Lapping a heat sink: Why?

[darrin]

Well I read (most) of the cooling article and I have a question as to why so many folks lap their heat sinks.

The reason I am asking this is that it seems to be similar to the acceleration issue raised many years ago. You see, a lot of folks seemed to think that no one would be able to break a certain speed limit (I think it was 200MPH, but I could be wrong) in the quarter mile because you could just get so much traction out of a set of tires (1g). Of course we all know that speeds in excess of 300 are common now and the limited traction theory was incorrect since hot rubber (tires) act like a liquid and conform to the road surface.

Now, my point in this is that, given a GOOD thermal paste (the Artic Silver III comes to mind), shouldn't the heat transfer from a lapped heat sink be even less than one that is just slightly rough?

For example, say we have a CPU surface that looks like the following: ----------- (ie, flat...very flat)

Now compare that to another surface that is just slightly rough like this: -_-_-_-_- (ie has very small valleys in it).

The surface area of the second example is greater than the first even though the size (WxH) is the same. Given the larger area, I would think that such a surface, provided that the thermal paste is pretty efficient, would actually be able to conduct more heat.

Does this make since? Does practice show that lapping a heat sink even when using a good thermal paste actually improves heat transfer?

 

[Aesik]

Thermal interface materials do not improve heat transfer. They are only there to bridge the gap (where air would normally be) between the core and the HSF or water block. These materials *at best* have conductivities around 9W/m-K whereas aluminum is 160-210 and copper 340-390. The thicker the TIM is, the higher the temperature drop across the interface will be and the less efficient the heat transfer is.

 

[nikhsub1]

First off lapping is done to make the surface of the heatsink FLAT, not mirror-like. You don't want the HSF to be concave or convex, a few scratches are ok, that is what the thermal paste is for.

 

[darrin]

So you are saying that the thermal paste, even a good one, acts more like an insulator than a conductor of heat in comparison to an alluminum or copper heat sink?

So, even though the surface area is increased which would tend to help the heat transfer, the fact that the material has to be thicker more than counteracts this correct?

 

[Aesik]

I think you are getting confused about when added surface area is good for heat transfer. This is the case in convection, where a fluid is moving over a solid surface and removing heat.

With a CPU-->TIM-->HSF setup, conduction is the means of heat transfer. The main factors that concern conduction are length of heat path, conductivity of the materials, temperature differential and cross-sectional area perpendicular to the heat flow. The cross sectional area is not being increased by having more paste, but the heat path through a lesser conductive material is being increased. This will in turn decrease the overall efiiciency of the system.

 

[darrin]

So you can increase the surface area all you want but it is only the cross-sectional area that makes any difference then?

I don't understand why an increase in the surface area wouldn't help if the thermal paste was a very good conductor (I know that it isn't in comparison...but if it were).

For example...if you had a large steel block and all you had to cool it with was a 1/4" heat sink. Wouldn't that cool off more slowly than if the same heat sink had it's surface area increased by having projections into the tub akin to what the paste would be doing?

Ahh....I guess the point is mute since the thermal pastes are so poor of a cunductor

Thanks for the info!

 

[Aesik]

Water in tub=fluid which means you are back to looking at convection again.

 

[darrin]

I edited it to be steel abrout two seconds after I sent it realizing it was an example by convection as you pointed out!

 

[vandersl]

Well, technically thermal paste DOES increase the cross-sectional area, in that any locations where the surfaces don't mate directly are available for heat transfer through the thermal material, rather than through air. So the effective cross-section is increased.

Just to put in in perspective. Assume 60W through a paste with TC of 9W/mK, 0.01mm thick (very thin), and 1cm square. The temperature drop would be 0.67 degrees C. Now if that were copper at around 401W/mK, the drop would be only 0.015 degrees C.

The real question is, when is the increase in effective cross-sectional area from using a paste offset by the increase in thermal resistance of the interface due to the paste being between the mating surfaces?

Hmm, I think I am going to try lapping to a mirror finish and not use any paste at all and see what happens

 

[Aesik]

Well, technically thermal paste DOES increase the cross-sectional area, in that any locations where the surfaces don't mate directly are available for heat transfer through the thermal material, rather than through air. So the effective cross-section is increased.

I think we were assuming that the paste did fill all voids, but yes, technically correct. The real issue is that by using more paste to fill larger gaps the distance through which the heat must travel through a lesser conducting material is increased.

Hmm, I think I am going to try lapping to a mirror finish and not use any paste at all and see what happens

You'd better be lapping the die as well

 

[darrin]

Doh!

Now I'm confused. So, the fact that there are surface irregularities on the die and the heatsink increases the cross-sectional area? I thought that was just WxH. I knew the surface area went up because of this, but how are you defining the cross-sectional area?

Also, are you now saying that an increase in the surface area does or does not make a difference with conduction?

 

[SunTzu]

Direct metal to die contact is better than thermal paste. But, because at the microscopic level, neither surface is particularly flat, if you attempt to mount a heatsink directly on a die, you end up with a lot of air pockets. Air, as we know, is a terrible heat conductor and acts more like an insulator when compared to direct die to metal contact. Paste helps alleviate this problem by filling in those air gaps.

I guess theorhetically, if we could perfectly match the mountains and valleys of one surface to another, it would probably aid in heat transfer as you suggested; however, doing so is far beyond the consumer technology we have today. Hence,it is better to get rid of the mountains and valleys as much as possible (lapping) and just applying thermal grease.

I know that there is a way to weld two separate objects together simply by pushing together two EXTREMELY flat objects at a very high pressure. At the atomic level, the high amount of exposure of both materials causes to atoms to start atract each other to such a degree that the materials actually become stuck to each other. Obviously, we won't be seeing this kind of technology being done at home anytime soon, so in the meantime, we'll have to stick to our good ole' thermal goop

 

[SunTzu]

Hmm, I think I am going to try lapping to a mirror finish and not use any paste at all and see what happens

Unless you have god-like precision with your hands and are using 3,000,000,000,000,000 grit sandpaper, I don't think this will work very well .

 

[Aesik]

Now I'm confused. So, the fact that there are surface irregularities on the die and the heatsink increases the cross-sectional area? I thought that was just WxH. I knew the surface area went up because of this, but how are you defining the cross-sectional area?

Also, are you now saying that an increase in the surface area does or does not make a difference with conduction?

No, the irregularities do not increases the cross sectional area. As you say, the area is simply WxH. An increase in surface area will not help to increase conduction. However, an increase in cross-sectional area will help move more heat.

vandersl was just saying that the paste increases the EFFECTIVE cross sectional area because it fills in gaps that would otherwise be populated with air.

To make things overly simple:

Conduction: cross sectional area increases the amount of heat that can be moved.

Convection: surface area increases the amount of heat that can be moved.

Now I know that doesn't even begin to cover the basics of heat transfer, but I think it should answer your question if only in a very general manner.

 

[darrin]

Ahh! I'm happy again

It's all clear now. All things else being equal, the bottleneck for conduction transfer is the cross-sectional area. For convection, it is the surface area. I suppose that makes since now that I think about it.

Thanks!