FLATNESS - anyone interested ?

[koslov]

Bill: Sounds like something of use to testers and mfgs, but perhaps not the OCer. Even most hardcore OCers don't get much past lapping their HSs and WBs with 1500 grit sandpaper... I myself go up to 2000, but then again, its probably futile since I don't have a perfectly flat sanding surface, just a glass table.

N8: Well, I know lapping the IHS is popular, but bare dies?? I've never heard of an OCer doing such a thing. Well, maybe more people will try it depending on the results of BillA's tests.

[gmat]

Bill, you might take into consideration the mechanical factor as well.
When one mates a HS or a waterblock with the CPU, quite an amount of pressure is applied. Since we're dealing with metals, i'm 100% sure that they 'flex' to a certain point.
With AMD CPUs, i think only the CPU core will flex (the HS will too, but to what amounts to zero).
With newer Intels (P4), a waterblock with a rather thin baseplate will maybe flex just the same (approx.) as the P4 IHS.
Of course the amounts have to be checked out. But in the case of 'imperfect' cores, the mechanical pressure of the mounting system can compensate, as long as the TIM is squished out properly...

[BillA]

koslov

I am headings in a direction that is not (yet) apparent to many people
I am pretty sure that the lapping 'procedures' used by OCers - are all wrong
sure, the surface gets nice and shiny, but it is not flat
- that 'problem' is my goal
then the question becomes 'how much flatness is worthwhile'

re CPUs
I will know eventually how flat they are, and if hand lapping (by any procedure) can make them flatter
- I am not talking about IHS flatness, or lapping

does the risk of lapping a CPU warrant the risk ?
we do not know the potential gain (if any ??), nor really the actual risk
- but if this were a favorable ratio, why would the mfgrs not be doing such ?

[BillA]

gmat

the IHS is a completely different situation
it is convex by design, intended to deform under load and to apply that load to the TIM joint on the opposite side

my (quite individual) opinion is that before I would lap an Intel IHS, I would simply remove it
- but I have heard the P4 chips are not so robust as the AMD ones, the latter being designed for the direct application of the load

the IHS is not what its name suggests, it is a crush protection package

and you are quite correct about the 'issue' of bp flexing wrt all these 'thin' designs becoming popular
keep an eye on JoeC's "Roundup" on OC, some info on this will be coming up in a couple of weeks

[gmat]

Originally posted by unregistered

does the risk of lapping a CPU warrant the risk ?

To me it's not worth the risk for an AMD CPU. Like i said i'm convinced the amount of flex of the core under pressure will negate any *small* imperfection. And the HS base will deform as well - especially if it's made out of copper.

[kai]

What about , if you buy a brand new say pure copper heatsink test it , to see if its flat when you first get it . Use it in a say 1.4ghz tbird the hottest you can find. To see if the constant heat and pressure from the clip and chip warp the heatsink. Ive always wondered this .

[myv65]

Originally posted by kai
What about , if you buy a brand new say pure copper heatsink test it , to see if its flat when you first get it . Use it in a say 1.4ghz tbird the hottest you can find. To see if the constant heat and pressure from the clip and chip warp the heatsink. Ive always wondered this .

The softness of copper isn't dramatically affected by increasing the temperature to ~90°C, which is higher than you could get the copper without crashing the PC. Along these lines, however, I've seen the AMD die engraving quite clearly in the base of heat sinks after a few weeks of running. This leads me to believe that there is a threshold beyond which additional "zero stress state" flatness yields no benefit. I don't bother with P4s and AMD doesn't yet use an IHS, so I'll not comment on the relative flatness or lack thereof on IHS plates.

[N8]

There are two more 'sub'-aspects of flatness that most people do not consider and/or need to start considering.

1. Polishing is a sub-aspect of flatness (smaller scale/surface roughness). There is a optimal surface polish/roughness that is determined by the TIM material. Just getting the surfaces ultra-polished doesn't mean it is still flat, and doesn't mean your TIM will perform well. You have to keep the faces flat and you have to have some porosity to the surface roughness for the TIM to fill in, or you have to make sure that the TIM will squeeze out. OTHERWISE you are forming a layer of TIM, and this is NOT how you transfer heat efficiently. Ideally you want a perfect metal to metal interface with no TIM and no air gaps. This is 'almost' impossible for 99%+ of people. Like BillA stated, how flat or how polished is good enough? That depends mostly on the TIM and its application.

2. The tilt or trueness of the mating faces - keeping the face of the die (or whatever you are lapping/polishing) parallel to the socket surface (since the socket surface is the base from which everything is attached.) You can lap and polish something very flat, but if you put some sort of angle into the face of it so that it is not applying an even force when attached, your flatness job will be compromised by the differential pressure from the heatsinks/coldplate/whatever you attach. This is another thing that is overlooked, but probably has some degree of significance, but how much?

Here at work, I routinely have to get all sorts of metals similar to what we use on computers (aluminum, copper, brass, +others) flat AND polished to within 1/4 wavelength of light and with the faces parallel within 1 micron max deviation. A lot of layers are stacked and epoxied together and my epoxy bonds need to be 5 microns or less in thickness, usually 1 micron or less. I really wish we (as in all of us reading), could have a TIM that was as runny as epoxy, because 1 micron bonds are really neglible. For less critical surfaces that are not polished, the bond thickness is almost always 1 micron or less, sometimes negative. Now you might think, "how the hell do you get a negative bond?". Just think of two lapped surfaces (not polished.) They both have surface roughness. When pressed together, the epoxy flows into the surface pores and allows the two surfaces to meet. If the surface roughness allows, the points of the 'hills' will actually embed into the 'valleys' of the opposing surface, and you get a negative reading when you measure the two pieces after bonding. The key thing here is that you want the surface roughness to be equal on both parts so that they can mate together, with a VERY LOW deviation in the actual roughness. Meaning the hills and valleys are consistant height/depth across the whole surface(s).

If you polish out all of the roughness, and can still keep the part(s) flat, you now need much less TIM. There are no pores for the TIM to flow into and you now need the TIM to actually flow out the sides or it makes a thicker layer than if you left the part unpolished.

That is enough of my $0.03 for now.

[Brians256]

I know very little about how to PROPERLY lap a metal surface, so I'd like a better understanding of how to do it right.

What I've been doing is using sandpaper that is taped to a granite block (a reference block used at work to planarize chuck surfaces and other mechanical pieces for semiconductor equip). The granite block itself is very flat, but there is a certain amount of flex in the sandpaper as I apply pressure when lapping. Also, even with a figure-eight motion, I get rounding on the edges. I know that the block is fairly flat, since I can lift plate glass with a wet block (surface tension of water). But, I don't know if there is a better way.

Perhaps I'm blowing all this out of proportion, since only the central 3/4" square needs to be coplanar with the CPU on an AMD cooling solution.

[BillA]

nice N8

I believe much of your 'finish' commentary bears directly on the highly angular, but apparently flat, surface of the AMD CPUs

I will lap these things to learn about them, but have little reason to believe that a 'system' designed by BillA is going to be better than that of AMD
- quite laughable

I have considered the 'off axis' yet flat surface, but suspect that the spring loading is adequate to compensate for the small angle
and the center loading from a clip eliminates such concerns completely

yes, the viscosity of the grease can be quite relevant
I have had a 9 month internship on goop application techniques, and it ain't over yet

I have found, hang onto your hats, that jiggling the mobo studs, in pairs to and away and diagonally, 2 times, 2 hours after mounting; will produce a drop of between 0.1 and 0.3°C in the die temp

the temp will initially rise, then fall ever more slowly over the next hour or so
(remember that I have 0.01°C resolution)
why does this occur ?
to my mind it is clearly related to the disturbing of the thermal grease 'joint', and the subsequent extrusion of additional grease beyond that which would have occurred without the disruption

I use this 'technique' as a documented part of my testing procedure
- controlling the mounting variables is extraordinarily difficult

Brians256
despide all the twaddle on the 'net, that is a terrible method
guaranteed to produce the 'grassy knoll' in the first image

my 'better way' ??
ues a PSA film to eliminate the gross 'lifting' of the paper (better is a loose grit but much more difficult)
try to lock your arm and shoulder in a fixed position
'sand' only on the forward stroke, lifting the piece clear to return again
do not rotate the piece

[gmat]

About TIM viscosity: to me it's obvious that a "better" finish (=higher grade) and polishing induces using a more 'viscous' TIM. Polishing mating surfaces to a (flat) mirror and yet using the relatively thick Arctic Silver is a nonsense, not to mention the size of silver particles in that paste, that may be bigger than ridges & grooves of a well finished surface.
I'm now using a quite liquid 'white stuff' under my lapped & polished waterblock. I tend to think its results are better (off by 1-1.5°C in regular situation) than those of AS, in that situation (TBird core). I'm using a razor blade to apply it, but still i think it leaves too much paste... After dozens of mountings / remountings i've experimented quite a few solutions but yet only the 'white liquid goop' does the job properly. AS always 'stack up' somewhere between the core and the WB, though 'white goop' nearly disappears...

And everytime i remove my WB (after a few weeks of use, at minimum) i see the imprint of the TBird core in the copper, the full square *and* the markings ("AMD Athlon" etc., mirrored of course) !
If the core was unbalanced or 'not flat' i wouldn't see such a clear picture, and my TIM would stack up where pressure is lower. I apply quite a heavy pressure, to the point where the mobo flexes slightly.

So where i'm going is, with cores like those of AMD CPUs i do not think that one can get any improvement out of their surface. IHS is another debate altogether...
Anyway HS / Waterblock bottom flatness (for the area that mates the core, at least) is what is really important. Elasticity (and even plasticity) of both surfaces will do the rest.

[BillA]

gmat
help me with this
here, there, and everywhere I see references to people 'imprinting' their bp or hsf with the AMD chip engraving

I'm searching to understand how this is done
the yield strength of rather pure (soft) copper is nominally 9000psi (62MPa) at 0.2% offset, or 0.5% extension under load
a 120mm² die (~0.186in²) with a 24lbf load results in an applied load of ~130psi (900kPa)

now I understand that you fellows are FAR beyond the AMD 24# spec,
but where is the force coming from to yield copper ?

BTW, I too gave up AS a long time ago

[nicozeg]

I think there's an easy explanation for that Bill: AMD markings rise over the silicon surface. When positioning the block, at first the only direct contact surfaces are the high point of the leters, so whith little force the copper will deform. Only when the markings are engraved in the copper you get a full contact surface.

That suggest me that the ideal surface roughness of a WB to be used on an AMD cpu would match the roughness of the markings.

N8: It seems to me that you could be able to measure that roughness and tell us what sandpaper grit could be used to produce a matching surface?

[BillA]

noooo

the AMD markings are engraved (laser cut in actuality, I've been told) into the surface
and I do believe this is so
'cause when I lap the green oxide will slowly be removed and the lettering remains

no, I don't have a surface profilometer (sp ?); what it would take to do that
and I suspect the very high angularity (promoting substrate-to-substrate contact) would be difficult to reproduce
(I do not know how that surface is produced - anyone ??)

[myv65]

Bill,

I agree pretty much with nicozeg on this one. And I'd add that if you've worked with AMD CPUs and haven't seen this happen, then there's something wrong on your end. It is roughly analogous to hardness measurement when you commonly use a small ball or diamond to dent a metal surface. You don't need much force when the area of contact is quite small. The net load need only be on the order of 20 lbf for the average pressure to be 100 psi. The area of the letters is much less than 1%, so if they are the only things to contact (initially) then the pressure is >10 ksi.

As an aside, there is also a time aspect to this. I tend to think this is due to imbalance in the fan and the minute reversing force this presents. Same thing could be said for water cooling and pulsations from the pump. I have never bothered (and never will) trying to characterize this versus TIM application, load, die size, etc.

And since your previous post just arrived after I'd written most of this, I'll add that it's my belief that laser engraving leaves a bit of a lip around the perimeter.

[bigben2k]

I'm with Dave on this one: time.

Not to speculate too much... but the combination of pump vibration, and thermal cycling (cpu load) may very well explain this phenomenon.

[nicozeg]

I thought that it were some kind of paint.

But it still could match my theory: The laser engraver melts the silicon surface, and some irregular spills at the border of the pit rises over the surface.

In this case a lapped core could show better results, or at least will require less clamping force to get them.

I bet that with a lapped core it's going to be imposible to get a core imprint at WB base.

Edit: corrected typo.

[BillA]

G33k provided this link to a pretty good AMD CPU surface

a question:
when the wb or hsf bp surface is lapped - is the AMD lettering raised ? or 'embedded' ?

this should be easy to tell, no ?

Dave, I don't see such 'cause I use cheapo hsfs with clips
whoa
does this imprinting occur also with the force from rinky-dink clips ?

[myv65]

That's a pretty good picture, but not good enough to learn much about the surface profile of the die (uh, yeah, it looks grainy ).

In answer to your question, I've seen this on an SK6 that I lapped (somewhat poorly, but ya do what you can with what ya got), a Swiftech MCX370 (factory finish), and a couple of retail AMD HSFs (factory finish aluminum jobbies). The SK6 uses a single center spring clip, probably toward the lower end of AMD's recommended force range. The MCX uses a rather goofy compression spring arrangement and I'm not sure where it falls in the range. The retail HSF uses a spring clip that I would hazard falls around AMD's "optimum" recommended load of 16 lbf.

I didn't look really close, but don't recall seeing this on another SK6 with the (lousy, very lousy) factory finish.

[nicozeg]

I've noticed that with a careful block mounting and unmounting (I mean only verical load at the center) the letters imprints are less notorious. Only after some side movement under pressure (or maybe long term vibration) are easier to see.

My reasoning is that the protuding edges are very little compared to letters size, and only after they scratch the surface arround due to movement are more easily noticed.

[N8]

Hey BillA, I found my photocopies from the article of reading optic interference lines, but they suck photo-quality-wise. I will see if I can dig out the original manual to scan it tomorrow.

I have access to a surface profilometer, but it cost some $ to get time on it. I will check to see how much $ and what resolution it has. Maybe I could take in samples of copper or aluminum that have different surface finishes to get a comparison.....hmmmm.

[MadDogMe]

I have seen the imprint everytime I've taken off my block/HSF, my unlapped maze3 has them on it, my poorly lapped SK6 had it. the imprints are so perfect that I first beleived the stuff was painted/etched on somehow...

I always have used artic silver. do people who use non~silver greases see this as well?...

If the numbers/letters are engraved by lazer that'd mean they are burnt/melted in which would allow for speeling/burring would'nt it?, which would explain the raised aspects...


PS. My lettering has never been as regular/definate as that in the pic', mine looks like it was written with a blotty ink pen, some aspects of letters/numbers are thicker (prouder as well?) than others. if you run your finger nail over it you can feel its 'irregularities'...

If someone has a naked Athlon could they run a razorblade slowly upto and across the lettering to see if it 'snags' on the lettering, to see if it's raised as I think it is please?...

[Puzzdre]

Before I got my hands on AS3, I used ordinary stuff like titan silver grease and such, it left the imprint on the bottom of my block. Now with AS3, I also have the imprint, blotty but readable, think that happens when tightening the bolts on the block, it wriggles a bit.

Just tried a razor on my spare duron, and it 'snags' a little, but I just cannot see if the letters are raised or not. My guess is they are not, the precision of the small letters (esp. TM mark) is great, and if this is done by laser, laser evaporates the material and leaves sharp edges. Dunno if the edges are left something raised over the surface as a result of burning...

I just tried to smear some paint over the core and print it on the paper, but didn't have much success (prolly cos of the paint and the pressure applied), it leaves just a whole smear in rectangular shape of the core. Think that the letters are very shallow to leave the imprint with this method used...

I think that the letters are lower than core surface (laser engraved) and considering the pressure applied on the block and the temperature on the TIM joint, the letters just are baked onto the surface of the block (somewhat like in laser printers). The TIM joint is very very thin, and with high temp and pressure it just bakes the letters onto the base of the block. That's the reason (IMO) why the imprinted letters cannot be smeared by finger after removing the block, think they're baked in very thin layer onto the surface...

Dunno...:shrug:

[Cathar]

I can say with 100% certainty that the AMD CPU markings are engraved/etched into the surface of the due. That is, they are concave.

Think about it. From a manufacturing point of view, what's easier? Have a flat surface and laser etch some thing onto it. Or have a flat surface and remove everything but that which you want there (the writing) and drop everything else down to a uniform depth.

The second approach is silly. The markings are not raised.

I've lapped AMD CPU dies before (don't bother - no point). However, while doing so with 1200-grit paper, I noticed that the color sheen was first to go, revealing the bare silicon color of the die. Then very slowly the edges of the writing started to go, and then was left with just the deeper middle portions of the markings, and then upon further lapping, they too disappeared.

[gmat]

Yes i think the letters are shallow as well.
As for the imprint, it doesn't go off by cleaning it even with tissue. But, polishing the surface (with copper cleaner for example) makes it go away. That means, it's *very* thin.
I've been using AS then standard white goop. At first i thought that AS would 'etch' chemically the core picture into copper - through oxydation.
But i saw that with white goop as well.
Ah and i saw that under Alpha heatsinks as well... I had the imprint under my PAL 6035. It's even more noticeable on the Alpha 8045 with its strong 4-bolts mounting system.