Efficient reconstruction, benchmarking and validation of cross-talk models in readout noise in near-term quantum devices

TL;DR

This paper introduces an efficient quantum detector tomography protocol to characterize and model cross-talk effects in readout noise of large-scale quantum devices, improving error mitigation accuracy.

Contribution

The authors develop Quantum Detector Overlapping Tomography, enabling scalable characterization of cross-talk in readout noise with reduced sample complexity.

Findings

Successfully applied to 79-qubit Rigetti and 127-qubit IBM devices.

Achieved approximately 20% improvement in energy estimation accuracy.

Demonstrated the utility of correlated noise models over simpler ones.

Abstract

Readout errors contribute significantly to the overall noise affecting present-day quantum computers. However, the complete characterization of generic readout noise is infeasible for devices consisting of a large number of qubits. Here we introduce an appropriately tailored quantum detector tomography protocol, the so called Quantum Detector Overlapping Tomography, which enables efficient characterization of $k-$local cross-talk effects in the readout noise as the sample complexity of the protocol scales logarithmically with the total number of qubits. We show that QDOT data provides information about suitably defined reduced POVM operators, correlations and coherences in the readout noise, as well as allows to reconstruct the correlated clusters and neighbours readout noise model. Benchmarks are introduced to verify utility and accuracy of the reconstructed model. We apply our method to investigate cross-talk effects on 79 qubit Rigetti and 127 qubit IBM devices. We discuss their readout noise characteristics, and demonstrate effectiveness of our approach by showing superior performance of correlated clusters and neighbours over models without cross-talk in model-based readout error mitigation applied to energy estimation of MAX-2-SAT Hamiltonians, with the improvement on the order of 20% for both devices.