Reading an Andreev qubit through its own supercurrent, with no ancilla

A theory group in Brazil and China says a superconducting qubit can be read out through a current it produces on its own. Their scheme needs no auxiliary "ancilla" qubit. According to the paper in Physical Review B (volume 114, L080502, published 5 August 2026), the oscillating supercurrent of an Andreev qubit can probe the qubit itself, "arbitrarily close to the nondestructive limit", with currently available experimental capabilities. The work was led by José Carlos Egues at the University of São Paulo and Yugui Yao at the Beijing Institute of Technology. It is a proposal backed by simulation, and no device has been measured this way yet.

01The problem: reading a qubit changes it

A projective measurement forces a qubit into one of the eigenstates of the measured observable. Any superposition of 0 and 1 is lost. Quantum nondemolition (QND) protocols get around this by moving the information elsewhere. As the authors' preprint describes it, the encoded information is transferred to a well-prepared ancilla, which is then measured. The ancilla's readout may be destructive, but the original qubit state remains intact and retrievable.

The cost is overhead. The paper's abstract lists ancilla preparations, information transfers, and ancilla measurements as an intrinsic burden on information processing. In a many-qubit device, every extra element is another thing to fabricate, control and calibrate.

02How the supercurrent carries the state

The platform is an Andreev qubit. According to the preprint's Figure 1, a gate-defined quantum dot in an InAs nanowire is tunnel-coupled to proximitized InAs/Al superconducting leads. A phase bias φ is applied across the leads. The two qubit levels are Andreev states confined in the dot region between the superconductors. The qubit is prepared in an arbitrary superposition by resonant microwave driving. That drive produces Rabi oscillations between states |0⟩ and |1⟩.

The central result concerns what happens to the supercurrent once the qubit is in a superposition. The team time-evolves the system exactly. They find that the supercurrent of a state written with the two Bloch-sphere angles has both static and dynamic contributions. As the preprint summarizes it, the time-averaged part fixes the polar angle θ, which sets the 0/1 weights. In Egues's description, the oscillation amplitude also reflects those weights, and the moment the oscillation begins encodes the relative phase. Together the two angles specify the full qubit state.

The oscillation arises from quantum interference between the eigenstates in the superposition. It is not driven by an external source. This is where the scheme departs from the familiar AC Josephson effect, which normally needs a DC voltage across the junction. Here the authors predict an ac-like Josephson effect without an applied external voltage. The qubit's own coherent evolution supplies the time dependence.

A readout device would work as follows, according to the São Paulo research foundation FAPESP. The qubit sits in a large superconducting loop. The oscillating current in that loop generates a magnetic signal. A small coil coupled to a SQUID detects the signal, so the state can be inferred with minimal perturbation.

03How this differs from dispersive and projective readout

Three approaches are worth comparing.

• Projective readout. The qubit is measured directly in some basis and collapses. Only one outcome is obtained per preparation.
• Dispersive readout. The qubit is coupled to a microwave resonator whose frequency depends on the qubit state. This is the standard route for superconducting circuits, and it is also used for Andreev spin qubits, which can be read out dispersively by coupling to a readout resonator. It is nondemolition with respect to the qubit's energy eigenbasis, and the resonator acts as the probe. Strictly, it reports populations in that basis. Reading phase information requires a rotation before measurement, which changes the state.
• Ancilla-based QND. The information is copied onto a helper qubit, as described above.

The supercurrent scheme differs in what is measured. It is a weak, continuous record of a current operator whose time dependence contains the coherence between the levels. The state information is therefore already present in a classical-looking signal.

The authors' reasoning is that coupling a SQUID weakly to the loop keeps the disturbance small. A summary of the study reports that because the qubit–SQUID coupling is very weak, the measurement's disturbance on the qubit stays small.

My own reading, not a claim from the paper, is that this is a standard weak-measurement trade-off. Weaker coupling means less backaction but also a smaller signal, so more integration time is needed against noise. "Arbitrarily close" to the nondestructive limit describes a limit, not a free lunch. The paper's claim is that, with this protocol, the limit is reachable at available coupling strengths and noise levels.

04What it could mean for error correction

The abstract frames the work as relevant to quantum error correction. Repeated correction cycles depend on measurements that extract information without erasing the encoded data. The authors argue the approach avoids ancillas altogether and that no repetitive qubit reinitialization is needed. A simpler architecture would also be a smaller one, since there would be no separate helper qubits to couple and reset. FAPESP's release states this conditionally: if experiments confirm the prediction, the method could give a simpler architecture for reading Andreev qubits.

Several caveats apply, and I would flag them for readers.

• Syndrome extraction is not state readout. Conventional codes such as the surface code measure multi-qubit parities, not the full state of one data qubit. Nothing in the available material shows this scheme extracting such parity information. It reads out a single Andreev qubit's state.
• Single-qubit state estimation has limits. A single copy of an unknown state cannot be fully characterized in general. The way the reconstruction accuracy depends on measurement time, averaging and prior knowledge of the preparation is something to check in the full paper. I did not have access to that detail.
• The platform is specialized. The proposal applies to Andreev qubits in quantum-dot Josephson junctions. It does not carry over directly to transmons or spin qubits.

05Numbers, status and next steps

Secondary coverage of the study reports simulated reconstruction accuracies of 99.52%, 99.81%, and 99.85%, with noise and feedback included. The same summary notes that the setup has not been tested experimentally. I could not confirm from the sources the exact conditions behind each figure, such as noise levels or coupling strength. These figures should be read as model-dependent until the paper's parameters are checked.

The preprint first appeared on arXiv in September 2024. It was revised in December 2025 and then published as a Letter. The authors assert that their findings should have an unprecedented impact on research with Andreev dots. That is a strong claim for a theory result.

The obvious next step is an experiment. It would need to prepare a known superposition and detect the voltage-free AC current with a SQUID. It would then compare the reconstructed state against an independent measurement. It would also have to show that the qubit's coherence survives the readout. That means checking that the SQUID and the large loop do not add dephasing channels of their own. Until that is done, the supercurrent readout remains a well-motivated route to an ancilla-free measurement, not a demonstrated one.