backtop


Print E-mail del.icio.us 43 comment(s) - last by ViroMan.. on Jun 11 at 3:43 AM


IBM Single Layer 3D Cooling Prototype  (Source: IBM)

Water Cooled 3D Stacked Chip  (Source: IBM)
IBM scientists say stacked processors have higher power densities than nuclear reactors

As processors add processing cores, circuits, and other components, the amount of heat they generate increases exponentially. Researchers and chipmakers have found that the ability to dissipate heat will be one of the main challenges to making processors in the future.

Researchers from IBM Labs and the Fraunhofer Institute in Berlin have demonstrated a prototype 3D chip that has a liquid cooling system built-in to deliver water directly between each layer of the stacked processor.

The 3D chip stacks take components in a traditional chip that sit side-by-side and stack them atop one another in a layer. While the process allows chipmakers to create more powerful chips, with shortened interconnects between components for data to travel, the process presents significant cooling challenges. With 3D stacked chip design data has to move only 1/1000th of the distance data needs to travel on a traditional 2D chip. In addition to the shortened pathways for data to traverse, the 3D process also allows for 100 times the amount of pathways for data to flow. Both techniques together greatly increase the potential performance of a 3D stacked chip.

The issue for researchers designing these 3D chips is that the stacked chips produce a very high level of aggregated heat dissipation of close to 1 kilowatt in a volume of only half a cubic centimeter. IBM researchers point out that that level of heat dissipation is 10 times higher than any other human-made device and that power densities in stacked processor designs are higher than in both nuclear and plasma reactors.

Thomas Brunschwiler, project leader at IBM’s Zurich Research Laboratory said in a statement, “As we package chips on top of each other to significantly speed a processor's capability to process data, we have found that conventional coolers attached to the back of a chip don't scale. In order to exploit the potential of high-performance 3-D chip stacking, we need interlayer cooling. Until now, nobody has demonstrated viable solutions to this problem.”

The solution to the problem the IBM team devised is to pipe water through cooling structures, as thin as 50 microns, between the individual layers of the stacked chip. The scientists were able to show cooling performance of up to 180 W/cm2 per layer in a stack with a footprint of 4 cm2. One IBM researcher says that the cooling performance is a significant breakthrough and that without the breakthrough the stacking of two or more high-density power layers would be impossible.

In experiments, the researchers passed water into a 1 by 1 cm test device made up of a cooling layer between two heat sources. The cooling layer was only 100 microns in height (about twice the thickness of a human hair) and has 10,000 vertical interconnects per square cm.

The interconnects were hermetically sealed to prevent the water from causing electrical shorts within the chip. Individual layers were built using existing 3D packaging fabrication methods to etch the holes for signal transmission from one layer to the next. Each interconnect had a silicon wall around it and a fine layer of silicon oxide to insulate the electrical connections from the water. A new thin-film soldering technique was developed by the researchers to provide the precision and robustness needed to provide thermal contact for the cooling film.

IBM first announced its 3D chip stacking process in April of 2007.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Or they could...
By Cheapshot on 6/6/2008 11:55:47 AM , Rating: 2
...use Flourinert or develop something better to submerse the 3D chip into.

I think submersion into Non Conductive / Non Corrossive fluids is the logical direction for extreme heat processors and electrical devices.




RE: Or they could...
By FITCamaro on 6/6/08, Rating: 0
RE: Or they could...
By geddarkstorm on 6/6/08, Rating: 0
RE: Or they could...
By snownpaint on 6/6/2008 1:07:16 PM , Rating: 3
Like you car engine, you only need to drain the coolant if you are opening the engine, most components on a computer/engine are attached to it. (modular "maintenance" design) The same would be true if you needed to remove or open your CPU.. Which almost nobody opens. If you had to remove it, drain the system, pop it off and maybe replace the seals upon installing a new CPU. I wouldn't be surprised if the MoBo companies started building the coolant system into the MoBo for chips of this caliber.


RE: Or they could...
By bldckstark on 6/6/2008 12:27:30 PM , Rating: 4
Submersing the 3D chips in liquid will not create the cooling required. The liquid has to be pumped through the interconnect area in order to remove the heat from the chip at each level. Otherwise you would burn up the center chips because you only had flow across the top and bottom chip.


RE: Or they could...
By MrBlastman on 6/6/2008 12:44:24 PM , Rating: 4
You must also take into account the viscosity of the liquid being used to cool the chips.

Many conventional oils as was suggested - have a much higher viscosity than water which would severely hamper flow and heat dissipation.

You could also go the extreme of using something such as methyl alcohol but that has the dowside of a much lower boiling point (147 F/64.5C) which would also create serious problems.


RE: Or they could...
By masher2 (blog) on 6/6/2008 2:01:02 PM , Rating: 2
Sealed in an enclosure under pressure, the boiling point would be considerably higher.


RE: Or they could...
By MrBlastman on 6/6/2008 2:17:36 PM , Rating: 3
Absolutely it would - but just how much pressure can these sensitive structures handle before collapsing? I think you have to consider that as well.


RE: Or they could...
By bobsmith1492 on 6/6/2008 5:42:38 PM , Rating: 2
With the entire chip inside a sealed enclosure the strength of the chip wouldn't matter, just the strength of the enclosure - same way we don't feel any pressure from the atmosphere since all the gasses in our bodies are at the same pressure.


RE: Or they could...
By JonnyDough on 6/8/2008 6:59:50 AM , Rating: 2
2 words. Carbon nanotubes!


RE: Or they could...
By MrPoletski on 6/7/2008 3:31:01 AM , Rating: 2
you could design a cooler around methyl alcohol that uses the fact it boils at a lower temperature. If your flow across the layers is instead vertical then the action of heating can cause the fluid to rise through the device and boiling wouldgive you the benefit of using the latent heat. pipe the vaporised medium away from the chip and back tot he bottom again.

I beleive there is a CPU cooler that does something similar with itself on the market. Doing this in a processor would be rather difficult - but then so is passing water through such a small space...


RE: Or they could...
By pnyffeler on 6/6/2008 3:42:29 PM , Rating: 2
While water may have its problems, besides its abundance and cost (practically nothing in the 3rd world), water has one huge advantage over copper: enormous heat capacity.

Tossing all issues out the window, say you had a CPU under 1 kilogram of water and the same CPU under a kilogram of copper. If the CPU was operated for a time long enough to raise the temperature of the water 10 degrees Celsius, the block of copper, under the same conditions, would go up 108 degrees Celsius.

The reason everyone uses copper instead of water for most applications is that copper doesn't tend to "leak" and short out electronics. That said, it can't hold a candle to water.

Fluorinert has a heat capacity is about 1/4 that of water and is very expensive, compared to water (IBM wants entire server farms running in water, not just a lone CPU). That said, it's still in the same ballpark, can be pumped, and is still loads better than copper.


RE: Or they could...
By greylica on 6/6/2008 5:01:31 PM , Rating: 2
They probably will make a primary and a secondary system, one for the processor itself, with mechanical characrteristics to support pressure from outside coolant systems and the coolant system itself. The first will simply take the heat to a big dissipator plate, the second, can be a simple heat pipe cooler.
Intel uses silicone into its chips, for example.


RE: Or they could...
By Reclaimer77 on 6/6/2008 9:01:49 PM , Rating: 5
quote:
...use Flourinert or develop something better to submerse the 3D chip into.


Its IBM. Why question it ? Honestly.

If IBM ran NASA we wouldn't be going to Mars. We would be LIVING on it already.

If they used water then you can be sure water is the best solution. Debate over.


RE: Or they could...
By ViroMan on 6/7/2008 3:24:55 AM , Rating: 4
Im sure watter is the best solution on at least two ways that I can think of. Its VERY cost effective and contrary to popular belief, it is non conductive.

Distilled watter can NOT short out electronics. It is a non conductive liquid. You can place your entire computer(minus the HD's and CD Drives) in a tub full of Distilled watter and it will work perfectly fine(till it overheats, The fans are not strong enough to move watter well). There are a couple of tiny movies out there showing what im talking about. The bad side however, is that it is horrifically corrosive, which can be fixed by using some additives. I forget what the names of the additives are atm but, its the same ones used by the watter cooled computers(don't want those copper watter blocks to get corroded/clogged).


RE: Or they could...
By Sethanus on 6/7/2008 10:35:54 PM , Rating: 1
I'll look into it but until you can spell "Water" I'll have to take your opinion with a tonne of salt.


RE: Or they could...
By ViroMan on 6/7/2008 11:43:05 PM , Rating: 2
What? watter isn't speld that way?! NOOOO!!! I BENN WRONG ALL THIS TIME!

Seriously, I saw that after I posted but, you know... no edit button. I wonder why the spell check didn't pick that up?


RE: Or they could...
By ViroMan on 6/11/2008 3:43:23 AM , Rating: 2
well FFS I couldn't find the article I wanted to post here but here are two to show that I am not spewing crap.

http://www.wku.edu/news/releases/february/amlani.h...

This one is incredible.(must be the future)
http://infovista-beta.blogspot.com/2007/06/transis...


RE: Or they could...
By EE Pete on 6/7/2008 11:44:31 PM , Rating: 2
He's right you know.


RE: Or they could...
By freeagle on 6/7/2008 7:23:43 AM , Rating: 2
... think of something that does not generate so much heat. How about investing a lot in a room temperature superconductors research?


RE: Or they could...
By ViroMan on 6/7/2008 11:58:49 PM , Rating: 2
There is a architecture type for cpus out there that creates very little heat and uses less energy to achieve the same result. It did this incredible feat by not using transistors. Transistors are notorious for creating heat. How you say? Because it takes energy to switch. Energy is wasted/converted into heat. Unfortunately the end result of not using transistors is something that is 2-4 times the original size of something created now. God I wish I had the research paper I got this from. I will post it when I find it.


RE: Or they could...
By TheriusDrake on 6/9/2008 12:38:35 AM , Rating: 3
quote:
Researchers and chipmakers have found that the ability to dissipate heat will be one of the main challenges to making processors in the future.


No! Really?? That must be a new issue...


plugged up?
By wackypete on 6/6/2008 12:25:25 PM , Rating: 2
Wouldn't these tiny water pipes get clogged up pretty quick from minerals and sediment etc.? I guess you'd have to use demineralized water. But even then, over time, if the pipes clog up do you have to throw out the chip?




RE: plugged up?
By JoshuaBuss on 6/6/2008 12:48:03 PM , Rating: 2
my understanding of the system is that it would be completely enclosed, so as long as each material the water touches is chosen carefully, it should remain inert and therefore safe.

as someone else pointed out, this is simply about bringing the heat from the center of the chip to the outer edges, where i would assume it could then be removed in more conventional ways.


RE: plugged up?
By mmntech on 6/6/2008 1:18:37 PM , Rating: 2
Even with contemporary liquid cooling systems you're supposed to use demineralized water. I expect though that they type of water will be heavily filtered. A lot more than the stuff you buy at the auto parts store.


RE: plugged up?
By masher2 (blog) on 6/6/2008 2:00:04 PM , Rating: 3
Water, though -- especially demineralized water -- is one of the most powerful solvents known to man. It doesn't stay demineralized long, especially when flowing across metallic and/or silicate materials.


RE: plugged up?
By DeepBlue1975 on 6/6/2008 2:21:51 PM , Rating: 2
I think they used water in the prototype just for proof of concept, but liquid cooling can use something better than water, such as, like you said before, some alcohol solution, though those are more frequently used in below-ambient cooling solutions.
Anyway, they say the different layers are sealed, so I assume they considered the corrosive long term potential you mentioned.

Either way you look it, it's another cool solution essay by IBM. (two-pun combo intended)


RE: plugged up?
By Silver2k7 on 6/6/2008 7:42:48 PM , Rating: 2
There is some non-conductive water cooling fluid you can buy.. IIRC its called 'Fluid XP' might be worth it..


RE: plugged up?
By tastyratz on 6/6/2008 10:07:48 PM , Rating: 2
That's good stuff and there's handfuls of other ones like fesser. The problem of corrosion still exists however. Even through micro erosion sediment WILL be released. With such microchanneling I wonder how thick the layers are to counter that (say erosion eats through the boundary in just 1 year of use, etc.)
Although intriguing it seems to me in its current form nothing more than proof of concept.


RE: plugged up?
By JonnyDough on 6/8/2008 7:07:44 AM , Rating: 2
You hit the nail on the head. It really all hinges on what material they use to entrap the liquid. One would assume it'll be some sort of metal alloy, naturally. But other possibilities would include any type of element, IBM first and foremost is awesome when it comes to chemistry. They lead in computer technology research at a chemical level, not just a chip design level. Great chemistry research is what allows them to make forward leaps in electronic tech.


Water really necessary?
By timmyeatchips on 6/6/2008 12:42:19 PM , Rating: 2
Is this really much more effective than having something like copper heat pipes running through the layers in the same config, connecting to active cooling outside of the chip? Might even be able to design the structure to improve the electrical environment around the components.

Also - these chips really produce 1kW?! That seems crazy.




RE: Water really necessary?
By MrPickins on 6/6/2008 1:06:05 PM , Rating: 2
Have you ever seen a copper heatpipe 100 microns in width?

There is your answer.


RE: Water really necessary?
By JediJeb on 6/6/2008 1:14:11 PM , Rating: 2
exactly, its not like stacking one P4 on top of another P4, but stacking the components inside the chip that is inside the P4 on top of each other. Very tiny water passages there.


By StevoLincolnite on 6/7/2008 10:22:37 AM , Rating: 2
Mind you the Prescott could have probably used this, even though it did not use any form of "3D Packaging".


RE: Water really necessary?
By snownpaint on 6/6/2008 1:47:05 PM , Rating: 1
Nanotubes with supercooling liquids passing through them.

Carbon (diamond) is great at transferring heat, and the liquid could be pumped through them, of course under high pressures. Maybe the heat would cause a natural convection in the cooling pipes and fluid, eliminating the need for a pump/impeller.

While were in the SciFi reality. How about instead of making the chips smaller, we make then bigger with more open space in them -like window screens with transistors covering the outside surface. Then the CPU screens have cool liquid move through and past them.. It would be about the size of a baby food jar, but imagine what you could over clock to.. Require 3D printing


RE: Water really necessary?
By masher2 (blog) on 6/6/2008 2:04:12 PM , Rating: 2
> "How about instead of making the chips smaller, we make then bigger "

That begins to degrade performance, simply due to speed-of-light delays for signals to cross the chip itself. At 4Ghz, light (or the fastest electrical signal) only travels about 75m per clock cycle, so a chip that's more than a few cm across will start to have timing issues.


RE: Water really necessary?
By random git on 6/7/2008 2:23:18 PM , Rating: 2
It's 75 mm/clock cycle @4GHz. I hope for your sake that that was just a typo.

Btw the article has an error. Power consumption scales linearly with the amount of circuits and transistors, not exponentially.


RE: Water really necessary?
By EE Pete on 6/7/2008 11:57:52 PM , Rating: 2
With 3D expansion doesn't the power consumption also get affected by the increased resistivity of the semiconductor due to the black body radiation? And also the non linear expansion of volumetric space?


Nerds rejoice!
By chmilz on 6/6/2008 11:55:05 AM , Rating: 3
"My cubic d6 is smarter than your pathetic icosahedron d20!"




RE: Nerds rejoice!
By FITCamaro on 6/6/2008 12:08:26 PM , Rating: 2
Icosawasathingy?


RE: Nerds rejoice!
By DeepBlue1975 on 6/6/2008 12:56:22 PM , Rating: 2
Wait till I solve the tetrion interference problem within my cubic d6, and I'll have your icosacrapion totally pwned at warp 8.

Now I'm gonna leave you cos I've gotta fix the darn plasma conduit that's showing on a level 3 system diagnostic.


RE: Nerds rejoice!
By chmilz on 6/6/2008 8:08:03 PM , Rating: 2
Hold on a minute, my give-a-damn is broken.


By Jedi2155 on 6/9/2008 12:01:30 AM , Rating: 2
Why not harvest the waste heat into powering itself?!?!?

Of course I bet there are tons of technical reasons/hurdles why it can't but just a thought.....




"I'm an Internet expert too. It's all right to wire the industrial zone only, but there are many problems if other regions of the North are wired." -- North Korean Supreme Commander Kim Jong-il














botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki