"Why Heatsink Polishing Might Be A Bad Idea"
Simon Watkins - 5/14/01
I used to be involved in the manufacture of Acousto-Optic devices for use within high power laser cavities. One of the products was called a Q switch. It was basically a piece of quartz with polished ends through which the laser passed, and a piezoelectric transducer fitted on one face parallel and in line with the laser that could cause diffraction in the beam when it was energized.
Anyway to cut a long story short, the quartz block, because of optical losses and heating by the laser, not to mention high levels of RF energy being fed into the transducer would get very hot - we were pumping 50W CW of NdYag energy into the quartz, and the q-switching effect would result in peak powers of several thousand watts - so hot that the quartz would be destroyed in the laser cavity very quickly if it wasn't cooled.
I remember being involved in a number of experiments during the development of the product that investigated means to cool the device efficiently. In the end, water cooling blocks were used to sandwich the quartz. We experimented with improving the interface between the quartz and the aluminum blocks, and used exactly the same techniques people are using today with H/S and CPU. Heatsink compound, lapping etc.
The aluminum water blocks were made optically flat using the same lapping tools used to flatten the quartz. We are talking extreme flatness here - but flatness being key rather more than smoothness.
Anyway, one thing we found was that if we made the mating surface of the aluminum block too smooth, we lost cooling efficiency - similarly if it was too coarse. To the extent that if we polished the surface, we got significantly worse performance than when the heatsink surface was matte after lapping with, say, 400-600 grade carborundum.
Our conclusions were that the micro pits and valleys left behind after finishing with coarser grade carborundum left micro cavities within which the heatsink compound could fill, but the micro peaks of the aluminium would provide good physical contact to the quartz when the whole assembly was squeezed together under pressure; the micro peaks would flatten slightly providing millions of micro plateaus of contact surface.
On the polished blocks, there was nowhere for the heatsink compound to go, other than out the sides of course, but you could never exert sufficient pressure to squeeze ALL the heatsink compound out, so you got poor quality thermal transfer through the HS compound sandwich.
Whereas on the rougher surfaced blocks, given a sensibly thin application of heatsink compound, there would be a very high proportion of aluminium making contact with the quartz and any pits would be taken care of by the HS compound. If we went too coarse, there was too little of the aluminum in direct contact with the quartz and temperatures worsened again.
We found the results hard to believe at first, as we all believed somewhat blindly that the polished surface would be best, yet the flat surfaced, but coarser finished heatsinks outperformed the polished ones significantly.
Basically with a polished sink, you'll never put a thin enough coat of H/S compound on it for it all to be expelled, given the relatively limited pressure you can apply. So what is key to achieving optimum thermal conductivity is a high aluminium to CPU contact - micropeaks and perfect flatness will ensure that you achieve that. Try using only up to 600 paper on your heatsinks, no finer.
The manufacturers are possibly aware of this fact as well. Take an FOP38 heatsink for instance. Its base is very flat (I measured mine with an optical flat), but pretty coarse (too coarse I think). Now GlobalWin is perfectly capable of machining the thing somewhat smoother - the sides of the heatsink are finished to a much finer finish than the base.
I suspect the reason for their reasonably coarse base finish is to accommodate the somewhat viscous phase change material, as recommended by AMD. So lapping the heatsink to a finer degree and using heatsink compound instead will help in reducing temps to a degree (groan).
Additionally, the somewhat more viscous nature of Arctic Silver will also help with a coarsely finished interface, because it has a lower thermal resistance than silicon HS compound - that is, IF they are both applied so they are forming the filling of a sandwich!
In a more professional application of the heatsink compound, ideally you need a less viscous substance so that maximal metal to CPU contact can occur. If you can't achieve that because you are making a compound sandwich, you will never get optimum temps.
Anyway, take that info for what it cost you. I just thought it may be of interest, and provoke some discussion.
Simon Watkins