The cryogenics stack — dilution refrigerator class, vibration, EM shielding

Why millikelvin matters for a 100-qubit transmon system

A transmon qubit is, at its heart, a non-linear LC oscillator built from a Josephson junction shunted by a large capacitor. The transition frequency between the |0⟩ and |1⟩ computational states sits in the 4–6 GHz band, which corresponds to a thermal energy of roughly 200–300 mK if you naively equate hf with k_BT. To suppress thermal population of |1⟩ to the 10⁻³ level or better — the floor you need before single-qubit gate fidelities of 99.9% become meaningful — you want the qubit's electromagnetic environment held at least an order of magnitude below that. In practice, that means a base-plate temperature in the 10–15 mK band, with the qubit chip itself ideally thermalised to within a few mK of that floor.

This is the entire reason quantum cryogenics is not a vendor-of-convenience decision. The dilution refrigerator is not "cooling kit attached to the quantum computer". It is the computer's body. Every wire that enters it, every photon that leaks past a filter, every microvibration coupled in through the slab — all of it shows up as decoherence in T₁ and T₂ measurements on the chip.

Dilution refrigerator class and the Ireland Quantum 100 thermal budget

For a 100-physical-qubit superconducting machine on a heavy-hex topology, we are specifying a large-bore wet-style dilution refrigerator with a continuous ³He/⁴He circulation loop and a base-temperature target of ≤10 mK at the mixing chamber, with at least ~20 µW of cooling power at 20 mK. That figure is not arbitrary. It has to absorb the passive heat leak through approximately 200–300 coaxial input lines (drive plus flux plus readout in/out), the dissipation of cryogenic HEMT amplifiers stationed at the 4 K stage, and the dynamic load from microwave drive pulses arriving at the chip.

The standard staged architecture is preserved: a 50 K stage and 4 K stage cooled by a pulse-tube cryocooler, a still plate near 800 mK, a cold-plate intermediate stage near 100 mK, and the mixing chamber plate at base. Attenuators are distributed deliberately — typically 20 dB at 4 K, 10 dB at the still, and 20 dB at the mixing chamber on each drive line — so that 300 K Johnson noise is thermalised down to the photon temperature the qubit actually sees. Get this attenuation profile wrong and you can have a 10 mK plate temperature and still measure an effective qubit temperature of 60–80 mK because room-temperature noise is leaking down the coax.

Wiring, filtering, and the photon environment

For 100 qubits on heavy-hex you are looking at roughly 100 XY drive lines, ~70 flux bias lines for the tunable couplers, and shared readout feedlines multiplexed 6–8 qubits per line through Purcell-filtered resonator banks. Output paths use TWPAs (travelling-wave parametric amplifiers) at the mixing chamber feeding HEMTs at 4 K. Each input line carries a stack of infrared (IR) blackbody filters — typically eccosorb-loaded sections — to kill stray 10–100 GHz photons that conventional microwave attenuators pass straight through. Without IR blocking, quasiparticle generation in the aluminium of the Josephson junctions becomes a dominant T₁ limit.

Vibration isolation: pulse tubes are the enemy of T₁

The pulse-tube cryocooler that does the heavy lifting at 4 K is also the largest single source of mechanical noise in the system. It runs at ~1.4 Hz and injects vibration through the cold head into every plate below it. For flux-tunable transmons, mechanical vibration modulates SQUID-loop area and therefore qubit frequency — you see it directly as ramsey-fringe jitter and reduced T₂*.

The mitigation stack we are specifying for the Tipperary install is layered:

• Site-level: the cryostat sits on an isolated inertia block decoupled from the main building slab, with a target floor vibration spec below ~1 µm/s RMS in the 1–100 Hz band. Co. Tipperary helps here — there is no nearby heavy rail or motorway loading.
• Cryostat-level: bellows-decoupled pulse tube with the compressor located in a separate plant room, flexible He gas lines, and a soft-suspended cold-head mount that breaks the rigid mechanical path to the 4 K plate.
• Stage-level: spring-suspended inner can with copper-braid thermal links sized to give adequate thermal conductance without becoming mechanical short-circuits. This is always a trade-off — every braid you add to improve thermalisation adds a vibration path.
• Chip-level: the qubit package itself is mounted with damping interposers and operated, where possible, at flux sweet spots where ∂f/∂Φ → 0 so first-order flux noise (mechanical or magnetic) drops out.

Electromagnetic shielding from DC to daylight

EM hygiene is the other axis where most installations underperform their datasheet. We specify three concentric layers around the qubit package: a superconducting aluminium can (which expels residual field once cooled below T_c), a high-permeability Cryoperm or mu-metal can outside that to shield against DC and low-frequency magnetic fields, and a light-tight outer copper can. Residual magnetic field at the chip needs to be in the sub-µT range; otherwise you trap flux vortices in the superconducting films, which become two-level-system (TLS) loss centres and slash T₁.

At the room-temperature side, the cryostat sits inside a Faraday-screened bay. Mains power enters through filtered isolation transformers. Control electronics — the AWGs, the readout digitisers, the LO sources — sit in shielded racks with all signal exits via filtered feedthroughs. Ground loops are designed out, not patched out: a single star ground at the cryostat, isolated DC supplies on every flux line, and optical isolation on any digital trigger crossing into the RF domain.

Why this is buildable in Tipperary on the stated timeline

The honest engineering position is that nothing in this stack is research-grade unobtainium. Large-bore dilution refrigerators, TWPAs, cryogenic HEMTs, IR-filtered coax assemblies, and Cryoperm shielding are all commercially available with known lead times. The integration risk is in tuning — getting the attenuation profile right