Superconducting qubit decoherence correlated with detected radiation events

TL;DR

The paper addresses the challenge that quantum error-correction assumes uncorrelated decoherence, while cosmic radiation can cause spatially correlated errors in superconducting qubits. It introduces a platform with a qubit between two MKID arrays to detect radiation events in real time and correlates these events with single-shot $T_1$ and $T_2$ measurements, finding that dual MKID detections can cause up to a ~30% drop in coherence. The study links penetrating radiation, likely cosmic muons, to increased quasiparticle density and qubit decoherence, while providing a scalable framework for studying radiation effects, shielding, and error-correction strategies tailored to correlated noise. These results have practical implications for designing QEC and mitigation techniques in larger, radiation-exposed quantum processors.

Abstract

Most quantum error correction (QEC) protocols for superconducting qubits assume spatially and temporally uncorrelated decoherence events; however, recent evidence suggests that cosmic radiation induces spatially correlated errors. We present a platform that sandwiches a superconducting transmon qubit between two microwave kinetic inductance detector (MKID) arrays, enabling real-time detection of radiation-induced phonon bursts. By synchronizing MKID event detection with single-shot measurements of qubit energy relaxation ($T_1$) and phase coherence ($T_2$), we observe statistically significant reductions in both $T_1$ and $T_2$-up to 30.5%-immediately following dual MKID events attributed to penetrating muons. Our findings directly link radiating events to correlated qubit decoherence. Furthermore, our experimental platform provides a foundation for systematic studies of radiation effects, the development of shielding and mitigation techniques, and the refinement of error-correction algorithms tailored to correlated noise sources.

Figures (9)

• Figure 1: Experimental platform for correlated radiation event detection. (a) Schematic showing a superconducting qubit chip sandwiched between two microwave kinetic inductance detector (MKID) arrays. This geometry enables direct evaluation of correlations between qubit performance and incident penetrating radiation. The MKID array's opening angle ($113.9^\circ$) provides coverage of the expected muon flux distribution. Each chip is housed in a gold-coated copper box to ensure thermalization and microwave shielding, and all three layers are fixed to a common mounting bracket at the dilution refrigerator mixing chamber stage. (b) Circuit layouts for the top and bottom MKID arrays, each comprising a $3 \times 3$ grid of individual MKIDs (shown in red). Each MKID has a unique resonance frequency, adjusted by the arm length of the interdigitated capacitor. All MKIDs are read out via capacitive coupling to a shared $50~\Omega$ impedance transmission line (shown in blue), enabling simultaneous monitoring of the entire array via a single microwave feedline.
• Figure 2: MKID response to radiation events. (a) Cumulative in-phase (I) and quadrature (Q) data for a single MKID, recorded over 12 hours with acquisition triggered when an event was detected on any MKID in the array. As the local temperature rises, the increased quasiparticle population broadens and shifts the resonance frequency, causing the signal amplitude to decrease accordingly. The resulting traces form a characteristic arc in IQ space that terminates near the origin. (b) Time-resolved, temperature-calibrated response of the same MKID to a single event, illustrating the relaxation of the resonator as quasiparticles recombine into Cooper pairs. Temperature calibration enables quantification of the energy deposited by the event.
• Figure 3: MKID event energy distributions. (a) Probability density of the normalized energy deposited for ‘top,’ ‘bottom,’ and ‘dual’ events. The energy tail for dual-detector events is shorter than for top-only and bottom-only events, likely reflecting differences in the type of radiation involved. Dual-detector events typically arise from penetrating particles, which deposit less energy per detector on average, whereas top-only and bottom-only events are more likely caused by localized sources that have shorter stopping distances, resulting in greater energy deposition. (b) Scatter plot of total normalized energy in dual-detector events, where at least one MKID in both the top and bottom arrays registers an event. A total normalized energy of 1 indicates that every detector on a substrate recorded the maximum detectable energy. The distribution reveals three event types: those where energy is deposited primarily in the top array (orange), primarily in the bottom array (green), and distributed between both arrays (blue).
• Figure 4: Qubit coherence following MKID-detected events. Mean expectation value of single shot (a) $T_1$ and (b) $T_2$ measurements as a function of time after event start, separated by event class. Insets in each panel illustrate the delay time used for measurement, with a schematic of the average qubit population expected from the pulse sequence. Qubit measurements triggered by dual-detector events (blue triangles) exhibit statistically significant changes in both coherence times. The smaller effects observed for top-only (orange squares) and bottom-only (green diamonds) events are likely due to incomplete detection of penetrating events (see discussion). Error bars represent $\pm1$ standard deviation.
• Figure A1: Typical pulse configuration for MKID/qubit correlation measurements. MKIDs are continuously measured with a 5 $\mu$s readout pulse in every 8 $\mu$s duty cycle. The delay between readout pulses is mostly due to buffering time on the FPGA. When an event is detected ($\Delta I$ and/or $\Delta Q$$> 8\sigma$), the pulse train shown after the dashed line is executed fifty times (about 800 $\mu$s for $T_1$ and 560 $\mu$s for $T_2$) to obtain time-resolved event data, capturing both qubit coherence and the full MKID array response to the event. This approach ensures efficient data buffering and prevents processing overflows.