It’s flatter than you think! – HighFlowRod
After reading one of Bill Adams’ latest article “P4 Lapping – Comments” , I tended to not agree with some of the things he wrote. First off, I believe that the IHS (Integrated Heat Spreader) is for even heat distribution and not crush resistance (my opinion, which is based on later question answered, may be incorrect) and a question which needed to be answered.
The question that needed to be answered was “So what is known about the P4’s IHS installed flatness when it is observed ‘unloaded’? Nothing, thank you.”
Not necessarily: With Intel ® being a very large company, one would tend to think that they conform to standard GD&T (Geometric Dimensioning & Tolerancing) specifications. One hint of this is the fact that the datasheets have length, width, and height specifications with tolerances, and I’m quite sure that they have IHS flatness tolerances also.
After doing some research, the flatness specification is right at the bottom of the Package Mechanical Specifications section in Intel’s ® Datasheets (go
HERE for a link to the datasheets (p 34, Table 3.1).
That flatness specification is .050mm – that’s in millimeters, so that’s 50 microns or 50 x 10^-6 meters! At work, I’ve conformed to flatter tolerances on some of my project parts, but that is a pretty flat IHS. That specification tells us that the IHS cannot vary from one corner to the other or one side to the other by 50 microns in linear height. Otherwise, the IHS would, be reworked or scraped (I’m guessing?).
But, in order to back up this claim, I will measure some samples. I work for a very large timing gear and sprocket company and have access to many great measuring devices. I will measure flatness (with a CMM (Coordinate Measuring Machine) and a flatness machine with qualified granite surface and gauge) and surface roughness (with a profilometer) before and after lapping on the CPU’s IHS.
I also will measure these same features on the bottom of the heat sink. These three machines were made by Mitutoyo, and are calibrated every 90 days. In addition to this, I will measure temperatures before and after lapping.
Right now I only have 2 specimens (2.0A and 1.7 Celeron). So, OC community, please send my your dead P4s or dead Socket 478 Celerons. I would like to have five more to measure (not lapped). Thanks!
If you ship me a dead processor, you will not get payment for the shipping charges and you won’t get the processor back. Though, I greatly appreciate anybody who sends them.
ed note: Dead P4s also appreciated at OC.com – we can use them for IHS experiments – please email me if you can send one – Email Joe
Before reading this article, please read my other article. After seeing some constructive criticism on some of the forums, I think I need to address some issues and apply some perspective on how ‘flat’ is ‘flat’.
First off, what needs to be kept in mind is that the specification for the flatness of the integrated heat spreader is .050 millimeters MAXIMUM. The .050 millimeters can not be exceeded or the IHS is out of specification and being ‘in’ specification doesn’t always mean you have the capability to produce a product.
Let’s say Intel ® is producing in-specification parts, but the flatness is running on the high side of tolerance. Say all of a certain number (for a capability study, usually 30 or more) of processor IHS’s measured .049 millimeters; this in no way means Intel ® has capability to produce this product. In order for Intel ® to have capability to produce this product, a Cpk (Google it for more info) of over 1 would be a must (I would think that Intel ® is a QS-9001 or ISO-9000 certified company – most cases, 1.33 Cpk or 1.67 Cpk).
Cpk is based on a position and a grouping under a bell curve. It defines the ability to produce parts in specification, with a tight grouping of repeatability of the featured specification. In a shorter definition, in order for Intel ® to be capable to produce this product, the flatness has to be the same measurement every time and at the nominal of tolerance (more or less, the flatness specification has to be .025 millimeters, every time).
With that out of the way, I saw where someone wrote that 50 microns wasn’t flat at all, AND for a surface area that small, 0.5 microns should be the MAXIMUM. Let’s think about that: 0.5 microns, or .0005 millimeters, is roughly 2 hundred thousands of an inch (.0000196″). At this point, I would like to write about something that everybody can relate to, a piece of copier paper.
A piece of copier paper is roughly 90 microns (.090 millimeters or .00354″), I can’t imagine holding a flatness tolerance that is roughly 2/10ths the thickness of a piece of copier paper. That is more of a tolerance than would probably apply in the laying of the circuitry inside the processor (if not tighter). FYI, the tightest flatness tolerance (at work) I’ve seen to date is 8 microns.
This actually brings up a good comment to mind, which should bring up some good questions. With the flatness being say, half the tolerance (25 microns, what it actually should be) on the IHS and the same on the mating surface of the heatsink, the MAXIMUM variation of ‘valley to valley’ microstructure of the copper (heatsink) to the nickel (IHS) should be only 50 microns approximately.
50 microns is roughly 2 thousandths of an inch (.002″) and the thickness that Arctic Silver 3 gives a thermal resistance for is .001″ (which is 1/4th of a piece of copier paper). I’ve always used a razor blade to spread my AS3 and I would think that the thickness that I would be closer to would be .002″.
In conclusion, I did some measuring on the 1.7 GHz Celeron I mentioned in the other article. The flatness turned out to be .00078″ which is 19.8 microns (variation was almost 1/4th of the thickness of a piece of copier paper). Surface roughness on the unlapped IHS was 7-10 on the Ra scale (copier paper roughness = 75-120 on this scale). Thanks for reading and Thanks, Victor.