Computationally Efficient Design Optimization of Superconducting Quantum Devices

Superconducting quantum devices are computationally intense to simulate and model, with some finite-element solvers requiring hours to converge. Accurate modeling of these devices is important for designing qubit charge sensitivity, minimizing frequency collisions, and optimizing two-qubit interaction strengths. However, designing these devices by hand can be a cumbersome process due to these long simulation times. In this work, we present a multi- fidelity Bayesian optimization process which finds optimal transmon design parameters to achieve a target Hamiltonian. This process integrates with Qiskit Metal to automate single- and multi-qubit device design. Additionally, we present a risk-averse Bayesian optimization process which evaluates sensitivity of Hamiltonian parameters to design parameters, and optimizes the design for maximal device yield. As a use case, we will highlight the application of this Bayesian optimization technique to the design of a tunable coupler system.