooling challenges
The challenges of hot processors, memory subsystem, and fabric interconnect components is not limited to air-cooled systems. Conduction-cooled systems are also nearing the limits of cooling capabilities within the confines of current specifications. Current standards-based, conduction-cooled systems are limited to approximately 100 watts per slot 75°C to 85°C card-edge temperatures.
The emerging VITA 48 standard will address many of the constraints now found in the use of IEEE 1101.2 specification for high-power density designs. For applications that require more performance within a strictly limited physical space, IEEE 1101.2 has become restrictive in many applications.
Goals of the VITA 48 standard include making the back side of the board usable for more components and increasing the thermal transfer capacity from the card edge to the side rails of the chassis. The VITA 48 specification is also expected to create the first COTS standard for liquid-flow-through (LFT) cooling, spray cooling, and similar designs, and to do it under a unified specification for advanced cooling with air-cooled and conduction-cooled designs.
Making the back side usable
The VITA 48 specification defines one method of placing boards on a one-inch spacing, with significant parts of the space allocated to the back side of the board. For high-performance systems, the back side of an IEEE 1101.2 board, which has on 0.8-inch spacing, has become almost unusable for components requiring thermal management.
One reason is the need to increase the printed wiring board thickness, whether to accommodate more layers for routing or to provide increased thermal conductivity through heavier copper layers or even copper cores, without resorting to exotic board materials. The routing pressure comes from relatively high-density components in fine-pitch ball-grid-array (BGA) packages that provide no room between the balls for routing signals. The majority of components are now in fine-pitch BGA packages including memory devices.
The need for heavy copper layers derives directly from the higher power dissipation of current components. Even though most component package types have the die facing away from the board so that only a small percentage of the power dissipates toward the board, the absolute value of the power is great enough to make the board a significant potential conductor of the heat.
Heavy copper traces help dissipate the heat, and cores of copper or other materials help even more. Because this additional thickness reduces the available height of the secondary side of the board, the back side of many board designs has become unusable.
A second factor making the secondary side of the board unusable for significant numbers of components in conduction-cooled designs is the lack of a cold plate for that side of the board. In IEEE 1101.2 boards, adding a cold plate appears to be counterproductive as it further reduces the height available for components. By applying a wide spacing of VITA 48 modules, a thermally significant cold plate can conduct heat from the components on the secondary side of the board.
Increasing thermal transfer capacity
The primary method of transferring heat from the card to the chassis is from the card edge to the chassis sidewalls. The thermal transfer capacity can improve by increasing the area in contact or by increasing the contact coefficient, which is a function of pressure and surface properties such as finish.
The IEEE 1101.2 defines wedge locks measuring one-quarter inch on a side. Compared to larger wedge locks, the IEEE 1101.2 wedge locks provide a small percentage of the thermal transfer capacity, mostly due to the lower pressure maintained at the primary conduction path contact. Increasing the slot spacing allows for increasing the wedge lock size, which in turn makes way for a larger center screw to expand the wedge lock. The larger center screw can increase pressure on the surface contact, resulting in an increase of the thermal transfer capacity of up to two times.
Higher performance density
The ManagedAir solution can provide ample cooling to support as much as 50 to 100 percent more processing power in a small-volume chassis. A good deal of these gains comes from the increased cooling ability of finely managed air. Many gains also come from additional performance density allowed on the board by using board covers that reduce the number of pads and size of keep-out areas previously reserved for heat-sink mounting.
The conduction-cooled solution reclaims the back side of the board for many component types. Depending on the board design, this typically increases the total usable area on a 6U board by at least 25 to 50 percent. The increase thermal transfer capacity from the larger wedge locks and increased pressure can see gains that double previous solutions.
Liquid-flow-through cooling or spray-cooling should further increase the per-slot capabilities to the 400-to-600-watt-per-slot range. Having the back side of the board available to place high-performance components and their resulting power dissipation is absolutely necessary.
There is another important effect of this cover methodology for air-cooled, and the cold plates that surround the conduction- or liquid-flow-through boards-the provision for electrostatic discharge (ESD) protection to satisfy a key requirement for two-level maintenance.
This too has been an impediment to using COTS products in deployed military applications. When used in combination with an appropriate ESD protected connector system, a complete COTS infrastructure for two-level maintenance can be provided.
The increasing power dissipation of processors and interconnects also drives more applications across the line from air-cooled or conduction-cooled to spray-cooled or liquid-flow-through designs. A design goal of VITA 48 is to create the first standard for spray-cooling and liquid-flow-through designs, and to do it under a unified specification with air-cooled and conduction-cooled designs.
Randy Banton is vice president of Defense Electronics Engineering & Applications at Mercury Computer Systems in Chelmsford, Mass., where he is responsible for product development and applications insight for the company’s defense electronics engineering and applications area.