Heralded Leakage Detection with Preserved Computational-State Coherence in a Fixed-Frequency Transmon

Superconducting quantum processors often suffer from leakage, where the qubit population escapes the computational subspace (the ground and first excited states) into higher energy levels. Because standard dispersive readout collapses all states indiscriminately, it is difficult to detect this leakage without destroying the encoded quantum information. The authors seek a way to flag leakage states while keeping the computational subspace intact and coherent.

The researchers apply a continuous Rabi drive to the qubit's computational transition during the dispersive readout process. This drive causes the ground and first excited states to oscillate rapidly, effectively time-averaging their dispersive shifts and making them appear identical to the readout resonator. In contrast, the second excited state (the leakage state) remains distinct and detectable. By carefully tuning the drive frequency and power, the team suppresses measurement-induced dephasing, allowing them to distinguish between the computational subspace and the leakage state.

The experiment achieved a leakage-detection fidelity of 97.1% within an 80-ns window. The false-flag rate—where a computational state is incorrectly identified as leakage—was 2.3%. When testing the protocol on a mixed state containing both computational and leakage populations, the team successfully performed post-selection, retaining an average state fidelity of 92.9% for the computational subspace. The protocol requires no additional hardware, making it highly compatible with existing superconducting quantum processor designs.

Leakage is a significant error mechanism that can persist through multiple error-correction cycles and generate correlated errors that degrade code performance. This heralded detection scheme provides a hardware-efficient way to flag these errors, enabling decoders to perform post-selective removal of leakage. This capability is particularly valuable for improving the performance of surface codes and is a promising tool for emerging erasure-qubit architectures.