They Broke a 165-Year-Old Law of Physics and Made Heat Programmable

In 1860 a rule was written down that said heat must be fair: whatever a surface is good at absorbing, it must be equally good at giving off. A team in Osaka just built a surface that breaks it, steers warmth in one direction, and remembers the setting with the power switched off.

Gustav Kirchhoff wrote down a rule in 1860 that was so obviously, boringly true that almost nobody has bothered to challenge it since.

It says heat has to be fair. Whatever a surface is good at absorbing, it is exactly as good at emitting, at the same wavelength, in the same direction. Warmth in, warmth out, symmetrical. You cannot build a one-way street for heat.

A team in Osaka has built a one-way street for heat.

The technical name for the rule is reciprocity, and for 165 years it has quietly set the ceiling on what engineers can do with warmth.

Think about a black surface in sunlight. It is good at soaking up heat. Kirchhoff's law says it is therefore also good at radiating heat away, and it will do so in the same directions it was willing to absorb from. That is a package deal. You do not get to choose.

Which means every thermal engineer who has ever wanted a surface that greedily drinks heat from one side and dumps it out of another has been told, politely, by physics: no. Absorption and emission are chained together. Break the chain and you could steer heat. Nobody could break the chain.

The Osaka-led team, publishing in Laser & Photonics Reviews with coverage landing on 7 to 8 July, broke it by combining two materials that have no business being in the same sentence.

Put them together as a patterned surface, a "metasurface," and you get a device that does three things nothing else does at once.

That last property is why the researchers reach for the word "programmable." A surface that can be set, that stays set, and that requires no energy to maintain its state is not behaving like a material. It is behaving like memory. Heat that stores a setting the way a chip stores a bit.

Here is where an abstract result collides with the least glamorous crisis in technology.

The bottleneck for artificial intelligence is no longer clever algorithms, and increasingly it is not even chips. It is heat. A modern AI data centre takes in enormous quantities of electricity and converts almost all of it, eventually, into waste warmth that has to be dragged out of the building before the machines cook themselves. Cooling now eats a serious fraction of the power bill and dictates where data centres can physically be built, which is why they cluster near cold air, cold water and cheap electricity.

Every one of those cooling systems is fighting Kirchhoff's law. They are trying to move heat in a preferred direction using surfaces that, by the rules, are not allowed to prefer a direction. So we compensate with fans, pumps, water and enormous quantities of energy.

Now imagine a surface that simply prefers to radiate heat outward, away from the chip, and not back at it. Or a roof that dumps warmth to the cold sky in summer and refuses to in winter, and holds that setting without being powered. That is what "steering heat" means in practice, and it is the difference between fighting thermodynamics and cooperating with it.

There is a second reason to care, and it is the one the researchers keep returning to.

Almost every clever thermal system we build has an appetite. A fan needs power. A pump needs power. An actively switched material usually needs to be held in its switched state, which means paying an energy toll for as long as you want the effect to persist. Over the lifetime of a building or a data centre, that toll is the whole cost.

This surface does not have one. Because GST holds its state, you spend energy once to set the configuration and then nothing, forever, to keep it. A roof that has been told to dump heat to the cold night sky simply keeps doing so, with no supply, no controller and no standby draw.

That is a genuinely unusual property for a thermal device, and it is the difference between a clever trick and something an engineer would actually put on a building.

This is a laboratory result, and we are going to be blunt about how early it is.

So do not expect a cooler data centre this year, or this decade, on the strength of this. What you should take away is narrower and more interesting: a constraint everyone assumed was permanent turns out to be negotiable.

It is a rule from 1860 stating that a surface must emit heat exactly as well as it absorbs it, at the same wavelength and in the same direction. This property is called reciprocity. It means absorption and emission are locked together, so engineers cannot design a surface that takes heat in from one direction and releases it in another. The Osaka team's device breaks that link.

The team combined a magneto-optical material, which behaves differently depending on the direction heat travels, with GST, the phase-change compound used in rewritable discs, which holds whichever state you set it to. The result is a surface that can steer heat in a chosen direction, be switched on and off, and retain its setting with no power at all. Because it stores a state without energy, it behaves less like a material and more like memory.

Not yet, and possibly not for a long time. The work is an early-stage laboratory result, and the applications, including thermal management, radiative cooling and infrared sensors, are proposed rather than demonstrated at scale. Its relevance is that cooling has become one of the largest costs and constraints in AI computing, and every cooling system today works against reciprocity. A surface that can steer heat instead of fighting it attacks that problem at the root.

Related from Frontier Signal: this week's deep dive on the free download that drives 20 different robots. Frontier Signal explains frontier technology in plain English.