A proposed microwave-cavity bus could let distant neutral-atom qubits interact without being physically moved across an array. In simulations, a new two-qubit gate reached 0.997 fidelity, while the same long-range connection shortened selected quantum error-correction rounds by factors ranging from 1.7 to 4.8.
Neutral-atom processors usually entangle nearby atoms through the Rydberg blockade, an interaction that works only over a short distance. Optical tweezers can rearrange atoms to create longer-range connections, but that motion takes time. The delay becomes important for quantum error-correcting codes whose checks connect qubits that are far apart on a flat physical array.
The University College London proposal places an optical-tweezer array inside a three-dimensional microwave cavity. Selected atoms are temporarily coupled through Rydberg states to a shared microwave field, allowing the cavity to carry an interaction across the register. A large microwave mode can be made more spatially uniform than an optical-cavity field, relaxing the need to position every atom at a narrow field maximum.
A gate designed to tolerate a warm cavity
Microwave cavities introduce a different problem: at cryogenic temperatures they can still contain thermal photons, which shift atomic transitions and disturb a gate. The researchers designed a bichromatic Raman gate that drives two excitation paths with balanced detunings. The unwanted dispersive shifts cancel to leading order, while the Rydberg states and cavity photons are populated only virtually.
For a simulated 30-gigahertz cavity with one-megahertz atom-cavity coupling, finite cavity loss and Rydberg decay, the strong-drive operating point produced a lossy two-qubit fidelity of 0.997. That value is a model result, not an experimental gate measurement. It also depends on the assumed cavity quality, mode volume, drive strengths and cryogenic environment. The paper notes that combining very high quality factors with the small mode volume and optical access needed for a tweezer array remains an engineering target.
The shared field can connect many atoms, but it does not make all operations simultaneous. The authors therefore tested a hybrid schedule: ordinary local gates and atom rearrangement handle nearby checks, while sequential cavity gates close the long links that wrap around the edge of a planar layout.
The speed-up belongs to specific code layouts
For a toric-code array with length 50 and an assumed 2.5-microsecond cavity gate, the model reduced the boundary-closure step from 1.3 milliseconds to 0.5 milliseconds. Including the rest of syndrome extraction, one full error-correction round was 2.3 times faster than an all-shuttling schedule.
The comparison was then extended to bivariate bicycle codes, which encode more logical qubits but require non-local stabilizer checks. Across the four smaller codes studied, the hybrid schedule shortened a round by factors of 2.6 to 4.8. The gain fell to 1.7 for the largest 288-qubit example because that layout needed fewer shuttling rounds and more sequential cavity operations.
The error model is also incomplete. Thermal-photon jumps can create time-correlated phase errors across several gates, while the threshold comparison in the paper assumes a stochastic-noise model. The authors identify that mismatch and do not claim fault-tolerant operation. Greater parallelism would also require multiple cavity modes, overlapping cavities or additional Rydberg levels, each adding control and fabrication demands.
Sources
- Paper record and abstract, submitted September 21, 2026
- Full paper with gate model, error analysis and code-level timing comparisons
- Paper PDF
- Lead image: Figure 1 from Kendall et al., illustrating the four-level atom-cavity schemes and array geometry, reused from the paper under CC BY 4.0.