Analysis of metal TIMs – Joe
SUMMARY: Low Melt Alloys as TIMs are strong competitors to “traditional” thermal greases, although with some caveats.
I mentioned earlier that I was taking a look at some LMA (Low Melt Alloy) TIM material, in this instance from TherMax Korea. This material is quite different from the “traditional” thermal grease we’ve seen on the market, in that it is metallic (and conductive) and not oil based. If you think of something akin to mercury, but solid, you get the idea.
Thermal grease manufacturers have added metals to grease mixtures to increase thermal conductivity – a notable example is Arctic Silver. Logically, if you can get rid of the non-metallic carrier and just have a metallic TIM that fills the micro-voids between the heatsink and CPU, you’ll have a superior TIM – nothing like pure metal.
One approach would be to solder the heatsink to the CPU – the drawback is that the melting point of solder will destroy the CPU – not a good move. Hence the Low Melt Alloys – these are alloys that melt at temperatures below those that will kill the CPU; the TherMax Korea HiFlux TIM HF-60110BT starts to flow at 62ºC – TherMax recommends that you run up to 65ºC to ensure melting.
The impact of the TIM joint between the CPU die and the IHS might surprise some readers – the following passage is interesting reading:
“The importance of a TIM is not proportional to its volume in the overall thermal solution. About 40% of the thermal resistance in the overall solution lies in the TIM1 layer between the die and the lid [ie, IHS]:
Many current designs are thermally inferior, exacerbating the problems of designing a feasible back-end (heat sink and fan) thermal solution. Steve Pawlowski, Director of Intel’s Microprocessor Lab stated at a developers’ conference in Taipei recently: ‘There’s a certain [thermal] transfer coefficient between the die and the package. If you can’t come up with a better package that will allow that heat transfer to occur from the die, no matter what you do on the outside, it’s going to be difficult.'”
This of us cooling high watt CPUs know this well enough.
I don’t want readers to get the impression that LMA TIMs are the perfect solution – there are drawbacks, as outlined in this paper, Performance, Reliability, and Approaches Using a Low Melt Alloy as a Thermal Interface Material. The following issues are pertinent:
- “Corrosion, considered to be the main failure
mechanism in LMA materials, is driven primarily
by moisture, oxygen and heat. The liquid phase
of these alloys at operating temperatures
facilitates rapid diffusion, accelerating the
- …degraded thermal performance was
observed after temperature cycling and
ageing with and without elevated humidity.
These results corroborate the susceptibility of at
least some LMA systems to temperature and
moisture induced failure mechanisms.
- It is conceivable that if excess
LMA were incorporated during assembly, this
excess could end up airborne from shock or
vibration. Therefore, LMA TIM should be
deployed in closed cavities where no
opportunities for shorts or adverse reactions with
other metals exist.”
- …degraded thermal performance was
I should stress that I have no first-hand knowledge that the TherMax HiFlux TIM has any of these characteristics.
TherMax HiFlux is a defined as a “Two-Phase Metal Alloy” with the characteristic that the “…thermal contact resistance is equivalent to the value of the soldered joints.” TherMax claims a 0.005°C in²/w – this is comparable to what is claimed for Arctic Silver – 0.0045°C in²/w (0.001 inch layer).
In order to get the TherMax to flow, you have to run the CPU up to 65ºC. Once this is done, the material is set and you achieve consistent results at all temperatures, as shown below:
Also note that clamping pressure is NOT a big factor in results – once the material is set, consistent results are obtained over a wide range of pressures.
The first test I did was to place a piece of TherMax on a copper based heatsink and, using a hair dryer, heat it up enough to get the LMA material to flow. As the picture below shows,
TherMax is delicate – it is very easy to tear; it is held to the heatsink by adhesive tape.
First thing I learned is a hair dryer doesn’t generate enough heat to melt it. I then used a heat gun I used to melt paint, and this did the trick:
The heatsink’s copper base retained enough heat so that the TherMax flowed a bit like solder – not as fluid, but enough to see the metal move over the surface in a small wave. After it cooled, I used a razor blade to lift the TherMax off the heatsink’s base:
I expected it to be tougher to remove the material, but it lifted very easily off the base. Note that this is a single use material – once you remove the heatsink from the CPU, you have to use a new piece of Thermax – you can’t re-use it.
The next test I tried was to see how a CPU would mate to the heatsink’s surface. I had a dead AMD Duron that I used – I again heated up a new patch of TherMax on the heatsink’s base, placed the CPU on the TherMax, placed a weight on it and waited for it to cool. After it cooled, I removed the CPU without any problem and found this on the heatsink:
The TherMax flowed quite nicely and left an impression of the printing on the CPU on the heatsink’s base. I again found it very easy to peel the TherMax off the heatsink’s base. At this point, I was fairly well satisfied that the TherMax would not be difficult to remove – one concern I had before further testing.
I then decided to try the TherMax on my die tester; I rooted around and decided to use a Glaciator Lite because the clipping is very easy (20 pounds) – I wanted to try a heatsink with minimal pressure. As a comparison, I used Arctic Silver 5¹ first, cleaned the heatsink and die, and then applied the TherMax – results below:
|Arctic Silver 5|
¹ I ran Arctic Silver over a 24 hour period and numerous temperature cycles. However, Arctic Silver clearly states “Due to the unique shape and sizes of the particles in Arctic Silver 5’s conductive matrix, it will take a up to 200 hours and several thermal cycles to achieve maximum particle to particle thermal conduction and for the heatsink to CPU interface to reach maximum conductivity. … On systems measuring actual internal core temperatures via the CPU’s internal diode, the measured temperature will often drop 2C to 5C over this “break-in” period.” Readers should evaluate test results with this caveat in mind.
The TherMax proved to be significantly better in this test.
After the test runs were completed, I removed the heatsink for the die tester and found this on the heatsink’s base:
Note that the excess material did not fully adhere to the base:
This is a concern, as this material is conductive – any excess material that might shake loose will not do any good to a motherboard. I lifted the patch and found this:
Even though I ran the die up to 65ºC, it looks like the TherMax did not melt as much as I thought. Even so, the performance was quite good. I decided to re-run the test, this time running the die temp between 65ºC to 75ºC three times to melt the TherMax, with the following results:
|TherMax – First Test|
|TherMax – Second Test|
After the higher initial thermal cycles, I did get somewhat better results. I also noticed that the contact patch was more defined:
It also appeared that the TherMax was thinner:
However, it was nowhere near what I achieved using a heat gun – I’m sure temps were well in excess of what I ran on the die tester.
LMAs are an interesting approach to TIMs – the notion of a metallic TIM is compelling. However, the TherMax I tested did not liquify at the temperatures I ran on the die tester. I am not comfortable running any CPU over 65ºC, and considering what I found, I’m not sure I’d be comfortable running the CPU at temps high enough to liquify the TherMax. Even so, the TherMax LMA performed very well.
The one drawback that gives me some pause is its conductivity – any loose material could cause real problems inside a PC or laptop. If I were to use it, I’d cut a piece just the size of the die and install the heatsink over it – very carefully so as not to move it.
Overall, TherMax Korea’s HiFlux TIM HF-60110BT is a very interesting TIM approach with excellent performance, but perhaps not for the “faint of heart”.
I managed to secure additional samples of Thermax’s LMA TIM and send samples to some of our Forum seniors to try out. First results have been encouraging, with some showing high single digit performance gains in cooling.
I decided to further test this TIM on a couple of systems, one being an Asus K8V Deluxe with an AMD Athlon 64 3200+; at spec, it radiates 89 watts.
There is a plastic sticky tape that runs along the border – this is used to hold the TIM onto the heatsink. The problem is that for AMD’s IHS, the plastic tape will sit on the IHS – not good! I cut the plastic tape off the TIM, cleaned the heatsink base and CPU with acetone and placed the TIM on the CPU. I did not burn it in, just ran it in place with Prime 95. The delta (CPU temp minus ambient temp, CPU temp per MBM) I got this way was 22ºC.
Then I ran the heatsink (Tuniq Tower 120) without the fan on till the CPU temp registered 68ºC. Flipped the fan on and the temp dropped to 30ºC for a delta to ambient of 11ºC.
I measured the edge temp of the IHS and found it was 6ºC less than the CPU temp, which means that I probably did not achieve the 65ºC at the case to fully melt the TIM. However, I ran Prime 95 on this system all night and in the morning it was showing a delta of 9ºC, performance that was better than any other grease I tried.
The burn-in is unsettling – I do not like to run anywhere near these temps. According to the manufacturer, you should see the same result if you ran the TIM without burning it in over a period of time – the thermal cycling should do it.
As a second test, my friend Andy used it on his HP PC with a XP 2800, a bare die CPU. Before using the LMA TIM, this PC would run about 45ºC at idle and 55ºC running “normal” apps. The heatsink is a non-descript aluminum OEM unit – nothing fancy – with a 70mm fan.
After installing the TIM, he rebooted; it hung for a minute, and when Windows XP popped up, he saw temps at idle to 30ºC, and “normal” app temps of 45ºC; this is all without burning in the TIM.
Based on these early tests, it would seem that a big factor is the watt density of the CPU. A bare die (no IHS) CPU, such as Intel Pentium Ms or AMD XPs, may yield results faster due to the higher watt densities at the interface between the heatsink and CPU. A CPU with an IHS may have lower watt densities at the interface and may take longer to show results.
We’ll continue to test this material and report back with results – so far, intriguing!