Short answer: Crosstalk is unwanted interaction between qubits or their control lines, so an operation on one qubit disturbs another. It is a severe error source in superconducting processors, and it is easy to miss because standard calibration tests qubits one at a time. Papers from 2021 to 2024 show ways to measure it with randomized benchmarking, to schedule the measurements efficiently, and to reduce its effects with control optimization or noise-aware qubit routing. The results come from simulations and from experiments on specific chips.
Why standard numbers can mislead
A 2023 arXiv paper (2303.14103) explains that standard calibration data on an IBM device includes single-qubit and two-qubit average error rates measured with randomized benchmarking, run either one after another on isolated qubits or qubit pairs, or in parallel. The authors say that when characterization experiments are run in isolation, crosstalk between them is assumed negligible, so the reported average error rates in general underestimate the errors seen when operations run together. They show how to characterize the size of crosstalk on an entire chip with randomized benchmarking, run the protocol on IBM hardware, and observe significant alterations in gate fidelities caused by crosstalk. They use the results to propose a crosstalk-aware noise model for more accurate simulation of noisy hardware.
Characterizing and routing around it
A 2024 arXiv paper (2401.06423) describes crosstalk as a severe error source in, for example, cross-resonance based superconducting processors. The authors develop a simplified measurement experiment for characterization and analyse the optimal scheduling of those experiments for common hardware architectures. They then mitigate the noise with a noise-aware qubit routing algorithm that extends earlier swap-insertion work using integer programming, adding measured crosstalk errors to other easily available noise data in the objective function. They also prove a convex hull result about an associated class of polytopes, which they say has applications beyond the paper. They characterized crosstalk on two chips with up to 127 qubits and used the data to improve the approximation ratio of the Quantum Approximate Optimization Algorithm by up to 10 percent compared with other established noise-aware routing methods. The authors say this shows the gains of including noise data when mapping abstract circuits onto hardware.
Fixing it with control software
A 2021 Physical Review Letters paper describes a scalable framework for modelling crosstalk effects on quantum processors. By applying optimal control techniques, the authors show how to tune up arbitrary high-fidelity parallel operations on systems with substantial local and nonlocal crosstalk. They give a simulated example, a two-dimensional square array of superconducting transmon qubits, where optimized control parameters improve single-qubit and two-qubit gate infidelities compared with the same crosstalk without optimization. They conclude that, rather than striving to engineer away undesirable interactions during fabrication, one can largely mitigate such effects with software through careful characterization and control optimization. That is the authors' conclusion from simulation.
What this does not tell you
The control-optimization result is from simulation, and the routing gain applies to the QAOA benchmark on the chips tested. These papers do not show that crosstalk is solved, and the right method depends on the hardware architecture and the type of crosstalk.
FAQ
What is crosstalk?
Unwanted interaction between qubits or control lines that makes one operation disturb another.
Why is it hard to see?
Standard calibration often measures qubits in isolation, which underestimates errors when operations run together.
How is it measured?
One paper uses randomized benchmarking across a whole chip.
How can it be reduced?
Through noise-aware qubit routing or optimal control, as the 2024 and 2021 papers show.
Is the control result an experiment?
No. It is a simulation.
All field notes