Short answer: Dispersive readout of a superconducting qubit uses a resonator whose frequency shifts with the qubit state. Making it faster means a stronger resonator decay rate, which can also drain the qubit, so designers add Purcell filters. Published experiments report 40 ns readouts at about 99% fidelity. All are small laboratory devices.
The readout trade-off
A 2017 Physical Review Applied paper notes that the speed of gates and measurements is a decisive factor in the fidelity of protocols on qubits with finite coherence time, and that state-of-the-art readout of superconducting qubits uses the dispersive interaction with a readout resonator. A 2022 arXiv paper (2202.06202) explains the catch: the speed of resonator-based operations is limited by the resonator's external decay rate, but increasing that rate also increases the rate of qubit decay through the resonator, which limits qubit lifetime.
Careful parameter design, 2017
The 2017 paper brings dispersive readout to what the authors call its current limit. They increase the dispersive interaction strength, choose an optimal resonator linewidth, use a Purcell filter and use phase-sensitive amplification, and report fast, high-fidelity single-shot measurements without affecting qubit coherence.
An intrinsic Purcell filter, 2022
The 2022 paper suppresses resonator-mediated qubit decay by using the distributed-element, multi-mode nature of the resonator itself. The authors show suppression exceeding two orders of magnitude over a 600 MHz bandwidth. With this intrinsic Purcell filter they demonstrate a 40 ns readout with 99.1% fidelity and a 100 ns reset with residual excitation below 1.7%.
A nonlinear Purcell filter, 2023
A 2023 arXiv paper (2309.04315) addresses a different problem: residual noise photons in the readout resonator become a major source of qubit dephasing when the resonator is optimised for fast readout. The authors propose and demonstrate a nonlinear Purcell filter. When a readout pulse is applied, the filter automatically reduces the effective linewidth of the resonator and so increases sensitivity to the input field. They report noise tolerance improved by a factor of 3 compared with a linear filter, and measurement rate improved by another factor of 3 using bifurcation of the filter. They report a readout fidelity of 99.4% and a quantum nondemolition fidelity of 99.2% with a 40 ns pulse.
Reading the results
The three papers use different devices, amplification chains and fidelity definitions, so their numbers describe each design rather than a ranking. They are small-device experiments, and none addresses reading out many qubits at once.
What each design buys
The 2017 paper shows that careful choice of dispersive shift, resonator linewidth, filter and amplifier is enough for fast readout, with no new element beyond a standard Purcell filter. The 2022 design removes the need for a separate filter by using the resonator's own multi-mode structure, which also helps reset. The 2023 design targets dephasing from noise photons, a problem that appears once the resonator is made fast. Read together, they show the design problem moving from speed, to speed without qubit decay, to speed without dephasing. Each paper still reports a small test device and its own measurement settings.
FAQ
What is a Purcell filter?
A circuit element that lets the readout resonator decay quickly without letting the qubit decay through it.
How fast was readout in these papers?
40 ns in both the 2022 and 2023 papers, with fidelities of 99.1% and 99.4% respectively.
What does the nonlinear filter add?
Tolerance to residual noise photons, which dephase the qubit.
What is a QND fidelity?
How well the measurement leaves the qubit state undisturbed; the 2023 paper reports 99.2%.
Is this standard in commercial devices?
These papers do not address that.
All field notes