Inter-qubit Correlation of Readout Noise in Near-Term Quantum Devices

Near-term quantum devices are often prone to cross-talk errors which are relatively difficult to mitigate than local errors. In this work, we benchmark the performance of the readout error mitigation (REM) technique for a correlated noise model against a local noise model. In the former case, we consider the full calibration matrix with no assumptions about the noise present in the device, while in the latter, we choose the uncorrelated-noise calibration matrix where the cross-talk effects are ignored. While the full calibration matrix is a more general approach, it comes with an overhead of exponentially increasing run time. In contrast, the uncorrelated-noise model, under certain strict assumptions, allows far more computationally efficient construction of the calibration matrix and thus can tackle a larger number of qubits. We found that the use of the full calibration matrix does not significantly increase the performance of REM as compared to the local-noise model and therefore conclude that the effect of cross-talk is not of the same order of magnitude as the local noise error.