Slashdot Mirror
New Semiconductor Coolers
An anonymous reader writes: "A new thermoelectric material is 2.4X as efficient as
best existing materials. The new solid state heat pumps
can provide 700 watts of cooling (nearly one horsepower)
with just one square centimeter. These new materials have the
potential to replace current heat sinks, thermoelectric
generators and mechanical heat pumps.
You can also read an article in nature."
There was a brief bit on NPR about this a few days ago. NPR recording
a couple of them in fact. (look to the bottom of the page)
"It is a greater offense to steal men's labor, than their clothes"
The problem with these heat pumps (and Peltier coolers) is that the cooler sucks heat away from the processor side and pushes it to the exposed side of the cooler. As an unfortunate side effect, the cooler GENERATES additional heat in the process.
As an example, if your processor generates 50 watts of heat output, the cooler might generate an additional 50. The processor itself would stay cool, but you're dumping a lot of extra heat into your case, requiring even more case ventilation.
Not very practical for most users.
-atrowe: Card-carrying Mensa member. I have no toleranse for stupidity.
Yes, good, the rate at which energy is carried away from the other side is a limiting factor. The reason you might want to apply it to only specific areas is that this is very strange material, and very expensive to produce.
The key in these materials is that they conduct electricity very well, but conduct heat poorly. This is weird, as the two are usually linked. The electrons carry heat energy with them as they move through the crystal, and the random motions of the atoms transfer heat through the crystal as well. The electrons and the vibrations (phonons) interact, hence the link between the two kinds of heat conduction. You usually only hear about the atomic vibrations because that effect is many thousands of times stronger than the electronic heat conduction.
However, we can control the motion of the electrons. We cannot control the flow of the heat transfered by the random motion of atoms. The big idea is to create a material that impedes the flow of heat, but allows us to control the flow of electrons. As bizarre as this sounds, there are some naturally occurring minerals that have this property (skutterudites). These are exremely rare, and harder to synthesize than diamonds. There are strategies involving alternating layers of semiconductor, and that sounds like the plan in this article.
The point is that these materials are hard to make, and very expensive (high purity, many production steps). It turns out that only some parts of an IC generate huge amounts of heat (this is an issue when we mount optical devices on ICs). The dot idea is a clever trick to save on production costs. Those clever engineers.
I agree with your conclusions.
This seems like a great way to quickly remove heat from a small area and spread it to a large area. You'll still have a lot of waste heat on the hot side of this and I'm sure you'll need a heatsink on there. Large than before in fact because this appears to be a powered thermocouple like a Peltier cooler which means it should generate waste heat as well.
The benefit though is that heatsinks become more efficient as the temperature gradient goes up, so we should still be able to get the heat into the air and then out of the case. And because this thermocouple maintains a rather large gradient we should be able to keep the CPU that much cooler.
As for the little dots of it, etc... I think what they mean is that inside the CPU core you'd have little dots of this being used to pump heat away from the main heat generating areas directly into the heat-spreader on top of the chip. The only other way to do it is let the heat diffuse through the whole core and then into the heat spreader.
So this would be a lot better at putting heat in manageable areas (the heatsink) but it isn't magic, you couldn't put a bit in a sealed package and have heat magically disappear.
1) A lot of people have posted that this material is 2.4X as efficient as existing materials. This is not correct. According to the paper, they have achieved a 'Figure of Merit' (which is calculated in some fairly complex way) 2.4x greater than previous state of the art. Since they also note that there is no absolute theoretical limit to this metric, I think we can assume that it not only is not efficiency, it doesn't map to efficiency in any linear way.
2) No one seems to have commented yet on the extraordinary thinness of these devices. They achieved a 70K thermal gradient across a 5 micron thickness when running in power conversion mode; this corresponds as they state to 134,000K per cm (!). It's also the root of their extremely fast thermal response (20,000 times faster than previous SoA). And it helps explain their very high watts-per-cm2 figures... in fact I'd say the microthickness, rather than the 2.4x 'efficiency' gain, is the real story here.