Future Processors Could Circulate Coolant Between 3D Chip Stacks

AlphaAtlas

[H]ard|Gawd
Staff member
Joined
Mar 3, 2018
Messages
1,713
As we've noted in previous articles, manufacturers are quickly hitting the limits of chip shrinking, which means cutting edge processors will increasingly have to turn to advanced packaging solutions (like Intel's "Foveros" and EMIB designs or AMD's chiplet-based products) for performance gains. But in 3D stacked designs, cooling becomes a serious issue, as the top layers of chips effectively insulate the lower layers from heatsinks, and power density increases as more silicon is squeezed into a smaller area. To get around that issue, a group of researchers claim that chip makers could pump distilled water directly through the 3D stacks of future processor designs. The researchers say the approach makes use of gaps that would exist between 3D stacked chips anyway, and that valves could be used to direct more flow to particular chips or zones, while tiny patches of "themoelectric materials" (which typically use electricity to actively pump heat) could thermally link particularly hot areas of the chip to the fluid.

The tier-specific cooling approach, compared with conventional microfluidic cooling, can reduce the pumping power by 37.5%, preventing overcooling, when an operating temperature is specified.

For anyone who's interested in further reading, the paper also rounds up, and links to, previous microfluidic cooling efforts.
 

viscountalpha

2[H]4U
Joined
Oct 16, 2011
Messages
2,549
Distilled water? Ha. No. That's a terrible idea. A non conductive liquid? Possibly. There's a better solution though entirely.
 
Joined
Jun 1, 2018
Messages
691
the games kids play these days are just too fast, make the computers hot. thats the real problem. make the games slow down, maybe the kids take a break go play outside. let the computer cool off. IMHO.
 

exiled350

[H]ard|Gawd
Joined
Jun 26, 2013
Messages
1,633
Distilled water? Ha. No. That's a terrible idea. A non conductive liquid? Possibly. There's a better solution though entirely.

Pure water is quite a poor conductor of electricity. https://www.lenntech.com/applications/ultrapure/conductivity/water-conductivity.htm

It is however a pretty aggressive solvent, which would be a much greater concern. Also adding to the fact it becomes more conductive as it adds various disolved solids.

So realistically you end up with a nice runaway reaction. As it starts stripping away the metals in the die, it becomes electrically conductive...
 

viscountalpha

2[H]4U
Joined
Oct 16, 2011
Messages
2,549
Pure water is quite a poor conductor of electricity. https://www.lenntech.com/applications/ultrapure/conductivity/water-conductivity.htm

It is however a pretty aggressive solvent, which would be a much greater concern. Also adding to the fact it becomes more conductive as it adds various disolved solids.

So realistically you end up with a nice runaway reaction. As it starts stripping away the metals in the die, it becomes electrically conductive...

Pure water likes to attach itself to other molecules. That's what makes century long erosion happen. Pure water is a TERRIBLE insulated coolant. The boiling point is far too low for its application as well.
 
Last edited:

mesyn191

2[H]4U
Joined
Jun 28, 2004
Messages
2,983
This is a oooold idea that gets brought up every few years. IBM was working something like this over a decade ago and Intel put in some effort too.

https://thefutureofthings.com/3547-ibms-nano-water-cooled-chips/

It'll probably never happen, at least in consumer CPU's, because cooling with water is a pretty difficult and expensive thing to do reliably in the long run (3yr+) even in a enclosed system.

It'll be much easier to do something like getting rid of the heatspreader and just enclose the entire top of the package in one huge vapor chamber so that all the dies on the package can benefit from the increased heat removal abilities it'll allow. No need for pumps of any sort either and it'll last a very long time before it starts to break down and loose its effectiveness.
 

Galvin

2[H]4U
Joined
Jan 22, 2002
Messages
2,697
We just need to switch to chips that run off light. Since light doesn't generate heat, cooling solved
 

BitMaster

Limp Gawd
Joined
Nov 10, 2016
Messages
367
We just need to switch to chips that run off light. Since light doesn't generate heat, cooling solved


Ehhhhh...NO !


and you still need to make I's to 0's...and that causes a lot of heat. Where to with all the charge from all those I's that are now 0's ? One could say to forget causes a lot of energy to go somewhere else.
 

DNMock

Limp Gawd
Joined
Apr 16, 2015
Messages
399
We just need to switch to chips that run off light. Since light doesn't generate heat, cooling solved

8smlp3tlnysz.gif



Anyway, distilled water in the chip is just inviting all manner of problems. Are they just not going to ship CPU's in the winter months? Order a new cpu from amazon only to find out it froze on the truck and the expanded ice cracked it.
 
Last edited:

Arcygenical

Fully [H]
Joined
Jun 10, 2005
Messages
25,056
I'll take things that will never happen for 2000, Alex.

We just need to switch to chips that run off light. Since light doesn't generate heat, cooling solved

I mean, things that generate light, by definition, create heat. Ever run a 100w COB LED? Easily dumps 70w to the heatsink back... Or for that matter a 1w+ Semiconductor laser source?
 

DeeFrag

Supreme [H]ardness
Joined
Jan 14, 2005
Messages
6,412
Bring back the slotket with a processor and water block mounted to each side of the slot.

Or even a cube slotket with processors on 5 sides.
 

Qthulu

Weaksauce
Joined
Apr 19, 2018
Messages
103
It's better to not stack the chips on each other.....stack the chips on the heatsink and actually.have the chip substrate act as the heatsink and transfer the heat to air or liquid.
 

mesyn191

2[H]4U
Joined
Jun 28, 2004
Messages
2,983
Chip substrates are relatively poor conductors of heat.

Much worse than a IHS. The chip substrate also has to have the pins or pads on the opposite side to connect to the socket as well which will reduce its ability to transfer heat even more.

So that idea is unworkable even for single dies.

For die stacking it'd be even worse since now the heat from the stacked dies has to transfer through the entire stack as well as the substrate.

Die stacking as some significant potential performance advantages to it so saying "just don't do it" isn't useful here either.
 

tangoseal

[H]F Junkie
Joined
Dec 18, 2010
Messages
8,925
Pure water likes to attach itself to other molecules. That's what makes century long erosion happen. Pure water is a TERRIBLE insulated coolant. The boiling point is far too low for its application as well.

That's right. In the absence of solute water will have in a sense a boiling point depression instead of the normal elevated state.

Couple with the fact that water molecules can also act as an acid or a base depending on the pH of the whole system so anything that can contribute protons or accept protons has to be absolutely mitigated out of the processors material build. Otherwise we just create a flowing solution that could end up being NaOH- if there is any sodium ions and sodium hydroxide is hella corrosive to aluminum and other metals. Even the slightest amount can wreck a system.
 

viscountalpha

2[H]4U
Joined
Oct 16, 2011
Messages
2,549
I will admit, it's not so easy to say "just add a coolant binder" it's quite a bit more complex but I think it's doable. Intel might already be looking at it.
 

power666

Weaksauce
Joined
Jun 23, 2018
Messages
110
[H]ard had an interview with someone from AMD a couple weeks back and they explained its to allow these cards to be used form medium/high level gaming, which wouldn't necessarily need the Vram, but then also be a low cost solution for those who can't afford a high end workstation card to do compute or render tasks that would need the vram.

Ehhhhh...NO !


and you still need to make I's to 0's...and that causes a lot of heat. Where to with all the charge from all those I's that are now 0's ? One could say to forget causes a lot of energy to go somewhere else.

Actually designs for optical logic gates exist today in labs. Conceptually it is possible to build an optical based processor. The catch isn't power consumption that is preventing their deployment but is two fold with the precision design necessary to manufacture even a single optical processor and secondly the mass production required for the current processor market.


The paper below is a good starting point as it summarizes much of the existing research into this topic. Diving into the citations gets into more specifics.
https://www.researchgate.net/public...c_Gates_Designs_Classification_and_Comparison
 
Top