The Ireland Quantum 100 machine — qubit, controller, cryostat, room

The qubit: fixed-frequency transmons on a heavy-hex lattice

The Ireland Quantum 100 machine is built around 100 physical superconducting transmon qubits fabricated on high-resistivity silicon. The transmon is the workhorse of utility-scale superconducting quantum hardware for a reason: it sits in the regime where the Josephson energy E_J dominates the charging energy E_C by a ratio of roughly 50–80, which suppresses charge-noise sensitivity exponentially while keeping anharmonicity in the 200–350 MHz range — enough to address the 0→1 transition cleanly without leaking into the 1→2 manifold during a 30–60 ns single-qubit gate.

Each qubit is a shunted Josephson junction — a thin aluminium–aluminium-oxide–aluminium sandwich, junction area on the order of 0.01 µm², in parallel with a large coplanar shunt capacitor that fixes the qubit frequency in the 4.5–5.5 GHz band. We are deliberately choosing fixed-frequency transmons over flux-tunable variants for the first-light device. Tunable qubits give you faster two-qubit gates but they buy that speed with extra flux-noise channels and an order-of-magnitude more control wiring at the mixing-chamber stage. For a sovereign machine that has to run reliably for climate-chemistry workloads rather than chase headline gate speeds, fixed-frequency wins.

Topology is heavy-hex — the same connectivity graph that has proven itself elsewhere in the industry — chosen specifically because it suppresses frequency collisions and spectator errors during cross-resonance two-qubit gates. Each data qubit has degree two or three, with ancilla qubits sitting on the edges. That layout maps cleanly onto the surface code we intend to run for error-corrected logical qubits in the post-2027 phase, even though Ireland Quantum 100 itself is a noisy intermediate-scale (NISQ) device at first light.

Coherence targets and what they buy you

Our coherence-time targets at first multi-qubit operation are T₁ ≥ 100 µs and T₂^echo ≥ 80 µs across the median qubit, with single-qubit gate fidelities above 99.9% and two-qubit cross-resonance gate fidelities above 99.0% on calibrated pairs. These are not aspirational marketing numbers — they are the floor for running variational quantum eigensolver (VQE) circuits deep enough to be useful on small molecules.

To put that in chemistry terms: a hardware-efficient ansatz for the active space of an amine-CO₂ adduct — the kind of system that matters for post-combustion carbon capture — needs roughly 12–20 logical qubits and circuit depths of several hundred two-qubit gates after compilation onto heavy-hex. At 99.0% two-qubit fidelity you are right at the edge of getting a meaningful energy estimate before noise washes out the signal; every extra nine on that fidelity number doubles the useful circuit depth.

The cryostat: getting to 12 mK and staying there

The qubit chip lives at the mixing-chamber plate of a dry dilution refrigerator. We are specifying a base temperature below 15 mK with at least 400 µW of cooling power at 100 mK and a sample space large enough to host the full 100-qubit package with on-chip Purcell filters, readout resonators, and the associated input/output lines without thermal compromise.

The thermal stack matters as much as the qubit design. From room temperature down, the stages run at roughly 50 K, 4 K, 800 mK (still plate), 100 mK (cold plate), and finally the 10–15 mK mixing chamber. Every coaxial line into the fridge has to be attenuated and filtered at each stage so that 300 K Johnson noise does not drive the qubit. A typical drive line carries 20 dB of attenuation at the 4 K stage, another 20 dB at 100 mK, and 20–30 dB at the mixing chamber, with infrared-blocking eccosorb filters and low-pass filters before the chip. Readout output lines run through cryogenic HEMT amplifiers at 4 K, preceded by isolators or circulators and Josephson travelling-wave parametric amplifiers (TWPAs) or impedance-matched parametric amplifiers (IMPAs) at the mixing chamber for near-quantum-limited readout.

For 100 qubits with multiplexed readout you are looking at on the order of 200–300 individual coaxial lines threading the fridge. Cable management, thermalisation, and microwave hygiene at this density are the unglamorous engineering problems that decide whether the machine actually works.

Control electronics and the classical stack

Each qubit needs an XY drive (microwave at the qubit frequency, generated by IQ-modulated arbitrary waveform generators with sub-nanosecond timing resolution) and a dispersive readout pulse on its readout resonator. Two-qubit gates on fixed-frequency transmons are driven by cross-resonance — driving the control qubit at the target qubit's frequency — with active cancellation tones to suppress unwanted single-qubit terms in the effective Hamiltonian.

The control system is a room-temperature rack of FPGA-based AWGs and digitisers running a real-time pulse sequencer with feedback latency low enough to support mid-circuit measurement and conditional gates — a precondition for any future error-corrected operation. The classical-quantum interface speaks OpenQASM 3 at the gate level, with Qiskit, PennyLane, and Cirq supported as front-end SDKs. Pulse-level access via OpenPulse is exposed for users who need to do their own gate calibration or run optimal-control experiments.

The room: Clonmel, magnetic shielding, and grid posture

Quantum hardware in Ireland has to contend with the same physical environment as anywhere else: vibration, RF interference, magnetic fields from the local grid, and the thermal load of the pulse tubes that pre-cool the dilution unit. The Co. Tipperary site is being fit out with a vibration-isolated slab decoupled from the building structure, a Faraday-cage-grade RF-shielded room, and mu-metal magnetic shielding at the fridge to suppress stray fields below 1 µT at the qubit plane. The pulse-tube compressors — the loudest and most vibrationally aggressive part of the system — sit in a separate plant room with flexible helium lines damped to attenuate the 1.4 Hz pulse-tube cycle that otherwise modulates qubit frequencies.

Site power is sized for roughly 50 kW continuous draw, d