The Heat Is On
Some readers may remember my post from AUSA, where I reported that the biggest buzz on the GE stand was created by the thermal technology being showcased by our colleagues from the GE Research Center. As it turns out (not to make it sound like it just happened; it was planned this way), that was just one piece of a bigger puzzle that we have been addressing for some time and are now starting to roll out in real product innovation.
Skipping over the irony that it is snowing here in New England as I write this, one of the biggest design issues we face today as a manufacturer of high-performance computing, networking and interface products for harsh environments is how to manage heat.
As a board designer, we are concerned with how to get heat from where it is generated—primarily the dies of the highest power devices on the board—to the periphery of the board. There are several bottlenecks to be addressed in this path. The heat must be extracted from the die to the heat sink of the board. This tends to be the point of highest heat flux, and the problem gets trickier each year. The power of processors tends to remain the same, but the die geometry shrinks with every new manufacturing process introduction. Therefore, the same heat must be pulled off a smaller and smaller surface area—making an efficient path critical.
But traditional approaches have their flaws. Heat spreaders with thermal grease or other materials often suffer reduced efficiency due to manufacturing variations in height and alignment of assemblies.
As a system designer, we also need to consider how to move the heat from the card frame to the chassis and how to reject it from the chassis to the environment. Traditional wedgelocks remain the primary way of anchoring the boards to the chassis and providing a path to get the heat off the board. Once the heat is in the chassis sidewalls, we need to find an efficient way to dissipate it to the outside world. This is often achieved via conduction to a larger body mass or by convection to the surrounding air.
Four GE innovations address different parts of the transfer path:
- Nano Thermal Interface (NTI)
- Thermal Management Technology Bridge
- Thermal Ground Plane (TGP)
- Dual Cool Jets (DCJ)
Nano Thermal Interface
This is a thermal interface material (TIM) designed to improve the thermal resistance from a chip to a heat sink by at least a factor of 10 over prior interface materials. One issue addressed is the presence of tiny gaps in the contact between the surfaces of the chip and the TIM, and between the TIM and heat sink. Up to now, this has often meant a compromise—materials with high compliance but low thermal conductivity, or materials with low compliance and high thermal conductivity. GE’s new interface uses an array of copper nanosprings that are sandwiched between two materials (such as silicon and copper). The nanospring layer is much, much more compliant than typical layers such as solder, so thermal stresses are carried by the nanosprings rather than the solder layers.
Thermal Management Technology Bridge
A typical conduction-cooled board assembly uses a metallic heat frame that is attached to heat-producing devices using a stack-up of TIMs, including thermal grease, metallic heat spreaders and gap pads. There are a lot of parts and manufacturing steps involved—with scope for misalignment. This can lead to undue stress under operation, and can even crack solder joints or the devices themselves.
GE’s Thermal Management Technology Bridge is designed to address these issues. It is a self-contained cartridge attached to a heat sink during the manufacturing process. The package contains a low melting point solder and a high-performance heat spreader with a metallic spring providing pressure. The part is assembled, and then goes through a one-time heat process to allow the solder to melt and spread under the pressure of the spring. This accounts for variations in die height and coplanarity that would otherwise serve to greatly degrade the thermal path.
Thermal Ground Plane (TGP)
Today, most boards move the heat to the sides using heat frames of a solid material, typically aluminum or copper. This is starting to limit the thermal load that can be carried by boards based on today’s high-power processors and GPUs. Under a DARPA contract, GRC developed TGP-based heat spreader technology that replaces the traditional heat frames and offers 2x or better the performance of the traditional frames.
TGP technology achieves its high thermal conductivity by utilizing two-phase heat transfer instead of the purely conductive paths through aluminum or copper heat frames. If this sounds familiar, it is because it has its roots in the heat pipes that have been used in PCs for many years. The difference is that those pipes would not work in the environments our systems are exposed to—high levels of shock, vibration and acceleration. Picture the processor in a tactical radar mounted in a fast jet pulling 10 Gs—not much chance of a liquid flowing back to the heat source there. The innovation comes in the use of nanostructures that use capillary action instead of gravity as the method of liquid transport—and in making the devices manufacturable and affordable.
Dual Cool Jets (DCJs)
Many chassis rely on convection to move the heat from the box to the surrounding air, using fins to greatly increase the available surface area for transfer. This is not a very effective mechanism if just left to natural convection, as a layer of air on the surface can actually insulate the chassis from cooler air. Fans can be used to break up this layer, but the lifetime of fans in sandy and dusty environments is less than may be desired. Enter DCJs, a piece of piezoelectric technology borrowed from our colleagues in GE Lighting. By acting as mini bellows, these little pieces of magic can improve heat flow by a factor of greater than three.
By combining one or more of these innovations, we can bring to market boards and systems that can have higher clock rates and, therefore, higher performance than would be possible otherwise. They can maintain device dies at lower temperatures, increasing long-term reliability. We can optimize system SWaP-C to do more stuff in a smaller volume.
See our white paper for more information. Or, better yet, contact your local GE Intelligent Platforms representative to discuss how we can bring these technologies to bear on your program.