The phenomenal improvements in IC processes, reduction in per-function dissipation, and decrease in feature size have led to a never-ending treadmill cycle: chips get smaller, are expected to do more, putting more features in a smaller die size, and then the dissipation issues get worse. It's a corollary to the law of unintended consequences, as improvement in IC technology leads to more thermal issues.

There are many passive and active ways of trying to deal with the increased dissipation requirements. In the case of a few ICs or devices which are at risk while the overall system is cool enough, the problem is usually that the thermal resistance between these devices and their immediate surroundings is too high. Solutions include micro heat sinks (with a nearby fan, in many cases), heat pipes, cold plates, and using PC-board copper to pull the heat away (transferred via the IC's pins or even a thermal pad on its underside). The latter works only if the nearby ICs are also not trying to dump their heat into the same copper; you have to be sure that your overheating IC is the only one in the neighborhood with this problem.

Researchers at Lockheed Martin are further exploring and exploiting an idea which has been talked about and tried, but with limited success: microspraying those hot ICs to cool them. The efforts are focused on gallium-nitride (GaN) power amplifiers and associated monolithic microwave ICs (MMICs) which are increasingly used in RF designs. GaN offer many operating advantages including high efficiency, but there is still a lot of heat to dissipate.

The Lockheed Martin effort, part of the Interchip/Intrachip Enhanced Cooling (ICECool) program from the Defense Advanced Research Projects Agency (DARPA) Microsystems Technology Office, sprays tiny drops of water on the bottom of microchips, thus dissipating large amounts of heat via this cold plate. It's not clear to me how they source and replenish the water, even though it's a miniscule amount.

EETA 2016JUN28 blog This super-tiny liquid-cooled micro-cooler contains less than a drop of water at any time, yet is capable of cooling extremely hot IC die.

These liquid-cooling systems are tiny: 250 microns thick, 5 mm long, and 2.5 mm wide. Lockheed is working with major GaN vendor Qorvo, Inc. to integrate this thermal solution with Qorvo's high-performance GaN devices. You can read more about the Lockheed Martin work at Cool Technology Turns Down the Heat on High-Tech Equipment and Hot Electronics? Just Add Water.

The numbers associated with this experimental microfluidic approach are impressive, beginning with a 4× reduction in thermal resistance. They also cooled a demonstration die which dissipated 1 kW/cm2 and had several local hot spots running over 30 kW/cm2.

It's obviously too early to know how successful and viable this approach will be in practice, as there is a always long road from lab experiment to commercial practicality. Still, it's an interesting example of how microchips bring on new problems and also set the stage for new solutions.

What's been your experience with "leading edge" cooling approaches, beyond basic heat sinks, fans, heat pipes, and cold plates? Have you ever been involved with a more complex technique—and how did that work out?