A dilution refrigerator cools a superconducting-qubit chip through cascaded stages (50 K → 4 K → still ≈0.8 K → cold plate ≈100 mK → mixing chamber ≈10 mK). Every qubit needs its own coax lines in and out (drive, flux bias, readout), and each line is a physical thermal short between stages: heat flows down it by conduction,
Q_stage = N_lines × q(material, stage) [µW per stage]
where q is the conducted heat per line, tabulated per stage and cable type below. Stainless-steel semi-rigid coax conducts far more heat than NbTi, which becomes superconducting (and a poor phonon conductor) below its ~9.2 K critical temperature — real cryostats mix both, using NbTi only for the coldest runs.
Each stage has a finite cooling power budget (its refrigerator's capacity at that temperature). If the wiring's heat load exceeds the mixing-chamber budget Q₀ at its design temperature T₀, the stage settles at a higher equilibrium temperature, since a dilution unit's cooling power scales roughly as T²:
T_MXC = T₀ · √(max(1, Q_load / Q₀))
A hotter mixing chamber means more thermal photons in each qubit's readout resonator, given by the Bose–Einstein occupation at the qubit frequency f (here 5 GHz):
n̄(T) = 1 / (exp(hf / k_BT) − 1)
Once n̄ climbs past roughly 0.005–0.05 the qubit's effective T1/T2 collapses — this is the real "wiring bottleneck" that limits how many physical qubits a single fridge can host, and why frequency-multiplexed readout (packing several qubits onto one output line) and NbTi harnesses are standard scaling techniques.
- Qubit count / lines per qubit — sets the total coax line count threading the fridge.
- Stainless steel vs. NbTi — swaps the conduction table for stages below 4 K.
- Freq-mux readout — cuts effective line count ~60%, modelling shared readout lines.