There was a time when computer CPU chips didn’t come with fans, or even with heat sinks. They just got pressed into little sockets mounted on the motherboard, and the passing air would be enough to keep them within their operating temperature range.
Those days are long gone as we pack more cores into the chips, and have a matrix of tiny solder points to connect these silicon brains to our computing machinery. And as this power increases, so does the heat that they generate. And that heat has to go somewhere else so that the chip doesn’t overheat and start performing erratically. A small increase in the efficiency of this heat removal can pay huge dividends in saved energy, because you don’t have to provide as much cooling to the whole system.
One point of interest that you might not think of is the mating of the chip to its heat sink. Just putting them together is not sufficient. While both sides might look flat, on a nano-scale the surfaces are incredibly uneven. Where the two sides don’t touch, you get inefficient heat transfer. (Keep in mind that air is actually a heat insulator and does not conduct heat well.)
Anyone who has built their own desktop computer probably has used “thermal paste” between the CPU and its heat sink. The purpose of this material is to fill in the gaps between the two sides. As you might imagine, researchers have studied how to make this paste work more efficiently.
The work of the heat transfer is done by filler particles (typically aluminum oxide or zinc oxide) that average about 5 microns in diameter, ranging up to as much as 40 microns. The variety of sizes makes it more likely that you’ll get a solid path of contact between the two sides. The particles are suspended in silicon oil, but the mixture is very thick to maximize the contact of the filler particles.
Peanut Butter and Crackers
So how do you get this paste to spread evenly? Imagine putting a blob of peanut butter between two crackers and squeezing the together. You want as thin a layer of peanut butter as possible (for the most effective heat transfer) but without any voids. And you don’t want to press so hard that you break the crackers.
As it turns out, when you squeeze this paste between the two “smooth” surfaces, the filler particles can tend to clump, resulting in a thicker and less efficient layer. How do you prevent it from clumping?
The answer lies in the use of an intermediate layer: a copper cap between the chip and the heat sink. If you scribe the copper layer with two diagonal lines, it helps the thermal paste flow more easily. But this does not produce the best fill factor of the particles. So you then divide those four regions with smaller lines, the fill factor improves. If you then divide those regions with even smaller lines, you get a thinner but more reliable paste layer.
These lines are known as Hierarchical Nested Channels (HNC), and they create thinner bond lines between the two mating parts. These are more efficient at conducting heat from one side to the other, while requiring less force to assemble. If you want to read more about this technology, the IBM Zurich Research Laboratory has a good presentation posted on the Internet.
Whether it is a single desktop computer or a huge server farm that powers the Cloud, heat management is an important issue. And sometimes all it takes to improve cooling efficiency is knowing where to draw the lines.