Cavity-Mediated Interconnects: Theory & Platforms

• Cavity-mediated interconnects are engineered channels that use shared confined modes to couple spatially separate quantum systems, replacing direct wiring.
• They are implemented in platforms such as optical cavities with atoms, quantum dots, superconducting qubits, and mechanical resonators to achieve coherent state transfer and entanglement.
• Tuning factors like detuning, coupling strengths, and cavity geometry dictate effective non-local interactions while mitigating unwanted decoherence.

Cavity-mediated interconnects are engineered interaction channels in which a shared cavity mode, a multimode cavity manifold, or a cavity-dressed effective bus couples spatially separated degrees of freedom while suppressing direct short-range wiring. Across the current literature, the coupled objects include Λ-type atomic spins in multimode optical cavities, quantum dots linked by fermionic whispering-gallery modes in a two-dimensional electron gas, Kittel modes in YIG spheres, antiferromagnetic spin-wave oscillators in hematite slabs, levitated nanoparticles, and modular trapped-ion, neutral-atom, and superconducting-qubit processors (Gopalakrishnan et al., 2011, Nicolí et al., 2017, Lambert et al., 2015, Białek et al., 2022, Vijayan et al., 2023, Ramette et al., 2021, Hwang et al., 9 Feb 2026).

1. Effective mechanisms and theoretical descriptions

The defining feature of a cavity-mediated interconnect is the replacement of direct local coupling by a shared confined mode whose elimination or coherent occupation generates an effective non-local Hamiltonian. In multimode optical cavities with Λ-type atoms, adiabatic elimination of the excited state and cavity photons yields an XY Hamiltonian

$$H=H_{\mathrm{at}}+\sum_{i<j}J_{ij}\left(\sigma_x^i\sigma_x^j+\sigma_y^i\sigma_y^j\right),$$

with

$$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$

In dispersive magnon-cavity systems the corresponding effective interaction is a beam-splitter exchange,

$$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$

with

$$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$

whereas in a dot–cavity–dot molecule the off-resonant long-distance tunneling scales as $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$ (Gopalakrishnan et al., 2011, Lambert et al., 2015, Nicolí et al., 2017).

A closely related formulation appears in circuit QED and cavity-network models. For resonant qubits dispersively detuned from a multimode coplanar-waveguide cavity,

$J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$

and the parity relation $g_{1k}=(-1)^{k+1}g_{2k}$ causes higher harmonics to alternately reinforce and cancel the exchange. In short mirror-terminated fiber networks, rapid bouncing inside the linear-optics network dynamically decouples the cavities from the fiber while leaving an effective bilinear hopping

$$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$

with the two-cavity result $J_{\mathrm{eff}}=-|g|^2/J$. In analytical studies of chip-to-chip state transfer, a single electromagnetic waveguide mode is treated in the same Jaynes–Cummings single-mode picture; the model is mathematically identical to a single-mode cavity bus (Filipp et al., 2010, Kyoseva et al., 2011, Zaragoza et al., 30 Apr 2026).

These descriptions are not limited to exchange-type interactions. In cavity-modified mesoscopic transport, adiabatic elimination of a single cavity mode generates effective long-range hoppings $\tilde t_{ij}$ between distant sites of a Hall bar, QPC, or Aharonov–Bohm interferometer. In driven two-band Hubbard systems, projection onto the low-energy sector produces a cavity-mediated forward-scattering density–density interaction whose magnitude is controlled by screened interband denominators and excitonic resonances (Boriçi et al., 2024, Wang et al., 2023).

2. Platform realizations

Platform	Mediating cavity or bus	Representative scale
Semiconductor quantum dots	Spin-degenerate fermionic whispering-gallery modes in a GaAs/AlGaAs 2DEG central reservoir	$$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$0 at a dot separation of $$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$1; next-nearest-neighbor coupling at $$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$2 (Nicolí et al., 2017)
Ferrimagnetic YIG spheres	Microwave cavity virtual photons	splitting $$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$3 at $$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$4 and sphere separation $$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$5 (Lambert et al., 2015)
Hematite slabs	THz cavity fields in a two-slab Fabry–Pérot-like cavity	cooperative bright-mode splitting near $$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$6; resolvable coupling up to $$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$7 (Białek et al., 2022)
Levitated nanoparticles	Coherent scattering into a TEM$$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$8 cavity mode	$$J_{ij}=\sum_{\alpha} \frac{|g_\alpha \Omega|^2}{\Delta^2}\,\frac{1}{\delta_\alpha}\,\Xi_\alpha(\mathbf{r}_i)\,\Xi_\alpha(\mathbf{r}_j).$$9; maximal splitting $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$0 at $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$1 (Vijayan et al., 2023)
Superconducting transmons	Shared $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$2 CPW resonator with tunable qubit-cavity couplers	iSWAP and CZ within $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$3; residual $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$4 below a few kilohertz (Hwang et al., 9 Feb 2026)

The semiconductor implementation is notable because the cavity is fully on-chip and all-electronic: spin-degenerate fermionic whispering-gallery modes embedded in a Fermi sea mediate dot–dot hybridization while screening suppresses cavity charging and direct cross-talk. In contrast, microwave and THz magnetic implementations use electromagnetic standing waves as the bus, with coherent avoided crossings and bright/dark collective modes as the primary indicators of mediated coupling (Nicolí et al., 2017, Lambert et al., 2015, Białek et al., 2022).

Mechanical variants divide into two distinct classes. In levitated optomechanics, the optical cavity itself is the interconnect: photons coherently scattered by spatially separated nanoparticles generate a coupling that “does not decay with distance within the cavity mode volume.” By contrast, in coupled flexural optomechanical cavities the optical cavity is a local transducer, while complex-band-engineered serpentine links provide the interconnect; the measured splitting then decays exponentially with the number of serpentine cells (Vijayan et al., 2023, Alonso-Tomás et al., 15 Jun 2026).

3. Interaction topology, programmability, and collective phases

In multimode optical cavities, topology is set jointly by cavity geometry, mode content, and the atomic position distribution. A single-mode standing-wave cavity gives a factorized all-to-all interaction weighted by the local mode intensity; a ring cavity gives the translation-invariant oscillatory form

$$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$5

and confocal or concentric cavities sum many near-degenerate Hermite–Gaussian modes. When atom positions are random relative to the antinode structure and sufficiently many modes contribute, the coupling matrix becomes random in sign and magnitude. The resulting multimode XY model maps onto the $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$6 generalization of the Hopfield/Cook neural-network model: for roughly $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$7 the system is associative and retrieves a stored mode pattern, whereas for $$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$8 metastable states proliferate and a spin-glass phase emerges. Under the hard-core-boson mapping, the same platform realizes a Bose–Hubbard model with strongly disordered hopping but no on-site disorder, allowing random-singlet glass and Mott glass phases absent in conventional optical-lattice realizations (Gopalakrishnan et al., 2011).

Ring cavities with forward-diffracted sidebands extend this programmability into momentum space. In the three-mode case the intracavity field is

$$H_{\mathrm{int,eff}} \approx J (b_1 b_2^\dagger + b_1^\dagger b_2),$$9

and elimination of the cavity yields the atom-only kernel

$$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$0

Fourier filtering of the allowed transverse wavevectors $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$1 therefore directly sculpts the interaction spectrum. In the three-mode cavity, the self-organized stripe phase breaks continuous translational symmetry and supports a gapless Goldstone mode together with a finite-gap Higgs mode. For multimode filtering, commensurate sets of $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$2 produce droplet arrays, while incommensurate sets favor a single droplet BEC state (Krešić, 2024).

A solid-state analogue appears in the driven two-band Hubbard model coupled to an optical cavity. There the effective cavity-mediated interaction is

$$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$3

and coupling to an excitonic transition strongly enhances both the interaction magnitude and its momentum support. The reported enhancement factors are $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$4 near optimal $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$5 at modest detunings, and the interaction broadens across extended regions of the Brillouin zone because the Frenkel exciton resonance is dispersionless in $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$6 (Wang et al., 2023).

4. Modular quantum communication and state transfer

For modular quantum computing, cavity-mediated interconnects are used less as equilibrium couplers than as entanglement-distribution channels. A single optical cavity containing many local processors can route heralded single-photon transfers between arbitrary communication qubits, after which teleported gates connect the corresponding memory qubits. In the trapped-ion instantiation, a single cavity with realistic parameters transfers photons every few $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$7s and enables the any-to-any entanglement of 20 ion chains containing a total of 500 qubits in 200 $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$8s. In the neutral-atom instantiation, the same architecture fully connects a large array of thousands of neutral atoms, and with multiple overlapping cavities the connectivity is extendable to tens of thousands of qubits (Ramette et al., 2021).

Remote entanglement fidelity is not set only by optical loss. In cavity-assisted photonic links, photon recoil can entangle the photonic qubit with motion and reduce the path indistinguishability quantified by $$J \approx \frac{g_1 g_2}{2}\left(\frac{1}{\Delta_1}+\frac{1}{\Delta_2}\right),$$9. An analytical kick-operator formalism shows that operating in the bad-cavity regime with cavity decay rate exceeding atom-photon coupling rate, together with near-ground-state cooling, suppresses motion-induced infidelity below the $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$0 level required for efficient quantum networking. The recommended regime is

$t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$1

and detection-time filtering trades heralding rate for lower recoil error (Kikura et al., 20 Feb 2025).

Deterministic state-transfer studies sharpen the fidelity–latency trade-off. In a 5-node chain with STIRAP, using $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$2, $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$3, and $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$4, the reported fidelity is $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$5 after one hop and $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$6 after four hops, exceeding the naive multiplicative lower bound $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$7. A complementary analytical treatment of a single-mode waveguide bus derives closed-form occupation probabilities and a latency predictor

$t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$8

while identifying systematic low-fidelity regions when $t_{\rm eff}\sim \Omega_1\Omega_2/\epsilon_{\rm c}$9 is a ratio of odd integers (Rached et al., 19 Aug 2025, Zaragoza et al., 30 Apr 2026).

5. Control knobs and design-space exploration

The principal control parameters are detunings, coupling strengths, geometry, and mode selection. In multimode optical cavities, the global scale

$J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$0

is tuned by pump intensity and detuning, while cavity geometry determines whether the network behaves as a deterministic few-mode bus or a frustrated multimode graph. The same platform provides explicit recipes for either suppressing or leveraging frustration: reduce $J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$1, tailor atom positions to even antinodes, increase $J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$2 to narrow the active modal set, or choose ring geometry for structured XY order; conversely, use confocal or concentric degeneracy, many transverse modes, and positional randomness to enter the spin-glass or random-singlet-glass regimes (Gopalakrishnan et al., 2011).

Magnetic and photonic buses obey analogous but platform-specific rules. For YIG spheres in a microwave cavity, strong dispersive exchange with minimal cavity population is obtained for detunings several times larger than the largest $J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$3, specifically $J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$4–$J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$5, with the magnets placed at magnetic-field antinodes and the drive symmetry chosen to avoid dark states. In fiber-mediated cavity networks, the analogous decoupling condition is a separation of time scales,

$J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$6

which suppresses fiber population as $J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$7 while preserving the effective interaction $J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$8 (Lambert et al., 2015, Kyoseva et al., 2011).

Architectural design-space studies turn these controls into explicit operating targets. One benchmark defines

$J_{12}=\sum_k \frac{g_{1k}g_{2k}}{\Delta_k},$9

with $g_{1k}=(-1)^{k+1}g_{2k}$0, together with the cooperativity

$g_{1k}=(-1)^{k+1}g_{2k}$1

The corresponding sweeps favor strong coupling with $g_{1k}=(-1)^{k+1}g_{2k}$2, identify optimal regions when $g_{1k}=(-1)^{k+1}g_{2k}$3 exceeds $g_{1k}=(-1)^{k+1}g_{2k}$4 and $g_{1k}=(-1)^{k+1}g_{2k}$5 remains below $g_{1k}=(-1)^{k+1}g_{2k}$6, and show narrow $g_{1k}=(-1)^{k+1}g_{2k}$7 regions around $g_{1k}=(-1)^{k+1}g_{2k}$8 where success probability contours and infidelity contours simultaneously improve. In superconducting implementations, the same design logic is realized circuit-wise by tuning each coupler frequency so that the indirect qubit–coupler–resonator path destructively interferes with the residual direct qubit–resonator path at idle, then lowering only selected couplers into the operating band to activate a target pair (Rached et al., 2024, Hwang et al., 9 Feb 2026).

6. Signatures, limitations, and interpretive boundaries

Experimental identification of cavity-mediated interconnects is usually spectroscopic. In electronic dot–cavity–dot devices, three transport categories are resolved: clean right-angle crossings for independent dot transport, diagonal lines for single dot–cavity hybrids, and avoided crossings in all currents for dot–cavity–dot hybrids. In microwave magnonics, well-resolved avoided crossings persist even when both magnets are detuned far from cavity resonance, and one hybrid branch can become dark because the drive symmetry does not match the collective mode symmetry. In multimode atomic cavities, associative phases are marked by bright mode-selective cavity emission, whereas the spin-glass regime is characterized by no macroscopic cavity emission, slow non-ergodic relaxation, and a broad distribution of relaxation times (Nicolí et al., 2017, Lambert et al., 2015, Gopalakrishnan et al., 2011).

Spatially resolved diagnostics are equally important. In levitated optomechanics, heterodyne spectrograms directly extract the minimal splitting $g_{1k}=(-1)^{k+1}g_{2k}$9, verify its periodic phase dependence with inter-particle separation, and show no decay with distance inside the cavity mode volume. In THz hematite cavities, reflection maps versus temperature and slab separation reveal cooperative anticrossings and the square-root density rule

$$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$0

with $$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$1. In cavity-modified mesoscopic transport, the signatures are finite longitudinal resistance on integer Hall plateaus, cavity-induced inter-edge backscattering in current-density maps, and reshaped Aharonov–Bohm visibility $$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$2 at fixed flux periodicity $$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$3 (Vijayan et al., 2023, Białek et al., 2022, Boriçi et al., 2024).

A recurring misconception is that a single metric, usually cavity $$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$4, universally ranks interconnect performance. The literature instead shows platform dependence. In the fermionic whispering-gallery bus, a quality factor $$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$5 and photon decay $$H_{\mathrm{eff}} \approx \sum_{i<j} J_{ij} (a_i^\dagger a_j + a_j^\dagger a_i),$$6 are not the relevant figures of merit; coherent hybridization and tunnel splittings in transport spectra quantify performance. In remote atom-photon networking, by contrast, the bad-cavity regime can be preferable because it suppresses recoil-induced which-way information. In multimode atomic and solid-state settings, disorder, multimode crowding, and sign changes are not merely parasitic; they can be the intended resource for associative memory, glassiness, broadened momentum-space coupling, or frustrated many-body simulation. This suggests that “cavity-mediated interconnects” is best understood not as a single hardware class, but as a design paradigm for engineering non-local coupling kernels through shared confined modes (Nicolí et al., 2017, Kikura et al., 20 Feb 2025, Wang et al., 2023).