Direct-Die Cooling

[koslov]

http://www.3dnet.hr/dr-ice11-en-01.html

This guy is claiming some pretty good numbers, but as usual his testing methods are rather primitive.

Is it possible that some form of direct-die cooling could match the performance of some of the top blocks out there now? I know it has been tried a couple times here, but I don't think we've ever seen definitive testing (due to the fact nobody can really test direct-die setups yet).

[BillA]

no, no real 'testing' problem, more like a stupid application

the package is organic, and permeable
which means that the sum total of the application is a 'time to failure' test

read around, they ALL fail

dumb, big time

[koslov]

Hmm, never thought of the permeability issue. But sealing it up shouldn't be a big difficulty...?

Wouldn't there be problems testing direct-die blocks on a die simulator since the blocks must be made specific to the shape of the CPU?

Another thought: Is it possible to erode the die at high flow rates?

[BillA]

Originally posted by koslov
. . . But sealing it up shouldn't be a big difficulty...?
. . . .

??
organic = permeable

erosion is a concern at high velocity

[koslov]

Sealing it up = coating package in epoxy

[bigben2k]

I think that the best point made is on page three: if the flow stops for any reason, it'll fry.

I think everyone should drop the direct die cooling idea: the only purpose of doing it is for some thermal numbers and properties of different CPUs. It is not meant to be an alternative to a waterblock.

[koslov]

OK, I am still confused. Are you saying the die itself is permeable? Or the organic substrate? If it was only the substrate that was permeable, then one could put a small layer of epoxy over the exposed substrate, right?

With some good jet impingement, why wouldn't it yield good results?

[BillA]

epoxy is permeable, this is the problem

[koslov]

I am quite tempted to bring out my Duron 700 and experiment with direct-die. It seems there has to be *some* way to do it safely and reliably, perhaps using a nonconductive coolant.

But unless the DIE itself is truly permeable, it shouldn't even matter. If not epoxy, then plumber's goop to seal the organic substrate.

You all say it is so unfeasible, yet Intel and Stanford engineers have done some serious research on it: http://www.mit.edu/afs/athena/org/m/...ms/paper12.pdf .

Their results were quite impressive.

Maybe I am just stubborn, but it seems foolish to ignore the possibilities of direct-die jet impingement without solid empirical data to support its impracticability.

[BillA]

"solid empirical data to support its impracticability"
HUH ???
suggest searching the various forums,
LOTS of failures - and NO long term successes
and the goop has been tried too

the die is silicon, quite impermeable; but the underfill too is organic

don't bandy words about with someone who has not done it (me);
go ahead and do it, keep us posted

be cool

[gone_fishin]

The die can be as impermiable as possible but if the real estate where the traces are eventually get soppin wet, what's the point?
The horse is black and blue, let's beat it some more

Edit,

More thoughts.
It has been shown to work short term, but so has LN2.

[koslov]

I will try it, but I don't have the chemical / technical know-how to do it alone. So if you help me by answering my questions, and it fails, there will be rock-solid proof that despite anyone's best efforts, direct-die always fails.


Why do I think I am the only one who can make it work?

Simple: I can run a nonconductive coolant in my loop: deionized water. My stainless steel radiator was *designed* for it, and my pump has a noryl housing and stainless steel bearings (designed for saltwater aquariums).


One concern however: coolant purity. This pump has previously been used in an aquarium setup, so there is some gunk inside. Can I treat it to remove this? Will running it with deionized water for awhile get rid of the impurities?

Will the ceramic (Duron 700, remember) substrate or CPU die ionize with the water?

And I will check those other threads.

[jaydee]

Originally posted by koslov

Will the ceramic (Duron 700, remember) substrate or CPU die ionize with the water?

And I will check those other threads.

I got a Duron 700 sitting right here. Testing it will give usless data to modern CPU's as they are not made the same but at least it is a cheap CPU. The hard part is creating a water proof seal between the metal die and the CPU body. This will be near impossible as the CPU die with expand andcontract differetly from the temperature that the CPU body. How much I do not knw as I have no clue what the thermal properties of the CPU body material is.:shrug: BUT if you use some type of silicone that will flex a little it may work up till the core errodes. Remember there is not much room under the surface of the core before it is toast.

Direct die is certainly not reliable, hell plain water cooling with a water block is not even all that reliable yet. Still see quite a few failures in the forums. Mostly due to the user, but with direct die there is a lot less room for error. If the pump shut off the thermal protection may not act fast enough. Also the pump MUST be running BEFORE the power switch is hit on the computer. I do not think it would be wise to have a relay that turns the pump on at the same time. I fried a CPU in about a second once. Always unplug/shutoff the damn power supply when removing a HS/WB in case you acidentally hit the power switch!!! Not one of my better moments. But if that water fails to circulate fast enough the core will heat up rapidy. Maybe not toasting it but it maybe enough to break the seal.

Tuff challenge. I think more work from the CPU manufactuer would be needed to make it work reliably. Which I doubt is in their best intrest right now.

[bigben2k]

This won't help a dead pump scenario, but you could add a thin copper foil over the core: that should solve the permeable and wear issues. Of course you'd need a TIM joint... and so you're back to the equivalence of a waterblock...:shrug:

[airspirit]

It can be done if someone can manufacture a direct die block that is almost exactly the size of the die and then seals the sides well enough. The problem is finding that material to use to seal the sides so there is NO leakage to the packaging. The potential is there to do this, it is just a matter of finding a way to overcome that one hurdle. There has to be SOME material that will keep a watertight seal under high pressure and mid-range temperatures ... but I'm not going to be the one to find out what it is because I'm not crazy enough.

[satanicoo]

ok so no one tested it.

but with all the problems, i still have this question:

teoricaly, is direct die cooling better then a water cooling block?

[redleader]

If you can seal it well enough that its just water permating slowly throught the seal, it might be ok. People use silicone tube all the time even though its permiable.

Still its a stupid idea.

[koslov]

The deionized water is a stupid idea, I don't know what I was thinking... too hard to maintain purity if there is any metal at all in the cooling loop.

If one COULD manage to keep deionized water pure in a WC setup, sealing the die would not be an issue. Otherwise, any metal particles would corrode the traces connecting the die to the substrate.

I know next to nothing about sealants and adhesives, but I would suppose the problem is maintaining the seal between the die and substrate under heat and pressure. Sealing the package itself should be no problem; since it is organic (porous), almost any sealant is likely to adhere. But (I am assuming) the silicon die is not porous, making it hard for a sealant to adhere.

What about a mechanical sealant? An O-ring, shaped to fit around the die?

A precisely made top could fit flush over the die, and not leak if the gap between the die and top was small enough.

Here's another one: Machine out a hole in a baseplate just barely smaller then the CPU die, to be positioned over the die. (A small?) clamping pressure would keep any water from flowing between the die and baseplate, yet water could still be impinged in most of the die's surface.

I can make a diagram if there are any questions as to what I am describing.

I am still determined to test this concept myself, I just need advice on the appropriate course of actions to take. As demonstrated, there are many possible ways to accomplish direct die cooling (theoretically), but which is most feasible?

[jamicon]

what about if it was a p4 or similar with a heatspreader, you could use the heatspreader as what gets cooled (since intel do a good job at mateing it with the die). That would stop the problems with makeing a tight seal around a small die, and it would not matter if the water impingment wore thru - it would wear thru to the die (but i highly doubt it would get thru all that copper).


And yes, I realise this would not be 'true' direct die cooling.

[hara]

Direct die cooling has hit the wall. This is because water has a too low SHC to keep up with that small area and high heat output. It may regain acceptance if another coolant is used.

[koslov]

Originally posted by hara
Direct die cooling has hit the wall. This is because water has a too low SHC to keep up with that small area and high heat output. It may regain acceptance if another coolant is used.

Data? SHC??? Water has a specific heat capacity over TEN TIMES that of copper (Water: 4180 J/(kg*K) , Copper: 390 J/(kg*K) src: http://yesican.yorku.ca/home/sh_table.html ) . SHC is irrelevant to continuous cooling systems anyway.

Look at the pdf I linked to earlier: http://www.mit.edu/afs/athena/org/m/...ms/paper12.pdf

You will notice they even compared direct-die microjets to microchannels (REAL microchannels) and found microjets to be "more effective because they can remove larger heat fluxes at lower surface chip temperatures."

And their results are using flow rates measured in single digit milliliters per minute. "Higher flow rates... will facilitate even higher heat flux removal and heat transfer coefficients." If even 1 LPM can be attained using these microjets, you can decrease their measured C/W by at least one order of magnitude.

I think there should be no doubt that direct-die microjets can have much lower thermal resistance than any heatsinks or conventional waterblocks.

The question now is how to engineer one that will work.

edit: presented -> linked to. Ben: "present" as in "To offer for observation, examination, or consideration." Do they use a different dictionary in Texas?

[bigben2k]

Originally posted by koslov
Look at the pdf I presented: http://www.mit.edu/afs/athena/org/m/...ms/paper12.pdf

Now when you say "presented", do you mean "linked" or do you mean that you had a part in it?

[redleader]

Data? SHC??? Water has a specific heat capacity over TEN TIMES that of copper (Water: 4180 J/(kg*K) , Copper: 390 J/(kg*K) src: http://yesican.yorku.ca/home/sh_table.html ) . SHC is irrelevant to continuous cooling systems anyway.

Maybe. Think about how much coolant actually comes into contact compared to how much you pump. Only a small fraction probably. The rest is indirect- though other molecules. At 4.186 j/gK, you need a good deal of contact to cool a 100w CPU. Not saying this is impossible, just that specific heat capacity is probably a lot more important then usual due to the tiny surface area.

Its here that copper has the advantage at almost 400w/m*k; you can have a huge surface area to meet the coolant.

[koslov]

Parylene-C

This is the ideal conformal coating I have been searching for. It is already being used in micro-spray cooling experiments. In fact, I heard of it in an article on micro-spray experiments being done at UCLA. It has outstanding dielectric properties, and can be deposited so thinly it is thermally transparent. Also, it is extremely wear- and impact-resistant, so high-velocity microjets won't damage the coating.

The only price I have seen for Parylene deposition is $500 for a batch of four 100mm wafers. The deposition machines are made by Specialty Coating System. Here is a description of the process. cached version on google, other link appears broken

Unless there are more affordable deposition services, I guess it is up to AMD and Intel to start producing Parylene-coated chips before direct-die cooling is a real possibility.

[Volenti]

Look at the pdf I linked to earlier: http://www.mit.edu/afs/athena/org/m...ems/paper12.pdf

The method that they're using appears to be different from "conventional" direct die water cooling, since their using a coolant that vaporizes on contact, not simply jetted water.