Electromagnetic shielding for superconducting qubits

A transmon qubit is, electrically, a very sensitive antenna pretending to be a computer. Anything that couples energy into it that you didn't put there yourself shows up as decoherence, leakage, or just unexplained drift in your calibration data. The hardest part of building a 100-qubit machine in Tipperary isn't the dilution fridge or the control electronics — those you can buy. The hardest part is convincing a few cubic metres of space, deep underground in a country with two major broadcast towers and an aggressive 4G/5G estate, to behave as if the twentieth century had never happened. This article is about how that's actually done.

What "noise" actually means at 15 millikelvin

When people say a qubit is "noisy", they usually picture audible hiss. The reality is more specific. A transmon sits at the bottom of a dilution refrigerator at sub-15 mK, with a transition frequency typically between 4 and 8 GHz. Anything in that band — or anything that can mix down into that band through a non-linearity in the readout chain — is a problem. Below the transition frequency, low-frequency magnetic fields shift the qubit through flux noise on the SQUID loop. Above it, microwave photons from the room can be absorbed directly. And then there's a third category nobody likes talking about: ionising radiation, which we'll come back to.

The practical consequence is that the shielding problem splits into three regimes. DC and low-frequency magnetic fields, where you need high-permeability material. RF and microwave fields, where you need high-conductivity material and well-designed seams. And quasi-particle generation from cosmic rays and ambient gammas, which needs mass and depth. No single layer solves all three.

Magnetic shielding: why mu-metal is non-negotiable

Mu-metal is a nickel-iron alloy with a relative permeability that, when properly annealed, sits in the tens to hundreds of thousands. The trick is that you can't just bend it into a can and expect performance — any cold work, drilling, or sharp deformation collapses the permeability locally. Real mu metal shielding for a quantum stack means parts that are formed first and then hydrogen-annealed at high temperature in their final geometry. Once annealed, you handle them as if they're crystal.

Around the mixing chamber plate you typically see nested mu-metal cans, sometimes with a cryoperm layer (a similar alloy with permeability that holds up better at cryogenic temperatures — ordinary mu-metal loses some of its magic below about 10 K). The geometry matters: open ends, seams, and cable feedthroughs are leakage paths. A factor-of-1000 attenuation on paper becomes a factor-of-50 in practice if your seam alignment is poor or your lid sits proud by half a millimetre.

Outside the fridge, the building itself contributes. Reinforcing steel, lift motors, HVAC fans, and the Earth's own field as you walk past the cryostat all produce slow magnetic transients. Some facilities go as far as Helmholtz coils around the lab to actively null the ambient field; for a 100-qubit machine, passive cancellation through careful site selection and material choice is usually sufficient, with active trim only on specific axes.

RF and microwave: the conductive cage

For EM shielding quantum work above a few megahertz, you stop caring about permeability and start caring about conductivity and skin depth. The standard pattern is a Faraday cage at the room level — typically copper mesh or solid copper sheet bonded carefully at every seam — and then nested superconducting cans (aluminium or niobium) on the cold stages inside the fridge. Aluminium goes superconducting at 1.2 K, niobium at 9.2 K, and once they do, they expel magnetic flux (the Meissner effect) and present essentially zero resistance to induced currents. That makes them excellent at both pinning the local magnetic environment and absorbing stray microwave photons.

The detail that catches people out is field-cooling. If you cool a superconducting can through its transition while a magnetic field is present, the field gets trapped inside as flux vortices, and now your "shield" is a permanent local source of noise. The procedure is to degauss thoroughly, then cool — and to design your warm magnetic shielding well enough that the residual field at the superconducting layer is below the trapping threshold of the material.

Cable penetrations are the other classic failure mode. Every coaxial line going into the fridge is a potential antenna. The standard mitigation is a stack of attenuators at successive temperature stages — you're not just thermalising the line, you're also reducing the room-temperature blackbody photon flux that would otherwise reach the qubit. Eccosorb filters in the readout chain absorb infrared photons that conventional low-pass filters miss. None of this is exotic; it's just unforgiving if you skip a stage.

The facility layer: what gets missed when you only think about the fridge

This is where a lot of academic groups quietly suffer. A beautifully shielded cryostat sitting in a building with poorly grounded three-phase power, a variable-speed drive on the chiller, and a Wi-Fi access point ten metres away will give you noisy data and you won't know why. Quantum facility EMI is a building-services problem as much as a physics problem.

The non-negotiables we're designing around for the Tipperary site:

• A dedicated single-point ground for the cryostat and its rack, isolated from the building's general protective earth, with no ground loops back through the control room.
• Variable-frequency drives — on chillers, compressors, water pumps — placed in a separate plant room with their own filtered supply, never sharing a feeder with the qubit electronics.
• Lighting on DC or filtered LED drivers; standard switched-mode LED lighting is a surprisingly aggressive broadband noise source.
• No wireless inside the lab. Wired comms only. Phones in a Faraday pouch at the door.
• HVAC fans selected for low magnetic signature, soft-mounted, and ducted away from the cryostat axis.
• UPS topology that doesn't inject switching transients into the qubit-side supply — typically a double-conversion online UPS with a downstream isolation transformer.

The site survey before any concrete is poured is the most valuable week of work in the whole project. You walk the building with a fluxgate magnetometer and a spectrum analyser, you map the diurnal variation, and you find the lift, the railway, and the neighbour's welder before they find you. For more on how this fits into the wider build, the Ireland Quantum 100 programme overview sets out the timeline.

Cosmic rays, gammas, and the limits of shielding

This is the regime nobody could solve until recently because nobody had measured it cleanly. Ionising radiation — cosmic ray muons, ambient gammas from concrete and granite, neutrons from the atmosphere — deposits energy in the silicon substrate of the qubit chip. That energy breaks Cooper pairs into quasi-particles, which then tunnel across the Josephson junction and cause correlated errors across multiple qubits at once. Correlated errors are the worst kind for surface-code error correction, because the code assumes errors are local and independent.

You can't shield muons with mu-metal. You shield them with mass — concrete, lead, or depth underground. Below-grade siting helps; lead-lined enclosures around the chip stage help; phonon traps and gap engineering on the chip itself help. The full mitigation stack is still an open research problem, and for a first-generation 100-qubit machine the honest answer is that we'll measure the error budget contribution carefully and design the next iteration around what we find. If you want to go deeper on the trade-offs between passive shielding, gap engineering, and code-level mitigation, the piece on error correction at scale covers the surface-code side.

How shielding shows up in the calibration data

The way you know your superconducting shielding is working isn't by looking at the shielding — it's by looking at the qubits. T1 (energy relaxation) and T2 (dephasing) times are the headline numbers, but the more diagnostic signal is the variance of those numbers across a long calibration run. A well-shielded qubit has T1 that drifts slowly and predictably with fridge temperature; a poorly shielded one shows step changes correlated with whatever was happening in the building at the time.

You also watch the readout resonator for unexpected sidebands, the qubit spectrum for two-level-system defects activating under stray fields, and — increasingly — the rate of correlated error events across spatially separated qubits. That last metric is the cleanest probe for radiation-induced errors and is what the field is converging on as the right test for the bottom of the noise stack.

Where to start

If you're scoping a quantum facility this week, do the site survey before you do anything else. Borrow or rent a fluxgate magnetometer and a spectrum analyser covering DC to 10 GHz, walk the candidate room over a full 24-hour cycle, and log everything. Identify the ten worst noise sources in the building and decide which ones you can move, which you can shield, and which mean you need a different room. The cryostat vendor will sell you a fridge that works on paper; whether it works in your actual building is a question only the survey can answer, and answering it late is what kills timelines.