How to check flatness of CPUs and coolers - INK and OPTICAL INTERFERENCE methods

[BoggledBeagle]

Following the discussion in another thread, I would like to post some practical information how to check flatness of the CPU heatspreades or cooler base.


The INK METHOD

Simply place a drop of ink (I used black fountain pen ink) on the surface and press against a glass. I used a 10mm thick plate of glass, which I use for lapping, which is really stiff and does not deform, you may get some useful info even with a lot thinner glass.

You can see the results:






The lapped surface in nearly perfectly flat, except the two right corners.
The CPU excluding a few milimetres on the edges and upper corners is also acceptably flat and I would not feel any need to lap it for normal use. You can even clearly see the grinding marks after it has been ground flat.


The OPTICAL INTERFERENCE METHOD

Under certain conditions you can see a rainbow pattern (with broad spectrum light) or light and dark pattern (with monochromatic light) between a reflective surface of the object and flat glass.

If you had a green light with wavelenght of 0.5 μm (micrometer), each neighboring dark (or light) stripe would mean, that the distance between the two surfaces changed by 0.25 μm between these stripes.

I quickly lapped and polished the above photographed CPU (and destroyed a valuable historic artefact) and got this picture:



You can see, that in the indicated area there is something like 8 stripes from top to bottom. That should mean that in this area the heatspreader is flat within 8 x 0.25 μm, that is two micrometers! (two thousands of a milimeter)

• Post the results using ink method of your brand new CPU and cooler base, of the CPU before and after spending some time mounted in the socket with normal mounting mechanism.
• Post your results before and after lapping cooler base and CPUs.
• If you had a small piece of glass, you can easily test the CPU while being mounted in the socket.
• Post the difference between stock mounting mechanism of the LGA1700 socket and the mounting frame. (In my experience an LGA1700 CPU remains slightly bent even after being removed from the socket with stock mounting mechanism, which may affect your results.)

WARNING: Lapping a CPU voids the warranty.

ANOTHER WARNING: Do not decide to lap the cpu only after seeing in out of the socket, you need to see what is happening with it when it is mounted in the socket!

EDIT:

If somebody wondered, why the interference pattern does not look very nice, it is because the surface itself in less than perfect - the plating started to wear off and how I said I lapped it just very quickly. I also did not have a proper monochromatic light and no special equipment. This sort of stuff can require special lens to capture it, special shape of the light beam etc.

However it proves clearly, that you can even with a very dirty quick lapping job get to 2 micrometer flatness on a large portion of the CPU.

Here is a picture of how the surface really looks:



EDIT 2:

I should add that I put the CPU under the middle of the glass piece and left it press against the CPU only by its weight (2.44kg). In reality coolers press the CPU with much larger force. While changing the force you can observe changing of the pattern, it also depends on how you hold or support the CPU. So if the shape of the CPU looks like "an unimpressive hill", this shape could improve under the pressure of the cooler.

I do not believe this optical interference method has much practical use for a normal consumer, the ink method is so much more easy and convenient...



The overall result - the thickness of the gap between the CPU and cooler base and resulting thermal resistance of this thermal interface depends on many factors: shape of the mating surfaces, flexibility of the mating bodies, contact force, mounting of the CPU and cooler, TIM, etc.

EDIT 3: For anyone wanting to post pictures in this thread: can you PLEASE INPUT THUMBNAILS in the text, not whole pictures? I have no idea why the forum inputs large pictures as a default.

 

[BoggledBeagle]

I adjusted the interference picture to see more in the extremities, because I was thinking about making a height graph along the heatspreader diagonal, but then I got lazy and decided not to do it, so I am just posting the picture. You can see some additional stripes in the corness close together as the height falls off quickly.

 

[ThrashZone]

Hi,
Straight edge and a light usually works well

 

[BoggledBeagle]

Hi,
Straight edge and a light usually works well

Well yes, but with the ink method you can quickly get an information about the whole area you are assessing. I have been using if for ages now... When you use just a small drop of ink (the drops in the first picture are too large), it is actually a pretty clean process and you may not get dirty ar all.

 

[Wirko]

Polarized light, even if it's white, might be even better for that purpose than monochrome light. LCD monitors emit light that's at least partially polarized, try and see if you can get good results with it.

 

[Shrek]

How does polarized light help?

 

[BoggledBeagle]

Polarized light, even if it's white, might be even better for that purpose than monochrome light. LCD monitors emit light that's at least partially polarized, try and see if you can get good results with it.

The interference method calls for a monochrome light for a good reason. I also wrote that I do not believe that this method is very useful for a normal consumer and also that the shape of the CPU or/and cooler surfaces are just one or two factors of many, which result in the thermal resistance of the interface.

Examining a CPU heatspreader with a resolution of 250 nm has a limited practical value.

I just remembered that AMD (or their supplier) lapps the CPU and cache chips so perfectly flat, that they just lay them on top of each other and they connect just by physical contact.

Whoever does this could possibly tell a lot of interesting things about lapping things flat, although I believe that some of that info are probably closely guarded secrets.

 

[Wirko]

How does polarized light help?

Can't really explain. A thin film polalarizes the reflected light (to no surprise, as common window glass does that too). But apparently the effect depends on the film's thickness. So viewing or casting light through a polarizer makes the interference pattern more pronounced.

Polarized light from soap bubbles | Soapbubble.dk

I just remembered that AMD (or their supplier) lapps the CPU and cache chips so perfectly flat, that they just lay them on top of each other and they connect just by physical contact.

Yeah, they (TSMC) are doing that. Copper-to-copper bonding. AMD has showed it on many slides, but I haven't yet seen a good explanation how it works. And lapped silicon dies don't just stay flat to a nanometer forever, there are temperature gradients and external forces that deform them ever so little.

Edit:
Another thought. It's worth checking with an ohmmeter if the ink is not conductive. Any dissolved mineral or acid content will make it conductive, and if that's the case, take care to avoid any spillage on the CPU substrate.

 

[BoggledBeagle]

Another thought. It's worth checking with an ohmmeter if the ink is not conductive. Any dissolved mineral or acid content will make it conductive, and if that's the case, take care to avoid any spillage on the CPU substrate.

That is a very good remark, but I believe that most normal fountain pen inks are chemical dyes, not pigment based (black pigment ink would probably contain conductive carbon), because the pigment could cause clogging.

Also, how I mentioned, if you apply just a small drop (SMALLER THAN THOSE IN THE FIRST PHOTOS), you will not get any spillage and you just wipe the small amount of ink from the surface without causing any mess.

 

[taka]

Question about flatness of the test object with ink, is there a chance that the regular glass is not that flat as for a test object and instead use a lens from a magnifying glass?
I guess that flatness tolerances for the lens are better than regular glass.

Found this 90mm (3.54 inch) one side is flat and the other convex.

 

[gurusmi]

Following the discussion in another thread, I would like to post some practical information how to check flatness of the CPU heatspreades or cooler base.

If you would have had a good education in metal working you wouldn't need all your techniques. All you need is a protrector or a lineal (i.e. metal). Also the surface you are testing won't be damaged or get dirty.

 

[Vayra86]

Learned things here

 

[joemama]

is there a chance that the regular glass is not that flat as for a test object and instead use a lens from a magnifying glass?
I guess that flatness tolerances for the lens are better than regular glass.

Glass usually have a very high flatness due to its manufacturing process floating above molten tin so its pretty much enough.

 

[BoggledBeagle]

Glass usually have a very high flatness due to its manufacturing process floating above molten tin so its pretty much enough.

I am not an expert but I believe that virtually ALL of current normal flat glass sheets are float and therefore flat. Unless you use some scrap glass from an old building, it it pretty certain that the glass you get is flat.

I believe that for the ink flatness checking you could get by using a 4mm thick glass, for lapping it is not enough, it is too flexible.

I do not understand that previous remark about magnifying glass at all, because lenses are usually very far from flat on both sides. I think I have never in my life seen any lens which was flat on one side.

 

[taka]

The lens i talk about is a type b in this image:

 

[BoggledBeagle]

If you check flatness page on Wikipedia: Optical flat - Wikipedia

there are some glass flatness standards by Mitutoyo pictured, I wonder how much they cost. I am not buying them, I am fine with a piece of a normal float glass.

EDIT: $140 a piece...

 

[FoulOnWhite]

Following the discussion in another thread, I would like to post some practical information how to check flatness of the CPU heatspreades or cooler base.


The INK METHOD

Simply place a drop of ink (I used black fountain pen ink) on the surface and press against a glass. I used a 10mm thick plate of glass, which I use for lapping, which is really stiff and does not deform, you may get some useful info even with a lot thinner glass.

You can see the results:

View attachment 330476

View attachment 330477


The lapped surface in nearly perfectly flat, except the two right corners.
The CPU excluding a few milimetres on the edges and upper corners is also acceptably flat and I would not feel any need to lap it for normal use. You can even clearly see the grinding marks after it has been ground flat.


The OPTICAL INTERFERENCE METHOD

Under certain conditions you can see a rainbow pattern (with broad spectrum light) or light and dark pattern (with monochromatic light) between a reflective surface of the object and flat glass.

If you had a green light with wavelenght of 0.5 μm (micrometer), each neighboring dark (or light) stripe would mean, that the distance between the two surfaces changed by 0.25 μm between these stripes.

I quickly lapped and polished the above photographed CPU (and destroyed a valuable historic artefact) and got this picture:

View attachment 330501View attachment 330502View attachment 330574

You can see, that in the indicated area there is something like 8 stripes from top to bottom. That should mean that in this area the heatspreader is flat within 8 x 0.25 μm, that is two micrometers! (two thousands of a milimeter)

• Post the results using ink method of your brand new CPU and cooler base, of the CPU before and after spending some time mounted in the socket with normal mounting mechanism.
• Post your results before and after lapping cooler base and CPUs.
• If you had a small piece of glass, you can easily test the CPU while being mounted in the socket.
• Post the difference between stock mounting mechanism of the LGA1700 socket and the mounting frame. (In my experience an LGA1700 CPU remains slightly bent even after being removed from the socket with stock mounting mechanism, which may affect your results.)

WARNING: Lapping a CPU voids the warranty.

ANOTHER WARNING: Do not decide to lap the cpu only after seeing in out of the socket, you need to see what is happening with it when it is mounted in the socket!

EDIT:

If somebody wondered, why the interference pattern does not look very nice, it is because the surface itself in less than perfect - the plating started to wear off and how I said I lapped it just very quickly. I also did not have a proper monochromatic light and no special equipment. This sort of stuff can require special lens to capture it, special shape of the light beam etc.

However it proves clearly, that you can even with a very dirty quick lapping job get to 2 micrometer flatness on a large portion of the CPU.

Here is a picture of how the surface really looks:

View attachment 330566

EDIT 2:

I should add that I put the CPU under the middle of the glass piece and left it press against the CPU only by its weight (2.44kg). In reality coolers press the CPU with much larger force. While changing the force you can observe changing of the pattern, it also depends on how you hold or support the CPU. So if the shape of the CPU looks like "an unimpressive hill", this shape could improve under the pressure of the cooler.

I do not believe this optical interference method has much practical use for a normal consumer, the ink method is so much more easy and convenient...



The overall result - the thickness of the gap between the CPU and cooler base and resulting thermal resistance of this thermal interface depends on many factors: shape of the mating surfaces, flexibility of the mating bodies, contact force, mounting of the CPU and cooler, TIM, etc.

EDIT 3: For anyone wanting to post pictures in this thread: can you PLEASE INPUT THUMBNAILS in the text, not whole pictures? I have no idea why the forum inputs large pictures as a default.

Very interesting, thx

 

[Shrek]

How flat is flat enough? given that the surface will warp due to uneven heating.

 

[BoggledBeagle]

How flat is flat enough? given that the surface will warp due to uneven heating.

I do not think this is the problem. Two main problems are:

1) the deformation of the heatpreaders in the socket due to the mounting mechanism.

2) the cooler mounting mechanism which applies too much force and is not possible to mount gently and with the correct force.

When you prepare the two mating surfaces, you should be able to apply just the right contact force. You should be able to feel how tight are the screws. You cannot use coolers that have screws on the springs - you cannot feel anything. You can count number of revolutions of the screws, but it is not as good as feeling how tight are the screws.

Flatness is not your goal, your goal is to minimise thermal resistance of the interface: you need as thin layer of the thermal paste as possible in the places with the most intensive heat flow. And with that flatness can help sometimes.

 

[delshay]

I learnt a lot when lapping three Vapor Chamber. There's actually a art to lapping & I could have had better cooling performance, but it was to late for two of the three vapor chamber. Anyway I discovered it's better to use a single finger & always aim for the center. The strain on a single finger is painful after a short time & naturally you want to use two fingers. I found using two fingers while lapping will give slightly worse result than using a single finger from start to finish,

 

[Noci]

If you void the warranty anyway, why not skip the whole process of painstaking work of precision lapping the IHS to an ultimate level (which imo is only measurable by specialists if were talking about tenths of microns) and go for direct die cooling.

Yes I know the risk of damaging and sending your CPU to heaven is much higher with delidding, but the ultimate goal is to significantly improve cooling.

Even with a perfectely lapped CPU IHS and cooler coldplate, the TIM between the die & IHS is the limiting factor (heat transfer wise). With proper lapping you'll only increase the efficiency of heat transfer between the IHS and cooler coldplate and than it is up to the other variables in the cooling solution on how much you gain in heat dissipiation.


But for those who think lapping is enough risk, this ink method of visualizing the flatness (or lack thereoff ) is indeed a nice tool so you don't have to fool around with a laser.
And if you use the thermal paste imprint method, you don't have to clean off the paste over and over again.

 

[BoggledBeagle]

... go for direct die cooling.

The block for direct die cooling should be flat too!

 

[Noci]

The block for direct die cooling should be flat too!

Yup, I fully agree but up to a certain level, as die's are not perfectly flat also (plenty of other threads on TPU where this is a point of discussion).
It's about which TIM and how good/fast the heat transfers. Preferrable something with a vapour chamber or direct contact heatpipes in case of air cooling. That is atm the fastests- way to transport heat as far as I know.


But were going off topic, it's about the visual indicator to show the flatness of a surface after lapping

 

[delshay]

But were going off topic, it's about the visual indicator to show the flatness of a surface after lapping

Well if you want to go along those lines, this is how I check for flatness.

Get something that is flat. stand it on edge on the coldplate. Now shine a torch behind it. Now tell me how much light is leaking though between edge of object & coldplate.

 

[Noci]

@ delshay, yup a proven method and in my opinion easier and sufficient accurate.