GPU-Accelerated, Differentiable Fitting of Transient Correlation Functions for Resonator Characterization

Many superconducting circuits are based on resonators. Hence, fitting them requires well-established routines. Most used approaches rely on steady-state data obtained via spectroscopy measurements. Those may fail to precisely capture some degenerate parameters or require extensive sweeps of controls.

In this work, we exploit informative time traces of transient data to fit the parameters of a resonator, enabling fast, accurate estimation of all the parameters characterizing the resonator from a compact experiment.

Technically, we fit time-domain correlation functions within an explicit, differentiable model. Our pipeline is end-to-end differentiable, GPU-amenable, and supports batched fits, which substantially reduces sweep time and improves robustness.

This lays the groundwork for fitting more complex systems with more informative experiments.