Researchers demonstrate a much faster way to measure some atoms in a neutral‑atom quantum register during a running computation and to act on the outcomes in real time. They read out four of five atomic qubits with error rates below one percent, while keeping the unmeasured qubit’s coherence largely intact. The whole measurement‑and‑feedforward cycle time falls from the typical millisecond scale to below 100 microseconds, in some cases as low as 45 µs.

The experiment uses a one‑dimensional array of five rubidium‑87 atoms held in optical tweezers inside a high‑finesse optical cavity. The cavity is tuned near the atoms’ 780 nm D2 transition so that photons emitted by a measured atom are collected much more quickly than in free space (a phenomenon known as Purcell enhancement). To avoid disturbing the other atoms, the team shines a second, off‑resonant laser on each unmeasured atom. That laser shifts the excited‑state energies of those atoms (an AC Stark shift), moving their optical resonance away from the cavity and making them effectively invisible to the probe light. Acousto‑optic deflectors steer the probe and shielding light to chosen sites.

Readout works by sending a near‑resonant probe to one atom at a time and counting photons that come out of the cavity. The authors use an adaptive gate implemented with a field‑programmable gate array (FPGA) to stop probing as soon as two photons are detected. This adaptive gating keeps the average bright‑state detection time down to about 9.5 µs and reduces atom loss during readout from about 95% to 8% on average. With a two‑photon threshold and 40 µs maximum windows, the reported per‑atom detection infidelities averaged across the array are roughly 0.9% for the bright state and 0.2% for the dark state.

They also demonstrate real‑time feedforward based on these mid‑circuit measurements. In one test, four atoms are measured in the middle of a Ramsey sequence on a fifth atom. The measurements induce phase shifts on the unmeasured qubit; the team applies corrective microwave phases in real time and recovers Ramsey contrast. They further implement an adaptive circuit for improved quantum‑state discrimination and for preparing states conditionally on measurement results. Overall, the combination of site‑selective shielding, cavity‑enhanced collection, and fast electronics reduces the measurement‑and‑feedforward cycle to below 100 µs, an order of magnitude improvement over free‑space neutral‑atom approaches.