Thursday, October 1, 2015
Cool it! Heat Relief for RF Power Chips
Current cooling systems for high power density chips in today’s radar, electronic warfare and avionics systems are large, heavy and costly. The inadequacy of the state of the art in cooling threatens the growth of processor performance that the aerospace industry has relied on for the last three decades. The U.S. Defense Advanced Research Agency (DARPA) is driving research in advanced microelectronics cooling technology ultimately to design chips in which cooling is built in from the start.
Many radar, electronic warfare, communications and avionics systems are thermally limited. These systems cannot remove the heat generated by their components efficiently enough to reach the potential that might be expected from their material properties, designs, and architectures. And, as the chips that convert electrical power into Radiofrequency (RF) energy get hotter, their thermal solutions get larger, heavier and more costly.
The U.S. Defense Advanced Research Projects Agency (DARPA) is leading efforts to solve these problems by incorporating “microfluidic” cooling directly into chips or components, using an “embedded cooling” paradigm. The agency is challenging industry to come up with new ways to deal with the increased heat levels that would be generated in order to crank more energy out of Monolithic Microwave Integrated Circuit (MMIC)-based power amplifiers.
The agency’s latest program calls for the development of miniaturized cooling systems that can extract die-level heat flux —¬ the level of waste heat generated per unit area in a die — of 1000 Watts per square centimeter (W/cm2) from power amplifier chips. That is a major increase from the 100 to 300 W/cm2 level of waste heat that can be removed from state-of-the-art MMIC power amplifier chips today.
DARPA’s Intrachip/Interchip Enhanced Cooling (ICECool) Applications program aims to spur the development of these thermal management solutions in a short time. “We are generally expecting that, when the program is over, we will be at Technology Readiness Level [TRL] 4 and [that a possible] follow-on program will take us to TRL 6 or 7,” says Avram Bar-Cohen, DARPA program manager.
Contractors on the RF side of the program — Northrop Grumman, Raytheon, Lockheed Martin and BAE Systems — are working on MMIC-based power amplifiers. The companies have identified thermally limited military systems as targets for potential technology insertions.
DARPA considers that today’s remote cooling approach — in which heat is conducted through successive layers of material away from the heat source to a cold plate — is inadequate for future demand. As the devices get smaller, they generate more heat, and the thermal solutions get larger, Bar-Cohen says.
Many defense systems, such as high-end radars, are already liquid-cooled, but the liquid only comes to the cold plate, Bar-Cohen says. “We are trying to take the liquid the last millimeter from the cold plate into the chip and, by doing that, to drastically improve our heat removal capability. We are not introducing liquid cooling but using the liquid much more effectively,” he adds.
Any microwave system needs to generate microwave energy from electrical energy, Bar-Cohen says. But, in the conversion process, typically 50 percent of the electrical power fed into a radar system ends up as heat. The hotter the chip, the less reliable it becomes. Reliability drops by about an order of magnitude with every 20-to-25-degree-centigrade increment at any temperature, he says.
Today’s Gallium Nitride (GaN) RF semiconductors are limited by the ability to get heat out, agrees Dave Altman, senior principal mechanical engineer with Raytheon’s Integrated Defense Systems unit. By improving the cooling technology we should be able to move toward the electromagnetic limitations of the material, he says. Improved cooling could also enable the production of higher-performance devices at lower cost, because the chips would use less wafer area.
There is a large margin for improvement, for example, in High Electron Mobility Transistors (HEMTs) — key elements in MMIC power amplifiers. Engineers have shown that they can put out 40 Watts per millimeter of HEMT periphery, the total length of a HEMT gate structure, at least theoretically, Altman says. By comparison, these devices today output 5 Watts or less per millimeter, he says.
DARPA is pushing an embedded cooling paradigm. The program gets to what may be the ultimate destination of the “inward migration” of thermal management technology — from the air-conditioned room to the rack, and now into the component or the chip itself.
Under the embedded cooling paradigm an RF chip might be driven at higher power or might incorporate more HEMTs on the die to generate more energy out. But the path from the heat source to the heat sink would be shorter than in current designs. That means performance could increase while any rise in the temperature of the active device would still be within an acceptable range for chip reliability. The program announcement stipulates 1 million hours mean time to failure.
In phase one of the ICECool applications program the contractors demonstrated their thermal management technologies with test vehicles that simulate HEMT hot spots. In phase two they will demonstrate actual RF circuitry using their respective heat extraction technologies, while proving that components have not been damaged and that they perform as expected. These electrical demonstration vehicles could be small chips or small areas on a chip, Bar-Cohen says. The companies are changing the design of an existing chip or designing a new chip that outputs more power or produces equivalent power in a smaller footprint.
As heat loads continue to increase, thermal management design can no longer be an add-on, says Carl Creamer, technology development manager at BAE Systems Technology Solutions Group. That means thermal and electrical co-design so that appropriate trades can be made as work progresses. He cites the use of techniques such as electromagnetic analysis of liquid moving in channels that are close to active devices.
Several participants in the DARPA program are using high-thermal-conductivity synthetic diamond material either to line microfluidic channels or to form the substrate of the RF chips.
Northrop Grumman lines the micro channels etched into the bottom of its GaN-on-Silicon Carbide (SiC) chip with diamond coatings in order to move heat away from the source and reduce the heat flux by spreading it over a larger area, explains Vincent Gambin, ICECool lead at Northrop Grumman’s Aerospace Systems unit.
The company uses “submerged impingement jets” — tiny openings in the underlying silicon “manifold” chip — which operate like high-pressure hoses to spray coolant to points directly beneath the heat-producing HEMTs. The impingement jets help transfer heat between the diamond and the fluid.
The liquid coolant sprayed through the submerged impingement jets impacts the undersurface of the HEMT substrate at a high velocity in order to overcome thermal boundary resistance and achieve a good heat transfer from solid to liquid, Gambin says. Coolant travels through the jet at up to 30 meters per second with pressure in the order of 100 to 150 pounds per square inch (psi). The idea in the projects is to tap the coolant and the pumps already present in the larger systems.
Impingement jets transfer heat well. They cause a lot of turbulence in the fluid, so that they don’t form a significant boundary layer, Altman says, and fresh coolant is constantly replacing the recently heated coolant, sweeping the heat away. The downside, however, is the risk of erosion and corrosion. High-velocity fluids could wear away the walls of the micro channels as well the bottom surfaces of the dies being cooled. To address this issue, Northrop Grumman carefully controlled factors, such as micro channel size and fluid pressure, as well as used diamond, a very hard material, to coat the channels.
Raytheon relies on creating more surface area to spread the heat via a diamond substrate and providing more, smaller channels etched into the substrate. The diamond “does the real work of heat rejection,” Altman says. The silicon manifold structure bonded to the chip just distributes the fluid to the various regions in the diamond, he explains.
Raytheon distinguishes its “manifolded micro channel impingement” cooling, which uses lower pressure, from impingement jet technology. The rule of thumb is that if you are over 6 to 7 meters per second, you can run into erosion effects, according to Altman.
Nevertheless the proximity of the coolant to the active electronics must be considered in chip design, Altman says. “The behavior of the transistors is influenced by the presence of the channels which have conductive fluid in them,” he says. The solution has to do with the location of the micro channels relative to the active features of the device. The channels need to be spaced far away enough so that they don’t detract from performance, Altman adds.
The ideal solution would be to rely on the thickness of its diamond substrate, about 300 microns in Raytheon’s case, but Raytheon is also concerned with the correct placement of the channels.
|Integrated Circuit Enhancement through Intrachip Microfluidic MMIC Cooling (ICE MMIC). Image courtesy of Raytheon.|
BAE Systems uses a diamond substrate but employs a low-temperature approach to bond the active layer to the substrate. This results in high-quality, large-grain diamond in close proximity to the transistor junction, Creamer says. The approach also mitigates the inherent stress due to the coefficient of thermal expansion mismatch between the GaN and the diamond, he adds.
BAE Systems uses a single-phase cooler incorporated into the ground plane located just below the diamond substrate. The design doesn’t involve etching micro channels into the diamond substrate, but it does involve etching via connections into the diamond in order to ground the circuit.
The prototype ground plane/fluid router is fabricated using a proprietary method developed by BAE partner, Science Research Laboratories. The process involves etching a series of copper-based foils to form micro channels that are bonded together to form the desired geometry, Creamer says.
Although coolant is pumped into the manifold structure, BAE does not use a high-velocity impingement approach, according to Creamer. “Fluid velocity is not the determining factor in our design,” he says. “Rather, we count on effectively coupling the heat from the heat source through the diamond substrate into the walls of the cooler and then into the fluid, while achieving low pressure loss.” The company has also designed a new chip that provides three times the power in half the area of a baseline chip.
Promise of Palladium
|Measured Infrared (IR) Thermal Images, Demonstrating the Effectiveness of ICECool Technology versus Conventional Cooling Methods.
Image courtesy of Lockheed Martin.
The company uses impingement heat transfer at velocities ranging from 10 to 20 meters per second. Multiple jets of coolant are aligned with the hotspots on the RF device and elsewhere along the underside of the GaN on SiC die to handle the die background heat flux. “There can be as many as 20 times the number of jets beneath the die as there are hot spots on its upper surface,” Ditri says. “We don’t just cool the HEMTs, we cool the whole underside.”
This stand-alone micro cooler is manufactured using an additive process rather than conventional etching processes. “Our microfluidic manifold contains a labyrinth of micron-scale internal fluid passages, which could not have been produced in a single, monolithic piece of material using conventional machining techniques,” Ditri says. He considers the additive approach a potential manufacturing and cost-effectiveness edge.
Under a separate DARPA contract, Lockheed Martin is working to extend its ICECool technology to TRL 5 or 6 by integrating embedded cooling into a high-fidelity transmit antenna prototype and demonstrating the benefits that its improved thermal management technology would bring. The demonstration is planned for the 2017 timeframe.
DARPA expects that the embedded microfluidic thermal management paradigm will transform the electronic system architecture of next-generation weaponry. Its contractors agree. Greater power output could translate into longer range for microwave systems, Gambin says. Designers might be able to run chips at higher power with comparable reliability to what’s available today or to put additional HEMTs on a die. They might be able to think about building stacked configurations, with which you could have lighter-weight, lower-cost, higher-performance systems.
You might even be able to do without a thermal ground plane, or heat sink, which is now interposed between the amplifier and the digital electronics. If the thermal ground plane is eliminated, the digital electronics could be moved nearer to the analog electronics. There could be closer pitches between the elements and you start to sense the possibility of an evolution to three-dimensional integration, according to Gambin.
Lockheed Martin envisions the possibility of chip stacks. “We see a good opportunity for thermal management of system-on-chip ICs and heterogeneously integrated chip stacks, both of which offer great electrical performance and cost opportunities but present significant thermal challenges,” Ditri says. Perhaps the most important achievement of the program so far, however, is the confidence that it has inspired in finding a solution to current problems. The program “has educated the community that thermal limits are not divinely imposed like the speed of light or sound,” Bar-Cohen says. “It is a domain that can be engineered. We have a family of solutions that have as yet to be applied and could go a long way to removing these thermal limitations,” he says.