Myth Busted: Thermal Paste Beats Thermal Tape

archer6817j

archer6817j

Enlightened
Joined
Aug 9, 2010
Messages
830
Location
Mountain View, CA
Hi all, let's get science-y!

There is an ongoing debate over the superiority of thermal paste vs thermal tape for securing an MCPCB to a heat sink. The majority of opinions I've seen say (hands down) that a "screw down" configuration with thermal paste is far superior to using pressure sensitive thermal tape...in terms of overall thermal performance.

Now if you look at the numbers on paper, thermal paste is the clear winner. It (in theory) conducts heat much more efficiently than thermal tape. I knew those figures once, and I'm in a bit of a hurry, so I don't have them here. Feel free to look them up and post them here.

Before we look at the data:

One thing that sets my lights apart from most others is the head design. I do not use a pill. I use a one-piece, solid head. This means the best possible thermal transfer away from the LED because there is no threaded connection between the pill and the head. In other words, most other lights use the pill as the primary heat sink. I use the entire head and the 3/8" thick piece of aluminum directly behind the LED, as you can see in the photo below.

head.jpg


The driver is mounted to a thick walled copper cylinder (this can be seen both above and below) with sufficient thermal mass to cool the driver without transferring as much heat to the head itself. The copper cylinder is press fit into the head only at the very bottom, otherwise it "floats" inside the body (see below). This is very different from mounting the driver directly to the head of the light.

head%2Bsection.jpg


In the next image below you can see what that looks like in practice. On the left is the prototype copper sleeve and on the right, the production sleeve. The production version has twice the thickness/mass. The prototype sleeve is depicted in the cutaway (first photo above).

PCB-collar.jpg


I've had a few critics say that if I had any clue what I was doing (and if I wanted to be taken seriously) I would ditch the tape and get with the paste. I wanted to find out if this was true. I've been in the business of making things long enough to know that what looks good on paper doesn't always pan out in the real world. I'm not an engineer that specializes in thermal design. Given that, my best recourse to understand a given situation is through applied research: make something and test it in the real world. Blah blah blah, let's get down to it. Here is what I found...

The Contenders: Arctic Silver AS5 thermal paste vs Bergquist Bond-Ply 100 thermal tape

Experimental Concept:

  1. The key for me was to compare the two materials on the relative performance of thermal transfer, not the absolute performance in terms of lumens. In plain english I'm asking, "Which thermal material moves heat more efficiently by observing the trend of output over time." (not sure that was plain english). As the LED heats up it will become less efficient and hence less bright.
Hypothesis:
  1. If thermal paste is a better conductor of heat, the output graph (over time) should be relatively flat. Thermal tape (being a worse conductor) should show a relatively more downward sloping line over time.
Experimental design:
  1. Arctic Silver AS5 is the undisputed king of thermal paste. I won't go into it here. It's good.
  2. Bond-Ply 100 is specifically designed for LED interface and Berquist (the manufacturer) specialized in thermal interface materials. It's also best in class.
  3. Both materials were applied per the manufacturer specifications.
  4. Both LEDs were installed in "open" heads with no optic or reflector. All I need is a reference number, not actual lumen output.
  5. LEDs were selected from the same batch and bin.
  6. Fully charged batteries were used for each test.
  7. Over a period of days, five samples were taken with each light. Each day I would measure the AS5 light once and the Thermal Tape light once.
  8. The test began with each light at room temperature.
  9. The output was measured every thirty seconds (including turn on) for the course of 3 minutes
  10. The five values, for each configuration, were averaged and put into an excel chart.
Raw Data:
BP%20vs%20AS5%20RAW.jpg
BP%20vs%20AS5%20RAW.jpg
BP%20vs%20AS5%20RAW.jpg

________________________________________________________________________________________
BP vs AS5 RAW.jpg

So each data point is the average of 5 values taken over 5 days. At first, it looks like AS-5 is the clear winner because it shows higher values. However, we are not looking at absolute brightness because LED and every other component are a little different. No two LEDs will be the same, and no two flashlights will be the same. What we are looking for is a difference in the slope of the two lines. Flatter for AS-5 and steeper for Bond-Ply. Let's move on to the next chart below.

Corrected Data:
BP%20vs%20AS5%20CORECTED.jpg
BP%20vs%20AS5%20CORECTED.jpg

_________________________________________________________________________________________
BP vs AS5 CORECTED.jpg

The first step was to remove the lumen values for "turn on" measured at time zero. The remaining data in the chart above begins at 30 seconds. The next step was to "correct" the values by moving the entire blue line up so it matches at the first value at 30 seconds. Now we can directly compare the slope of the two lines and the relative performance of the two thermal interface materials.

Anecdotally, I have another data set (using and XML instead of MCE) where the thermal tape appears to perform better than thermal paste.

Conclusion:

The only conclusion I can draw from the data is that (over 3 minutes) Thermal Paste and Thermal Tape have almost exactly the same performance. Remarkably identical in fact. This (apparently) "busts" the myth that thermal paste is far superior to thermal tape. There is some hint, at time 180 seconds, that thermal paste might be gaining an advantage, but it is incredibly slight. A longer study, maybe out to 10 minutes, would need to be conducted.

I don't think this closes the book on the issue, but IMHO the data clearly debunks the thermal paste trumps thermal tape myth.
 
Last edited:
hellokitty[hk]

hellokitty[hk]

Enlightened
Joined
Jan 29, 2007
Messages
645
Why are you showing relative output starting at 30 seconds? I think it should be on turn on, that would show how much light output decreases relative to the room temperature output, which would be at zero.

I'm also not sure about the slope. Theoretically, both lights will come to an equilibrium point where the host is going to dissipate heat at the same rate that the LED is making heat, and the slope will be the same for both compounds. However, the one with the superior thermal compound is going to have a better thermal path, less temperature difference between the outside of the host and the air LED junction, and a lower equilibrium temperature.
 
Last edited:
archer6817j

archer6817j

Enlightened
Joined
Aug 9, 2010
Messages
830
Location
Mountain View, CA
Why are you showing relative output starting at 30 seconds? I think it should be on turn on, that would show how much light output decreases relative to the room temperature output, which would be at zero.

I see your point. If we then look at the drop in the interval from turn on to 30 seconds (the first chart):

AS5: 422-396 = 26
BondPly: 406-367 = 39

This makes the difference 13 lumens: advantage goes to AS5

That's still an incredibly small difference, something like 3%. Based on that, I would still say "busted." :)
 
hellokitty[hk]

hellokitty[hk]

Enlightened
Joined
Jan 29, 2007
Messages
645
Interesting testing nonetheless. The most important factor IMO is dissipation ability of the host.
 
fyrstormer

fyrstormer

Banned
Joined
Jul 24, 2009
Messages
6,620
Location
Maryland, Near DC, USA
That's still an incredibly small difference, something like 3%. Based on that, I would still say "busted." :)
I think most people's experiences with thermal compounds come from much higher-wattage applications. Sure, an XM-L emitter can handle...what is it, 5 watts, or thereabouts? Well, your average computer CPU can handle 100 watts, so much power that without a heatsink it will shut itself down before the screen on the computer has a chance to turn on. In that scenario, the quality of the interface between the die and the primary heatsink is far more important. Heck, I've seen CPUs cool down by over 10 degrees C simply by having old thermal paste replaced with new thermal paste of the exact same type.
 
badtziscool

badtziscool

Flashlight Enthusiast
CPF Supporter
Joined
Oct 13, 2006
Messages
1,655
Awesome data and a good approach to the experiment. I think sometimes people use thermal paste for something other than it's intended purpose. It's only meant to fill in the minute gaps and crevices of a metal to metal interface to allow for maximum thermal transfer of an imperfect surface. If you can get a perfectly flat surface mate with another perfectly flat surface, then there's no need for thermal paste/tape, etc. Of course that's almost impossible in an environment that can't be precisely controlled and hence the need for the filler medium.

So the myth that paste > tape should depend on how it's used. To fill in minute gaps and spaces? Yes, paste > tape obviously. But to fill in large gaps or spaces? Probably not.
 
N

notrefined

Enlightened
Joined
Sep 4, 2004
Messages
383
Location
Illinois
two things I'd like to see if you do this again...run the lights off a bench power supply to eliminate the confounder of battery performance (which should improve as the battery is warmed, favoring also the better thermal path, but the system is complex so better to simply eliminate that aspect), and compare both to a soldered thermal path if feasible
 
fyrstormer

fyrstormer

Banned
Joined
Jul 24, 2009
Messages
6,620
Location
Maryland, Near DC, USA
Awesome data and a good approach to the experiment. I think sometimes people use thermal paste for something other than it's intended purpose. It's only meant to fill in the minute gaps and crevices of a metal to metal interface to allow for maximum thermal transfer of an imperfect surface. If you can get a perfectly flat surface mate with another perfectly flat surface, then there's no need for thermal paste/tape, etc. Of course that's almost impossible in an environment that can't be precisely controlled and hence the need for the filler medium.

So the myth that paste > tape should depend on how it's used. To fill in minute gaps and spaces? Yes, paste > tape obviously. But to fill in large gaps or spaces? Probably not.
Which is why solder works better than both thermal paste and thermal tape, if you have the option to solder.
 
archer6817j

archer6817j

Enlightened
Joined
Aug 9, 2010
Messages
830
Location
Mountain View, CA
Which is why solder works better than both thermal paste and thermal tape, if you have the option to solder.

100% in agreement. I've done some testing with a few members here and confirmed the assumption...direct solder is very superior to any other bonding solution. I don't have any real numbers but it "seemed" like around a 15-20% increase in output and much less drop over time...until the sink gets saturated at least.

I've been meaning to machine some copper stars and do the same test to find out if that actually makes a measurable difference. My hypothesis is that it might have a significant effect at < 30 seconds, but above that the interface material will be come the limiting barrier. The most superior solution would be soldering directly to a copper pill...or solid copper head and skip the star all together. Unfortunately none of those solutions are feasible at this stage in terms of logistics and implementation (for me). That's pretty "bleeding edge" though. Not uncommon in modding, but I'm not aware of any manufacturer that direct solders their LEDs. If you know of one feel free to bring it to our attention :) I'd love to move in that direction but I'm going to need a lot more volume to justify the capital needed to make that happen.
 
badtziscool

badtziscool

Flashlight Enthusiast
CPF Supporter
Joined
Oct 13, 2006
Messages
1,655
I believe LamdaLights are the only ones who do direct led-to-copper bonding and yeah, that is pretty bleeding edge technology. Probably not something easily justifiable in terms of a manufacturing and cost standpoint.
 
fyrstormer

fyrstormer

Banned
Joined
Jul 24, 2009
Messages
6,620
Location
Maryland, Near DC, USA
100% in agreement. I've done some testing with a few members here and confirmed the assumption...direct solder is very superior to any other bonding solution. I don't have any real numbers but it "seemed" like around a 15-20% increase in output and much less drop over time...until the sink gets saturated at least.

I've been meaning to machine some copper stars and do the same test to find out if that actually makes a measurable difference. My hypothesis is that it might have a significant effect at < 30 seconds, but above that the interface material will be come the limiting barrier. The most superior solution would be soldering directly to a copper pill...or solid copper head and skip the star all together. Unfortunately none of those solutions are feasible at this stage in terms of logistics and implementation (for me). That's pretty "bleeding edge" though. Not uncommon in modding, but I'm not aware of any manufacturer that direct solders their LEDs. If you know of one feel free to bring it to our attention :) I'd love to move in that direction but I'm going to need a lot more volume to justify the capital needed to make that happen.
If you machine the copper heatsink with a tiny pedestal in the center that can reach up to attach to the underside of the emitter, while leaving room for the power wires to attach to the emitter on either side of the pedestal, that would be ideal. Less ideal but probably just as effective is to solder a tiny copper puck to the non-conductive part of the emitter base, solder the wires next to it, then solder the copper puck to the copper heatsink.

When I modded an Arc6 to use a Cree MC-E, I used the soldered-copper-puck solution to produce a usable thermal interface on the bottom of the MC-E emitter. I had to thermal-epoxy the copper puck to the heatsink inside the Arc6 head, but that's because the Arc6 head was anodized throughout and solder wouldn't stick. I would've soldered the emitter to the puck and the puck to the heatsink if I'd had the option.

McGizmo emitters are direct-soldered to their heatsink boards. They're not direct-soldered to the inside of the head, though, because his lights use a removable-pill design. The emitters themselves are direct-soldered though.
 
Last edited:
Cavi Mike

Cavi Mike

Banned
Joined
Dec 24, 2011
Messages
29
Location
Rochester, NY
I've never heard of this discussion on a computer over-clocker board and I don't think I ever will. Paste will always win because tape is too thick.
 
T

The_bad_Frag

Enlightened
Joined
Mar 13, 2011
Messages
233
Location
Germany
I think the tape still looses over all.

After 5-8 years its conducting heat very badly. I disassembled a lot of old computer hardware and every time there is a thermal pad its hardened and dry. The thermal paste on the other side is still good after 10 years or more.

And by the way AS5 is not really the best you can get. You can use a liquid metal. I think the easiest way to use this is not in liquid phase in a syringe like AS5 but as a pad. http://www.coollaboratory.com/en/products/liquid-metalpad/ . Thats the real high end stuff. :D


But I think it would be a much much bigger improvement to use these led pcbs where the led is soldered directly to the copper. http://www.candlepowerforums.com/vb...ated-copper-star!-*update*-XM-L-T5&highlight= Thats the best in thermal management you can do to your lights! ;)
 
saabluster

saabluster

Flashlight Enthusiast
CPF Supporter
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
I believe LamdaLights are the only ones who do direct led-to-copper bonding and yeah, that is pretty bleeding edge technology. Probably not something easily justifiable in terms of a manufacturing and cost standpoint.

I've been doing it for years in my lights. It is not as important when running these lights at lower intensity but overdriving requires the use of the best of the best.


I've had a few critics say that if I had any clue what I was doing (and if I wanted to be taken seriously) I would ditch the tape and get with the paste. I wanted to find out if this was true.
Archer I applaud your quest for knowledge but it is a bit much to say myth busted. It is not busted I assure you. But then it all depends on application. What you are doing however is the best way to show everyone that at least in your application the tape works just fine. There have been some good suggestions from some here on how to make the results of your test more accurate but hopefully putting it all out there for everyone will help them have confidence your product which to my eyes seems very nice.
 
archer6817j

archer6817j

Enlightened
Joined
Aug 9, 2010
Messages
830
Location
Mountain View, CA
I think the tape still looses over all.

After 5-8 years its conducting heat very badly. I disassembled a lot of old computer hardware and every time there is a thermal pad its hardened and dry. The thermal paste on the other side is still good after 10 years or more.

That's true but I'm assuming you mean the foam-type pads? Bondply-100 is a .005 sheet of acrylic and fiberglass, from it's composition I can't see anything that would dry out.

And by the way AS5 is not really the best you can get. You can use a liquid metal. I think the easiest way to use this is not in liquid phase in a syringe like AS5 but as a pad. http://www.coollaboratory.com/en/products/liquid-metalpad/ . Thats the real high end stuff. :D

Sounds like interesting stuff I couldn't find any technical performance data. Let me know if you know where to find it. If I used it in my lights I have to rename it to the T-1000 :)
 
archer6817j

archer6817j

Enlightened
Joined
Aug 9, 2010
Messages
830
Location
Mountain View, CA
I've been doing it for years in my lights. It is not as important when running these lights at lower intensity but overdriving requires the use of the best of the best.

Archer I applaud your quest for knowledge but it is a bit much to say myth busted. It is not busted I assure you. But then it all depends on application. What you are doing however is the best way to show everyone that at least in your application the tape works just fine. There have been some good suggestions from some here on how to make the results of your test more accurate but hopefully putting it all out there for everyone will help them have confidence your product which to my eyes seems very nice.

Hey saabluster, thanks for weighing in :) Nice to have some real experts around.

With respect to being "busted" or not, based on the data I looked at, I can't draw any other conclusion. Inherent in saying "data I looked at" is the assumption that the data is specific to my application. However, I think it represents any application where the LED is mounted to a "standard" MCPCB and driven at or below (but not above) maximum spec. I agree that over-driving probably requires a "direct solder" attachment to keep the system in balance. I don't think paste would be adequate for that application either, but that's pure speculation on my part. I should clarify that by "adequate" I also mean removing heat quickly enough to protect the LED from catastrophic failure or otherwise drastically reducing it's life span because of excessive heat.

I think your design goals are the same as mine: optimize the system for balanced performance, given the engineering limitations (drive current, thermal mass, life span, etc.) So in my application, and virtually all others that use a standard MCPCB and drive the LED within spec, I'd still say there is no clear advantage to thermal paste...and hence the myth is still busted.

If I might be allowed more conjecture, the data may further indicate that the MCPCB thermal interface is actually the limiting thermal factor, not the thermal interface between the star and the heat sink. The next step would be to get some of those direct-solder stars in The Bad Frag's post and conduct the same test.

Another thing I try to keep in mind is what kind of performance gains are "worth it" in a real world sense. To use an analogy, I'm trying to make a BMW M3, not a formula 1 car. I put your lights in the latter category :) The liquid-metal pad company claims a 4-7 degree difference in CPU temperature with their product. Say your CPU temp is 50 degrees...that's a 10% difference. Quite a bit actually, but is it worth putting a $10 thermal pad in a light to get 10% increase in performance? At some point each lumen become far more expensive to squeeze out, and the practical difference (real world use) is not that great :) ...unless you are making a race car of course!
 
lolzertank

lolzertank

Enlightened
Joined
Dec 29, 2008
Messages
555
Location
The Land of Silicon
I see your point. If we then look at the drop in the interval from turn on to 30 seconds (the first chart):

AS5: 422-396 = 26
BondPly: 406-367 = 39

This makes the difference 13 lumens: advantage goes to AS5

That's still an incredibly small difference, something like 3%. Based on that, I would still say "busted." :)

Not busted, I think... A 3% difference indicates that the die of the LED with thermal tape is almost 20C hotter than the one with grease, based on the junction temperature vs flux curve given in the Cree MC-E datasheet. Considering the MCPCB should be the main bottleneck, not the tape or grease, the tape is a substantial increase in thermal resistance.

Besides, a normally driven MC-E doesn't need a fantastic thermal interface material if mounted on a half-decent star, since the star will spread the heat around. But with directly mounted (because the MCPCB itself becomes the problem) LEDs like the SST-90 things like this can happen: http://www.candlepowerforums.com/vb/showthread.php?289522-SST-90-smokes
 
saabluster

saabluster

Flashlight Enthusiast
CPF Supporter
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
Not busted, I think... A 3% difference indicates that the die of the LED with thermal tape is almost 20C hotter than the one with grease, based on the junction temperature vs flux curve given in the Cree MC-E datasheet. Considering the MCPCB should be the main bottleneck, not the tape or grease, the tape is a substantial increase in thermal resistance.

Besides, a normally driven MC-E doesn't need a fantastic thermal interface material if mounted on a half-decent star, since the star will spread the heat around. But with directly mounted (because the MCPCB itself becomes the problem) LEDs like the SST-90 things like this can happen: http://www.candlepowerforums.com/vb/showthread.php?289522-SST-90-smokes

You are quite right that 3% is significant. More so than it would seem at first glance. Since there is only a 30% drop in performance from 25ºC to the max allowed at 150ºC it can be viewed with that perspective. A 3% drop in output due to heat represents a 10% consumption of the allowable 30% ceiling. And as you say a not so insignificant 20ºC increase. That right there busts the myth that the myth is busted. That said.... it doesn't necessarily matter. There are always design compromises that need to be made or otherwise all of our lights would be $1K or more. Knowing where to make compromises and where not to is an art and has to be taken in to consideration with the end goal in mind. As such I don't necessarily see anything wrong with what he wants to do with his lights. Although that seems like a lot of current to run in a light of that size. I'm not intimately familiar with his design as to whether or not there is some electronic controls on overheating or he relies on the user to have good judgement. At any rate I love the aesthetics of his light.
 
M

MikeAusC

Enlightened
Joined
Jul 8, 2010
Messages
995
Location
Sydney, Australia
. . . . . The liquid-metal pad company claims a 4-7 degree difference in CPU temperature with their product. Say your CPU temp is 50 degrees...that's a 10% difference. . . . .

Actually it's a 7/323 = 2% difference in temperature.

Zero Deg C is the melting point of ice, an irrelevant reference point here. The CPU is 323 Degrees above Zero Temperature.
 
easilyled

easilyled

Flashaholic
Joined
Jun 25, 2004
Messages
7,254
Location
Middlesex, UK
You are quite right that 3% is significant. More so than it would seem at first glance. Since there is only a 30% drop in performance from 25ºC to the max allowed at 150ºC it can be viewed with that perspective. A 3% drop in output due to heat represents a 10% consumption of the allowable 30% ceiling. And as you say a not so insignificant 20ºC increase. That right there busts the myth that the myth is busted. That said.... it doesn't necessarily matter. There are always design compromises that need to be made or otherwise all of our lights would be $1K or more. Knowing where to make compromises and where not to is an art and has to be taken in to consideration with the end goal in mind. As such I don't necessarily see anything wrong with what he wants to do with his lights. Although that seems like a lot of current to run in a light of that size. I'm not intimately familiar with his design as to whether or not there is some electronic controls on overheating or he relies on the user to have good judgement. At any rate I love the aesthetics of his light.

One thing I can tell you about the design of archer6817j's lights is that the heat-sink is integral to the head. (all one unit)

In my all copper light, that's one massive heat-sink. So that once the heat has passed the theoretical barrier posed by the tape, it should flow out to the external surface quickly. This is certainly born-out by how quickly my all-copper light heats up (almost immediately on the highest level of 2.8A)
 
Last edited:
Top