Single-shot readout of charge and spin states by charge sensors such as quantum point contacts and quantum dots are essential technologies for the operation of semiconductor spin qubits. The fidelity of the single-shot readout depends both on experimental conditions such as signal-to-noise ratio, system temperature, and numerical parameters such as threshold values. Accurate charge sensing schemes that are robust under noisy environments are indispensable for developing a scalable fault-tolerant quantum computation architecture. In this study, we present a novel single-shot readout classification method that is robust to noises using a deep neural network (DNN). Importantly, the DNN classifier is automatically configured for spin-up and spin-down traces in any noise environment by tuning the trainable parameters using the datasets of charge transition signals experimentally obtained at a charging line. Moreover, we verify that our DNN classification is robust under noisy environment in comparison to the two conventional classification methods used for charge and spin state measurements in various quantum dot experiments.

通过电荷传感器（例如量子点接触和量子点）对电荷和自旋状态的单次读出是半导体自旋量子位操作的基本技术。单次读数的保真度取决于实验条件（如信噪比、系统温度）和数值参数（如阈值）。在嘈杂环境下稳健的准确电荷传感方案对于开发可扩展的容错量子计算架构是必不可少的。在这项研究中，我们提出了一种新颖的单次读出分类方法，该方法使用深度神经网络 (DNN) 对噪声具有鲁棒性。重要的，通过使用在充电线上实验获得的电荷转移信号数据集调整可训练参数，DNN 分类器自动配置为在任何噪声环境中的自旋向上和自旋向下轨迹。此外，与在各种量子点实验中用于电荷和自旋态测量的两种传统分类方法相比，我们验证了我们的 DNN 分类在嘈杂环境下是稳健的。