TUM team's Mamba-based qubit readout cuts logical error rate up to 26% in simulation
Researchers at the Technical University of Munich (Luca Otting, Xiaorang Guo, Emmanouil Giortamis, Benjamin Lienhard, Pramod Bhatotia and Martin Schulz) describe a multi-stage discriminator for superconducting qubit readout built on the Mamba sequence model. It combines qubit trace classification and relaxation-error correction in one pipeline, cuts model parameter size by 49.6% versus existing methods, and holds a readout fidelity of 0.893 at a 500 nanosecond trace length. The authors report a geometric mean readout fidelity of 0.906 for the lightweight model and 0.911 for the optimized model, and a logical error rate up to 26% lower at the 500 ns readout duration. The work is posted on arXiv (2607.11668v1) and is a preprint, not yet peer reviewed.