149
Part Three · E · Hardware, simulation and practice

Control systems, cryogenics, and scaling

Photographs of quantum computers show a golden chandelier: a stack of plates hanging inside a cylinder, tangled with coaxial cables. That chandelier is a dilution refrigerator, and the cables carry the microwave pulses that control each qubit. A million qubits would mean millions of cables running into a space colder than intergalactic space, each one leaking heat. Scaling a quantum computer is, to a large extent, a problem of plumbing, wiring and electronics.

On the plate: the heat budget of a refrigerator, stage by stage.

How a dilution refrigerator works

Below about 0.87 K, a mixture of helium-3 and helium-4 separates into two phases, one rich in helium-3 and one dilute. Helium-3 atoms crossing from the rich phase into the dilute phase absorb heat, like evaporation, and keep doing so down to the lowest temperatures because the dilute phase retains about 6.6% helium-3 even near absolute zero. A pump circulates helium-3 continuously through the mixing chamber, reaching 10 mK. The cooling power falls as T2T^2 (Chapter 18), so it is tiny where it matters most: around 20 µW at 20 mK. Above the mixing chamber sit stages at about 50 K, 4 K, 1 K (the “still”) and 100 mK (the cold plate), each intercepting heat before it reaches the stage below. Frank Pobell’s book is the standard reference. Established

The wiring bottleneck

Each coaxial cable conducts heat from warmer stages. Sebastian Krinner and colleagues at ETH Zurich measured the passive heat load of each line type in 2019. Using their numbers for a drive line, choose 100 qubits with 2.5 lines each. The plate compares heat load with cooling power at every stage. The 100 mK cold plate saturates first, at a few hundred lines; the authors concluded that their fridge could support about 150 qubits as configured, perhaps a thousand with modifications. Active heating from the pulses themselves adds more. Established

Ways out, and the decoding clock

Remedies include multiplexing many readout signals on one line, flexible superconducting ribbon cables, and putting control electronics inside the fridge: cryo-CMOS chips at 4 K, or single-flux-quantum logic that switches with picosecond pulses and dissipates little heat. Each must avoid disturbing the qubits it controls. Another clock constrains everything: error correction produces syndromes every cycle, about 1.1 µs on Google’s Willow, and the decoder must keep up. Google reported an average decoder latency of 63 µs in 2025, fine for a memory experiment but too slow if every logical gate had to wait for it. The reaction time in resource estimates (Chapter 143) is exactly this latency. Established

What you now have

How dilution refrigeration reaches millikelvin temperatures, the T² fall of cooling power, stage-by-stage heat budgets, the wiring bottleneck with measured line loads, cryo-CMOS and SFQ control, and decoder latency against cycle time.

· · ·

Three questions to carry with you.

Why does dilution cooling keep working near absolute zero?tap
The dilute phase retains about 6.6% ³He, so atoms can always cross and absorb heat.
Which stage limits wiring in Krinner’s fridge?tap
The roughly 100 mK cold plate, at a few hundred lines.
Why does decoder latency matter?tap
Logical operations that depend on measurement results must wait for decoding.

And one to work out. A 63 µs decoder latency against a 1.1 µs cycle: how many cycles behind is the decoder?

cycles ≈

Sources: F. Pobell, Matter and Methods at Low Temperatures, 3rd ed. (Springer, 2007). S. Krinner et al., EPJ Quantum Technol. 6, 2 (2019). Google Quantum AI and collaborators, Nature 638, 920 (2025). Per-line loads and stage cooling powers from Krinner et al., drive line (stainless coax). Last checked 18 November 2026.

Next · Chapter 150
Benchmarking and evaluating advantage claims

Since 2019 the industry has claimed quantum advantage many times. Classical computers have taken several of those claims back.

Turn the page