German-Japanese researchers invent electricity-free tech that could cool datacenters Datacenter cooling could get a lot more affordable, and sustainable, if a joint German-Japanese research project can be scaled beyond the laboratory. Researchers from the Karlsruhe Institute of Technology and the University of Tsukuba published a paper in Nature Energy on Friday discussing the success they had with a solid- …
Where does the heat go? Pretty basic physics says it has to go somewhere. With air and water cooling, the heat goes out with the now-hot air or water, and so into the environment that way. You could hook a thermoelectric generator to turn some into power, but that would be happening on a large scale already if it was practical to do.
I’m also enjoying the thought experiment of working this one out, but it seems to be fairly simple - like any aircon or fridge - it’s just moving energy from one zone to another. The two zones are on different sides of a metal plate (or a metal laminate, more properly), and yes clearly if the cooler side wasn’t able to radiate it away you’d hit equilibrium pretty quickly, like a hotel mini-fridge running in a cupboard.
Now I’m wondering what happens if you stack these laminates? Do you get 4deg active cooling per layer? Because that would get very interesting very quickly.
You get 4degC cooling at 86C, which is already in the unhappy zone for silicon. Presumably if you wanted 8 degrees of cooling with two layers, then the first layer needs to be at 90C?
If you could stack the layers to move more heat and create a thermal differential out of nothing, could you produce free energy by strapping it to a thermoelectric generator?
I'm still very confused at how this is supposed to work. Has anyone asked Martin Fleischmann?
86 °C "unhappy region" depends on what you cool. x86 CPUs today don't like it since they are so big, but thermal throttling usually kicks in at 90°C to near 100°C. GPUs usually don't like it since they are so big, though I've seen > 100°C there quite often as normal temperature at GPUs which are smaller. And that is the key: Smaller silicon chips can get much hotter without breaking.
But I would have loved to see them testing with 70 °C region as well. They might do that soon.
I understand that these are logic devices. The silicon I'm using is thermally limited for 175C, but it's dimensioned differently. The stuff I use that *doesn't* have thermal throttling has the more typical 150C limit, but you understand that's just a nice round number.
The Intel stuff normally throttles at 100+ C. Unlike analog electronics, most of the heat dissipation is not actually at the switch junction: it's bulk resistance.
It’s definitely not breaking thermodynamics or creating free energy—it’s just a super-efficient heat spreader, not an air conditioner.
It works because at 86°C, the temperature gap between the hot silicon and the ambient air is huge. The stacked layers just give the heat a path of much lower resistance to radiate away fast.
Stacking more layers doesn't "multiply" the cooling or create a thermal differential from nothing—it just spreads the heat over a larger area. Once the temp evens out, the heat flow stops, so a TEG wouldn't generate free power either. No Fleischmann magic needed, just clever materials science!
That was exactly my question at first. The author seemed to imply the heat just disappeared since this was cited as a sustainable solution to global warming issues. Don't the data centers still generate just as much heat?
I suppose, if the cooling device is not using electricity to power it, it must therefore be using less energy as a system overall and is thus generating less heat (or better for mother earth and our resource depletion and contribution to CO2 and waste heat, etc.)
Mother Gaia has done just fine for most of geological (and biological) time with CO2 up to 9,000ppm, and no permanent ice anywhere on the globe. If you want to talk about what's comfortable for monke, that's one thing, but the biosphere will do just fine with us unsequestering carbon and putting it back into the desperately denuded biosphere, and keeping the glaciers at bay. Ice is not a biosphere, no matter what Minecraft tells you.
Of course to the hot side! As article states (K = Kelvin, I write °C when they used K since in that context it IS the same):
- Even if the "outside" temperature was 130°C, i.e. over 50°C above the 86°C "to cool" side it managed to transfer 2 °C overall.
- It managed 4 °C performance with "normal" temperatures, I assume 25°C room temperature here.
- The actual cooling of the thin film was over 12 °C.
Read the nature PDF at https://www.nature.com/articles/s41560-026-02122-6 , linked in the article, has nice pictures of the actual tests.
The TheReg Article is, IMHO, clear here: It is heat transfer, not heat disappearing.
I *think* it's this:
One material, when heated, doesn't increase in temperature: instead it changes shape. Think of it as a third kind of phase-change, like melting or boiling which consume energy without temperature change.
That shape change can exert a force on something else - in this case the two are closely bound together.
The second thing has a strange property: when you bend it, it gets much warmer than a passive material would. Also like a phase-change, like condensation or freezing, which dump energy out.
Thus energy moves from the first material to the second - just once, sure - but it's like a sheet where you blow a flame on one side, and it stays cool (for a while), but the other side of the sheet gets hot instead, and radiates/convects the heat away to air.
This does suggest they can stack, getting much larger temperate deltas, but you'd still need active cooling or radiators on the hot side. But because it's like a phase-change, I don't get how they're thinking you can provide continuous cooling, without replacing the material repeatedly - like sitting an ice cube on it: you need another one soon.
If "the other side" is colder than the chip, then they've invented a low thermal resistance layer - a bit like having larger fins on your heatsink. But the heat still has to go somewhere, whether that's into air or water or some other medium; and that heat still has to vent to the environment somehow.
If "the other side" is hotter than the chip, then they've invented a perpetual motion machine.
"If "the other side" is hotter than the chip, then they've invented a perpetual motion machine."
That's been my thought here. It's a movement of entropy and doing that without the creation of even more entropy is perpetual motion.
Even a heat pump needs some sort of energy input to be able to move heat from one place to another. It also has limitations based on ∆T.
I'd like to see a demonstration of it working inside the thermal-vac chamber at JPL (cooled and pumped down, of course).
I am shocked, shocked to find that some folks still seem to believe in quaint 19th century concepts like Conservation of Energy*. Come on folks. This is the 21st Century. Reality is negotiable. Have these neoluddites** paused for even a moment to reflect that were Conservation of Energy valid, exciting technical advances such as this could have few, if any, practical uses. </SARC>
* Since Einstein (1905) -- Conservation of Mass-Energy
** The Luddites were right y'know. The weaving machines did eventually destroy their cozy, if inefficient, cottage weaving industry.
… and jobs at t’mill also destroyed as closed as it was offshored abroad as part of global capitalism’s pernicious trickle down economics.
Doing an internet search…
‘ Where Fabric is Woven???
… While raw cotton is grown across tropical and subtropical regions, the actual processing—spinning fiber into yarn and weaving or knitting it into fabric—is heavily concentrated in countries with large textile industries like China, India, and Pakistan.’.
So not Oldham, Manchester, Bolton, Nottingham, Burnley, Derwent Valley.
"The Luddites were right y'know. The weaving machines did eventually destroy their cozy, if inefficient, cottage weaving industry."
The inefficient industry also made cloth expensive. With the introduction of power looms, common cloth became much less expensive and more people had the ability to own more than two shirts. It did displace a load of people as change often does. The skills changed as well. What must be appreciated is that the power looms magnified what a single person could produce in a single day. Now the push seems to be on to get humans out completely of jobs that pay more than minimums. Keeping the oil and grease topped up isn't skilled labor.
Indeed. We have known this for some time. As Larry Niven pointed out in 1970 in his SF novel Ringworld:
""Heat is produced as a waste product of civilization. ... Most kinds of cooling systems only pump heat around, and produce more heat for power. ... Do you understand, then, that the heat of our civilization was making our world uninhabitable?"
"Do you understand, then, that the heat of our civilization was making our world uninhabitable?"
So we need to locate the Outsiders, buy a planetary drive or 6 and move Earth back from the sun? The moon is pretty handy to have around too so we'll need to keep that.
I get the feeling that we couldn't pay for all of that with AI generated cat videos even though it's the DC's doing that work that may contribute to the need to make a move.
Power stations generate their heat over a large surface. A CPU/GPU/NPU/whateverPU ist hostspots only. Imagine a 200 W halogen bulb 1cm by 2cm big, which you have to cool to 70°C, evenly or else the glass breaks. Much easier than a chunky transformer generating 200 W heat with its large surface. And transformers can be run fine at 200°C or more - but you have to stay below the curie temperature.
Yeah, it's Figure 10 in Supplementary Materials of the "paper" in Nature Energy (TFA link) that explains potential applications ("future potential" mentioned in the main paper). Their kit could be thought of as a heatspreader extension in some ways, but it's more fun to think about it as a mechanical heat-turbo device imho. Where an ICE Turbo harvests the flow of waste exhaust gas to boost the flow of inlet fuel, this here gizmo would harvest waste heat from a Power Supply Module (PSM) to pull more heat out of a CPU (their example, iiuc). The PSM would then need a bigger heatsink but the CPU a smaller one -- possibly a better fit to each's physical size. And the energy transfer between the two would occur mechanically, without fluids or electricity.
The stress-induced martensitic transformations behind such elastocalorics have also been proposed for making thermal batteries that could relieve parts of the daytime AC loadings in applicable summertime scenarios (afaics) -- like a supercapacitor of cool! ;)
The art of damping out mechanical resonances is an ancient and sophisticated one. Have no fear there. Most likely the inherent hysteresis in heating a mechanical mass will damp out any process-loop type cycles in the thermal load anyway, so unlikely to raise itself as an issue.
Surely the input requires a temperature difference to extract the energy to power the heat pump part of the device. This implies a cold sink, which isn't mentioned.
To keep pumping heat is going to require some sort of repeating cycle of operation, which also isn't mentioned.
More a case of, we moved the heat from the chip to the mounting. Now, how are we going to get the heat out of the mounting? Radiative fins are never enough. I know, let's pump a cooling fluid through. Not those awful CFCs, just plain water. Oh, look, the farm already has water cooling.
So all this is really doing is creating a thermal bridge between the chip to the water. Which might actually be useful?
"Phase changes induced by heat, or actually size changes."
There's a point where there's saturation. Heating liquid water gets to a temperature where adding more energy accelerates the phase change until there's no more liquid.
A material that's changing size would tear itself apart at some point and the process would be irreversible.
I've got a project sitting on a back burner to store heat (a thermal battery) using a low-melt metal alloy so I've been learning about phase change processes. I don't think I'd hit the thermal limit, but the alloy would "come apart" at a high enough temperature so there is a limit. I'll be using water as the transport medium so I have to stay within the limits it imposes. The alloy melts at 70C and I need to keep the upper temp under 100C in the battery for safety. I couldn't keep putting energy into the battery unless I was also taking it away at the same time.
First and Second Law of thermodynamics
As recalled by Flanders and Swann
https://youtube.com/watch?v=VnbiVw_1FNs
Song at around 1'48"
Honestly, electricity-free and water-free cooling sounds like a dream for the tech industry right now. A 4K to 12K drop without feeding a single watt into the cooling unit itself is huge for a proof-of-concept.
The real challenge will be scaling these NiTi foils for massive server racks, but using the processor's own heat as the actuator is a genuinely clever piece of engineering.
This looks like "modern quack science" to me. I mean, it works, indeed. But the practical effects are so limited, the cost is so high, and the physics behind it clearly tells us that there is no way to make it at least 10 times more efficient (or to be precise, make it work so that it gives us 10 times the temperature gradient it gives us now, and it works). Because a 2K difference at device level is not much, and simply stacking layers of this will not work as a sum (so it's not that 10 layers = 20 K difference).
This is the "reproducible proof of concept", much more than those solar roadways/walkways/bikeways ever managed. And you throw it away so quickly? Is this the classical "we ignore it, and <some other country> drives it to success, and then we envy them for it and start a economic war with taxes?" Would not be the first time, happened very often in the last > 200 years.
Comparison: The first heat pipes came up around 1830, and it took a little while to reach the current state. It will probably be faster since research speed has changed a lot since 1830s.
I think you're right about the article putting this in the context of data centres – other systems with far greater effeciencies are already available to handle the aggregate heat of a data centre, though we're still short of a step to bind the thermal energy into chemical bonds to make it reusable – but I think the real point is that it's solid state, which means you can use it where you can't run fluids, at least directly. In which case it's a better heat sink.
This appears to be a "one shot" system. It moves a tiny bit of heat from one location to another. If the location is very very small this gives a moderate temperature change. As it doesn't seem to have a mechanism for repeating and doing it again, that all you get. A little cooling once.
There are two sides to this:
On the cooling side there is the chip to be cooled, and elastocaloric cooler an a heat sink stacked vertically. A cycle consists of: 1 crush the cooler to make it cold. 2 lower it onto the chip to absorb heat. 3 raise the cooler off the chip. 4 stretch the cooler to increase the temperature. 5 raise the cooler to dump heat into the heat sink. 6 lower the cooler to mid way between chip and heat sink to complete the cycle.
On the actuator side is a heat source (I will use a burning coal), a shape memory actuator and a heat sink stacked vertically. A cycle consists of: 1 The SMA starts in contact with its heat sink causing it to expand horizontally (crush cooler). 2 lower SMA to the mid point. 4 lower SMA to the coal so it shrinks horizontally (stretch cooler). 5 raise SMA to mid point. 6 raise SMA to its heat sink to restart cycle.
The good bit is that using an SMA actuator creates large forces over a short stroke which is a better match for an elastocaloric cooler than pneumatic/hydraulic/electric actuators. This gives better efficiency and reduced size/mass.
The disappointing bit is that in order to pump heat from a cold chip to a hot heat sink other heat has to flow from a hot coal to a cold heatsink. The laws of thermodynamics have not been repealed today.
Er, so while it is indeed "electricity free", that only applies if your 'cooling device crusher' is not electrically powered..
If one were to make an automatic machine for cycling this cooler, i wonder what the system efficiency compared to compressors or peltier elements would be
One of the key concepts here is a Carnot engine. A Carnot engine is a theoretical device that outputs the maximum possible mechanical energy when a given quantity of thermal energy flows through it from a hot place to a cold place. Run it in reverse and it pumps the maximum amount of thermal energy from a cold place to a hot place for a given input of mechanical energy. The efficiency of a Carnot engine depends on the hot and cold temperatures.
The paper quoted efficiency numbers for various technologies compared to a reverse Carnot engine. Peltier elements typically achieve 15% of the best possible efficiency (last resort for when you really cannot have any moving parts). IIRC the paper quoted 84% for their device and about 60% for compressing and expanding a gas.
The key feature is that shape memory alloys closely match the force / stroke length for elastocaloric cooling. Other actuators are possible but require a mechanical linkage that levers the supplied low force / long stroke into the required very high force / very short stroke. It can be done but loses energy to friction and bending of the components along with the expense of additional precision components.
Years ago I saw a video of a giga press in operation. On one level it is an impressive piece of brute force engineering. If you look more closely you will see it uses continuously variable leverage. It starts with low force with fast motion and smoothly transitions to enormous force for the last millimetres of compression. Someone thought carefully about how to achieve that and have it work reliably all day every day. The compressor in your fridge is partially powered by the expansion side of the cycle. Again impressive mechanical engineering went into making that work in a mass produced device that runs smoothly for decades. Elastocaloric cooling with a miss matched actuator would also require impressive mechanical engineering to work well. The task is simplified significantly by using a shape memory alloy actuator instead.
"Closing the curtain to generate dark." — Works for me. :)
"Instead of compressing and expanding a gas to generate cold"
Not the most scientifically literate statement. "To expel heat" might have been better.
Compressing a gas (which requires work) raises its temperature ("generates hot") which might cause heat to flow into cooler surroundings and should that compressed gas be permitted to expand (which potentially can do work) its temperature will decrease ("generating cold" :) which could allow heat from a hotter (than the gas) environment to flow into the gas.
I imagine Sadi Carnot is spinning his grave.
Curiously Dark on Discworld, just the plain ordinary dark - not your exotics like the Summoning Dark, was a separate phenomenom from light.
Also known as Electrolux Refrigeration System which I think was also used by the old kerosene fridges used in the remoter areas of AU until the ~1960s.
Absorption Refrigeration has a long history. Einstein and Szilárd patented a version using butane instead of hydrogen.
Obviously if you don't put any extra energy† into the system the hot source cannot be cooled below the temperature of cold sink - it acts as a complicated heat pipe.
If you had a large high temperature heat source you could use this tech to refrigerate your kit. eg a solar furnace on your roof.
† actually power (J/s) as the source would be continuously producing heat.
When I was a boy - a looong time ago - my folks had a 'fridge which ran on gas.
(Town gas in those days, before North sea methane.)
I never found how it worked. I was too young for such investigation.
No electricity though.
At 86 C the device is out of the commercial temperature range of 0-70. The device performance will be reducing.
Exactly how much contact pressure is needed for the cooling to worj? Plastic packages will tolerate 14 pounds per square inch. WLCSP (maily silicon) are quite fragile and will shatter.
How is this process mechanically controlled so the cool plate contacts the device without damaging the device or the pcb? Imagine a pcb with a high density of hbm4 and gpus.
These devices cool without using electricity.
Before the arrival of electricity and cooling fans, this British colonials employed Punkah Wallahs. to operate these
https://en.wikipedia.org/wiki/Punkah
https://thefridaytimes.com/28-Feb-2020/punkah-wallahs-1900
All those who've been made redundant by AI could be gainfully employed as Punkah Wallahs at Data Centers
1) You can't win.
2) You can't break even.
3) You can't get out of the game.
But... If this thing can move heat around in the way they describe, we can build a perpetual motion machine which will still run if we extract energy from it.
By the way, you wouldn't be interested in buying a nice bridge in New York, would you...?
Like the say, hell is full... oh, why belabor the obvious?
Follow the link from the article, it leads to the nature published article, and to the PDFs (several) and videos how it works in their test setup. Thermodynamics are not broken, they are applied as an integral part of the system to make it work, and there is no perpetual motion since the heat is the source of the energy to move the heat.
Since that article has been posted over four days ago: I can imagine several setups in compact flat style, like the vapour chambers, operating at an oscillation frequency of hundreds per second instead of 2 per second like in that proof of concept setup.
