Short answer: Erasure conversion means arranging for most errors to be of a kind that can be detected and located, so they are easier to correct than unknown errors. Three published neutral atom experiments report doing this: two with ytterbium-171 and one with alkaline-earth atoms in a Rydberg array. They report gate and Bell state fidelities of 0.98 to above 0.997 and small logical-qubit demonstrations. All are laboratory results.
Why erasures help
A 2023 Nature paper on ytterbium-171 notes that for eventual quantum error correction it is advantageous to have qubits with structured error models, such as biased Pauli errors or conversion of errors into detectable erasures. A 2026 Nature Physics paper adds that large-scale algorithms with practical advantage will need fault tolerance through error correction, but that the overhead is prohibitive, and can be reduced by building physical qubits with fewer errors and by shaping residual errors so they are easier to correct.
A metastable ytterbium qubit, 2023
The 2023 paper demonstrates a neutral atom qubit based on the nuclear spin of a long-lived metastable state in ytterbium-171. Its long coherence time and fast excitation to the Rydberg state allow one-qubit and two-qubit gates with fidelities of 0.9990(1) and 0.980(1). A large fraction of gate errors result in decays out of the qubit subspace to the ground state. By fast mid-circuit detection of these errors, the authors convert them into erasure errors, with an induced error probability below 10^-5 on the qubits remaining in the computational space during detection.
Logical qubits with erasure information, 2026
The 2026 paper demonstrates error-correcting codes and logical qubit circuits in the same kind of metastable ytterbium-171 qubit, with a noise bias towards erasure errors. These errors can be located separately from any syndrome information, and the authors demonstrate adaptive circuit execution based on the erasure information. They report strong suppression of dephasing errors during coherent transport and entangling gates that keep a high fidelity despite gate-beam inhomogeneity or pointing errors. They encode a logical qubit in the [[4, 2, 2]] code, with error correction at decoding based on mid-circuit erasure measurements even though the code is too small to correct any Pauli errors. They also demonstrate logical qubit teleportation between several code blocks with ancillas selected conditionally on mid-circuit erasure checks.
Erasure conversion in a Rydberg array
A 2023 Nature paper says erasure conversion has been proposed for Rydberg arrays, by detecting leakage errors with single-atom resolution, but had not yet been realised for matter-based qubits in general. The authors demonstrate both erasure conversion and high-fidelity Bell state generation in a Rydberg quantum simulator, using fast imaging of alkaline-earth atoms. After excising data with observed erasure errors, they report a Bell state fidelity of at least 0.9971, improving to at least 0.9985 when correcting for remaining state-preparation errors (both with the uncertainties given in the paper). They also apply erasure conversion in a quantum simulation experiment on preparing long-range order across a quantum phase transition.
Limits
These are single-group experiments on small numbers of atoms, and the fidelities depend on each setup and on excluding detected erasures. None of the papers demonstrates a full fault-tolerant computation.
FAQ
What is an erasure error?
An error whose location is known, which makes it easier to correct than an error at an unknown place.
How are errors converted into erasures?
By fast mid-circuit detection of decays out of the qubit subspace, as in the ytterbium-171 and Rydberg experiments.
What gate fidelities were reported?
In the 2023 ytterbium paper, 0.9990(1) for one-qubit and 0.980(1) for two-qubit gates.
What was the logical qubit demonstration?
A [[4, 2, 2]] encoding with decoding that uses mid-circuit erasure measurements, plus teleportation between code blocks.
Does this mean neutral atoms are fault tolerant?
No. The papers report small demonstrations.
All field notes