Presentation Information
[18p-S2_202-4]Machine-Learning Classification for IQ-Mapped Multi-Qubit Measurements
〇(M2)Duanlian Zhang1, Naoya Negami1, Raisei Mizokuchi1, Shunsuke Ota1, Riku Wada1, Tetsuo Kodera1 (1.Science Tokyo)
Keywords:
IQ-encoded readout,Machine learning
Semiconductor spin qubits are a leading candidate for building large scale, fault tolerant quantum computers due to their compatibility with advanced semiconductor fabrication and their potential for dense integration. A key remaining challenge, however, is achieving reliable and scalable qubit readout. IQ encoded readout provides a promising route toward simultaneous measurement of multiple qubits by mapping distinct multi qubit states onto locations in the complex in-phase (I) - and quadrature (Q) plane. In practical experiments, the fidelity of such mapping is degraded by measurement noise and drift, where both white noise and pink noise can blur state dependent signal distributions and complicate robust discrimination.To address this issue, we apply machine learning to state classification using synthetic IQ encoded datasets that incorporate realistic noise conditions. We benchmark representative deep learning models, including convolutional neural networks (CNN) and Conformer based architectures, against a conventional threshold-based readout approach to quantify robustness across noise levels and signal variations. This comparative study clarifies when and how data driven classifiers outperform hand crafted decision rules, and it provides practical guidance for deploying machine learning assisted readout pipelines for scalable semiconductor spin qubit systems.