This article is a slightly edited version of the article found at ZZZ Online Reissued on this website by request and permission of the author.
The website is dedicated to talking about all kinds of bleeding-edge technology, computer and otherwise. Imagine a science newsletter with less jargon and more humor, and you got it. No plug was requested, but the site deserves it. Worth more than a look.
Any of you who have ever even thought about overclocking are probably familiar
with thermal paste and its function. Lots of OEM or low end cooling setups use
either a thermal interface pad (TIM) or that white goop you get at Radio Shack.
The fact is that neither of those does a great job of transferring heat from the
processor to the heatsink. While they work ok, they don’t exactly assist
Moore’s law in fulfilling itself by limiting clock speeds with heat.
are still getting faster, but one needs only to look at overclocking results
with stock cooling versus those achieved with a good heatsink and good thermal
paste to realize that better cooling = faster computers. Heck, if we’d stuck
with the tiny old anodized fanless heatsinks on 486es, we might not be past 1GHz
Current high end pastes range widely in composition, but in
terms of performance they all
fall within a very small degree range.
Silver has been a longtime favorite among many, simply because they were the
first company to release a competitive paste – one which was actually well
suited to the task of transferring heat. Nanotherm is another big name these
days, and lots of people are talking about PCM+, their upcoming metal-free
product. But as I said,
all of these products still get very similar results.
A degree or two at most is all you can hope for in moving from one brand to another.
So you might
resign yourself to defeat saying, “Pastes have achieved perfection, so the
bottleneck must now be the heatsink and the die of the CPU itself.” And that
would seem to me a very reasonable thing to say…the fact that several companies
are putting lots of resources into the development of more efficient thermal
transfer and seeing diminishing returns is fairly strong evidence to support
such a statement.
But according to materials engineer Dr. Deborah Chung,
current thermal pastes are rubbish, hooey, and applesauce, and she has something
In tests comparing it to
solder (a method of thermal transfer not typically used with electronic
components because of the temperature required to bond it to both surfaces), the
carbon paste surpassed the pure metal bond in thermal conductivity by
It was also superior to diamond and carbon nanotube based pastes currently
undergoing development. Even if the carbon paste were to merely match
the diamond and nanotube pastes, it would be a significant improvement because
of the cost differences.
Why does it work so well? Spreadability.
The problem with
current commercial pastes is that they have focused too long on the thermal
conductivity of the material, and not on the fundamental principle of a thermal
paste, which is gap filling. Silicone based ‘goop’ from Radio Shack is fairly
thermally conductive, but the size of the particles and the terrible
spreadability cause it to be more of an insulator than a conductor.
other hand, using something entirely liquid such as mineral oil doesn’t cool
well either because it isn’t conductive enough, so the key is to find something
with the right balance of conductivity and spreadability.
The diamond paste
contained particles 25 microns wide, about as small as they can be ground down
to. This is what keeps the incredibly conductive material from winning – the
diamonds actually end up pushing the two surfaces apart.
To check for the
importance of particle size in the opposite direction (smaller than the carbon
black particles), a paste based on .1 micron diameter carbon filament was also
tested. One would think that the carbon filament, being only .1 micron
thickness would perform better – but this is not the case.
The reason, Dr.
Chung believes, is because of the porosity and compressibility of carbon
black. Porosity plays a big role because when pushed with significant pressure
against the two mating surfaces, it actually allows the surfaces to penetrate
into the carbon black particle. Picture that in contrast to the diamond meeting
the heatsink/cpu surface only at one small point.
This discovery points out a few very important things about
thermal conductivity. First, simple conductivity measurements are insufficient
to make an effective paste. One must also take into account porosity of the
filler particles, spreadability, and compressibility.
Second, as the data tables show, testing found a strong pressure bias – increases in pressure lead to great increases
in thermal conductance:
18.94 ± 0.60
24.87 ± 1.00
25.74 ± 1.20
Graphite (5 mm)
3.03 ± 0.09
3.67 ± 0.08
4.02 ± 0.12
Graphite (1 mm)
1.52 ± 0.03
1.77 ± 0.04
2.04 ± 0.05
Nickel (3 mm)
1.85 ± 0.05
2.14 ± 0.02
2.84 ± 0.04
Nickel (1 mm)
0.91 ± 0.07
2.03 ± 0.10
2.66 ± 0.03
Diamond (25 mm)
1.15 ± 0.02
1.21 ± 0.09
1.54 ± 0.03
Carbon filaments (0.1
1.09 ± 0.03
1.32 ± 0.02
1.51 ± 0.03
Single-walled carbon nanotubes*
13.5 ± 0.2
13.8 ± 0.3
14.1 ± 0.4
The top row, measured in MPa, indicates the pressure used
to push together the two copper discs used to calculate conductance. As you can
see, conductance improves dramatically with pressure, and carbon black simply
smashes all other interfaces.
Ed. Note and Disclaimer: If what you get from this is “All I have to do is increase
pressure with my current setup”, please keep in mind that the ceramic core of certain CPUs may
have something to say about this, like “Crunch.” If you decide to make a batch of this stuff (see below),
please exercise caution, or the only thing you’ll end up smashing is your CPU.
The single walled nanotubes(*) mentioned in the
bottom row, which are essentially impossible to manufacture at this point in
time, still fall behind the carbon black – the most likely reason for this being
compressibility, as the nanotubes don’t compress at all (as seen by the lack of
improvement in conductance with increasing pressure).
What does this mean for the computer industry?
Possibilities of a whole new level of thermal design. Carbon black is cheap
and abundant – so implementing it in every Dell, HP, Gateway and Apple won’t
cost a cent more than plain old white Radio Shack goop.
The full text and results by Dr. Chung are available
here. Ed. note: This link is for
the research paper submitted, which also describes how these materials were made. It’s not
for the first-timer.