TECHNOLOGY IN SYSTEMS
Optimizing Size, Weight and Power
Getting the Heat Out of Ever Smaller Systems
As systems shrink in size and gain in computing power, all the techniques in the book are needed to handle the heat. Interestingly, there are design techniques that can increase efficiency and reliability, while reducing system size and increasing performance.
MIKE JONES, ADLINK TECHNOLOGY
Page 1 of 1
Demand for performance in computer systems continues to grow, and the embedded systems used in mobile or exposed applications are no exception. What is exceptional in these applications is the environment in which they operate. When temperatures are high, or weight and size are costly, cooling these computer systems becomes more difficult. In small form factor (SFF) designs, building systems that focus on external rather than internal standardization allows for more flexibility in achieving these thermal efficiencies. Better cooling can be achieved by using the latest, lower power processing components and by orienting them, as well as other heat-generating components, so they connect directly to the dissipating surface—and paradoxically, by reducing system size. These mechanical steps enable lowered thermal resistance via shorter paths and fewer thermal junctions.
Smaller, Faster . . . Hotter
Onboard processors in transportation systems for control and public safety, and roadside installations of monitoring equipment as well as unmanned vehicles for commercial, industrial, and military applications, all have increasing demands for wide, high-speed data, more and higher resolution sensors and greater processing power. The ability of these platforms to provide the cooling necessary for reliable high-performance operation is critical. Unmanned military vehicles are a good example of the need to improve computer system efficiency. These vehicles can be much smaller, so the systems that comprise their capabilities need to be smaller, too.
Systems need to provide connectivity as well as the wide, high-speed I/O required to support visible spectrum and infrared (IR) cameras, radar and other fast, high-definition sensors. At the heart of these systems is the processing power (CPUs, GPGPUs, FPGAs) required to process that data for object detection, classification and tracking. In applications like these, every bit of weight and volume that can be removed has the potential to improve the range, capabilities or cost of a deployed unit. To this end, engineers must consider the function of every cubic centimeter of space and each gram of weight. All unnecessary space should be squeezed from the system to reduce its size, and any unnecessary mass should be eliminated. Because dissipation of the thermal energy these systems produce requires space, making the electronics portion of a system smaller leaves more room available for thermal dissipation.
At the same time, there are standards that define the entire scope of system architectures. These standards provide uniformity and modularity at various levels of a system. They describe the size and shape of internal components, as well as the outer size and shape of the system itself. All of these requirements put pressure on the system designer to improve efficiency. Considering what improvements can be made to keep up with increasing performance demands in smaller platforms while keeping the benefits of physical modularity for serviceability, would seem to be a tall order, but we find that there are multiple opportunities to improve efficiency. System size and weight can be reduced while improving thermal efficiency. Standards from legacy systems—which offer no advantage in modern applications, yet still have a cost in size and weight—can be replaced by more efficient versions to better align with current requirements, allowing for more efficient system design.
Smaller Size Demands More Efficiency
The increasing pressure for lower COTS cost and the continuing desire to reduce size, weight, and power comes without a reduction in the importance of reliability or performance. Taken as a whole, the array of design criteria (SWaP2C2) can be reduced to just a few key elements, with all others being derived from them.
Size reduction is the number one goal of SWaP. Happily, when you find a way to put the same functions in a smaller package, the weight, and to some degree the cost, will go down. The value of all parts of the design must be considered. Any component of the system that can be eliminated or made smaller without reducing performance should be eliminated. One part often overlooked is empty space. While it can have a purpose, that purpose should be well understood so, where possible, empty space that doesn’t serve a purpose can be removed.
Another consideration is the material that makes up the system: the structure of the enclosure, the thermal shunting, stiffeners, mounting plates, covers, etc. These parts contribute to the weight of the system. Improvements in electrical efficiency will either increase performance or reduce power requirements or both.
Electrical efficiency is chiefly driven by the evolution of processors. Decreasing junction sizes and lowering operating voltages provide continuous improvement in power consumption for a given level of performance (Gigaflops per Joule).
As for thermal efficiency, ΔT between the critical heat producing components and the system’s dissipation interface must be minimized. This can be done by using conduction materials with lower thermal conductivity, reducing or eliminating thermal junctions, and/or reducing the length of the conduction pathways. Improvements in thermal efficiency allow for the use of higher performance processors or improvements in reliability resulting from a reduction in component operating temperatures.
So, because weight, cost, power requirements, cooling, reliability and performance all hinge on size, electrical efficiency and thermal efficiency as the root drivers of SWaP, as a whole (or, SWaP2C2E2R), when we achieve these goals, the other elements of SWaP will follow (Figure 1).
In SWaP2C2E2R, electrical efficiency, thermal efficiency and size are the true basis for the equation, with all remaining properties derived from these three core elements.
As we have said, many standards work together to provide uniformity to the infrastructure of a system. The popular 3U VPX (VITA 46) systems are a good example. The suite of specifications defining VPX covers every detail. This includes internal connections, connector types, and the form factor and clearance of PC boards and thermal shunting. These relatively compact solutions are currently providing high-performance computing in mobile platforms. Systems built on the 3U VPX form factor have evolved to keep up with the latest bus speeds of modern CPUs while providing good SWaP performance.
VPX specifies the infrastructure in which the critical, high-speed electronics operate, including a robust bus scheme featuring rugged bus connectors. VPX also provides an infrastructure of bladed expansion cards. This modular bladed architecture has the ability to be easily expanded and reconfigured to accommodate changing requirements. With removable and reconfigurable cards, VPX systems are ideal for in-lab system design and prototyping. This approach makes heavy use of internal thermal conduction. Boards use thermal shunts, which guide the cards into card cage slots, and provide a thermal pathway from the components on the board to the enclosure body (Figure 2). These shunts are made from aluminum or copper alloys with high thermal conductivity. Choosing these materials is important, because in a 3U VPX system, the path from the components through the thermal shunts and out to the dissipating surfaces is fairly long.
VPX systems use a fairly long thermal pathway to guide heat off of components and boards.
This is not ideal, but rather the price for modularity. It’s not ideal because heat is more efficiently removed by conduction, when the heat source is placed as close as possible to the dissipating surface. This forces a smaller ΔT because a much greater portion of the heat flux is present at the dissipating surface. When the heat flux is directly coupled to the dissipating surface, not only is the heat transfer more efficient, but there is also less radiation of heat back into the system. Heat re-radiating from large internal thermal shunts creates an oven for your components. So paradoxically, when internal conduction is reduced, ΔT is reduced because total thermal resistance is lowered, and internally radiated heat is minimized. Giving heat the bum’s rush in this way lowers component temperatures. Lower component temperatures translate directly into improved reliability and/or higher performance. Therefore, heat shunting, while required in a bladed design such as VPX, can be drastically reduced in a VITA-75 form factor SFF design (Figure 3).
VITA-75 can drastically reduce the length of heat shunting required over that of VPX.
Contrary to the popular saying that newer processors demand better cooling, the latest processors actually reduce thermal requirements. This erroneous thinking comes from the fact that modern processors provide the ability to operate above thermal design power (TDP) when the thermal environment allows. Because of this, they are capable of much higher performance, which produces higher thermal loads. This “allows” for designs capable of dissipating more heat to take advantage of peak performance, but the thermal loads of processors should be comparable for a given level of performance. In addition to the ability to operate above 100% TDP, modern processors continue to shrink in size. Intel’s 3rd Generation Core i7 CPUs use a 22nm process. That combined with lower voltage operation provides reduced thermal dissipation requirements for any performance level while operating at normal TDP. When compared properly, modern processors continue to improve on efficiency, and because of these improvements, for any given performance requirement, modern processors actually reduce the thermal requirements of a platform and increase reliability.
So, even if your applications don’t require higher compute performance, moving to the latest CPUs and GPUs will keep your system cooler and more reliable. The thermal efficiency improvements of eliminating thermal shunts come from the lowered thermal resistance of shorter paths and fewer thermal junctions. In SFF designs, this is accomplished by orienting heat-generating components so they connect directly to the dissipating surface. By externally conforming to the VITA-75 form factor, but eliminating the internal card cage and backplane, ADLINK’s HPERC eliminates the SWaP costs caused by over-constraining standardization internal to the system.
Compact SFF solutions often use internal boards based on COM Express, QSeven, PC/104, or custom form factors. Standardizing on external form factors provides the correct solution to mobile platforms where replacement, upgrade and reconfiguration are best accomplished at the box level. This simple redesign directly improves the size, weight, reliability, performance and even cost of system over a VPX solution.
San Jose, CA