A built-in energy drain could help superconducting qubits reset quickly

A quantum computer cannot reliably run a calculation if its qubits begin in unpredictable states. During error correction, qubits may need to be reset repeatedly. Superconducting qubits can also stray into higher energy levels beyond the two states used to represent information, a problem called leakage. This experiment tests a way to give qubits a local path for losing unwanted energy, rather than relying on dissipation through external microwave circuitry.

The design gives each qubit a dedicated resonator, connected to a shared on-chip transmission line ending in thin-film resistors. A resonator responds strongly at a particular frequency. During ordinary operation, the qubit is kept away from its resonator’s frequency. For a reset, a control pulse briefly tunes the qubit into resonance. Energy can then flow from the qubit into the resonator and onward to the resistive load, where it is dissipated. The architecture is intended to allow rapid energy loss during reset while limiting unwanted loss during other operations, a protection known as Purcell protection.

The team measured 20 working transmon qubits on a 72-qubit test processor. To assess reset, they prepared qubits in either the ground state, |g⟩, or excited state, |e⟩, applied a 400-nanosecond pulse, and used single-shot readout to assign the resulting state. They also measured how quickly excited-state assignments decayed at different qubit frequencies, and tested a pulse intended to remove leakage from the higher-energy |f⟩ state. Measurements were made in groups to reduce crosstalk; the study does not show that crosstalk was absent.

Across the 20 qubits, the median fitted reset time constant was 48.1 nanoseconds. A time constant is the characteristic time in an exponential decay: it describes how quickly the measured excited-state signal falls, rather than specifying the exact time every qubit takes to reach its target state. After a 400-nanosecond pulse, the median residual excited-state assignment probability was 0.29% for qubits prepared in |g⟩ and 0.67% for those prepared in |e⟩. These are readout assignment results following different initial preparations, not identical measures of reset under the two conditions.

The leakage-removal test was performed on one qubit, Q0. After preparation in |f⟩, the probability of assigning the result to |f⟩ was 77.54% without the pulse and 0.51% after a 400-nanosecond pulse. The pulse was designed to promote the |f⟩→|e⟩ transition, so this reduction does not mean that all the assignment probability transferred directly to the ground state. The authors note that assignments to |g⟩ also increased, potentially because of additional relaxation during the pulse sequence. This is a single-qubit demonstration, not a processor-wide leakage-removal result.

The authors also infer an effective bath temperature of about 30 millikelvin from measured state-assignment probabilities, using assumptions about thermal populations and equilibration. It is not a direct thermometer reading. The paper says the inferred temperature could reflect reduced exposure to external noise, while residual excitation may indicate imperfect thermalization or radiation.

There are uncertainties in how the reset dynamics are interpreted. The measured time constants come from exponential fits, and the paper notes that qubit dephasing and distortions in flux-control pulses may affect the dynamics. For the same qubits, the median measured time constant was 48.1 nanoseconds, compared with a model prediction of 28.9 nanoseconds. The authors suggest dephasing and pulse distortions could contribute to this gap. They also identify further optimization of the resistive terminations and the balance between reset speed and frequency selectivity as possible next steps.

The experiment demonstrates an integrated reset approach on a multi-qubit test processor; it does not show that the design improves a complete quantum error-correction system or will scale without complications. Reset performance was assessed on 20 operational qubits, and leakage removal on one. The report is an arXiv preprint, so its findings should be understood as a research result rather than proof of independent replication or broad agreement. Source: arXiv:2611.14391v1.