Cryogenics

Dilution Refrigerator

Helium-3/Helium-4 phase mixing reaches the coldest temperatures in the universe

A dilution refrigerator is a cryogenic device that cools to 5–10 millikelvin (0.005–0.010 K) — colder than interstellar space (~2.7 K) by a factor of ~300. It exploits the Bose-Fermi-mixture phase transition between helium-3 and helium-4 below 870 mK: He-3 atoms "dissolve" from a concentrated phase into a dilute phase, absorbing heat in the process — the only continuous cooling method that works below ~250 mK. A typical "fridge" has 5 nested vacuum stages: 50 K, 4 K, still (~700 mK), cold plate (~50 mK), and mixing chamber (10 mK). IBM Quantum, Google Quantum AI, IQM, Quantinuum, and Rigetti all rely on dilution refrigerators to keep superconducting qubits coherent.

  • Base temperature5–10 mK (record: ~1.75 mK)
  • Cooling power at 100 mK~400 µW (lab), ~10 mW (BlueFors LD400)
  • Stages50 K, 4 K, still, CP, MC
  • CycleContinuous (unlike adiabatic demag)
  • He-3 cost~$3,000/L (rare isotope)
  • Cooldown time12–48 hours

Interactive visualization

Press play, or step through manually. The visualization is yours to drive — try it before reading on.

Open visualization fullscreen ↗

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Why dilution fridges matter

  • Every superconducting quantum computer. IBM Quantum's Condor (1,121 qubits, 2023), Google's Sycamore (54 qubits) and Willow (105 qubits, 2024), Rigetti, IQM, Quantinuum — all live inside dilution fridges at 10–20 mK.
  • Dark matter detectors. SuperCDMS, EDELWEISS, and CRESST hunt weakly-interacting massive particles using millikelvin bolometers; thermal noise floors below 10 mK are mandatory to detect single-keV recoils.
  • Low-temperature condensed matter physics. The fractional quantum Hall effect, heavy-fermion superconductivity, and unconventional pairing states only manifest below 100 mK and are mapped using dilution-cooled samples.
  • Quantum Hall metrology. The von Klitzing constant resistance standard is traceable to roughly 1 part in 1e10 using quantum Hall plateaus measured at 100 mK.
  • Microkelvin physics platforms. Helsinki's ROTA-3, the Aalto LTL, and Lancaster's ULT facility extend dilution to microkelvin via additional adiabatic nuclear demagnetization stages.
  • Single-photon calorimetry. Transition-edge sensors and microwave kinetic-inductance detectors used in CMB telescopes (Simons Observatory, CMB-S4) ride on dilution fridges or sub-Kelvin sorption coolers.
  • Foundational tests. Casimir-force measurements, levitated-nanosphere quantum experiments, and dark-photon searches all depend on millikelvin environments.

Common misconceptions

  • "Absolute zero." No fridge reaches 0 K. The third law of thermodynamics forbids it. Best continuous dilution: ~1.75 mK; best demagnetization: nuclear-spin temperatures of microkelvin and below, but only single-shot.
  • "Any fridge can do it." Only He-3/He-4 dilution and adiabatic nuclear demagnetization reach below 10 mK continuously. Pulse-tube and Gifford-McMahon cryocoolers stop at about 2.5 K. He-3 sorption fridges reach 250 mK but only single-shot for hours.
  • "The qubit chip is at 10 mK in air." No — the inside of a dilution fridge is held at ultra-high vacuum (10⁻⁶ to 10⁻⁹ mbar). Air would freeze solid; even residual gas would convect heat from warmer stages.
  • "It's like a kitchen fridge but colder." A kitchen fridge uses single-phase compression-evaporation of a refrigerant. Dilution fridges exploit a quantum phase separation that has no analogue at room temperature. There's no compressor on the cold side — the work is done by gas-handling pumps at room temperature.
  • "He-3 is just rare He." He-3 is a separate isotope, distinct from common He-4. Roughly 1 atom in a million in natural helium is He-3; modern supply comes from tritium decay in nuclear weapons stockpiles, costing about 3,000 USD per liquid liter and constituting a strategic supply concern.
  • "Bigger fridge = colder fridge." Bigger fridge means more cooling power (more wattage at a given temperature) and more sample space, but the base temperature is set by parasitic heat leaks vs T²-scaling cooling power, not by overall size.
  • "Cooling is instant." Cooldown from 300 K to 10 mK takes 12–48 hours: pulse tube down to 4 K (12 h), He-3/He-4 mixture condensation (4–8 h), and circulation initiation to base (4–12 h). Warm-up to swap a sample takes a similar amount of time.

Frequently asked questions

Why does He-3/He-4 mixing cool?

Below 870 mK liquid helium separates into two phases: a concentrated phase that's nearly pure He-3 floating on top of a dilute phase that's about 6.6% He-3 in superfluid He-4. He-3 atoms have higher entropy in the dilute phase. When a He-3 atom crosses the phase boundary into the dilute side, it absorbs heat (analogous to evaporation) — except the He-3 'evaporates' into a quantum-mechanically distinct phase rather than into vapor. The cooling power scales as roughly T-squared, but it never freezes out, allowing continuous operation.

Why is 870 mK the magic number?

Above 870 mK He-3 and He-4 mix freely. Below 870 mK they undergo a Bose-Fermi mixture phase separation: He-3 (a fermion) and He-4 (a boson) settle into thermodynamically distinct phases. The exact temperature is set by the He-4 superfluid transition coupled to fermion-boson interactions. The 6.6% finite solubility of He-3 in He-4 even at absolute zero is what makes continuous dilution cooling possible — a normal evaporator stops working when the vapor pressure goes to zero.

What stops the fridge from going colder?

Three things. First, the He-3 circulation rate: more circulation means more cooling power but also more viscous heating in narrow capillaries. Second, residual heat leak from the still and warmer stages via thermal conduction in support struts and signal lines. Third, the ground state: at the lowest temperatures, the cooling power drops as T-squared, while parasitic heat leaks are roughly constant — they balance at the base temperature. Adiabatic nuclear demagnetization can punch below 1 mK but only for short single-shot experiments.

Why must qubits be this cold?

Superconducting qubits operate at gigahertz frequencies (typically 4 to 8 GHz). The Boltzmann temperature equivalent of 5 GHz is roughly 240 mK. To keep thermal photons from spuriously exciting qubits, you need k_B times T much smaller than h times f, which means below about 50 mK to keep thermal occupation under 1%. Quantum coherence times also degrade with thermal noise. Topological qubits and some trapped-ion modalities can run warmer (around 1 K and around 4 K respectively), but every superconducting platform needs millikelvin.

What pumps the helium?

A room-temperature gas-handling system. He-3 vapor evaporated from the still at around 700 mK is sucked away by an external turbomolecular and roots pump combination, compressed, cleaned through a liquid-nitrogen cold trap, and re-injected into the impedance line where it precools through heat exchangers before re-condensing. A modern dry fridge (like BlueFors LD400) uses a pulse tube cryocooler for the 50 K and 4 K stages, eliminating the need for a separate liquid-helium bath.

How do signal lines enter without warming the fridge?

Each microwave coax or DC line is anchored thermally to every plate (50 K, 4 K, still, CP, MC), with attenuators and filters strategically placed. A typical setup uses 20 dB attenuators at 4 K and 20 dB at MC for input lines to throw away thermal photons. Output lines use HEMT amplifiers at 4 K and isolators (circulators) at MC. NbTi superconducting coax minimizes conduction. A 1000-qubit system needs thousands of lines and significant rethinking — cryogenic CMOS multiplexing and photonic links are active research.