An arXiv preprint reports an operational readout duration of 97(1) nanoseconds for a superconducting transmon, while the 58-nanosecond measurement assigned the qubit’s two main states with 0.17(1)% error. The 97-nanosecond figure was defined by a subsequent pi pulse whose error fell below 10^-4, making it an operational benchmark tied to the next qubit operation. The experiment focused on one transmon from a 64-qubit processor. The authors conclude that fast, high-fidelity, low-leakage dispersive readout can be achieved in the small qubit-resonator-detuning regime.
The reported clock includes the wait
The readout pulse lasted 58 ns. Afterward, residual readout photons depleted passively over tens of nanoseconds, with no active depletion pulse. The total reached 97(1) ns at the point where the error of a subsequent pi pulse fell below 10^-4. In that sense, the reported duration is an operational surrogate for readout recovery, tied to the next qubit operation rather than only to the duration of the signal pulse.
The readout design paired a state-averaged effective resonator decay rate of κeff/2π = 30.8 MHz with a dispersive shift close to the condition that optimizes signal-to-noise per photon. The parameters were characterized with the chi-kappa-power method. The effective decay rates were 37.5 MHz and 24.1 MHz, with decay time constants of 4.2 ns and 6.6 ns.
Leakage stayed low, but not zero
The study also used a leakage-sensitive measurement, sorting records into |0⟩, |1⟩ and a pooled class called L for leakage. The classifier combined a depth-5 time-augmented log-signature with multinomial logistic regression. Pooling the results into L meant the analysis reported leakage as a class rather than assigning each higher level separately.
At the assignment optimum, mean leakage per measurement was 2.7(2) × 10^-5. With repeated measurements turned off, it was 1.3(3) × 10^-5. The on-setting value was approximately twice the off-setting value and more than two orders of magnitude below the measurement-induced relaxation rate. The leakage classifier had 0.87% assignment error, reached 95.97% accuracy for |2⟩ preparation, and recorded a 0.005% false-positive leakage rate.
In a separate measurement-on versus off/background comparison, the reported relaxation time was 12.7 microseconds with measurement on and 20.3 microseconds in the off/background condition. The estimated coherence-limited assignment error was 0.14%, accounting for 80% of the measured two-state error.
Repeated-readout behavior was assessed with the MIST benchmark, which used 28 repetition values from m = 1 to m = 650. At each value, the benchmark collected 1.5 × 10^4 repetitions and 4.2 × 10^5 shots for each qubit state. This provided a repeated-readout test alongside the single-operation timing benchmark.
Simulations found weak resonances
Simulations examined whether multiphoton resonances could connect the driven readout to higher qubit levels. At maximum photon numbers of 8.2 for |0⟩ and 10.1 for |1⟩, Floquet calculations showed very narrow resonances, indicating weak coupling under the tested readout conditions.
Across 100 evenly sampled gate charges, modeled mean leakage was 2 × 10^-8 for the |0⟩ to |10⟩ transition and 1 × 10^-7 for |1⟩ to |9⟩. The modeled maxima were 1 × 10^-6 and 2 × 10^-6, respectively. These figures are model estimates for the specified high-lying transitions, not direct measurements of every leakage level.
A result tied to one transmon
The evidence remains tied to the tested circuit and operating point. The experiment characterized one transmon from a 64-qubit processor, while the leakage analysis grouped leakage outcomes into L. The reported 97(1)-ns duration was an operational surrogate based on the error of a subsequent pi pulse. Those limits leave open whether the same speed, fidelity and leakage would hold across other qubits, devices or operating points.
The authors’ conclusion is a feasibility claim for fast, high-fidelity, low-leakage dispersive readout in the small-detuning regime. The supplied document is an arXiv version-1 preprint dated 26 August 2026. The work was supported in part by the MEXT Q-LEAP program, grant JPMXS0118068682, and the JST ASPIRE program, grant JPMJAP2513.
Paper data and sources
Original title: Low-leakage superconducting-qubit measurement with sub-100-ns total duration
Authors: Peter A. Spring, Adrian L. Hesse, Shiyu Wang et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text