Adaptive Spectroscopy of Fast Two-level-system Dynamics in Superconducting Qubits

Superconducting quantum processors are frequently limited by two-level-system (TLS) defects, which cause energy relaxation and temporal instability. While these defects are known to fluctuate, traditional spectroscopy methods are too slow to capture their fast dynamics. This paper asks whether a high-speed, adaptive Bayesian estimation protocol can resolve these rapid TLS fluctuations and determine if they contribute to gate-level errors in real-time.

The researchers implemented an adaptive relaxation-spectroscopy protocol on flux-tunable transmon qubits using an FPGA-based controller. By continuously updating a Bayesian probability distribution for the qubit decay rate (Γ1) based on single-shot measurement outcomes, the controller can dynamically adjust probe wait times. This allows for the rapid acquisition of frequency-resolved maps of TLS-induced loss. The team applied this technique to two different device architectures across two separate laboratories and interleaved the spectroscopy with randomized benchmarking to assess the impact on gate fidelity.

The adaptive protocol successfully resolved TLS dynamics on timescales of seconds, revealing telegraphic switching and spectral diffusion approximately 300 times faster than conventional nonadaptive methods. Specifically, the team observed telegraphic switching with a correlation time of roughly 2.2 seconds and spectral diffusion with a diffusivity of approximately 0.9 MHz²/s. Furthermore, the study demonstrated that these rapid fluctuations in the qubit relaxation rate are directly correlated with fluctuations in gate infidelity, confirming that TLS dynamics are a primary, measurable contributor to time-dependent gate errors.

This work provides a new, high-speed tool for characterizing and calibrating superconducting quantum processors. By enabling the identification of fast-fluctuating TLS defects, this approach allows for more effective error mitigation, such as dynamically avoiding resonant frequencies or pausing operations during periods of high noise. It shifts the paradigm from infrequent, slow recalibration to low-latency, adaptive control loops that can respond to the actual, time-varying state of the hardware.