Flux-Tunable Transmon Coupler

• Flux-tunable transmon couplers are superconducting circuit elements that modulate qubit coupling via a SQUID-based transmon whose Josephson energy is flux-adjustable.
• They enable dynamic control over two-qubit interactions by turning coupling on/off, which minimizes residual ZZ crosstalk and improves gate fidelities.
• Experimental results demonstrate >99.99% CZ gate fidelity, scalable integration, and effective suppression of spectator errors in both planar and 3D architectures.

A flux-tunable transmon coupler is a superconducting circuit element that enables controlled, high-fidelity two-qubit interactions by leveraging a transmon with a SQUID (Superconducting Quantum Interference Device) geometry. Its key feature is the ability to modulate its effective Josephson energy, and thus its transition frequency, via externally applied magnetic flux. This property enables dynamic “on/off” control of the coupling strength between adjacent quantum systems such as superconducting qubits (fluxonium, transmon, or Xmon), facilitating gate operations with minimized crosstalk and static interaction. Architectures based on this concept underpin state-of-the-art quantum processors in both planar and 3D circuit QED, providing robust, frequency-flexible mediated coupling, suppression of residual entangling rates (notably “ZZ” crosstalk), and compatibility with scalable, low-error quantum computing (Zwanenburg et al., 10 Mar 2026, Vallés-Sanclemente et al., 17 Mar 2025, Ding et al., 2023).

1. Circuit Architecture and Hamiltonian

The canonical architecture involves two data qubits (e.g., fluxoniums or transmons) coupled capacitively to a central, grounded, flux-tunable transmon coupler. The coupler is realized as a symmetric SQUID, i.e., two Josephson junctions in parallel, forming a loop subject to an external flux $\Phi_\text{ext}$. The basic Hamiltonian, after neglecting higher excited states, decomposes as:

$H = \sum_{k = 1,2,c} H_k + H_\text{int}$

$$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$

$$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$

where $E_C^{(k)}$, $E_J^{(k)}$, and $\hat{n}_k$, $\hat{\varphi}_k$ are the charging energy, Josephson energy (flux-tunable for the coupler), and conjugate charge/phase operators, respectively. The tunable Josephson energy of the SQUID transmon follows $$E_J^T(\Phi_\text{ext}) \simeq E_{J,\max}\cos(\pi\Phi_\text{ext}/\Phi_0)$$, where $\Phi_0 = h/2e$ is the flux quantum (Zwanenburg et al., 10 Mar 2026, Ding et al., 2023, Vallés-Sanclemente et al., 17 Mar 2025).

Capacitive couplings $H = \sum_{k = 1,2,c} H_k + H_\text{int}$0, $H = \sum_{k = 1,2,c} H_k + H_\text{int}$1 between data qubits and coupler, and $H = \sum_{k = 1,2,c} H_k + H_\text{int}$2 (direct qubit-qubit) mediate the relevant interactions. In large-scale arrays, additional terms may include direct capacitive couplings between couplers and weak parasitic interactions (Zwanenburg et al., 10 Mar 2026).

2. Flux-Mediated Coupling and Effective Interactions

In the dispersive regime (qubit-coupler detuning much larger than coupling), the coupler can be adiabatically eliminated using a Schrieffer–Wolff transformation to yield an effective exchange (XY) and longitudinal (ZZ) interaction between data qubits:

$H = \sum_{k = 1,2,c} H_k + H_\text{int}$3

with

$H = \sum_{k = 1,2,c} H_k + H_\text{int}$4

$H = \sum_{k = 1,2,c} H_k + H_\text{int}$5

$H = \sum_{k = 1,2,c} H_k + H_\text{int}$6

where $H = \sum_{k = 1,2,c} H_k + H_\text{int}$7, $H = \sum_{k = 1,2,c} H_k + H_\text{int}$8 are the data qubit and coupler transition frequencies; $H = \sum_{k = 1,2,c} H_k + H_\text{int}$9 are data qubit anharmonicities (Vallés-Sanclemente et al., 17 Mar 2025). Dynamic tuning of $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$0 via flux controls $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$1 and $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$2, allowing for the implementation of conditional logic gates and the suppression of residual entangling rates.

For fluxonium–transmon–fluxonium (FTF) designs, the mediated exchange typically dominates over direct capacitive coupling, enabling both fast two-qubit gates and strong static ZZ suppression (to kHz scale) without strict device parameter matching (Ding et al., 2023, Zwanenburg et al., 10 Mar 2026).

3. Gate Protocols, Tunability, and Spectator Error Mitigation

High-fidelity conditional-phase (CZ) gates are obtainable by pulsing the flux bias of the coupler alone. The coupler is parked far detuned during idle (off-state, e.g., $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$3300 MHz above the $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$4 transition for fluxonium), strongly suppressing virtual interactions and spectator-induced errors ($$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$5) (Zwanenburg et al., 10 Mar 2026).

During gate operation, the coupler is dynamically brought near resonance (e.g., 50–150 MHz detuned above a noncomputational manifold transition) to maximize matrix elements, then returned to off (Ding et al., 2023). The protocol is compatible with microwave-activated and flux-adiabatic gates, including DRAG-style pulse shaping for leakage suppression.

A central result is that parking inactive couplers at these “off” points drastically reduces spectator-qubit crosstalk and static ZZ, enabling 1D/2D processor layouts with $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$699.99% CZ fidelities even in the presence of up to six spectator qubits (Zwanenburg et al., 10 Mar 2026, Vallés-Sanclemente et al., 17 Mar 2025). Trade-offs arise because the nulls of $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$7, $$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$8, and higher-manifold exchanges do not in general coincide in flux, motivating dynamic repositioning of coupler bias depending on operation (gate, readout, idle) (Vallés-Sanclemente et al., 17 Mar 2025).

4. Performance Metrics and Experimental Results

Gate fidelities and residual errors in leading FTF architectures and related transmon-based processors are summarized in the following table:

Architecture	CZ Gate Fidelity	Residual ZZ	Spectator Error	Gate Duration
Fluxonium-TM-Fluxonium (FTF) [1D/2D]	≥99.99% [CZ]	kHz	$$H_k = 4E_{C}^{(k)} \hat{n}_k^2 - E_{J}^{(k)}(\Phi_k)\cos\hat{\varphi}_k$$9 ($$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$0 in 2D)	50–100 ns
Transmon-TC-TM [bipolar/adiabatic]	99.5–99.9%	<30 kHz	<0.3% RB	40–100 ns
Multi-TM CZ [SFQ]	>99.9%	tunable	memory savings	70–130 ns

Fidelities reflect performance under optimized pulse shaping, use of dynamic coupler parking, and spectator suppression (Zwanenburg et al., 10 Mar 2026, Ding et al., 2023, Torosov et al., 2024). The duration and pulse bandwidth are compatible with current AWG/reconfigurable coupler technology.

Microwave crosstalk and parasitic coupler–coupler interactions (up to $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$1 MHz) raise errors only at the $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$2 level, and these effects are correctable via routine gate recalibration (Zwanenburg et al., 10 Mar 2026, Smirnov et al., 5 Sep 2025).

5. Scalability, Frequency Allocation, and Integration into Large-Scale Architectures

Flux-tunable transmon couplers are designed for robust scalability:

• Frequency-Flexible Operation: Couplers provide 1–2 GHz of in situ frequency choice, relaxing static crowding constraints and enhancing spectral allocation freedom in large arrays (Ding et al., 2023).
• Robustness to Process Variations: The grounded transmon geometry ensures that static ZZ cancellation is insensitive to minor capacitance or fabrication shifts, increasing manufacturing yield (Ding et al., 2023).
• Cross-Architecture Compatibility: The core principles extend to transmon–transmon, Xmon, and hybrid-mode coupler topologies. Centimeter-scale couplers have demonstrated high modulation contrast (on/off ratios $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$3 for $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$4 and $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$5 for $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$6) at millisecond coherence times, enabling strong, switchable interactions over macroscopic distances (Xu et al., 17 Jun 2025).
• Multiplexed and Modular Designs: Double-transmon coupler topologies and tunable bus architectures support modular integration, with internal zero-coupling points independent of qubit frequency, favoring fault tolerance and modular hardware scaling (Campbell et al., 2022).

6. Comparison to Competing Coupler Designs

Flux-tunable transmon couplers are distinguished by:

• Large On-Off Ratio: Gate “off” states yield residual $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$7 kHz, matching or exceeding the best performance of single-JJ “gmon” or rf-SQUID couplers (Geller et al., 2014, Campbell et al., 2022).
• Minimal Parameter Matching Requirements: Unlike fixed-frequency capacitive couplers, no fine-tuning of capacitance ratios is needed to achieve strong gates and low crosstalk.
• Noise Mitigation: Operability at “sweet spots” in flux, absence of net DC flux transfer during net-zero gate protocols, and negligible static coupling in the idle state all jointly minimize susceptibility to $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$8 flux noise and slow drift (Smirnov et al., 5 Sep 2025).
• Hardware-Software Codesign: The architecture is compatible with SFQ-based digital flux control, enabling gradient-based and model-free optimization of gate sequences with deep memory savings (Torosov et al., 2024).

Double-transmon couplers offer internal zero-coupling points and improved isolation, but at the expense of circuit complexity. Gmon and rf-SQUID couplers are architecturally simpler but lack the robustness of an internally defined “dark” state or may suffer from larger flux noise exposure (Campbell et al., 2022).

7. Practical Considerations and Limitations

Design and operation of flux-tunable transmon couplers require:

• Careful calibrations of coupler parking points and per-gate recalibration in the presence of parasitic capacitive interactions.
• Shielding from environmental flux noise and crosstalk through dedicated bias lines, superconductor shielding, and filtering.
• Selection of coupler and data qubit $$H_\text{int} = \hbar g_{12} \hat{n}_1 \hat{n}_2 + \hbar g_{1c} \hat{n}_1 \hat{n}_c + \hbar g_{2c} \hat{n}_2 \hat{n}_c$$9 ratios to balance large achievable couplings, high coherence, and sufficient anharmonicity for leakage suppression.
• In multi-qubit arrays, dynamic retuning of coupler bias in error-correcting cycles to balance gate, readout, and single-qubit operation fidelity (Vallés-Sanclemente et al., 17 Mar 2025).

A plausible implication is that as processor sizes increase and connectivities become more elaborate (e.g., higher-degree surface code modules), dynamic coupler “repositioning” will become crucial for mitigating cumulative spectator-mediated crosstalk, further emphasizing the central role of flux-tunable coupler strategies in fault-tolerant superconducting architectures (Vallés-Sanclemente et al., 17 Mar 2025, Zwanenburg et al., 10 Mar 2026).