Efficient measurement of neutral-atom qubits with matched filters

TL;DR

The paper tackles the challenge of high-fidelity readout in tightly spaced neutral-atom qubit arrays by introducing two simple, scalable matched-filter algorithms, MF-Site and MF-Array, that significantly reduce readout infidelity due to crosstalk while vastly lowering computational requirements compared with CNNs. MF-Site learns per-qubit pixel weights; MF-Array extends this by incorporating neighboring-qubit information to mitigate crosstalk, both trained via linear least-squares regression. Empirically, they achieve up to 32% and 43% infidelity reductions relative to Gaussian readout and maintain competitive performance with far fewer trainable parameters and multiplications; MF-Array generally performs best at longer readout times. The approach offers physical interpretability of learned weights and supports on-chip, real-time deployment, suggesting a practical path to scalable, high-fidelity readout for large neutral-atom quantum processors.

Abstract

Quantum computers require high-fidelity measurement of many qubits to achieve a quantum advantage. Traditional approaches suffer from readout crosstalk for a neutral-atom quantum processor with a tightly spaced array. Although classical machine learning algorithms based on convolutional neural networks can improve fidelity, they are computationally expensive, making it difficult to scale them to large qubit counts. We present two simpler and scalable machine learning algorithms that realize matched filters for the readout problem. One is a local model that focuses on a single qubit, and the other uses information from neighboring qubits in the array to prevent crosstalk among the qubits. We demonstrate error reductions of up to 32% and 43% for the site and array models, respectively, compared to a conventional Gaussian threshold approach. Additionally, our array model uses two orders of magnitude fewer trainable parameters and four orders of magnitude fewer multiplications and nonlinear function evaluations than a recent convolutional neural network approach, with only a minor (3.5%) increase in error across different readout times. Another strength of our approach is its physical interpretability: the learned filter can be visualized to provide insights into experimental imperfections. We also show that a convolutional neural network model for improved can be pruned to have 70x and 4000x fewer parameters, respectively, while maintaining similar errors. Our work shows that simple machine learning approaches can achieve high-fidelity qubit measurements while remaining scalable to systems with larger qubit counts.

Figures (6)

• Figure 1: Matched filter designs for a single qubit with boundary size $s=6$. (a) MF-Site configuration. (b) MF-Array configuration.
• Figure 2: The measurement infidelity from different analysis methods at different readout times. The square, Gaussian, CNN (site) and CNN (array) infidelity values are taken from Ref. Phuttitarn_2024.
• Figure 3: Classification infidelity for each qubit (numbered in top left of each box) for a measurement time of 36 ms. The standard error for the matched filter values is below 0.0001 for each site.
• Figure 4: False-3D images of the learned weights for each qubit.
• Figure 5: Weights of neighbor qubits (numbered in the top left of each box) for the MF-Array model.