Machine Learning Approach to Characterization of Multiple Fluxonium Qubits

Three energy parameters characterize fluxonium qubits: the charging energy E_C, inductance energy E_L, and Josephson energy E_J . Fitting of the fluxonium spectral lines as functions of the external flux provides estimates for these parameters. The fitting method based on the exact
Hamiltonian diagonalization for each new set of energy parameters is viable for a small number of qubits. For a larger system of qubits, such fitting becomes inefficient. Here we propose a machine learning approach by training a neural network that takes the measured 0-1 transition energies at several external flux values and predicts parameters E_C, E_L, and E_J for each uxonium qubit. We demonstrate that such a prediction model can be generalized for many non-interacting or weakly interacting fluxonium qubits. This approach also allows us to identify transitions between higher energy states correctly. Analyzing energy parameters of a large number of fluxoniums on a chip will help to improve the fabrication of fluxonium processors.