Multi-Mode Tunable Coupler Overview

• Multi-mode tunable coupler is a reconfigurable coupling element that exploits interference between several modes to adjust interaction strengths dynamically.
• It leverages diverse physical implementations, including superconducting circuits and photonic cavities, to enable selective control over excitation manifolds.
• The design principle emphasizes mode-selective interaction engineering, balancing strong activation with clean isolation to minimize leakage and residual couplings.

A multi-mode tunable coupler is a coupling element or coupling architecture in which the effective interaction between subsystems is reconfigured while more than one relevant mode participates in the coupling physics. In the literature, however, the word mode is platform-dependent: it may denote internal coupler normal modes in superconducting circuits, selected longitudinal resonances in coupled optical cavities, multiple cavity modes coupled to a single waveguide, or mechanical normal modes in a mediated oscillator network. The term is therefore not uniform. Some devices are explicitly multi-mode, whereas others that appear under the same label are more accurately described as dual-polarization, distributed single-mode, or modular tunable couplers (Jiang et al., 30 Sep 2025, Tang et al., 2019).

1. Terminology and scope

The most useful way to read the term is as a family resemblance rather than a single device class. Across platforms, two features recur: a reconfigurable coupling strength and a mode structure richer than a single fixed bus oscillator.

Context	Meaning of “mode”	Representative characterization
Superconducting circuits	Internal coupler normal modes or excitation-manifold channels	Explicitly multi-mode in the three-mode coupler, double-transmon coupler, hybrid-mode long-distance coupler, and full-localization two-mode coupler (Egorova et al., 2024, Campbell et al., 2022, Xu et al., 17 Jun 2025, Jiang et al., 30 Sep 2025)
Integrated photonics	Longitudinal cavity resonances or cavity-waveguide modes	Localized resonance hybridization, inverse-designed multimode cavity coupling, or node-engineered cavity-waveguide decoupling (Gentry et al., 2014, Jin et al., 2018, Koshino, 2024)
Optofluidics	TE and TM polarization states rather than multiple spatial modes	Better described as a polarization-insensitive tunable coupler, not a spatially multimode coupler (Tang et al., 2019)
Levitated nanoparticles	Mechanical normal modes in a three-oscillator network	Ancillary particle mediates tunable coupling between two outer mechanical modes (Wu et al., 2024)

A central terminological distinction is between genuinely multimode devices and adjacent but non-multimode tunable couplers. The optofluidic directional coupler of 2019 is tunable and supports both TE and TM inputs with dynamic range above $45$ dB and excess loss below $0.06$ dB, but it is not presented as a multimode coupler in the spatial-mode sense; it is more accurately described as dual-polarization-capable or weakly polarization-sensitive (Tang et al., 2019). Similarly, the multi-chip floating tunable coupler for modular superconducting qubits is architecturally distributed across several chips, yet its working model remains a single tunable transmon-like coupler mode rather than an intentionally multimode coupler (Field et al., 2023). The quarter-wave resonator-based Xmon coupler begins from a sum over resonator modes, but its design logic reduces to the fundamental mode because the large low-mode spacing allows higher modes to be neglected (Wang et al., 2022).

2. Core mechanisms

Across implementations, multimode tunable coupling is usually realized by one of three mechanisms: interference between several virtual coupling paths, flux- or boundary-controlled hybridization of internal modes, or selective brightening and darkening of modes relative to an external channel.

In superconducting circuits, the generic single-mode baseline is the cancellation of a fixed direct interaction by a tunable virtual interaction. In the floating tunable coupler, the net exchange is written as

$g = g_{12} - g_{\rm eff},$

with

$$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$

so zero coupling is obtained when the static and mediated terms cancel (Sete et al., 2021). Multi-mode circuit couplers generalize this logic by replacing a single virtual path with a sum over several internal channels. In the centimeter-scale hybrid-mode coupler, the effective exchange is

$$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$

so tunability comes from co-tuning both the mode frequencies and the port amplitudes of two hybridized resonator-transmon modes (Xu et al., 17 Jun 2025).

The strongest recent formulation is the two-mode coupler that targets full localization. After diagonalizing the coupler sector, the decoupled point is defined by

$\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$

so each qubit couples only to its own orthogonal hybridized coupler mode. Under the symmetry condition $g_{a2}=-\alpha g_{a1}$ and $g_{b1}=\alpha g_{b2}$, the localization condition becomes

$$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$

This is stronger than perturbative cancellation of a reduced $J$: it block-diagonalizes the Hamiltonian into independent qubit-coupler subsystems and is proposed to eliminate residual qubit wavefunction delocalization at the off point (Jiang et al., 30 Sep 2025).

In photonics and microwave distributed structures, the same idea appears as mode-shape control. In the galvanically connected cavity-waveguide coupler, the tunable stub changes the boundary condition of a three-port transmission-line junction so that a cavity-mode node can be moved onto the branch point. At the exact condition $0.06$0, the cavity mode has zero amplitude in the semi-infinite waveguide and $0.06$1 in principle; away from that condition, coupling to the continuum is restored and can reach the GHz scale (Koshino, 2024). In dual-ring optical resonators, thermal tuning of one ring shifts one selected resonance into an avoided crossing, producing a local supermode splitting used to compensate resonance mismatch for four-wave mixing (Gentry et al., 2014).

A related mechanism is multichannel mediation through an ancillary oscillator. In levitated nanoparticle arrays, a third nanoparticle is inserted between two outer particles chosen to be effectively uncoupled in the relevant channel. The outer pair then interact only through the coupler particle, with the conservative and non-conservative parts of the dipole-dipole interaction controlled by optical phase, spacing, and geometry (Wu et al., 2024).

3. Superconducting-circuit realizations

Superconducting circuits contain the clearest explicit examples of multi-mode tunable couplers. The principal motivation is no longer merely on-off exchange control; it is the ability to shape interactions differently in the one- and two-excitation manifolds, reduce residual $0.06$2, suppress leakage, and retain large activated coupling.

A direct realization is the three-mode tunable coupler for superconducting two-qubit gates. Its coupler is a three-node structure terminated by Josephson elements, supporting two flux-tunable even modes and one odd flux-insensitive mode. The mediated interaction is shaped by three resonant denominators in the interqubit impedance, rather than one. Experimentally, this architecture produced a native CZ gate with $0.06$3 ns pulse duration and fidelity above $0.06$4, while full-circuit simulation predicted $0.06$5 at $0.06$6 ns (Egorova et al., 2024). The significance of this device is that the richer mode structure broadens the tunable $0.06$7 window without relying on large SQUID asymmetry.

A second line of work realizes multimode behavior through internal coupler composition rather than a larger network node count. The modular double-transmon coupler is built from two internal transmon-like modes linked by a flux-tunable three-junction loop. Its internal normal modes $0.06$8 mediate coupling to external modes through flux-controlled interference between capacitive and inductive exchange, $0.06$9. The authors emphasize that this yields an internally defined zero-coupling state, largely independent of the attached data modes, and discuss applications beyond two-qubit gates, including readout, quantum bus interfacing, and coupling to a waveguide or open quantum system (Campbell et al., 2022). In this sense the device is a two-port, internally multimode coupling module.

Fluxonium implementations make the multimode character even more explicit. In the fluxonium processor with a tunable coupler, the middle circuit $g = g_{12} - g_{\rm eff},$0 is described as a two-mode tunable coupler whose fluxonium mode $g = g_{12} - g_{\rm eff},$1 and harmonic mode $g = g_{12} - g_{\rm eff},$2 mediate the interaction between qubit modes $g = g_{12} - g_{\rm eff},$3 and $g = g_{12} - g_{\rm eff},$4. The experiment does not populate those coupler modes during the gate; instead it eliminates them into an effective Hamiltonian

$g = g_{12} - g_{\rm eff},$5

Using this architecture, the experiment demonstrated an fSim-type gate with $g = g_{12} - g_{\rm eff},$6 fidelity and a CZ gate with $g = g_{12} - g_{\rm eff},$7 fidelity, while residual $g = g_{12} - g_{\rm eff},$8 was suppressed to the few-kHz level and measured below $g = g_{12} - g_{\rm eff},$9 kHz over a wide range of coupler fluxes (Moskalenko et al., 2022).

A complementary fluxonium result is the tunable inductive coupler between heavy-fluxonium qubits. Here the strongest multi-mode aspect is not a general multimode bus but the coexistence of two internal coupler degrees of freedom and two parametrically addressable gate channels. The static interaction $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$0 can be tuned from $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$1 to $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$2 MHz, while the $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$3 interaction remains below $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$4 kHz across the coupler bias range and below $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$5 Hz at the off position. The same coupler supports a difference-frequency $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$6 gate in $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$7 ns with fidelity $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$8 and a sum-frequency $$g_{\rm eff} = \frac{g_{1c}g_{2c}}{2}\sum_{j=1}^2\left(\frac{1}{\Delta_j} + \frac{1}{\Sigma_j}\right),$$9 gate in $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$0 ns with fidelity $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$1 (Zhang et al., 2023). This is not a multimode bus in the resonator sense; rather, it is a single tunable interaction supporting multiple gate channels and two opposite-sign coupler-mediated pathways.

The long-distance hybrid-mode coupler pushes the multimode idea into the distributed regime. A $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$2-cm coplanar-waveguide bus with an embedded Josephson element is engineered so that its two lowest normal modes are both active. Flux changes their frequencies, field profiles, and port amplitudes, so the mediated interaction is the sum of two tunable channels. For a $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$3-cm coupler, the reported values are flux-tunable $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$4 up to about $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$5–$$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$6 MHz, $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$7 up to about $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$8 MHz, $$J_{12} = \sum_{m=1,2} g_{1m}g_{2m}\left(\Delta_{1m}^{-1}+\Delta_{2m}^{-1}\right),$$9 modulation contrast exceeding $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$0, and $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$1 modulation contrast exceeding $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$2 (Xu et al., 17 Jun 2025). This is one of the clearest examples in which “multi-mode tunable coupler” refers literally to several intentionally engineered coupler modes.

The most ambitious formulation is the multi-mode coupler enabling full localization. In the two-mode case, the authors argue that a conventional single-mode coupler cannot independently control interactions in the one- and two-excitation manifolds. Their two-mode design instead seeks large $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$3 together with suppressed $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$4 and $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$5, so that $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$6 exchange is strong while $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$7 and $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$8 channels are reduced. The numerical results show an off regime where $\tilde g_{a2}=0,\qquad \tilde g_{b1}=0,$9 and $g_{a2}=-\alpha g_{a1}$0 kHz, and an on regime where $g_{a2}=-\alpha g_{a1}$1 MHz while the selectivity ratio

$g_{a2}=-\alpha g_{a1}$2

exceeds $g_{a2}=-\alpha g_{a1}$3. An iSWAP gate time of about $g_{a2}=-\alpha g_{a1}$4 ns is reported for a representative operating point (Jiang et al., 30 Sep 2025).

Not all superconducting results that are adjacent to this literature are genuinely multimode. The floating tunable coupler and its multi-chip extension are best understood as single active coupler modes with sign-engineered capacitive pathways and modular packaging (Sete et al., 2021, Field et al., 2023). The shared tunable coupler used for controlled-controlled-phase gates is physically a single anharmonic mode, but functionally multi-channel because higher excitation manifolds produce several conditional shifts $g_{a2}=-\alpha g_{a1}$5, and $g_{a2}=-\alpha g_{a1}$6 (2206.12392). A plausible implication is that the multimode problem in superconducting circuits is as much about manifold engineering as about the number of explicit linear resonances.

4. Photonic, microwave-distributed, and optofluidic implementations

In photonics, the multimode tunable coupler appears primarily as a mode-selective spectral engineering device rather than a universal switchable bus. The canonical optical example is the dual-ring resonant structure for dispersion compensation. A primary silicon microring and an auxiliary ring with a different free spectral range are coupled so that only one selected longitudinal resonance hybridizes strongly. Thermal tuning of the auxiliary ring shifts that localized avoided crossing and compensates a $g_{a2}=-\alpha g_{a1}$7 GHz intrinsic resonance mismatch, producing an $g_{a2}=-\alpha g_{a1}$8 dB enhancement of seeded four-wave-mixing efficiency and a peak wavelength conversion efficiency of $g_{a2}=-\alpha g_{a1}$9 dB across an FSR of $g_{b1}=\alpha g_{b2}$0 THz ($g_{b1}=\alpha g_{b2}$1 nm) (Gentry et al., 2014). This is multi-mode in the sense of selected cavity resonances, not in the sense of simultaneous broadband independent control over many channels.

A more systematic realization is the inverse-designed multimode cavity coupler. There, a compact passive coupling region is optimized so that a single waveguide can critically or nearly critically couple to multiple cavity resonances simultaneously. The reported demonstrations target two wavelengths for SHG, three for SFG, and six resonances spanning roughly $g_{b1}=\alpha g_{b2}$2–$g_{b1}=\alpha g_{b2}$3 nm. The six-frequency example achieves near-critical coupling at six frequencies over an octave, with transmission plus reflection $g_{b1}=\alpha g_{b2}$4 (Jin et al., 2018). This work is genuinely multimode but only design-tunable: there is no post-fabrication active tuning mechanism.

The distributed microwave analog is the galvanically connected tunable cavity-waveguide coupler. Here Ports 2 and 3 form an effective cavity, Port 1 is a semi-infinite waveguide, and a SQUID-terminated stub changes the standing-wave pattern so that the cavity-mode node can be shifted to the branch point. The lowest branch resonance of Port 2 is $g_{b1}=\alpha g_{b2}$5 GHz, and the Port-3 resonance can be tuned over $g_{b1}=\alpha g_{b2}$6–$g_{b1}=\alpha g_{b2}$7 GHz. Exact decoupling occurs when $g_{b1}=\alpha g_{b2}$8, yielding $g_{b1}=\alpha g_{b2}$9 in principle, while the on state can reach cavity-waveguide coupling in the GHz range and even the ultrastrong regime in the broad sense that $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$0 becomes comparable to the cavity frequency (Koshino, 2024). This work is especially relevant because its mechanism—node-position control rather than a small inserted coupling element—naturally suggests multimode generalizations.

A neighboring but distinct literature concerns multimode interference couplers in silicon photonics. These devices are fundamentally multimode in the interference sense but not tunable. The SOI study of arbitrary coupling-ratio MMIs reports fixed-ratio $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$1 devices spanning target splits from $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$2 to $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$3, with coupling ratios close to design over $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$4–$$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$5 nm and excess loss conservatively below approximately $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$6 dB (Doménech et al., 2014). This is a geometric and fabrication baseline for future tunable multimode couplers rather than a demonstration of tunability itself.

The optofluidic directional coupler occupies an important boundary case. Its splitting ratio is tuned by changing the refractive index of a liquid mixture that acts as the upper cladding in the coupling region, and it shows dynamic range above $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$7 dB, excess loss smaller than $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$8 dB, low polarization dependence, and weak wavelength dependence over $$\frac{\lambda}{\Delta_{12}}=\frac{\alpha}{1-\alpha^2}.$$9–$J$0 nm (Tang et al., 2019). Yet the analysis addresses only TE and TM fundamental guided modes, not multiple spatial modes. It is therefore more accurate to describe it as a low-polarization-dependence tunable coupler than as a spatially multimode tunable coupler.

5. Mechanical and levitated-particle couplers

In levitated nanoparticle arrays, the multimode tunable-coupler idea appears as a three-mode mechanical mediation problem. Two outer nanoparticles are chosen so that their direct light-induced dipole-dipole interaction is negligible in the relevant channel, and a third nanoparticle is inserted as a coupler on the perpendicular bisector. The interaction depends on the coupler position angle $J$1, the outer-particle spacing $J$2, and the optical phase difference $J$3 (Wu et al., 2024).

For the particle pair $J$4–$J$5, the conservative and non-conservative couplings are

$J$6

$J$7

so the conservative part is maximal at $J$8, can be suppressed near $J$9 and $0.06$00, and changes sign with $0.06$01. Experimentally, the normal-mode splitting of the outer pair tracks the mediated conservative coupling, and the coupling can be switched between clearly split and unresolved regimes by changing phase, distance, or the angle of the middle particle (Wu et al., 2024).

This platform is not described with bus-resonator language, but it fits the general concept closely. The coupler is a distinct dynamical mode inserted between two target modes; the direct matrix element is designed to vanish; and the interaction graph becomes reconfigurable through local control of the ancillary mode. A plausible implication is that “multi-mode tunable coupler” in this context denotes a programmable coupling graph for oscillator networks rather than a specific hardware primitive.

6. Design tradeoffs, metrics, and recurring misconceptions

The central tradeoff in all of these systems is between strong activation and clean isolation. Multi-mode structure gives more interference channels and therefore more control, but it also introduces more avoided crossings, more spectator states, and more calibration burden.

In superconducting circuits, the chief motivation for moving beyond a single-mode tunable coupler is the inability of that architecture to control the one- and two-excitation manifolds independently. The full-localization proposal states this explicitly and uses the quantities $0.06$02 to separate desired exchange from leakage channels (Jiang et al., 30 Sep 2025). The long-distance hybrid-mode coupler reaches large $0.06$03 and $0.06$04 tunability, but the authors also note that as the coupler gets longer the mode spacing shrinks, so $0.06$05, $0.06$06, and $0.06$07 must be co-designed carefully (Xu et al., 17 Jun 2025). The double-transmon coupler extends naturally to qubit-readout, bus, and waveguide settings, but it introduces additional flux-control and accidental-resonance constraints (Campbell et al., 2022).

In photonics, a major tradeoff is between localized spectral correction and modal overlap penalty. In the dual-ring FWM coupler, splitting one resonance into supermodes delocalizes the idler field across both cavities, which can reduce the nonlinear overlap coefficient to $0.06$08 of the uncoupled value and the corresponding efficiency to $0.06$09 of that of a hypothetical single dispersionless resonator (Gentry et al., 2014). In inverse-designed multimode cavity couplers, adding more target frequencies increases footprint and fabrication complexity; the six-frequency example required minimum features of about $0.06$10 nm, whereas the simpler SHG and SFG examples remained above about $0.06$11 nm (Jin et al., 2018).

In optofluidics, the tradeoff is between tunability and geometry sensitivity. The directional coupler is most sensitive to the coupling gap $0.06$12, while width, rib height, and slab thickness variations have much smaller impact; the two microfluidic tapers are inferred to reduce abrupt transition loss and contribute to the very low excess loss (Tang et al., 2019).

Three misconceptions recur. First, multimode does not always mean multiple spatial modes. The optofluidic coupler is dual-polarization rather than spatially multimode (Tang et al., 2019). Second, a distributed or modular coupler is not automatically a multimode coupler. The multi-chip floating superconducting coupler spans several chips while remaining a single effective tunable mode (Field et al., 2023). Third, a device may be functionally multi-channel without containing several linear coupler modes. The shared tunable coupler for three-qubit superconducting gates is physically a single anharmonic mode, yet it mediates several pairwise conditional phases and a genuine three-body conditional phase through different excitation manifolds (2206.12392).

The long-term direction suggested by this literature is not merely stronger on-off ratios. It is mode-selective interaction engineering: internal multimode structure used to localize qubits at idle, separate coupling behavior across excitation manifolds, control spectral bright and dark states, and extend tunable coupling to heterogeneous circuit, photonic, and mechanical platforms. In that narrower and more technical sense, the modern multi-mode tunable coupler is less a single component than a design principle: reconfigurable interaction through deliberately structured mode interference (Jiang et al., 30 Sep 2025, Xu et al., 17 Jun 2025).