Intel Develops Micro-Refrigerator To Cool Chips

Spacedonkey writes "Researchers at Intel, RTI International of North Carolina, and Arizona State University have made ultra-thin 'micro-refrigerators' for computer chips. The device uses a thermoelectric cooler made from nanostructured thin-film superlattice that can reduce the temperature by 55C when a current passes through it. In testing, it reduced the temperature on part of a chip by 15C without impairing its performance. The researchers say the component could be particularly useful for cooling hot spots that frequently occur on multi-core chips."

Comments

Pelletier effect?

[Fastfwd]

Is this the same as a pelletier effect? I hate fans and definitely would pay a premium to get rid of them.

Re:Pelletier effect?

[PotatoFarmer]

Looks very similar. And, like Peltier devices, the same fundamental problem remains - you've moved the heat from one spot to another, but it doesn't just disappear. You're still going to have to get it out of the general vicinity of other temperature-sensitive components, and that probably means fans.

Re:Pelletier effect?

[goombah99]

The thing I don't full understand here is how a cooling device that is the same area as the chip itself accomplishes much. It moves the heat.

It seems to me that to be more than traniently effective you still need tansfer the heat to something with greater surface area. And if the attached heatsink fins have the same surface area as before, what has been accomplished?

Arguably, if you can make the fins hotter they will radiate faster, so that could be one strategy. Or one way to gain is if you could extract work (current) from the heat. then it really would give a net cool. Usually however peltier devices actually add their own heat loads in addition to the heat transfer. Don't know about these, but the second law puts a limit on how much heat you can convert to work.

So where is the gain coming from? moving more heat with less added heat? that won't bode well for future improvements. Is it hotter heat sinks. or is it somehow managing to increase the sufface area?

Re:Pelletier effect?

[Rakh]

It might be that the cooling element can withstand much higher temperatures than the chip itself can. Thus there is a benefit to decrease the temperature inside the chip, even if it that does make the other side of the cooling device much hotter, since the heat will not be doing any damage on that side.

Re:Pelletier effect?

[AKAImBatman]

It seems to me that to be more than traniently effective you still need tansfer the heat to something with greater surface area. And if the attached heatsink fins have the same surface area as before, what has been accomplished?

Usually when a chip is running, only certain parts receive heavy use. These parts of the chip are going to be dumping more heat than the parts of the chip that are lying idle. In result, the chips has a few hotspots that are cooking your most important circuitry.

These mini-refrigerators will remove these hot spots by dispersing heat to areas that are currently underutilized. This should give the chip a more even operating temp and thus provide a greater surface area with which to disperse heat in general. The end result is that chips become more reliable and can be run at higher wattages without melting a hole through your chip. Higher wattages means that they can be clocked higher without error and thus get more work done in less time.

Re:Pelletier effect?

[blueg3]

The purpose is to move the heat within the chip. You're thinking of thermal transfer from the surface of the chip to the environment. What Intel is concerned about is thermal transfer from the component inside the chip that is generating heat to the outside surface.

Currently, chips are limited (in part) by heat production within the chip -- the heat gets to the chip surface by simple conduction. It's the components inside, generating the heat, that are going to fail at high temperatures, though.

Fortunately removal of heat at the chip's surface is not a big issue. As you note, a thermoelectric cooler could push the heat to a set of hot fins and a fan. Water coolers have plenty of capacity as well.

Re:Pelletier effect?

[jhfry]

Imagine this...

circuit operates at 3GHz at 20C or at 4GHz at 10C. So you say, lets cool it then!

Well if you couple chip to copper heatsink and fan, you can't possibly drop temps to sub-ambient temps.

However, with a cooler, you can... so long as you can dump the extra heat fast enough.

The biggest reason peltier coolers are rarely used is that they tend to cause condensation, and they acutally generate additional heat, requiring even more cooling.

However, if you could create a small one, that was within the housing, cooled only the necessary areas, and didn't need to be sealed from the humidity... it would be this.

Re:Pelletier effect?

[goombah99]

It might be that the cooling element can withstand much higher temperatures than the chip itself can. Thus there is a benefit to decrease the temperature inside the chip, even if it that does make the other side of the cooling device much hotter, since the heat will not be doing any damage on that side.

Yes, absolutely, but that's why I said "to be more than transiently" effctive. You can only do that for so long before you are first limited by heat capacity, then conduction, and finally convection. Then you can't sustain the differential any longer. If we assume the heat capacity of the far side is roughly (in terms of being accessibly within a diffusion length) of the chip itself then the time it takes to heat it to saturation will be at most a handful of times longer than it took to heat the original chip. You can substitute what you like for "handful", say 2x or 10x, but since the original chip heats in seconds, where not talking much of anything except for transient imporvement.

at somepoint it has to expand the surface area or rate of convection.

Re:Pelletier effect?

[CaptainPatent]

It's not so much about dealing with the heat overall, it's dealing with the heat in the hottest places. The more heat bottlenecks you get rid of, the hotter you can run the chip stably.

Don't get me wrong, the implementation doesn't come without drawbacks. There's the higher expense for the extra circuitry, and the higher electrical requirements to run the coolers. It looks like the only need for this is on high-end chips and even there it's only absolute bleeding edge that'll need anything like this, however for the enthusiast, the CAD designer, the video editor or the programmer, this may just be a breath of fresh air.

Re:Pelletier effect?

[Bearhouse]

Yes

http://en.wikipedia.org/wiki/Thermoelectric_cooling

Re:Pelletier effect?

[gtbritishskull]

The rate of convection depends on the temperature difference. Heat is being diffused, constantly, so more heat can be diffused per unit time if the temperature differenctial is higher.

Re:Pelletier effect?

[bluefoxlucid]

Shut the fuck up, that has nothing to do with anything anyone said. I read your whole post to the end, wtf are you babbling about that's anywhere near relevant?

Intel is also planning...

[Anonymous Coward]

...micro-keggers for tiny little beers and a nano-couch backplane.

Finally an architecture without that lamo fsb that Intel can be proud of.

Re:Intel is also planning...

[Kingrames]

But the flying spaghetti bus is the most important part of the computer.

Re:Peltier Effect

[Hrungnir]

They are putting this between hot spots of the chip and the heat spreader that normally covers the chip and gives a surface for heatsinks to sit on. So the heat is still being extracted by the heat sink, this thing just helps keep the hottest spots cooler

Re:Peltier Effect

[yttrstein]

You do it with one of these:

http://www.instructables.com/id/SH8YISTFPPG0L4D/

The heat sink for a piezoelectric spot cooler. So really no, there's not a huge amount of point until someone figures out how to do heat exchange with something other than heat conductive metals who's efficiency depends directly on surface area.

Re:Peltier Effect

[CyprusBlue113]

The idea isn't to remove the heat from the chip, the idea is to remove the heat from this ONE SPOT on the chip.
Basically they are trying to keep the core cooler, and dump heat to the transfer plate more effectively.

Competition

[phorm]

While many have already mentioned the obvious drawbacks (heat may drop on the most-effected areas, but it still needs to get the heat *out* of the case), if this is still an effective and innovative method for cooling then I wonder how Intel would go about licensing it. Holding onto tech that would allow for a 15c drop in core temperature would probably give them quite a strong advantage over competitors such as AMD, etc, which might be worth more than the advantage of licensing it out...

Re:Peltier Effect

[jhfry]

Don't think of it as a peltier cooler... think of it as a way of instantly transporting the heat away from a particular portion of an IC. It is integrated into the IC itself, so it's not a cooler, but a heat transmitter.

So, for example, if I want to "over clock" a portion of my IC, but it keeps running to hot, I could use this to extract heat from the area and distribute it where it doesn't matter so much.

Essentially... the Watts of heat you pull from your CPU, aren't generated across the entire chip, but are commonly more localized. For example, cache doesn't generate much heat. If I can take heat from the FPU and move it to the cache area, I can clock the FPU higher, and have fewer heat-related failures.

So in summery... it's not a cooler!

Yes but ....

[Archangel Michael]

Does it reach -232 degrees Celsius?????

Re:Peltier Effect

[jhfry]

I retract my previous statement... I thought this was a different tech I read about somewhere.

This is a cooler, it's a thin cooler they are placing between the chip and it's housing.

So it's a peltier cooler after all.

It does not just move the heat

[EmbeddedJanitor]

When you move heat, you're concentrating the heat and making the hot side hotter. Heat sinks are rated in Watts/degree so a heat sink that is 10 degrees above ambient will dump heat 5 times as fast as a heat sink at 2 degrees above ambient. Thus, a Peltier device pumping heat into a heatsink will cause the heatsink to run hotter and work more effectively.

Oh, FFS

[R2.0]

It's NOT a refrigerator. Refrigerators use the refrigeration cycle to move hat from one place to another. This is basically a Peltier. That doesn't make it any less valuable for it's purpose, but why didn't they just call it a "cooler"? I mean, it's not like the audience for these types of announcements is tech-illiterate.

What's the carnot efficiency? acoustic cooling.

[John Sokol]

Sorry I couldn't fit what I want into the title.

Carnot efficiency is very important.

Peltier/Seebeck and Thomson effects are only 5% where compressor based systems are more like 50%

So Peltier thermoelectric coolers actually create almost as much heat as they remove. You also end up with condensation problems when the chip drops below room temperature.

We were able to reach -90C with a stack of Peltier cooler, but it was terrible efficiency.
Didn't really matter for overclocking anyhow.
But we had to hermetically sealed the computer and fill it with Dry gas and desiccant to prevent icing and condensation. We lost a few motherboards before we went to that level.

There is also Thermionic cooling, that promises to be much more efficient.

With my old company we experimented with many forms of cooling some passive (high thermal conductivity) and some active.

One of the ones I liked best was a Micro Acoustic Cooler we made. Never did get to do enough testing, but it also looked very promising, using
a gas in a very small tube that was hit with high powered ultrasonic sound waves. It was amazing to see it work.

Magnetic cooling was also interesting.

One very effective solution was a (active phase change) micro compressor based system that was equivalent to a continuously hitting the CPU with freeze spray.

Re:Peltier Effect

[jhfry]

Yeah, though I doubt any scientist would dispute that a thermoelectric device that uses the peltier effect is commonly called a Pletier cooler... even if the name makes little sense in conversations between physicists.

There are many scientific products who's common name makes little sense in the context of those concerned with the theory of the device.

For example, if your an American I would bet that you incorrectly call a voltaic cell a battery. That's the common name even if it's dead wrong in a technical sense. When electronic engineers talk, they discuss a 1.5V AA size cell; while my wife would ask for a AA battery.

Re:What's the carnot efficiency? acoustic cooling.

[zippthorne]

For refrigeration, you'd be concerned with coefficient of performance (COP), rather than efficiency. It's a related term, basically the inverse of efficiency, but it refers to how much energy you need to use to move a given amount of energy between two temperatures.

But your numbers are weird. A refrigerator at 50% (COP of 2) sounds reasonable for a small device or large temperature difference, but COP of 20 is really good.

A COP of 5 percent would be horrific. 20W required to move 1W, a modern processor would require more electricity than a two-burner electric range at full power... I'd only put up with that kind of number for very specific applications. (like, if I needed to recycle a small amount liquid nitrogen in a sealed, difficult to access device or something)

By the way, why didn't you just slather a layer of nonconductive lacquer over the motherboard? Surely that would've been cheaper than a complicated heat exchanger, desiccant and sealed box trick.

Re:Peltier Effect

[Anonymous Coward]

In accordance with thermodynamics, there is no such thing as a cooler -- there are only heat transmitters. Any refrigerating device is a fancy way of moving heat from one place to another, generating a little extra heat in the process.

There is nothing in the laws of thermodynamics that would imply there's no such thing as a cooler. You either badly misunderstand the laws of thermodynamics, or you badly misunderstand what the word "cooler" means. I'm guessing the latter. A cooler is something that cools something else down. There's nothing in the definition of "cooler" that dictates how it does this, so heating something else up in the process is perfectly allowed, and does not violate either the laws of thermodynamics or the definition of "cooler".

How about reusing that heat?

[blankaBrew]

This might be useful for concentrating the heat in one place. However, what about using that heat by attaching micro-sized stirling engines to generate electricity which could recharge the batteries of a laptop? That would be kinda like a hybrid laptop: recapturing the wasted energy from the inefficiencies of the processor. That's something I'd like to see.

Re:What's the carnot efficiency? acoustic cooling.

[Tenebrousedge]

The best idea I've heard for using Peltiers is in combination with mineral oil submersion, which handily takes care of both heat transfer and condensation. Power and efficiency issues remain.