Transmon Qubit Dispersive Readout

• Transmon qubit dispersive readout is a circuit-QED method that leverages off-resonant microwave cavities to induce state-dependent shifts for effective qubit discrimination.
• The technique employs weakly anharmonic transmons and a cross-Kerr Hamiltonian to optimize signal-to-noise ratios and measurement speed while mitigating backaction.
• Recent advances integrate calibration, drive power tuning, and Purcell filtering to balance dissipative effects and improve readout fidelity in multi-qubit architectures.

Transmon qubit dispersive readout is the circuit-QED measurement protocol in which a weakly anharmonic superconducting qubit is coupled off resonantly to a microwave resonator so that the resonator response acquires a qubit-state-dependent shift. In its standard form, the effective Hamiltonian is $$H/\hbar=\omega_r a^\dagger a + (\omega_q/2)\sigma_z + \chi a^\dagger a \sigma_z$$, with cavity pull $2\chi$ and coherent pointer states whose transmitted or reflected field is amplified and integrated for state assignment. For a weakly anharmonic transmon, the leading dispersive shift is written in the cited literature as $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$ and, with the sign convention $\alpha<0$, also as $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$, reflecting differing conventions for $\Delta$ and $\alpha$ (Salunkhe et al., 29 Jan 2025, Ye et al., 2024). Contemporary work places this textbook picture inside a broader framework in which measurement backaction is shaped not only by $g,\Delta,\alpha,\chi,\kappa$, and amplifier efficiency, but also by Purcell channels, measurement-induced state transitions, drive-renormalized spectra, TLS baths, and architectural choices such as sequential release, all-pass resonators, and nonperturbative cross-Kerr couplings (Huang et al., 14 Mar 2026).

1. Hamiltonian structure and the dispersive regime

The starting point is the Jaynes–Cummings model,

$$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$

or, for a multilevel transmon, its weakly anharmonic generalization in which adjacent transitions couple with matrix-element-dependent rates. In the dispersive regime, $|\Delta|=|\omega_q-\omega_r|\gg g$, the interaction is block-diagonalized to an effective cross-Kerr form,

$2\chi$0

so that the resonator resonance is shifted by $2\chi$1 depending on the qubit state. The cavity pull $2\chi$2 underlies linear phase or amplitude discrimination, while the strong-dispersive regime $2\chi$3 yields well-separated pointer states and faster discrimination (Dassonneville et al., 2019, Salunkhe et al., 29 Jan 2025).

A complementary first-principles description treats the transmon terminating a transmission line as a frequency-dependent boundary condition with a pole structure set by the qubit transitions. In that formulation, the dressed modes satisfy a transcendental equation $2\chi$4, where the state-dependent boundary function $2\chi$5 encodes the transmon admittance, the residue at the $2\chi$6 pole is $2\chi$7, and both the dispersive shift and the vacuum Rabi splitting emerge from the same eigenvalue problem. A level-repulsion theorem guarantees that no dressed mode frequency coincides with a transmon transition (Bakr, 30 Dec 2025).

2. Pointer states, measurement rate, and state discrimination

Under continuous resonator drive, the readout field is conditioned on the qubit state. In the asymmetric formulation used for a Purcell-filtered tunable transmon, the steady-state displacements are

$2\chi$8

with $2\chi$9, $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$0, and state-dependent linewidths $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$1. The mean intracavity photon numbers are $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$2, the measurement-induced dephasing rate is

$\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$3

and in the symmetric weak-pull limit this reduces to the familiar $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$4 (Huang et al., 14 Mar 2026).

These pointer states define the practical trade space of dispersive readout. A standard scaling for the signal-to-noise ratio is $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$5, while order-of-magnitude estimates often use $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$6. These relations explain why fast readout typically seeks $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$7 of order $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$8, moderate $\chi \approx g^2 \alpha / [\Delta(\Delta+\alpha)]$9, and a high-efficiency output chain, while remaining in a regime where the dispersive approximation is still reliable (Ye et al., 2024, Salunkhe et al., 29 Jan 2025).

The usual guardrail is the critical photon number $\alpha<0$0, or its multilevel generalization. In early transmon JBA experiments, reported parameters $\alpha<0$1 MHz and $\alpha<0$2 GHz gave $\alpha<0$3, while $\alpha<0$4 GHz gave $\alpha<0$5; bifurcation was observed near $\alpha<0$6–$\alpha<0$7 photons and a jump to $\alpha<0$8–$\alpha<0$9 photons (Mallet et al., 2010). More recent analyses show that $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$0 is only a partial diagnostic, because multi-photon avoided crossings and non-Lorentzian spectral reshaping can set stricter or qualitatively different limits.

3. Backaction, non-QND channels, and drive-renormalized spectra

A recurrent misconception is that a fixed measurement rate fully specifies backaction. Huang et al. showed that during dispersive readout the transmon emission spectrum can become non-Lorentzian and strongly dependent on the drive frequency even when $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$1 is held fixed. Their analytic correlation function

$\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$2

leads to a number-splitting spectrum

$\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$3

At $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$4, corresponding to drive at the ground-state resonator frequency $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$5, all poles coalesce and the spectrum is narrowest; at $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$6, corresponding to drive at $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$7, the spectrum becomes multi-peaked and overlaps more strongly with detuned TLSs. In the driven Fermi-golden-rule picture,

$\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$8

so $\chi \approx -g^2 \alpha / [\Delta(\Delta+\alpha)]$9 degradation is controlled by spectral overlap rather than by $\Delta$0 alone. Master-equation simulations and experiment on a flux-tunable transmon with $\Delta$1 MHz, $\Delta$2 MHz, and $\Delta$3 MHz confirmed that driving at $\Delta$4 gives the longest $\Delta$5 at fixed measurement rate (Huang et al., 14 Mar 2026).

The onset of measurement-induced state transitions is now understood in terms of multiphoton avoided crossings in the dressed transmon–resonator spectrum rather than by $\Delta$6 alone. Two universal diagnostics are the qubit purity $\Delta$7 under the actual ring-up protocol and the dressed matrix-element error

$\Delta$8

Both quantities remain small while the system follows a dressed coherent state within a single eigenladder, then increase sharply near avoided crossings associated with conditions $\Delta$9, often with $\alpha$0 (Nesterov et al., 2024).

In the regime $\alpha$1, another non-QND channel becomes dominant: Raman-like inelastic scattering of readout photons via the transmon nonlinearity. The measured transition rate is

$\alpha$2

so leakage grows linearly with AC Stark shift and is weighted by the dissipative environment at the lower scattered-photon frequency. In the reported device, $\alpha$3–$\alpha$4 GHz, $\alpha$5 MHz, $\alpha$6 MHz, and $\alpha$7 MHz at $\alpha$8 MHz; the parameter-free theory matched measured $\alpha$9 and $g,\Delta,\alpha,\chi,\kappa$0 to within about $g,\Delta,\alpha,\chi,\kappa$1 (Connolly et al., 5 Jun 2025).

In multi-qubit processors, the MIST threshold is not a single-qubit property. A Floquet-branch comparison between coupling-first and drive-first labelings shows that spectators and couplers create additional avoided crossings, lower the threshold, and can even conditionally suppress or restore transitions. For two coupled transmons with $g,\Delta,\alpha,\chi,\kappa$2 MHz, a spectator-induced swap appeared at $g,\Delta,\alpha,\chi,\kappa$3, far below the isolated-qubit thresholds $g,\Delta,\alpha,\chi,\kappa$4 and $g,\Delta,\alpha,\chi,\kappa$5. In a tunable-coupler architecture, matrix-element-cancellation bands can instead raise the threshold, but these bands do not coincide with zero-$g,\Delta,\alpha,\chi,\kappa$6 curves (Hoyau et al., 3 Jun 2026).

4. Experimental modalities and representative performance

Dispersive transmon readout has been implemented as linear homodyne detection, bifurcation-based sample-and-hold, sequential storage-and-release, transmission through all-pass resonators, and interferometrically enhanced measurement. Mallet et al. used a coplanar waveguide resonator operated as a Josephson Bifurcation Amplifier and reported $g,\Delta,\alpha,\chi,\kappa$7 contrast with electron shelving, $g,\Delta,\alpha,\chi,\kappa$8 Rabi visibility, and no increase in qubit relaxation from the readout itself (Mallet et al., 2010). Peronnin et al. separated probe loading, unitary interaction, and rapid release, obtaining $g,\Delta,\alpha,\chi,\kappa$9 average single-shot fidelity in $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$0 ns with a release time of about $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$1 ns and $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$2 (Peronnin et al., 2019). Modeling of SU(1,1) interferometry with parametric amplifiers predicts that for pulse durations of about $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$3–$$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$4 ns, assignment error probability can be reduced by about a factor of $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$5, and for longer pulses the reduction can be an order of magnitude or more if qubit $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$6 is sufficiently long (Barzanjeh et al., 2014).

Modality	Reported result	Reference
Non-perturbative polariton cross-Kerr	$$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$7 single-shot fidelity for $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$8 ns pulses; $$\frac{H}{\hbar}=\omega_r a^\dagger a+\frac{\omega_q}{2}\sigma_z+g(a\sigma_+ + a^\dagger \sigma_-),$$9	(Dassonneville et al., 2019)
Quantromon	$|\Delta|=|\omega_q-\omega_r|\gg g$0 single-shot readout fidelity without using a parametric amplifier	(Salunkhe et al., 29 Jan 2025)
All-pass transmission readout	$|\Delta|=|\omega_q-\omega_r|\gg g$1 in $|\Delta|=|\omega_q-\omega_r|\gg g$2 ns; $|\Delta|=|\omega_q-\omega_r|\gg g$3 in $|\Delta|=|\omega_q-\omega_r|\gg g$4 ns with shelving	(Yen et al., 2024)
Junction readout with bifurcation	$|\Delta|=|\omega_q-\omega_r|\gg g$5 assignment fidelity with $|\Delta|=|\omega_q-\omega_r|\gg g$6 ns integration; $|\Delta|=|\omega_q-\omega_r|\gg g$7 QND fidelity	(Beaulieu et al., 8 Jan 2026)
Quartonic readout	$|\Delta|=|\omega_q-\omega_r|\gg g$8 ns readout; $|\Delta|=|\omega_q-\omega_r|\gg g$9 assignment fidelity and $2\chi$00 QND fidelity in full master-equation simulations	(Ye et al., 2024)

This suggests that cross-platform comparisons should be made with care, because some results report contrast, some assignment fidelity, some QND-ness, and some are simulation-based. The same dispersive framework also supports sensing tasks beyond computational-state discrimination. In an offset-charge-sensitive transmon, direct single-shot dispersive monitoring of charge parity reached $2\chi$01 at $2\chi$02, measured $2\chi$03 ms, and, after added high-frequency filtering, achieved $2\chi$04 (Serniak et al., 2019).

5. Beyond perturbative transverse coupling

A major development is the replacement of perturbative $2\chi$05-based dispersive shifts by native or nonperturbative cross-Kerr interactions. In the polariton architecture, a “transmon molecule” realizes a native energy–energy coupling $2\chi$06 after strong ancilla–cavity hybridization. The measured parameters included $2\chi$07 MHz, $2\chi$08 MHz, $2\chi$09 MHz, and $2\chi$10. A conventional dispersive scheme producing the same lower-polariton shift at $2\chi$11 MHz would have required $2\chi$12 MHz and implied a Purcell-limited $2\chi$13, more than an order of magnitude below the measured lifetime (Dassonneville et al., 2019).

The quantromon pushes the same idea further by engineering orthogonal qubit and readout modes so that exchange coupling is suppressed by symmetry while an intrinsic four-wave-mixing cross-Kerr remains. In Sample A, $2\chi$14 was tuned from $2\chi$15 GHz to $2\chi$16 GHz while $2\chi$17 changed only from $2\chi$18 MHz to $2\chi$19 MHz, a total variation of about $2\chi$20 MHz. In Sample B, increasing $2\chi$21 by about $2\chi$22 still preserved Purcell protection once asymmetry was tuned near zero. In Sample C, the same architecture reached $2\chi$23 single-shot fidelity without a parametric amplifier at $2\chi$24 MHz, $2\chi$25 MHz, $2\chi$26, and $2\chi$27 (Salunkhe et al., 29 Jan 2025).

Junction readout uses a coupling capacitor and a Josephson junction in parallel, with interaction

$2\chi$28

The $2\chi$29 sector supplies a large nonperturbative cross-Kerr and resonator self-Kerr, while the capacitive term is tuned to suppress exchange by destructive interference. The same coupler creates an intrinsic LC notch at $2\chi$30, and the measured Purcell-limited lifetime increased by nearly four orders of magnitude near the notch. At the operating point used for readout, $2\chi$31 MHz, $2\chi$32 MHz, $2\chi$33, and the device was read out bifurcation-style with $2\chi$34 for $2\chi$35 and $2\chi$36 for $2\chi$37 (Beaulieu et al., 8 Jan 2026).

The quarton coupler represents the most aggressive version of this program. There the coupling is purely nonlinear at the linear level, the readout resonator is a linearized transmon mode, and the optimized point gives $2\chi$38 MHz with $2\chi$39 MHz and $2\chi$40. Full lab-frame master-equation and stochastic-trajectory simulations then yield $2\chi$41 ns readout with assignment fidelity $2\chi$42 for $2\chi$43, $2\chi$44 for $2\chi$45, and average QND fidelity $2\chi$46 (Ye et al., 2024).

6. Calibration, optimization, and scaling

The latest transmon data indicate that readout optimization should be performed in frequency space rather than in power space alone. A practical workflow is to characterize $2\chi$47, $2\chi$48, $2\chi$49, and any Purcell-filter asymmetry from $2\chi$50; calibrate $2\chi$51 from the AC Stark shift; set the power at each $2\chi$52 so that $2\chi$53 is constant across drive frequencies; then map $2\chi$54 versus $2\chi$55 at fixed $2\chi$56. In the Huang et al. transmon, this procedure identified $2\chi$57 as the optimal choice because $2\chi$58 coalesces the number-split poles and minimizes overlap with detuned TLSs. For near-resonant TLS hot spots, modestly increasing power at fixed $2\chi$59 can dilute spectral weight among more number-split peaks and increase $2\chi$60, but the same work explicitly warns against entering the MIST or ionization regime (Huang et al., 14 Mar 2026).

Scaling places additional constraints on the microwave environment. All-pass readout removes intentional mismatch and reduces linewidth spread from $2\chi$61 to $2\chi$62, thereby mitigating one source of multiplexing nonuniformity (Yen et al., 2024). The quantromon’s quasi-constant $2\chi$63 relaxes frequency-allocation constraints over multi-GHz detuning windows (Salunkhe et al., 29 Jan 2025). Boundary-condition analyses of multi-qubit dispersive readout add a cautionary point: when two qubits have matched dispersive shifts, odd-parity states become frequency-degenerate, but true parity-only measurement still requires engineered suppression of linear dispersive terms (Bakr, 30 Dec 2025).

Several design rules recur across architectures. One is to keep $2\chi$64 of order $2\chi$65 whenever the coupling mechanism permits it. Another is to operate below whichever threshold is stricter: $2\chi$66, the first multiphoton avoided crossing identified by purity or Floquet-branch diagnostics, or the first environment-assisted leakage channel identified by $2\chi$67. A third is to treat the readout drive frequency as a control parameter on equal footing with power. Future directions already identified in the literature include using frequency-resolved $2\chi$68 to characterize TLS baths more effectively, engineering pulse shapes and filters to reduce spectral overlap, extending quasi-constant-$2\chi$69 architectures to multiplexed multi-qubit processors, and integrating these constraints directly into quantum-error-correction readout design (Huang et al., 14 Mar 2026, Hoyau et al., 3 Jun 2026).