Photodiode readout could identify neutral-atom qubits within 100 microseconds

Measuring light against a reference

The proposed scheme distinguishes two internal states of an atom held in an optical tweezer, a tightly focused trapping beam. A detection laser makes one state fluoresce while the other remains dark. Rather than counting fluorescence photons directly, the researchers mix the collected light with a much stronger reference beam, called a local oscillator.

A beam splitter sends the combined light to two photodiodes. Subtracting their electrical signals produces a measurement sensitive to a selected component of the fluorescence field, determined by its phase relative to the reference. This technique is called balanced homodyne detection.

The reference and detection beams share a laser source, helping stabilize their relative phase. The approach uses fast differential photodiode measurements instead of serial camera readout. It does not eliminate the time needed to collect fluorescence, however, and the paper does not demonstrate a complete experimental readout system’s end-to-end latency.

Why waiting longer can hurt

The central difficulty is that the atom moves. As it oscillates in its trap, the fluorescence phase changes relative to the reference beam. Contributions collected at different moments can partially cancel, reducing the accumulated signal even while the atom continues emitting light.

Scattering also gives the atom momentum kicks. These photon recoils change its oscillation amplitude and phase, so different measurement attempts follow different trajectories. The result is both a smaller average signal and extra variability beyond the measurement’s shot noise.

The researchers derive an analytical model linking these effects to the signal-to-noise ratio. Initially, collecting light for longer improves discrimination between bright and dark states. Eventually, recoil-driven motion weakens the useful signal and broadens its distribution enough that errors increase again. There is therefore an optimal measurement duration rather than an unlimited benefit from waiting.

The model provides a striking length scale: for a detection wavelength of 780 nanometres, an axial oscillation amplitude near 149 nanometres can make the harmonic-motion-averaged signal vanish. Larger amplitudes can produce contributions with the opposite sign, helping explain a tail of weak signals from otherwise bright atoms.

Testing the proposal in simulations

The team compares its analytical description with three-dimensional Monte Carlo simulations of individual rubidium-87 atoms. These simulations sample thermal starting positions and velocities, then follow random scattering events and motion through the trapping potentials.

The initial temperature is 30 microkelvin. For the headline prediction, the optical trap has an energy depth equivalent to 7 millikelvin, and the overall detection efficiency is approximately 7.5%. That efficiency combines fluorescence collection, optical mode matching with the reference beam, and photodiode conversion efficiency.

Under optimized detection conditions, the simulations predict fidelity above 99% within 100 microseconds. Here, fidelity means one minus the average probability of confusing the two states, giving bright-state and dark-state errors equal weight. It is not a measure of overall quantum-computer performance.

The researchers also model tightly focused detection beams intended to address individual sites. Tight focusing introduces unwanted polarization components that can move population outside the desired fluorescence cycle. Nevertheless, with other parameters optimized, simulated beam radii from 0.6 to 3.0 micrometres retain the above-99% prediction in the 7-millikelvin trap.

What remains unproven

The agreement between the analytical model and the more detailed simulations supports the proposed explanation of motion-limited readout. It is not independent experimental validation. The calculation treats atomic motion semiclassically and makes assumptions about the trapping potentials and optical detection conditions.

Atom retention is also distinct from correct state identification. No losses occurred among 50,000 simulated trajectories for trap depths of at least 3 millikelvin under the reported 200-microsecond conditions, but a finite sample cannot establish zero loss probability.

The practical opportunity is to reduce camera-related delay without sacrificing comparable discrimination accuracy. Demonstrating that advantage will require achieving the assumed optical collection and mode overlap while maintaining phase-sensitive detection. Extending the approach to large arrays would additionally require multichannel balanced receivers; the present study models single-atom readout rather than demonstrating that parallel system.