Learning-based Calibration of Flux Crosstalk in Transmon Qubit Arrays*

Superconducting quantum processors comprising flux-tunable data and coupler qubits are a promising platform for analog quantum simulation and digital quantum computation. One challenge to scaling this platform is the magnetic flux crosstalk between flux-control lines and qubits, which impedes precision control of qubit frequencies. To implement high-fidelity quantum operations as processor sizes increase, we need an extensible approach to measure flux crosstalk and compensate for it. We demonstrate the experimental performance of a learning-based approach to DC-flux and fast-flux crosstalk calibration on an array of 16 flux-tunable transmon qubits. The overall calibration time for this approach empirically scales linearly with system size, while achieving a median qubit frequency error below 300 kHz.