Future cooling concepts for HPEC programs

In High Performance Computing (HPC) data centers, more powerful processors – with their greater heat output – are routinely accommodated by simply turning the air conditioning up a notch. However, high performance computing has now moved out of the data center and into the embedded world where cooling is not such a simple process.

High Performance Embedded Computing (HPEC) is at the heart of many current military programs; whether it’s mission-critical corporate computing or mission-critical military computing, heat is the enemy because it limits processing power and can cause system unreliability and even failure. Turning up the AC isn’t an option in the embedded world.

The problem is compounded on two fronts. Military embedded computing is becoming increasingly sophisticated and requires the most powerful processors. At the same time, those powerful processors are being deployed in environments that are constrained in terms of Size, Weight, and Power (SWaP) – and SWaP-constrained systems are notoriously difficult to cool.

That’s why there’s a growing focus in the military embedded systems world on next-generation cooling technology. Many embedded processing systems today use conduction cooling, where heat is conducted away from a component to the card edge and out to the chassis, which acts as a heat sink, dissipating the heat into the air. Forced-air convection cooling may speed the process, but fans add weight and increase the potential for failure. Although much thought has gone into the design of heat sinks and the development of the thermal interface materials that conduct heat from the component to the heat sink, the relentless growth in power density and heat output limits the effectiveness of traditional cooling methods.

Innovative cooling doesn’t just mean that faster processors can be deployed. The U.S. Defense Advanced Research Projects Agency (DARPA), which is driving much of the work in advanced cooling systems, envisions large reductions in SWaP if the problem can be solved by integrating cooling systems at the chip level. SWaP has transitioned to SWaP-C, where C stands for cooling. Some of DARPA’s research into new cooling technologies has been undertaken in collaboration with GE’s Global Research Center and has resulted in innovative approaches that could find their way into the military embedded computing systems of the future. More information on the outcome of that collaboration can be found at http://opsy.st/1dQHyUe

Researchers classify cooling technologies into “remote” or “embedded” paradigms. Remote cooling conducts heat out of a chip to a heat sink; embedded cooling designs cooling right into the chip. DARPA’s Thermal Management Technologies (TMT) program focused primarily on the first approach; the agency’s more recent Intrachip/Interchip Enhanced Cooling (ICECool) program advances the second approach. Both strategies apply micro- and nano-scale engineering to enhance heat dissipation at their respective levels.

Miniaturization

The latest prototypes take both conduction and convection cooling to their technological extremes, with tiny pumps, fans, pipes, and valves. On the remote cooling side, for example, a thermal interface has been developed that minimizes the heat effects of thermally mismatched materials used in the processor, such as silicon – and the heat sink, such as copper. One new material sandwiches high-conductivity, nano-scale copper “springs” between layers that match the thermal characteristics of the heat source and the heat sink, respectively. The copper’s conductivity and geometry reduce thermal stresses in the heat path.

On the embedded cooling side, companies have designed tiny pipes into chips for microfluidic cooling. Microchannels in the chip, for example, could take in a continuous flow of chilled fluid from a network built into a computer and carry away heat by evaporation and convection. *

Between the remote and embedded cooling poles, researchers have developed a two-phase “vapor chamber” heat-transfer system that can be squeezed into a cavity in a multichip module substrate. Fluid in this thermal ground plane cavity – like a miniature weather system – absorbs the heat, converts to vapor, condenses against a cold wall, and flows back to the hot section via capillary action induced by the micro-/nano-engineered internal surface of the case (see Figure 1).

21
Figure 1: A Thermal Ground Plane (TGP) is a micro-/nano-engineered “machine” that fits into a cavity in a multichip module.
(Click graphic to zoom by 1.9x)

While these devices are not in production yet – and some of them probably never will be – they represent significant R&D trends. Dealing with heat more efficiently is an urgent need, and these new cooling technologies could revolutionize military embedded computing.

defense.ge-ip.com

* “DARPA’s Intra/Interchip Enhanced Cooling (ICECool) Program,” by Avram Bar-Cohen, Joseph J. Maurer and Joanathan G. Felbinger, May, 2013, page 172, published in association with the CS MANTECH Conference in New Orleans.