Neural-Network-Assisted Bayesian Qubit Readout at the Single-Photon Level for Scalable Atomic Quantum Processors

Area of Science:

• Quantum Information Science
• Atomic Physics
• Machine Learning Applications

Background:

• Scalable quantum information processing requires efficient quantum state readout.
• Neutral atom arrays use fluorescence readout, necessitating short exposures to prevent atom loss.
• Conventional threshold discrimination fails in the single-photon regime due to overlapping state distributions.

Purpose of the Study:

• To develop a robust quantum state readout method for neutral atom arrays.
• To address the challenge of overlapping state distributions in fluorescence readout.
• To improve readout fidelity and enable extraction of quantum dynamics.

Main Methods:

• Implemented a neural-network-assisted Bayesian inference for fluorescence readout.
• Introduced a weakly anchored Bayesian scheme requiring calibration of only one state.
• Utilized a permutation-invariant neural network for a 100-fold speedup in Bayesian inference.

Main Results:

• Achieved relative readout fidelity above 99% for 61% histogram overlap and 98% for 72% overlap.
• Demonstrated reliable extraction of Rabi oscillations and Ramsey interference.
• Results surpass the capabilities of conventional threshold-based readout methods.

Conclusions:

• The neural-network-assisted Bayesian method significantly enhances quantum state readout fidelity in neutral atom arrays.
• This approach overcomes fundamental limitations of threshold discrimination in low photon count regimes.
• The developed technique is crucial for advancing scalable quantum information processing.