News Researchers develop two-phase cooling technique that dissipates 7X more heat

[Admin]

Researchers built a 3D capillary system that allows water and steam to flow freely within it to dissipate heat efficiently.

Researchers develop two-phase cooling technique that dissipates 7X more heat : Read more

 

[Notton]

It sounds like vapor chamber, but with an enhanced wicking structure?

 

[A Stoner]

Most chips currently run lower than the boiling point of water. So, unless they are going to start being run at >120C or will be under some degree of vacuum to lower the boiling point. I am not sure how this is going to work. And if these are going to work without active pumping and fans, it is still going to need to expel the energy outside the system somewhere.

 

[palladin9479]

Most chips currently run lower than the boiling point of water. So, unless they are going to start being run at >120C or will be under some degree of vacuum to lower the boiling point. I am not sure how this is going to work. And if these are going to work without active pumping and fans, it is still going to need to expel the energy outside the system somewhere.

They are using a closed system under partial vacuum. H2O under goes phase change at much less then 100C, it's how your body sweats. 100C is where H2O undergoes rapid phase change, much faster then regular evaporation.

If you run custom loop WC with a cylindrical reservoir you can see this as water droplets near the top.

 

[JRStern]

So what they are saying, maybe, is that if you have a warm surface, say 90c, and wet it very sparsely, and with room to expand, that would remove more heat than much more water at lower temperatures? Even without vacuum. Could be. Weird stuff like this happens, like you can freeze a cup of boiling water faster than you can freeze a cup of lukewarm water. Is it practical? Dunno.

 

[thestryker]

I'm guessing this is furthering research with regards to cooling channels being built in directly during manufacture. I think TSMC had mentioned significant additional cooling capacity when using channels directly in the chips. This seems like something which would certainly be a big deal at the datacenter type level, but I don't see it any time soon for client.

 

[CelicaGT]

Ah, yes, the latent heat energy of vapourization. They act like it's some new discovery...

 

[helper800]

They are using a closed system under partial vacuum. H2O under goes phase change at much less then 100C, it's how your body sweats. 100C is where H2O undergoes rapid phase change, much faster then regular evaporation.

If you run custom loop WC with a cylindrical reservoir you can see this as water droplets near the top.

That combined with the very high conductance of copper in contact with the water accelerates this process.

 

[GoldenGonchies]

Every cooler with heat pipes already takes advantage of this phase transition effect, so does your fridge and air conditioner, so the "news" here wouldn't be using phase transitions in a cooler.

Building cooling channels directly into the device would be cool though. Choosing water as the heat carrier would be weird, but maybe its temporary.

 

[A Stoner]

They are using a closed system under partial vacuum. H2O under goes phase change at much less then 100C, it's how your body sweats. 100C is where H2O undergoes rapid phase change, much faster then regular evaporation.

If you run custom loop WC with a cylindrical reservoir you can see this as water droplets near the top.

Yeah, and ice can turn into vapor as well. Sublimation. But the cooling effect needs a pretty extensive amount of evaporation to cool down 200 plus watt CPUs which is why I am assuming they have to have the water at boil to get enough cooling. Then what ever turns to vapor has to be cooled quickly and returned to the CPU.

Vacuum would explain how they could do that. To get water to boil at 70C the pressure would have to be 1/3 that of atmosphere. Not a hard vacuum, but a pretty significant one. The problem with vacuums is that they fail, always, and very few systems are designed to have the vacuum restored with a port.

 

[bit_user]

Folks, no need to speculate. It's an open access article, available for free viewing & PDF download, here.

https://www.cell.com/cell-reports-physical-science/fulltext/S2666-3864(25)00119-5
​

 

[uplink-svk]

So heatpipe/vapor-chamber effect with some extra steps and requirements...

 

[George³]

Hmm liquid cooling with microchannels in the chips is developed before 20 or more years or I am wrong?

 

[bit_user]

Hmm liquid cooling with microchannels in the chips is developed before 20 or more years or I am wrong?

One significant difference is that's single-phase (i.e. no boiling). Obviously, boiling creates new challenges. Even then, if you follow the link I posted, this new paper cites two prior efforts that involve 2-phase cooling. You'd have to read at least the abstract to see exactly what they claim is novel or innovative about their method.

BTW, the "Extended PDF" has some nice graphics that Tom's should've included, but maybe they wanted to publish before they'd gotten permission to reprint them.

 

[edzieba]

Folks, no need to speculate. It's an open access article, available for free viewing & PDF download, here.

​

https://www.cell.com/cell-reports-physical-science/fulltext/S2666-3864(25)00119-5​

​

The research here is to do with the microstructure of the evaporator being better at seprating phase flow (i.e. keeping unevaporated liquid water from exiting through the outlet and keeping gaseous water vapour from backing up into the inlet).

Reporting the research with several paragraphs on how phase-change cooling works (and has worked in every heatpipe and vapour chamber of many decades) is like reporting on a new engine fuel injection system by dedicating several paragraphs to "wow, having round things covered in rubber reduces rolling resistance!".

 

[palladin9479]

Ah, yes, the latent heat energy of vapourization. They act like it's some new discovery...

I think it's more about the geometry of the heat exchanger allowing for more surface area and a quick way for the water vapor to escape and make room for more liquid water.

 

[CelicaGT]

I think it's more about the geometry of the heat exchanger allowing for more surface area and a quick way for the water vapor to escape and make room for more liquid water.

It was more a jab at how the article title frames it. As you state correctly, the news here is the micro channel design, not the well understood, taught in grade school science class, but still somehow sensationalized in the article header, latent heat of vapourization. I'd also prefer Tom's provide a direct link to the paper, not to another article on another website that does include the link.

 

[TheOtherOne]

Time to put your PC in the kitchen and cook food without gas! 🔥🥵

 

[bit_user]

I'd also prefer Tom's provide a direct link to the paper, not to another article on another website that does include the link.

They often do, but I'm not sure this author covers a lot of research topics and might not be accustomed to doing that. Since it's a Sunday and a holiday, maybe the usual editors, who would've caught that, aren't online.

We could tag @PaulAlcorn , in case he wants to mention to Jowi to chase down the paper link, in future. It was easy for me to find, as the linked SciTech Daily article had a link to it. Even if it didn't, you can often find a link by searching keywords and author names on Google Scholar.

A side-benefit of that is that the authors might have some of their own graphics that Tom's can reprint. A lot of times, the graphics in the published paper are actually made by the journal editors, so it probably wouldn't be possible for other publications to use them.

 

[DS426]

Getting far better cooling capacity deep into XPU packaging and dies themselves is indeed the highlight here. Think about the improvement in cooling when using direct-die water blocks, like what Der8auer makes.

 

[ekio]

The focus should be about how to make chips produce less heat in the first place.
Better quality ISAs (not x86...), compute units directly within the memory, etc

 

[bit_user]

The focus should be about how to make chips produce less heat in the first place.
Better quality ISAs (not x86...),

Let's be clear, though. This is mainly about GPUs. Sure, CPUs are getting bigger and hotter, but I think not at the rate that they're adding cores. So, the per-core power consumption is going down.

For GPUs, the hardware is so expensive that the up-front purchase costs are way more than the lifetime energy costs, at least when I ran the numbers for Hopper. Like 10x, even when you factor in cooling. So, the incentives are all pushing power consumption ever higher.

Maybe, when Nvidia & TSMC don't have such a lock on the market, and if HBM ceases to be a limiting factor, then it'll make sense to dial back frequencies a bit and just deploy more units. Unless/until that happens, these AI chips are going to be pushed ever further. IIRC, Nvidia is even talking about 600 kW per rack, in the second gen Rubin parts. So... buckle up.

compute units directly within the memory, etc

This is happening, too. I forget which, but one of the upcoming HBM standards will stack the DRAM dies directly on the compute dies.

Others, like Tenstorrent and Cerebras, have gone the path of integrating lots of SRAM in their compute dies and just scaling horizontally.

 

[palladin9479]

Better quality ISAs (not x86...)

There hasn't been an x86 chip made in decades now.

The focus should be about how to make chips produce less heat in the first place.

There is a thing called physics, that is what generates heat.

https://en.wikipedia.org/wiki/Joule_heating

All non-superconducting materials produce heat when an electric current is moved through them.

(Energy dissipated per unit time) = (Charge passing through resistor per unit time) × (Energy dissipated per charge passing through resistor)

In laymen's terms, it's clock speed multiplied by voltage multiplied by material resistance. Some materials are more resistant them others with superconductors having zero resistance. Within known materials science we've pretty much maxed out electrical conductivity for semiconductors, we're down so small that quantum effects are starting to complicate things. Now we're just trying to pump up clock speed and hope the cooling system can deal with the additional heat created from that ohmic heating effect.

You can't have infinite increase in performance per unit of energy (heat), you slam into a wall where physics says no.

 

[jonaswox]

In laymen's terms, it's clock speed multiplied by voltage multiplied by material resistance. Some materials are more resistant them others with superconductors having zero resistance. Within known materials science we've pretty much maxed out electrical conductivity for semiconductors, we're down so small that quantum effects are starting to complicate things. Now we're just trying to pump up clock speed and hope the cooling system can deal with the additional heat created from that ohmic heating effect.

You can't have infinite increase in performance per unit of energy (heat), you slam into a wall where physics says no.

I don't think anyone was disputing that fact?? Improving efficiency is obviously always a tier 1 priority. Yes you cannot transmit energy "for free" but I have zero clue why you are down this path 😀 in today's chips more power are used moving data than doing compute , so we obviously have some innovation to do.

The article is beyond odd. I don't think Tom's usual readers are much interested in capillary channel design 😀 and its hard to ooze more ignorance than presenting this as something new.

Yeah, and ice can turn into vapor as well. Sublimation. But the cooling effect needs a pretty extensive amount of evaporation to cool down 200 plus watt CPUs which is why I am assuming they have to have the water at boil to get enough cooling. Then what ever turns to vapor has to be cooled quickly and returned to the CPU.

Vacuum would explain how they could do that. To get water to boil at 70C the pressure would have to be 1/3 that of atmosphere. Not a hard vacuum, but a pretty significant one. The problem with vacuums is that they fail, always, and very few systems are designed to have the vacuum restored with a port.

It seems to work beyond splendid in a heat pipe.

The limiting factor is the rate at which you can transfer the energy into the water. So for this design the amount of water in contact with the heat source is proportional to cooling power. So the challenge here is to expose the small chip , to a lot of low pressure water , in a small amount of time. , doing anything that will significantly move the bar compared to the current state requires innovation in terms of contact area - and looking into making the channels part of the chip structure itself is possibly one of the few feasible routes to take with this idea. , since it's hard imagine more contact area without doing something relatively drastic.

For fun info is that phase change coolers was conceived by Einstein and is just a fridge. And a fun fact about vaporizing water is that temp rises as you add energy (ofc) , but when you hit the phase transition , temperature stalls for a small time, and you need to saturate the new heat capacity of the now turned vapor , before temp rises again. So you have this odd moment where you are adding energy to the system continuously and suddenly the temperature stops rising. That is the materialisation of the expanded heat capacity.

Supercritical water is even better at moving energy.