Deep Dive: The Thermodynamics of Silence - How to Cool Supercomputers in a Vacuum
Myth Busting: Easy to cool things down in space?
It is often one of the most common misconceptions that keeping high-powered servers cool in space (where the temperature is approximately 2.7 Kelvin) would be an effortless activity. However, in reality, it is quite the opposite of that. The vacuum of space faces new challenges for engineers and designers.
The reality of vacuums is that they are extraordinary thermal insulators. This is because there are no atoms to vibrate and transfer heat energy to one another. This would mean there is no convection or conduction in space.
Who does this affect and why?
This targets datacentres. With more and more datacentres moving to space, they need to find a way to stay cool. This is because when datacentres work, they turn every watt of power consumed by their server into heat. This leads to datacentres getting very hot - basically acting like a furnace. This is a crucial problem because as temperatures spike, chips slow down or melt - not making the datacentre reach its full capability or potential. This issue would be magnified with the rate AI is increasing all over the world as each prompt gets sent through fibre-optic cables to a datacentre.
The current solution on Earth:
However, the solution on Earth often involves the movements of fluids such as cold air through the aisles in front of racks in a datacentre, sucking out the hot air produced. The reason why colder fluids are used are because of the second law of thermodynamics stating: “Heat can never pass from a colder to warmer body without some other change, connected therewith, occurring at the same time.” Also some other solutions also following this law include water going around the datacentre transferring the heat from the datacentre via convection outwards.
The Question:
This sets up the question: As there are no particles in space not allowing convection or conduction, how do you keep these super-computers cool in the vacuum?
The maths behind this:
To understand how this works, we need to look at a few maths and physics formulas, the first of which being The Radiative Cooling Law:
P rad = ϵσA(T^4 radiator - T^4 env)
In this case, the P rad is the radiated power measured in watts. It is the total thermal energy that must be dumped into space to match the server heat loads. Engineers need to make sure P rad is positive as that means it is losing heat to space, which allows datacentres to cool down.
The next variable is ϵ which is the emissivity. The emissivity is the measurement of the thermal efficiency of the radiator’s surface material measured from 0 to 1. To maximise computing powers, engineers must keep this value as close to 1 as possible. The more the material on the super-computer degrades on the surface, the more radiation that can seep through causing the datacentre to heat itself up even more. Normally, datacentres in space have materials equipped on them that like micro-porous ceramic coatings, carbon nanotube arrays, or optical solar reflectors (OSRs) that achieve between 0.95 to 0.92.
The next variable: σ is the Stefan-Boltzmann constant which is approximately 5.67 X 10^-8 and dictates radiative transfer limits by stating the energy capacity of a photon as well as saying how long it takes for the photons to move out of the way of the satellite/datacentre.
The A is the surface area of the cooling panels. The bigger A is, the easier it is too cool down the supercomputer/satellite.
The next variable is T^4 radiator which is the temperature of the radiator or the cooling panels of the orbital datacentre. The variable is increased to the fourth power as there are three photon states which must be times by the energy per photon getting you T^4. This variable is important as engineers must use internal components to transfer heat effectively so that T radiator stays hot enough to dump heat energy quickly, but not so hot that the electronics fail. Even a tiny increase in radiator temperature exponentially supercharges the cooling capacity suggesting why it is so important to get this right.
The final variable is T^4 env which is the temperature of space. The variable can spike when the supercomputer is facing the Earth or the Sun due to radiation.
How this all comes together:
This all comes together to show how the theoretical maximum limit of how much thermal energy an object can radiate into its environment based on its temperature and size. It tells the way how datacentres in space on satellites need to be designed with the correct surface area (A).
The analogy:
A way of thinking how datacentres in space onboard satellites are is like saying a datacentre on Earth is like putting a car engine in a wind tunnel. There, the wind cools down the engine, carrying away the heat through convection.
However, in space, an orbital datacentre is like a hot thermos bottle floating in a void. This would mean the heat inside the thermos bottle cannot escape, so engineers have to manufacture and approach so that heat can be released as an invisible infrared light. The approach could consist of internal liquid loops transferring the heat to the wings where it will be released.
Current Solutions and Start-ups:
Some companies have already started countering the challenges by using specific architecture such as deployable loop heat pipes (LHPs) which use internal fluids on the orbital datacentre to pump out heat from high-density GPU’s out to the radiator fins without the external mechanical power. A company that is doing this is Starcloud.
In addition there is modular thermal tiles which integrates solar powered generation and radiative cooling treating the entire structural shell of the datacentre as a heat exchanger. This was pioneered by Sophia Space.
There is also active space heat pumps which mechanically rise radiator operating temperatures so that there is no need for the massive surface area of heavy metal wings. This is said to be deployed later in the decade.
These are all innovations that are the direct answer of the radiative cooling law showing ways to keep the internal heat down and release energy from the thermos.
The future:
This sets up the future with engineers trying to work out the perfect balance between density in a datacentre as well as surface area and mass. This is because the chips need dense, compact processing power, countered by the cooling aids which currently need square kilometres of thin delicate surface area.
The future beholds an area of engineering development to see who can win this next-generation space race.



Comments