Thermal compound conductivity choice...rule of thumb? Need to do the maths?

[K5_489]

Went to buy more thermal compound today....it's been MANY years since I last had to restock.  I used to alway just go to Radio Shack or Frys electronics, look for the expensive silver stuff, and buy the biggest tube they had on the shelf. 

Today, I went to everyone's favorite first stop of Amazon, and realized there were many more choices.  I had absolutely no idea you could get this stuff in tub form vs just the syringe tubes and noticed there were different "levels"? of thermal conductivity in watts per meter kelvin.  Ok, Google time, where I learned that it's a rating of how much heat it will conduct in watts based on surface area of the compound.  Ok, simple enough. 

But then how do I apply that to the parts I need the compound for?  Such as the current project - replacing 2N3771 transistors on an Astron 30 amp power supply.  I suppose the simple answer is "buy the highest rating you can find", but then these things are rated to dissipate 150W through the collector..clearly the tiny smidgen of paste I'm going to put on the case that's rated to transfer 3.17W/m-k isn't going to handle that, nor will the highest rating I've seen so far of 6.something watts. 

So clearly I'm missing something here.  Or maybe looking at the wrong products altogether?  The goop I'm always scraping off transistors and heat sinks have looked like dried up white lithium grease (I'm assuming it isn't actually simple lithium grease, just has the same appearance), whereas the thermal compounds I've been buying have looked like the gray stuff commonly found on CPUs.  I was under the impression that the gray stuff is "better", but I wouldn't be able to tell you why. 

[T3sl4co1l]

Wakefield 120.  The basic, plain-yogurt, of thermal grease, and only a little less appetizing. (No please, don't actually eat it..)

Thermal conductivity doesn't matter much, because a properly prepared joint has very little air gap to fill, and excess grease squidges out, leaving a minimum height joint.  The transistor (RthJC), insulator (if applicable) and heatsink dominate overall thermal resistance.

Perspective shift: I would say If grease is adding notable resistance, the joint is wrong, not the grease.

AFAIK, lithium grease is a fairly ordinary petroleum or synthetic oil, gelled with a lithium soap (a low-solubility variant of the active ingredient in regular bar soap; basically, soap scum).  The soap particles partially dissolve, thickening the oil, but also disperse, remaining in suspension, forming a gel that stays in place, and giving something solid but non-abrasive for the bearing surfaces to rest on, when not in (hydrodynamic) motion.

Thermal grease is made much the same way, of course it isn't used under relative motion, heh.  As a result, more abrasive materials can be used as the filler; ZnO is typical.  (Indeed, wiping thermal grease off a bare aluminum heatsink, you'll find it leaving dark streaks, as the particles abrade and polish the metal.)  A silicone oil is usually used, being chemically inert, and perhaps better at staying in place.  Still, a common failure mode -- whether by improper formulation, application, overheating, thermal cycling, or it just does that -- is separation, leaving a dry powdery joint under the device, which needs to be cleaned and reapplied.

So, yes indeed -- it has a similar appearance, for similar reasons, but using different materials.

Anyway, high-K grease: the trouble is, this can only be done two ways: using higher-K filler, and using less oil.  You can very well end up worse off, because the grease doesn't squeeze out, and now you have, sure it's 15 W/(m.K) or whatever, 20 times better than the plain white stuff, but maybe the white stuff crushes down to 20µm average, while you've got a 0.5mm thick blob under the device, and it's not even fully spread out, maybe the contact area is half.  Or maybe the clamping force is so great, and uneven, that the device is cracked before it can sink any power at all(!).  Applying those super-thick compounds is nontrivial.

But, combining all of the above: if you're in a situation where the joints just aren't flat, there's big gaps, the faces are rounded, whatever, nothing fits right, and it's not practical to fix the surfaces; that might be a good place to use a higher-K grease, instead of the low-K stuff that'll end up filling large pockets and adding significant Rth.

Soft rubber gap pads might also be attractive: these are available in fairly high K nowadays too (5 and up), and the softness means they fill voids very well.  An excellent application is using them to clamp a PCB, components and all: they can fill in the gaps between SMT components, making contact all around, as well as to the board itself, vastly increasing power dissipation.  Downside: they are solid rubber products still, not puttylike (well, some of them maybe?), so start to crack and tear when strained too far; they aren't great for rigid (screw and strong spring-clamp) mounting.  On the upside, being soft rubber materials with some surface tack, they need hardly any holding force, and a soft spring clamp can do the job.  Self-adhesive (firmer, very tacky) materials are also available.

Tim

[K5_489]

Ok...maybe I'm misunderstanding the need for this stuff altogether?  Or was that a long way of saying "buy the cheap white stuff unless you have identified a specific need due to something like mating surfaces having a poor fit up"?

I've been under the impression that when attaching things like these high power transistors or CPUs to a heatsink, you absolutely positively 100% NEED to apply some sort of thermal compound to the mating surfaces, otherwise the magic smoke is gonna escape post haste? 

At least from an anecdotal standpoint, I've found that when I forget the paste on a CPU, the thing will throttle within seconds of starting up, sometimes not even completing POST due to overheating.  I don't have much experience with transistors, as I'm just now starting my adventure in learning electronic repair. 

As for the current project, I just popped the four transistors off the back wall of the power supply along with the mica insulators, found the dried up greasy goo, and thought it would be wise to clean it up and put fresh goo down.  If it wasn't for being out of greasy goo, I would have grabbed a tube of the same silver stuff I'd put on a computer CPU, put some on the transistors and been on my way.  It wasn't until I started looking at specs of different greasy goos that I realized there may far more to this, and I always like learning better ways of doing things. 

[Smokey]

Paste with diamond in it!  Look it up!

[IanB]

Ok...maybe I'm misunderstanding the need for this stuff altogether?  Or was that a long way of saying "buy the cheap white stuff unless you have identified a specific need due to something like mating surfaces having a poor fit up"?

I've been under the impression that when attaching things like these high power transistors or CPUs to a heatsink, you absolutely positively 100% NEED to apply some sort of thermal compound to the mating surfaces, otherwise the magic smoke is gonna escape post haste?

Broadly speaking, thermal grease is bad for heat conduction, except for the alternative, which is worse.

When you try to mate two metal surfaces together, if you look under a microscope, you will find they only touch at a few points, and most of the gap between them is filled with air, which is a very good thermal insulator. This is why not using thermal paste leads to bad results.

The job of thermal paste is to fill the gaps, and replace the air with something more thermally conductive. It only has to be more conductive than air, not be as conductive as metal (not likely). The reason it doesn't have to be as conductive as metal is that the space to be filled is very thin, and therefore not much thermal paste should be in the conduction path.

If you wanted to go totally crazy, the best thermal result would be to solder the part to the heatsink, but this is rarely practical or feasible. One time it is totally feasible is with surface mount parts that have a metal pad underneath to be soldered to the copper heat sink layer on the PCB.

The short summary is that any kind of filler is better than no filler, even toothpaste or plain oil has been proposed! But in reality, thermal paste is thermal paste, and good enough is good enough for most ordinary purposes.

[T3sl4co1l]

Yes, "cheap stuff is fine".

But also that, yes, there are reasons to use various products.

Without paste, there's only air (or perhaps fingerprint oils, dust and such) in the gap, a particularly poor conductor for one, but it gets worse than that: on a microscopic scale, both surfaces are rough, and make contact at very few points (surface asperities).  Heat can flow directly between surfaces where they touch, but these points are few and far between.  Around contact points, the separation might be some nm, and they can't "feel" each other anymore: no conduction.  In fact, since the mean free path of molecules in air itself is only about 50nm, for gaps less than this, there's effectively vacuum between faces -- only radiation carrying heat, between (probably reflective metal) surfaces.  But most of the gap, for average surface finish and flatness, will be in the 10s of µm, where a classical air filling is accurate enough.  Which, with so little air space, and temperature drop, convection is negligible, and it's like a single layer blanket.

So yeah, just a little filling -- anything at all, it could be plain oil, but it's better with a conductive (solid) filler, and stays in place better as a grease, paste, or rubbery material -- improves things dramatically, and further improvement (in grease properties) makes very little change in the average case.

Outside of the average case, this may be less true, and more benefit may be had from better greases.

Tim

[showman]

In fact, since the mean free path of molecules in air itself is only about 50nm, for gaps less than this, there's effectively vacuum between faces

Don't you think that the heat transfer could actually be better in that case? If the distance is smaller than the mean free path then effectively the energy can be directly transported from the hot to cold side, i.e. a molecule gains energy from the hot side and instead of a long process via random collisions with other air molecules, it is likely to hit the cold side next and lose some of its energy.

Also, the figures provided by thermal compound manufacturers who focus on computers are not to be trusted: https://www.arctic.de/en/faq/detail/why-doesn-t-arctic-communicate-thermal-conductivity-values

Why doesn’t ARCTIC communicate thermal conductivity values?

ARCTIC made a conscious decision not to specify any values for the thermal conductivity of its thermal paste and thermal pads, because many manufacturers invent, artificially inflate or embellish this value. Thermal paste has a thermal conductivity of 1 to 4 W/mK. Values outside of this range, such as 12.5 W/mK, are at odds with the truth.
Many competitors quote values above 4 W/mK to suggest better performance. This often leads to false expectations and dissatisfied users.
ARCTIC offers its customers innovative thermal interface materials at the best possible price-performance ratio instead of relying on manipulated performance data.

[T3sl4co1l]

Don't you think that the heat transfer could actually be better in that case? If the distance is smaller than the mean free path then effectively the energy can be directly transported from the hot to cold side, i.e. a molecule gains energy from the hot side and instead of a long process via random collisions with other air molecules, it is likely to hit the cold side next and lose some of its energy.

Honestly, I don't know the dynamics of heat transfer in such a regime -- that's an interesting question.

My (hand-waved) understanding of it is, there's simply far fewer molecules in the gap at all -- if there were more, they'd be colliding with each other as well, right?  Or put another way, think of how often the surfaces are hit by "bouncing balls", and where; the amount of area covered by those hits (over a given time frame), decreases as the surfaces come together and there's less gas in the gap.  At the diffusion limit (say 50nm, up to some ~mm where convection can maybe start to help), conductivity is as expected, but below, in this pseudo-vacuum limit, it drops...perhaps proportionally with separation? Until it goes back up again when the surfaces can "feel" each other (Van der Waals forces, overlap of atomic orbitals).

Also to be clear, heat transport in still air is mediated by diffusion, so, it's kinda not very effective, and conductivity scales with... sqrt(distance), probably?

(I don't think we ever covered thermal transport in gasses in school -- maybe chem eng would? -- and I don't recall working any problems about the interface between matter in Stat Mech.  Oh god, imagine how messy those problems must be...  And, not that I remember much of it these days, it's been so long, lol.  Can't say I've read much about transport in gas or (near) vacuum since then, either; it's not really a very important topic in general anyway, with applications probably like, outer space... or a curiosity like Crooke's radiometer.  Which come to think of it, that the latter has been such a mystery for centuries, is probably suggestive in part of how "rarefied" (if you'll excuse me ) coverage of this topic is, given that the currently-accepted explanation of adsorption-driven heat transport.  Neat, huh?)

Tim

[IanB]

I don't think we ever covered thermal transport in gasses in school -- maybe chem eng would?

As it happens I'm a chemical engineer, and for sure, heat transfer is one of the core topics. However, most industrial treatment of heat transfer addresses large scales, where the behavior of molecules can be averaged out to bulk phenomena. Much treatment of the subject where fluids are concerned is empirical, where investigators do lots of experiments, collate the results, and fit correlations to the data. This is because first principles, analytical solutions are hard to come by. The most accurate predictions come from CFD modeling, but this is still a numerical solution to the governing macroscopic equations.

If you reach low enough pressures, or small enough scales where the mean free path of gas molecules becomes significant, then different theory is needed, and I am not familiar with this area. It is unusual, and not normally encountered in everyday engineering.

[T3sl4co1l]

Yeah, we did thermo as well, basic fluids and heat stuff, but all very macroscopic (and empirical).

As you can see, most cases it's a bad thing easily avoided -- chemists usually want more pressure to shrink reactor size, raise boiling point, etc.; heat transfer you just use the grease and don't care about the air-gap case.  But that inevitably leaves the useless situation something of a mystery.

Tim

[coppercone2]

lower is better always, lets say a bolt loosens and contact gets worse. The better the compound the more durable the thermal interconnect will be. Lower thermal resistance means cooler parts.


You have the choice of cheap or expenisve bolts, cheap or expensive thermal pads and cheap or expenisve thermal compounds. It formulates into a quality decision

you can also measure it with a good micrometer of the correct type to see how thick your bond line is, the question is, is the design compatible with the correct micrometer? It may require a special fixture to verify in many standard PCB cases. A mechanical micrometer can do 0.0001 inches resolution, meaning 2.5 micrometers.

[Sorama]

It doesn’t really matter what soort of paste (or any liquid for that matter) is used.

There are only 3 ways of energy transfer:
Conduction
Radiation
Convection

So in this case, even radiation is no great because of the poor distance.
Conduction how ever is your best friend in this application and for that, the more space is covered with paste/compound ( more contact between component and heat sink) the better.

The energy flux depends of the temperature delta, the surface and the thermal resistance.
Even toothpaste works fine, as long is doesn’t get burned.

[Phil1977]

Having done many FEM simulations about that topic and also experimentally tested the results I can contribute:

- Thermal conductivity in W/mK of the thermal paste does NOT much influence the results as long as you use stuff that´s called thermal compound. I wouldn't advise to use toothpaste, but every cheap ZnO-filled grease can do this job.
- A thin interface is always better than anything thick. Avoid silicone-foils for that reason.
- Mechanical stability of the connection is super important. If the interface moves only a few um then even the silicone foil can be of advantage.
- A copper heat spreader inside an Al-heatsink reduces component temperature much more than all other experiments with different thermal compound. It seems to be counter-intuitive to improve cooling by introducing more material and an additional interface, but heat transfer is 3-dimensional and not 1-dimensional as in a resistance chain. FEM simulations well confirm this.

[Ice-Tea]

Unless I'm misreading your original post, it seems you're mixing up W/m.k with W. They don't have to be equal or higher or something. Here's an intuitive way to understand what W/m.K actually means.

A 1W/m.k rating means that if you take a cube of 1m x 1m x 1m of material and you inject 1W of heat on one side (over 1m²) that cube it will cause a delta T of 1K over that blocks thickness of 1m. After that, you can play around with that cube. Per example, you probably don't have an interface of 1m² but, say, 1cm². That means that for the same 1m thickness of material, you now have a delta T of 10 000K. But obviously, you won't be using a 1m thick material but, say, 1mm. So, the delta T becomes 10K.

Now, thermal paste is not supposed to have much thickness at all. It's there to fill microgaps and uneven surfaces. Per example, you might have a layer thickness of 25um. For the same 1W/mK material as above over 1cm² that would be a delta T of 0.25K. For the 150W you mentioned: 37.5K.

A few pointers:
- thermal conductivity for paste is important but controling surface roughness is at least as important
- often the importance of the paste is secundary to other parameters like die-to-case thermal resistance etc
- the 150W dissipation in datasheets should be taken with a considerable grain of salt. Ussually only achievable if half a dozen vestal virgins pee a continuous stream of liquid nitrogen on it.

[showman]

Don't you think that the heat transfer could actually be better in that case? If the distance is smaller than the mean free path then effectively the energy can be directly transported from the hot to cold side, i.e. a molecule gains energy from the hot side and instead of a long process via random collisions with other air molecules, it is likely to hit the cold side next and lose some of its energy.

Honestly, I don't know the dynamics of heat transfer in such a regime -- that's an interesting question.

My (hand-waved) understanding of it is, there's simply far fewer molecules in the gap at all -- if there were more, they'd be colliding with each other as well, right?  Or put another way, think of how often the surfaces are hit by "bouncing balls", and where; the amount of area covered by those hits (over a given time frame), decreases as the surfaces come together and there's less gas in the gap.  At the diffusion limit (say 50nm, up to some ~mm where convection can maybe start to help), conductivity is as expected, but below, in this pseudo-vacuum limit, it drops...perhaps proportionally with separation? Until it goes back up again when the surfaces can "feel" each other (Van der Waals forces, overlap of atomic orbitals).

Neither do I, but yes it is indeed interesting.

I need to think about your interpretation a bit, draw some figures, maybe write a small simulation. Yes, if there are fewer bounces then I think you could either directly or effectively interpret it as lower pressure and therefore as some type of vacuum and lower heat transfer, but I'm not totally convinced that it is the case. But neither am I convinced by my very simplified view mainly because those direct bounces from one surface to the other do not necessarily imply energy transfer. It does not help that most common literature I can find is about this subject is for nanopourous materials (aerogels etc), but in those cases they seem to be considering the gaseous heat transfer along the pores, not across those.

[TimFox]

To get a good result with a real thermal grease (if electrical insulation is not required) you need decent metal surfaces (watch out for burrs on drilled holes), complete coverage of the area, and a very thin layer of grease, with good compression force on the transistor package.
Insulating spacers (e.g. mica) should have a thermal spec and need two grease layers.

[IanB]

I need to think about your interpretation a bit, draw some figures, maybe write a small simulation. Yes, if there are fewer bounces then I think you could either directly or effectively interpret it as lower pressure and therefore as some type of vacuum and lower heat transfer, but I'm not totally convinced that it is the case. But neither am I convinced by my very simplified view mainly because those direct bounces from one surface to the other do not necessarily imply energy transfer. It does not help that most common literature I can find is about this subject is for nanopourous materials (aerogels etc), but in those cases they seem to be considering the gaseous heat transfer along the pores, not across those.

Temperature in an ideal gas corresponds exactly to the average kinetic energy (speed) of the gas molecules. What happens when a gas molecule bumps into a hot surface is that it gets a little kick from the vibrating molecules at the boundary, and bounces off faster than it arrived. In the process it takes a little energy away from the hot surface. If it then bumps into a colder surface, it imparts a little jolt to those surface molecules, losing some energy in the process.

When we look at the situation on a large scale, there are so many molecules that the different speeds of all the molecules average out, and heat transfer can be looked at as a continuous process, rather than a discrete process of individual kicks and jolts.

What we perceive as temperature does not actually exist on a quantum scale. It is just rotational, vibrational and kinetic energy of molecules. Thermodynamic temperature only becomes a thing when we average out the energy macroscopically.

So, if you had thinly spaced trapped molecules bouncing back and forth between hot and cold surfaces like tiny rubber balls, they would still be transferring heat from one surface to the other, and nothing would really change about what is observed. Air molecules are so small, and so numerous, that even in a vacuum, with a tiny gap, there is still an uncountably huge number of molecules present. So macroscopic thermodynamics will still apply, and heat transfer by conduction will still happen just the same (although the measured thermal conductivity may change).

[IanB]

To get a good result with a real thermal grease (if electrical insulation is not required) you need decent metal surfaces (watch out for burrs on drilled holes), complete coverage of the area, and a very thin layer of grease, with good compression force on the transistor package.
Insulating spacers (e.g. mica) should have a thermal spec and need two grease layers.

I think if you have a package like a TO-220, and if electrical insulation is not required, you can solder such packages to the heat sink. Especially if the heat sink is a large copper area on a PCB. You could even solder them to an aluminum heat sink, if you can overcome the challenge of soldering to aluminum, and getting solder to melt on a device purposely designed to suck the heat away 

[K5_489]

Unless I'm misreading your original post, it seems you're mixing up W/m.k with W. They don't have to be equal or higher or something. Here's an intuitive way to understand what W/m.K actually means.

All I know is that up until yesterday, all I ever knew of thermal greasy goo was that I looked for a tube that said silver on it on the shelf at the local electronics place, because I figured more expensive was better and for the tiny amount I was using the difference between the cheapest and the most expensive was insignificant.  I then smeared a thin layer on my part, occasionally wondered why the tiny bit of dried and crusty greasy goo I was removing was white and I was putting gray down, and went on my merry way. 

It wasn't until yesterday that I even knew there was such a rating of W/m-k on greasy goo, and I certainly didn't know just WTF W/m-k meant.  I also then learned I could get this stuff in tubs that were MUCH cheaper than the prefilled syringe type tubes on a per gram basis, and wondered if I was maybe using the wrong products or needlessly throwing money away on stuff that was much better than what I really needed, because I noticed I could get 200g of the cheap stuff for the same price as only 50g of the "better" stuff.  Though honestly, based on the rate that I've been using it, 50g would probably last me the rest of my life, lol. 

[darrellg]

I also then learned I could get this stuff in tubs that were MUCH cheaper than the prefilled syringe type tubes on a per gram basis, and wondered if I was maybe using the wrong products or needlessly throwing money away on stuff that was much better than what I really needed, because I noticed I could get 200g of the cheap stuff for the same price as only 50g of the "better" stuff.  Though honestly, based on the rate that I've been using it, 50g would probably last me the rest of my life, lol.

If you really want to "save" money, you can buy a 30 lb (13.6078 kg) bucket from Grainger.
https://www.grainger.com/product/SUPER-LUBE-Silicone-Heat-Sink-Compound-44N789

[TimFox]

Since it’s impracticable to control the exact thickness of the grease layer, its thermal conductivity value is only useful when comparing different greases to each other.  Mica washers don’t compress, so their conductance (not conductivity) value is relevant.

[K5_489]

[quote author=darrellg link=topic=441503.msg5665995#msg5665995 date=1728068827

If you really want to "save" money, you can buy a 30 lb (13.6078 kg) bucket from Grainger.
https://www.grainger.com/product/SUPER-LUBE-Silicone-Heat-Sink-Compound-44N789
[/quote]

Can that actually be used?  Funny enough, I have a few of the 3oz tubes of that stuff for welding and soldering to keep adjacent parts from getting too hot. 

[TimFox]

Grainger's page says: "These compounds dissipate heat during welding, brazing, and soldering. They are spread onto adjacent pipes and absorb heat so as to protect the equipment that they are plumbed up to."
As Hamlet said, “There are more things in Heaven and Earth, Horatio, than are dreamt of in your philosophy.”

[showman]

So, if you had thinly spaced trapped molecules bouncing back and forth between hot and cold surfaces like tiny rubber balls, they would still be transferring heat from one surface to the other, and nothing would really change about what is observed. Air molecules are so small, and so numerous, that even in a vacuum, with a tiny gap, there is still an uncountably huge number of molecules present. So macroscopic thermodynamics will still apply, and heat transfer by conduction will still happen just the same.

Well no. This logic definitely breaks down, for example aerogels that contain air can have significantly lower thermal conductivity than air itself https://en.wikipedia.org/wiki/Aerogel#Knudsen_effect https://en.wikipedia.org/wiki/Knudsen_diffusion As I said, I think that this is because the thermal conductivity decreases along the pores, not across it, but that might be my wrong interpretation of the Knudsen effect. Also, I'm much more familiar with solids and at least for those the interfaces themselves have thermal resistance. For example if you put two dissimilar metals in (perfect, as in atoms touch each other without any vacuum or gas) contact there is a sharp drop of temperature at the interface although the common logic would dictate that the temperature drop is continuous and depends only on the thermal conductivities of the two metals. That is in simplistic terms because some of the energy gets refected back because of the discontinuity.

[IanB]

Well no. This logic definitely breaks down, for example aerogels that contain air can have significantly lower thermal conductivity than air itself https://en.wikipedia.org/wiki/Aerogel#Knudsen_effect As I said, I think that this is because the thermal conductivity decreases along the pores, not across it, but that might be my wrong interpretation of the Knudsen effect. Also, I'm much more familiar with solids and at least for those the interfaces themselves have thermal resistance. For example if you put two dissimilar metals in (perfect) contact there is a sharp drop of temperature at the interface although the common logic would dictate that the temperature drop is continuous and depends only on the thermal conductivity. That is in simplistic terms because some of the energy gets refected back because of the discontinuity.

No, there is no breakdown in the logic.

Aerogel is an insulator in part because it traps air in pockets and prevents convection from occurring, which is another mechanism of heat transfer that is very important (probably most important) in free and open air. Heat sinks dissipate a lot of heat by convection, which is why fans (forced convection) make such a difference to the rate of heat transfer.

But in this thread we are talking about conduction, not convection. Conduction happens in small spaces where the air is not free to move. Thermal conductivity is a physical constant of air and does not change when the air is in pockets or otherwise (unless the pockets are microscopically small, apparently). In fact, aerogel typical air pocket insulators have a higher thermal conductivity than air because the thermal conductivity of the aerogel solid matrix material is greater than that of air. (For example, air has a thermal conductivity of 0.026 W/m-K, while typical solid materials are in the order of 50-100 times more conductive than that.)

If you have two dissimilar metals in perfect molecular contact (for example soldered or brazed), there is not, in fact, a sharp drop of temperature at the interface. The temperature gradient is smooth and continuous, although there will be a sharp change in the gradient at the interface due to the different thermal conductivities.

Edit: Updated to correct a misunderstanding about the special properties of aerogels.