Multi-bit state discrimination using machine learning for IQ -encoded spin-qubit readout

Abstract Semiconductor spin qubits are compatible with CMOS processes and support high integration density, making them promising for scalable quantum computing. Yet multiplexed readout remains a bottleneck. Frequency multiplexing is bandwidth-limited, while time multiplexing introduces sequential latency and extra power. We adopt in-phase ( I )–quadrature ( Q ) encoding, mapping readout signals to the complex ( I – Q ) plane to enable parallel multi-qubit discrimination and compact, digitally processable representations, compatible with both resonator-based and inductor-less electrical readout. We develop machine-learning classifiers trained on synthesized IQ datasets that emulate spin-qubit measurements with noise, drift, and varied constellation layouts. Across wide signal-to-noise ratios, the models outperform conventional thresholding and sustain high accuracy, demonstrating robust state discrimination under realistic conditions. By improving qubit-state decoding in a digitally scalable form, this work supports larger spin-qubit arrays and more efficient cryogenic readout architectures.