Distinguishing types of correlated errors in superconducting qubits

Abstract

Errors in superconducting qubits that are correlated in time and space can pose problems for quantum error correction codes. Radiation from cosmic and terrestrial sources can increase the quasiparticle (QP) density in a superconducting qubit device, resulting in an increased rate of QPs tunneling across proximal Josephson junctions (JJs) and causing correlated errors. Mechanical vibrations, such as those induced by the pulse tube in a dry dilution refrigerator, are also a known source of correlated errors. We present a method for distinguishing these two types of errors by their temporal, spatial, and frequency domain features, enabling physically motivated error-mitigation strategies. We also present accelerometer data to study the correlation between dilution refrigerator vibrations and the errors. We measure arrays of transmon qubits where the difference in superconducting gap across the JJ is less than the qubit energy, as well as those where the gap is greater than the qubit energy, which has been shown to mitigate radiation-induced errors. We show that these latter devices are also protected against vibration-induced errors.

Figures (14)

• Figure 1: (A) The two mechanisms of correlated error generation discussed in this paper: ionizing radiation and vibrations due to the pulse tube. (B) The superconducting gap profile of the three devices ($\mathrm{d_1}$, $\mathrm{d_2}$, and $\mathrm{d_3}$) under test. The $\delta\Delta_{\mathrm{JJ}}$ parameter refers to the difference in superconducting gap across the JJ. When $\delta\Delta_{\mathrm{JJ}}$ is greater than the qubit frequency $f_{\mathrm{q}}$, quasiparticle tunneling across the JJ is inhibited. The $\delta\Delta_{\mathrm{gnd}}$ parameter refers to the difference in superconducting gap between the JJ and the ground plane. Reproduced in part from Ref. pinckney2026characterizationradiationinducederrorssuperconducting. Further detail about the devices is given in Table \ref{['tab:devices']}.
• Figure 2: Details on the device mounting and measurement sequence. (A) A picture of device $\mathrm{d_1}$ mounted in a DR. The device package is mounted vertically, and the device is suspended by wirebonds from the package. The panel on the right shows an artificially colored layout of the qubit array, reproduced from Ref. pinckney2026characterizationradiationinducederrorssuperconducting, with the 10 qubits arranged in two rows of 5. The qubit number and junction orientation (S or F) are labeled. (B) A closeup of the black box in panel (A), showing the location of the JJ (in orange) for the two orientations. (C) The readout sequence for all measurements. A $\pi$-pulse is applied to change the qubit state. After 1 , the qubit is read out. The cadence of the full measurement is 6.95 .
• Figure 3: Distinguishing the two types of correlated qubit errors. (A) A histogram of filter score vs. decay lifetime for all error events detected in device $\mathrm{d_1}$ in DR1. The exponential filter used to identify error event candidates within the $N_r$ data is inset. The filter has an exponential decay lifetime of 5 ms, the expected recovery lifetime for S orientation qubits after a radiation-induced event harringtonSynchronousDetectionCosmic2024. This filter is time-reversed before convolving with $N_r$. Multiple error populations are visible. The red marker points to the filter score and decay lifetime of the radiation-induced event in (E), and the cyan marker points to the PT-induced event in (F). (B) Histogram of filter score vs. decay lifetime for device $\mathrm{d_1}$ in DR2, showing a reduction in PT-induced events. (C) Histogram of filter score vs. decay lifetime for device $\mathrm{d_2}$, showing a reduction in all events. (D) Histogram of filter score vs. decay lifetime for device $\mathrm{d_3}$, showing a reduction in all events. (E) A sample radiation-induced event. The left panel shows a moving average of the number of relaxations measured in each qubit, $N_{r, i} \ast \phi_{\mathrm{box}}$, for a radiation-induced error. The dashed line corresponds to the start time of the event. The right panel shows the maximum value of $N_{r, i} \ast \phi_{\mathrm{box}}$ for each of the qubits. S orientation qubits are shown in purple colors, and F orientation qubits are shown in orange colors. (F) A sample pulse tube-induced event.
• Figure 4: Number of relaxations averaged over 100 measurements as a function of time for device $\mathrm{d_1}$ in three measurement configurations. In the top panel (PT on, DR1) the PT frequency is clearly visible, corresponding to the gray dashed lines. In the other two panels, the PT frequency is not visible. The x axis is relative time; these measurements were not taken simultaneously.
• Figure 5: Histograms of the metric $A$ for the radiation- and PT-induced errors, showing that the location of the radiation-induced errors on the device varies across events, whereas the location of the PT-induced errors does not. The negative bias in $A$ for the PT-induced events is likely due to the inhomogeneous response of the qubits across the device. The area of each histogram integrates to one so that the shapes can be compared.