Liquid cooling directly on the chip

[numpad]

Summary:
 
Researchers at Georgia Institute of Technology have successfully built liquid cooling channels directly onto 28nm FPGA chips, eliminating the need for additional heatsinks and fans on top of the chips.

By etching cooling passages into the silicon and incorporating silicon cylinders for improved heat transfer the researchers pumped de-ionized water through the channels at 147 ml per minute, achieving a ΔT of 4°C, compared to an aircooled solution operating at 60°C.
 
Incorporating copper structures into the silicon would allow stacking multiple chips for cooler, denser, and consequently more energy-efficient systems.
 

[The researchers] etched cooling passages into the silicon, incorporating silicon cylinders approximately 100 microns in diameter to improve heat transmission into the liquid. A silicon layer was then placed over the flow passages, and ports were attached for the connection of water tubes.

 

This may open the door to stacking multiple chips, potentially multiple FPGA chips or FPGA chips with other chips that are high in power consumption.

 
Now for the exciting bit:

the same technology could also be used to cool CPUs, GPUs and other devices such as power amplifiers

 
[[source]]

 

The chips used for the research were stock chips that the researchers modified. If manufacturers integrated this into their designs the results could potentially be even better. We could also see drastic reductions in motherboard and expansion card sizes.

[VerticalDiscussions]

O-O

[BradH]

why arnt we funding this more!

[Kherm]

sounds very cool!

[-CLASSIFIED-]

[Enderman]

particle buildup would clog the channels and it would be nearly impossible to clean them

normal watercooling loops degrade over time leaving plastic or metal in thin waterblock fins, especially if metals are mixed or if the coolant is not replaced every few months

 

and you cant just put a CPU under boiling water to clean it

 

I dont see this becoming a practical solution any time soon...

even deliding is already extremely rare

[SurvivorNVL]

[bcredeur97]

This is really cool. My guess is you need less water flow in order to cool the hip if it's built into the die itself?

[Trik'Stari]

When can I have a full sized computer case filled with tiny ass transparent water cooling tubes, lit up to look like a bundle of fiber optic cables?

[Dingus]

<snip> achieving a ΔT of 4°C, compared to an aircooled solution operating at 60°C.

 

 

Does it seem to anyone else that these numbers are skewed to look more impressive?  What was the ambient temperature from which delta T of 4C was derived with an air cooled temp of 60C

[carolkarine]

Does it seem to anyone else that these numbers are skewed to look more impressive?  What was the ambient temperature from which delta T of 4C was derived with an air cooled temp of 60C

 

skewed or not, that's still damn impressive. only 4 degrees over idle is NUTS.

[theninja35]

Does it seem to anyone else that these numbers are skewed to look more impressive?  What was the ambient temperature from which delta T of 4C was derived with an air cooled temp of 60C

It is possible that they had an industrial sized radiator.

[Takata]

I thought IBM's Aquasar did this years ago, but after looking at it again, it looks like they might have only done microchannels in the copper, and not in the silicon.

[rtpb5642]

skewed or not, that's still damn impressive. only 4 degrees over idle is NUTS.

thus the room temperature was really low.

[SansVarnic]

<snip>

Very interesting, I would like to see were this goes.

 

particle buildup would clog the channels and it would be nearly impossible to clean them

normal watercooling loops degrade over time leaving plastic or metal in thin waterblock fins, especially if metals are mixed or if the coolant is not replaced every few months

 

and you cant just put a CPU under boiling water to clean it

 

I dont see this becoming a practical solution any time soon...

even deliding is already extremely rare

I agree but I would like to assume they are going to address this as they should be aware of this issue, hopefully they are able to come up with something.

[AHaskin14]

YESSSSSSS. Love the idea.

[BobbyG2]

More practical way of doing this would be to use vapor chambers, which you then use a normal heatsink or water block on top of, to better conduct heat between the cpu and heatsink/block. That way you don't have to worry about debris clogging up your cpu heatspreader

[GoldSrc]

Yeah, I have my doubts about this, it sounds like a pretty neat idea but I don't know.

[Dabombinable]

Fuuuuckk, my QX6850 needed this when it was launched. (Half of the CPU is binned far lower than the other and runs hotter-gotta love quad cores made the lazy way).

[cesrai]

What was the ambient ? what radiator did they use ? what fans ? I don't know but the ambient could've been very low, they could've used a huge ass radiator like the industrial ones, they could've used Delta fans or whatever it's just that results look off.

[SansVarnic]

More practical way of doing this would be to use vapor chambers, which you then use a normal heatsink or water block on top of, to better conduct heat between the cpu and heatsink/block. That way you don't have to worry about debris clogging up your cpu heatspreader

You make a good point.

[Stefan1024]

particle buildup would clog the channels and it would be nearly impossible to clean them

normal watercooling loops degrade over time leaving plastic or metal in thin waterblock fins, especially if metals are mixed or if the coolant is not replaced every few months

and you cant just put a CPU under boiling water to clean it

I dont see this becoming a practical solution any time soon...

even deliding is already extremely rare

You will need a dual loop systems. The liquid going through the chips will be a special mixture and the loop will be sealed. Than you cool the liqzud with water.

Also the idea of cooling the chip with embedded channels is floating around for years now. Nice to see they finally build it.

But IMOH it's not practical. Every mm^2 of the wafer in there high end processes is extremly expensive. But you can't shrink the channels to much so they goona eat up a lot of space makeing the chips bigger and super expensive.

Also without this cooling method the developpers are forced to save power. Therre is no point in makeing a 300 watt CPU.

[LukaP]

You will need a dual loop systems. The liquid going through the chips will be a special mixture and the loop will be sealed. Than you cool the liqzud with water.

Also the idea of cooling the chip with embedded channels is floating around for years now. Nice to see they finally build it.

But IMOH it's not practical. Every mm^2 of the wafer in there high end processes is extremly expensive. But you can't shrink the channels to much so they goona eat up a lot of space makeing the chips bigger and super expensive.

Also without this cooling method the developpers are forced to save power. Therre is no point in makeing a 300 watt CPU.

chip sizes have been staying around the same value for a while now (GPUs go up to TSMC's max size, 600mm2 iirc, and intels dies have been staying around 12-15mm in both dimensions for a while). This is (as far as i understand) embedded into the silicon wafer, so if you have a set (or setish) chip size, then the channels in the wafer can stay just the way they are. 

 

Also just fyi, wafers cost what they cost. and according to AMD its not that much, IIRC they said the losses from building big chips (and thus less per wafer) arent high

[Dabombinable]

chip sizes have been staying around the same value for a while now (GPUs go up to TSMC's max size, 600mm2 iirc, and intels dies have been staying around 12-15mm in both dimensions for a while). This is (as far as i understand) embedded into the silicon wafer, so if you have a set (or setish) chip size, then the channels in the wafer can stay just the way they are. 

 

Also just fyi, wafers cost what they cost. and according to AMD its not that much, IIRC they said the losses from building big chips (and thus less per wafer) arent high

Wait, what?: "I'm Out. I'll only post to update build/mod logs."

[LukaP]

Wait, what?: "I'm Out. I'll only post to update build/mod logs."

PS: before you yell at me I'll post to things that interest me...