Learning-based Calibration of Flux Crosstalk in a Transmon Qubit Array

Flux-tunable transmon qubit arrays are a promising platform for quantum computation and simulation. A challenge in scaling this platform, however, is the flux crosstalk in the system. In order to exercise precise control, we need to measure this flux crosstalk, and compensate for it. Brute force approaches to flux crosstalk calibration scale quadratically with the number of qubits in the array. As we strive towards chips with hundreds of qubits, we need an extensible approach. We propose an iterative learning-based procedure to calibrate flux crosstalk. Based on statistical tests on simulated data, we have demonstrated our calibration protocol to be accurate (less than 1 MHz mean frequency error for qubits with a maximum frequency of 5 GHz) and to scale approximately linearly with array size. We have experimentally implemented this calibration protocol on planar and 3D integrated flip-chip devices, enabling precise control of qubit frequencies with an approach that will extend naturally as we fabricate larger qubit arrays.