Qubit Count and the Cooling Curve
As processors grow from tens to millions of qubits, the wiring, control, and heat loads scale in ways that push against the limited cooling power of dilution refrigerators.
Cooling is a budget
A dilution refrigerator delivers only so much cooling power at each temperature: roughly hundreds of microwatts at 100 millikelvin and a few watts at 4 kelvin for a large system. Everything the processor and its wiring dissipate must fit inside that budget. As qubit count rises, the demands on the budget rise, and understanding how they scale is essential to planning large machines.
What scales with qubit number
Several loads grow with the number of qubits. Each qubit typically needs multiple control and readout lines, so the conducted heat down the wiring grows roughly linearly with qubit count. Cold attenuators and amplifiers add dissipation proportional to the number of signal chains. Cryo-CMOS control, if used, adds a per-qubit power draw at the cold stages. The aggregate can outgrow a single fridge.
Approaches to bend the curve
Several strategies reduce load per qubit. Frequency multiplexing lets many qubits share one readout line. Cryo-CMOS moves control local to the qubits and cuts the number of wires crossing stages. Larger and multiple refrigerators, and clustering processors into networked modules, spread the load. Each approach trades complexity, latency, or added cold dissipation against wiring count.
The frontier
Fault-tolerant computing may require on the order of a million physical qubits. No single dilution refrigerator built today cools that many with conventional wiring. The path forward combines cold control electronics, aggressive multiplexing, and modular architectures. Every one of these depends on a steady supply of helium-3 for the mixing-chamber circuits, which is why cooling capacity and isotope supply are linked concerns for the field.
- Cooling power per stage is fixed and small
- Wiring and control loads grow with qubit count
- Multiplexing and cryo-CMOS reduce load per qubit
- Fault tolerance may need around a million physical qubits