|
Coppermine Lapping GuideAs you can tell from other articles on my site, I am no stranger to CPU lapping and have practiced it on a number of occasions. Unlike most people I have no hesitation to apply the procedure to flip chip processors like the coppermine. The results have been worth it in my case, and I feel I have gathered enough experience to share it with other users on the web... |
The term "lapping" refers to the process of sanding and polishing CPUs and heatsinks to achieve a very flat and smooth surface. It makes sense to distuingish between lapping and polishing in the sense that lapping is to produce a very flat surface and polishing is to produce a very shiny surface. For the purpose of optimal heat transfer it is the very flat surface we are after. In fact there have been reports that a very shiny surface is not helpful at all and will even decrease the efficiency of the heat transfer.
To put it in a nutshell, lapping a CPU usually involves fixing sand paper on a very flat surface (like a glass table or some marble work tops), wetting it and moving the CPU over it in a figure-eight motion. You normally start with a coarse sand paper and work your way up. Depending on the amount of material to shift I would start with 400 grit, move on to 600 grit, then 800 grit, finally 1200 grit.
The process has been originally described for CPUs with a metal cap, like the original Pentium 2 or the Celeron 1 CPU. There are a number of articles on the web, including:
There used to be a lot more sites with lapping articles (Review News, OC Tools, All Games Network), but when I last checked the links were not working anymore.
For those type of CPUs it made perfect sense since the metal cap usually had a dent in the middle - exactly where most of the heat would be produced. Moreover, you could improve the heat transfer by removing the top layer of metal and lap it down to a copper layer which has superior heat conductivity.
When looking at flip chip CPUs you notice immediately that the flatness and the finish of the surface can hardly be improved. However, there is a blue plastic sealant on top of the chip which impairs heat transfer. So the aim of lapping a flip chip is to remove this top layer. Again, apart from the articles on my site, you will find a number of web pages describing the process:
I have measured an 8 degree drop under full load with two different P3 CPUs, as documented by this screen shot (compare left and right side of the curve). Assuming the CPU was putting out 25W at the time, this translates to an improvement of thermal resistance by 0.32K/W.
Let's have a look at the temperature distribution in a typical setup. I should say up front that the actual figures are completely fictional, they only serve to illustrate the point. OK, we see that the ambient has a temperature of 20 degrees.The transition from heatsink to ambient represents a certain thermal resistance, meaning the temperature of the heatsink is higher. Thermal resistance is usually measured in temperature difference per performance, specified in K/W (Kelvin per Watt). The same thing happens at the transition of CPU core to heatsink.
When lapping the heatsink or the CPU, or when using a better thermal compound we basically improve the thermal resistance of the transition from the CPU to the heatsink. Above all this will reduce the overall thermal resistance between CPU core and ambient, thus leading to a lower temperature of the CPU core. Since it lowers the thermal resistance of the primary heat path (through the heatsink), the primary heat path will transport more heat then before, at the expense of secondary heat paths. This leads to an increase in temperature of the heat sink. We encounter the seemingly paradox phenomenon that a warmer heatsink actually indicates better CPU cooling.
In reality, the temperature is not a constant within the heatsink and within the core since every material has a finite thermal resistance per volume. The distribution could look like this:
From this picture we can derive another beneficial aspect of lapping, namely the treatment of hot spots. The CPU core does not heat up evenly. Intel specify in their data sheets for the Coppermine CPU that almost all the thermal design power will be created in the core area excluding the cache. The core area accounts for roughly 63% of the die. And surely, depending on the application, different parts of the core generate a different level of heat.
The process described as deep lapping will also change the geometry of the die significantly, shaving half of it off. This will bring the hot spots closer to the heatsink, improving the primary heat path significantly. Following the same logic as before, by improving the primary heat path the hot spots will shed more heat via the primary heat path than before.
The picture also indicates the benefit of using a copper base plate in an aluminium heat sink: the overall thermal resistance of the heatsink may not be affected greatly or in some cases made worse due to the additional transition between copper and alu. However, temperatures in the base plate will level out more quickly, leading to lower temperatures adjacent to hot spots.
Before going any further I must point out to the reader that the lapping process inherently contains the risk of destroying your CPU. Only consider it
In order to gain the required experience you could lap heatsinks or an old CPU that you have lying around.
I can see two major sources of danger. For starters, you could inadvertently kill the CPU through electrostatic charges, particularly in sub zero temperatures or if you walk on plastic carpets. There is also the possibility of metal particles shorting the CPU when you first switch it on after the procedure. The second source of danger is inappropriate lapping results, as described further down in the section about deep lapping.
The process of "normal" coppermine lapping involves the removal of the blue plastic coating and nothing else. This method is relatively safe and positively recommended for beginners. All you need to do is to lap the CPU with very little force for 30 seconds on grit 1200 paper. However, it is absolutely critical to keep the CPU in an exact horizontal position. This can be achieved by touching the CPU with one finger only, pressing lightly on the precise geometric centre of the substrate.
I should also point out that the CPU should not be covered in any antistatic bags etc. in an attempt to protect it. This would make it more difficult to hold the CPU in horizontal position.
The result of the normal lapping procedure should look like this.
The next picture shows the result in real life. In fact this is a picture of the very P3-1GHz I used for the deep lapping experiment.
The key success factor is experience. If you screw up you might still be able to salvage the CPU by using the methods described in the following section.
It is certainly not recommended to lap any further using the technique presented in the previous chapter. The reason is that little errors in the orientation of the CPU will add up and lead to an unusable CPU. The two most common failures are the skew core and the dome shape.
One of the reasons why this is more likely to happen when lapping flip chips is the small surface area. This leads to a larger "wobble", and additionally a larger percentage of the surface area is close to an edge. The edges are prone to stronger erosion of material, particularly since the silicon die is quite soft, almost like graphite. So one way around the problem is to increase the surface area artificially, and also to introduce edges made from a harder material. Some people have reported good results using a copper shim. A good method I have found is the use of washers. I would fix a washer on each corner of the substrate and lap everything down to the desired thickness.
It is absolutely vital not to apply pressure to the centre of the substrate this time, because the substrate can bend through, and since the silicon die is very soft you would remove material unevenly.
The result of the procedure might look like this (compared to the image above).
The deep lapping procedure is also helpful in recovering a CPU after a failed attempt at normal lapping. That is what happened to me recently. I found that BurnP6 would send my CPU temperature above 70 degrees, crashing the CPU. I identified that the problem was dust that had accumulated in the heat sink. During the cleaning activities I had the brilliant idea to lap my P3-1GHz further down, just to see what the effect would be on the temperature. Using the normal lapping technique I basically screwed up the CPU. I then devised the deep lapping technique and managed to salvage the CPU. I reckon the core has lost more than half of its original tickness. It now looks like this:
Here is a couple of pictures of the CPU with and without the washers. The pictures taken after the lapping process. If you look closely you can see a fair amount of non-critical damage to the substrate. This damage occured during the original failed lapping attempt. The deep lapping technique (using the washers) did not cause any additional damage.
Finally a picture of the washers before and after deep lapping. The washers are delivered 0.8mm thick, which according to the intel datasheet is about the thickness of a coppermine core.
The deep lapping process is quite a dirty process, both due to the length of time that is required to work down the washers and due to the fact that metal shavings and other gunk are being produced. This stuff invariably ends up on the pins of the CPU as well as everywhere else. I found it did the CPU no harm whatsoever to hold it under a sharp beam of water occasionally, and to whash it down with methylated spirit after the ordeal.
According to my observations, deep lapping produces the following result: the CPU temperature when running 3DMark2001 is 3 degrees above mainboard sensor. Compare that with 4 degrees after normal lapping. Other applications are more demanding: Quake 3 and Stability Test run the CPU at 4 to 5 degrees above mainboard, BurnP6 runs at 8 to 9 degrees above mainboard.
These results have not been collected in a very scientific fashion, and the readings of Asus Probe are questionable in absolute terms at the best of times. However, it does give an indication as to how two thermal solutions compare.
My conclusion is that deep lapping does have an effect, but it is not really worth the effort compared to normal lapping. However, it is a good technique to salvage a CPU when normal lapping goes haywire.
Obvious candidates for CPU lapping are the hot running specimen from the AMD range of CPUs. Due to the components on the top of the ceramic body or the substrate, deep lapping is not advisable. Some people have reported that normal lapping will give them a 4 degrees improvement, which means the sealant has only around 0.08 K/W thermal resistance. This is less than the coppermine coating and makes lapping a lot less attractive.
Less obvious candidates are the intel CPUs with integrated heat spreader. Here the IHS needs to be removed first, before any lapping can be performed.
I happened to have another coppermine which I lapped a couple of years ago. It was the first flip chip I have ever lapped, and at the time I went into dome shape initially, although I got it flat eventually. This is how it looked like:
One interesting observation are the speckles on top of the CPU core. I have never noticed them on another lapped CPU of mine, but then again, this one has been running for two years now. I have no explanation for these dots. Whether it is a reaction with the Arctic Silver II thermal compound or something else I cannot say. At any rate they did not seem to penetrate the silicon. Maybe the blue coating is there for a reason after all? ;-)
Anyway, not perturbed by any of this I glued washers onto the substrate and continued with the deep lapping procedure as described above. I used 180 grit paper initially to shift some material fast. When I was happy with the thickness of the core I switched straight to 1200 grit paper. In retrospect it may have shortened the process had I used 600 or 800 grit paper in between, but it worked anyway, no worries. The picture has been taken in the final stages of using the 180 grit paper.
The following picture documents how dirty the CPU will get in the process.
But luckily a CPU is not made of sugar. So we can give it a wash...
and a rinse. I reckon a blow dry is also possible, but I didn't see the need on this occasion.
I am not sure whether I can now carry the title of OCAU Extreme Cooling Club H2O Member. :-D Anyway, this is the result:
And here is the mandatory picture of an object reflecting a mirror image on the CPU core. We are breaking with tradition here as it is common practice to use a coin for that purpose. ;-)
That concludes my article about coppermine lapping. I hope you enjoyed it as much as I did, and do not hesitate to drop me a line. Thanks,
Martin
| < OC report part VI | Home > |