Cavity and Circuit QED Frontiers

Status: coherent single-emitter cavity coupling, cavity-enhanced emission, dispersive circuit-QED control, waveguide-mediated collective radiance, ultrastrong light–matter coupling, and bosonic storage are established. Scalable multimode interfaces, delayed-feedback waveguide networks, error-corrected bosonic memories, microwave–optical transducers, and modular links are active. Loss-tolerant quantum transduction between remote processors, an operational advantage from ultrastrong-coupling ground-state correlations, and fault-tolerant processors built from present bosonic-memory devices are not established.

Last reviewed: 26 July 2026. Coupling ratios, memory gains, conversion efficiencies, added-noise values, and array sizes are date-sensitive. A resolved spectral anticrossing is not by itself a coherent gate; sub-one-photon added noise is not by itself a high-fidelity transducer; and a logical-memory gain is not by itself a fault-tolerant computer.

Core Question

When does an engineered electromagnetic environment become a useful quantum interface rather than merely a modified spectrum?

The question spans apparently different devices:

• one atom between optical mirrors;
• a quantum dot in a nanophotonic waveguide;
• a transmon coupled to a microwave resonator;
• many cyclotron transitions coupled to several terahertz modes;
• a bosonic cavity stabilized as a logical memory; and
• a mechanical, electro-optic, spin, or atomic converter between microwave and optical frequencies.

All of them redistribute excitations among matter, confined fields, propagating fields, pumps, and unobserved reservoirs. Their common evidence chain is

$$\begin{aligned} \text{Hamiltonian design} &\longrightarrow \text{calibrated coupling and loss} \\ &\longrightarrow \text{state preparation} \longrightarrow \text{dynamics or scattering} \\ &\longrightarrow \text{output record} \longrightarrow \text{task-level claim}. \end{aligned}$$

An anticrossing constrains the first two arrows. A time-domain swap tests more of the chain. A memory benchmark tests preparation, repeated control, storage, and recovery. A network claim additionally requires a channel, synchronization, and end-to-end verification.

The frontier therefore has six coupled questions:

1. How large can coherent coupling become relative to every relevant frequency and decay rate?
2. When must one replace a single-mode, rotating-wave description by a multimode, counter-rotating, gauge-consistent theory?
3. Can propagating continua mediate directional, delayed, and collective interactions without uncontrolled loss?
4. Can superconducting circuits turn resonators into long-lived logical degrees of freedom rather than passive hardware?
5. Can hybrid converters preserve quantum states while crossing frequency, temperature, and material boundaries?
6. Which benchmark demonstrates useful storage, transfer, or processing rather than an isolated component record?

Why It Matters

Confinement turns weak microscopic interactions into controllable ones

A resonator concentrates vacuum fluctuations into a selected mode and gives an emitter repeated opportunities to interact with the same field. A one-dimensional waveguide routes a large fraction of spontaneous emission into a monitored continuum. A Josephson junction supplies a strong, fabricated nonlinearity. These mechanisms make interactions at the single-excitation level observable and controllable.

That control underlies:

• nondestructive measurement and spin squeezing;
• deterministic emission and absorption of shaped photons;
• microwave and optical quantum-network nodes;
• nonlinear scattering at the few-photon level;
• analog simulation of driven, open, and long-range systems;
• dispersive measurement and feedback;
• bosonic quantum error correction; and
• interfaces between superconducting hardware and telecom fibre.

The same coupling creates new failure modes

Increasing coupling does not monotonically improve a device. Larger $g$ can invalidate the rotating-wave approximation, activate unwanted levels, couple to parasitic modes, enhance Purcell decay, or make a local Markov model inadequate. Stronger pumps can improve conversion while heating a mechanical or microwave mode. Increasing a cat-state amplitude suppresses one logical error while amplifying another. Adding waveguide ports increases connectivity while opening radiative loss channels.

Useful design is therefore multidimensional. A representative performance vector is

$$\mathcal P = \left( g,\kappa,\gamma,\gamma_\phi, \eta_{\rm ext},n_{\rm add},B,\tau, F_{\rm op},p_{\rm leak},d \right),$$

where the entries may denote coherent coupling, resonator loss, emitter relaxation, pure dephasing, external efficiency, added noise, bandwidth, latency, operation fidelity, leakage, and logical dimension. No single scalar orders all platforms.

Interfaces determine whether modular architectures are credible

Superconducting circuits are naturally local and microwave-frequency. Optical photons are naturally long-distance carriers. Matter memories can be long lived but difficult to network. A modular architecture becomes more than a diagram only when its interfaces have quantified:

• conversion or collection probability;
• added noise referred to a declared port;
• bandwidth and temporal-mode acceptance;
• phase stability;
• pump-induced heating and duty cycle;
• heralding or postselection;
• storage and retrieval fidelity; and
• an end-to-end entanglement or channel benchmark.

This is why a one-kilometre coherent signal link, a sub-one-photon-noise converter, and a beyond-break-even memory are important but logically different milestones.

Canonical Boundary

This page is the canonical home for dated frontier comparisons and open questions across cavity, circuit, multimode, waveguide, hybrid-interface, and bosonic-memory research.

It does not repeat the canonical derivations:

• Cavity QED develops vacuum coupling, cooperativity, Purcell enhancement, polaritons, and conventional strong-coupling diagnostics.
• Jaynes–Cummings Model derives the rotating-wave dressed spectrum and coherent exchange.
• Input–Output Theory Overview derives port boundary conditions and scattering amplitudes.
• Cavity-QED Platforms owns resonator geometry, emitter placement, stabilization, calibration, and platform engineering.
• Circuit QED Overview owns superconducting artificial atoms, dispersive control, readout, cryogenics, leakage, and hardware benchmarking.
• Open-System Cavity QED and Open-System Circuit QED own master equations, trajectories, measurement backaction, and receiver models.
• Strong Coupling and Memory Effects owns the distinction between coupling strength and non-Markovianity.

Results are summarized here only far enough to compare current claims. Follow the links for the canonical formalism.

Minimal Background

Strong coupling is a comparison of rates

For one nearly resonant emitter and one lossy cavity mode, the central rate set is

$$\left\{g,\kappa,\gamma,\gamma_\phi\right\}.$$

Here $g$ is the coherent vacuum coupling, $\kappa$ the cavity energy-decay rate, $\gamma$ the emitter population-decay rate into noncavity channels, and $\gamma_\phi$ the emitter pure-dephasing rate. Authors differ over whether quoted linewidths are half widths, full widths, field-decay rates, or energy-decay rates. A comparison is meaningful only after converting all quantities to one convention.

One common cooperativity convention is

$$C=\frac{4g^2}{\kappa\gamma}.$$

$C>1$ says that coherent coupling competes favorably with a particular pair of loss rates. It does not guarantee a clean normal-mode doublet, high-fidelity swaps, efficient extraction through a desired port, or small pure dephasing.

For a convention in which $\kappa$ is the cavity energy-decay rate and $\gamma_\perp$ is the emitter coherence-decay rate, a simple linear-response matrix is

$$M = \begin{pmatrix} \omega_c-i\kappa/2 & g \\ g & \omega_a-i\gamma_\perp \end{pmatrix}.$$

At resonance, its complex mode frequencies are

$$\widetilde\omega_\pm = \omega_0 -\frac{i}{2} \left( \frac{\kappa}{2}+\gamma_\perp \right) \pm \sqrt{ g^2 -\frac{1}{4} \left( \frac{\kappa}{2}-\gamma_\perp \right)^2 },$$

with $\omega_c=\omega_a=\omega_0$. The square-root structure explains why resolvable splitting, underdamped time-domain exchange, and $C>1$ are related but not equivalent criteria.

Strong, ultrastrong, and deep-strong answer different questions

Conventional strong coupling compares $g$ with losses. Ultrastrong coupling compares $g$ with bare oscillation frequencies. A commonly used normalized ratio is

$$\eta=\frac{g}{\omega_c},$$

or an analogous ratio using the matter frequency. Values around $\eta\gtrsim0.1$ are conventionally called ultrastrong, while $\eta\gtrsim1$ is often called deep strong. These are useful labels, not phase boundaries.

The quantum Rabi model retains counter-rotating terms:

$$\frac{H}{\hbar} = \omega_c a^\dagger a +\frac{\omega_q}{2}\sigma_z +g(a+a^\dagger)\sigma_x +\frac{H_{\rm sp}}{\hbar},$$

where $H_{\rm sp}$ denotes the diamagnetic or self-polarization contribution required by the chosen microscopic description. In this regime:

• excitation number is not conserved;
• the ground state is dressed and may contain virtual bare excitations;
• Bloch–Siegert shifts and non-RWA selection rules become important;
• a two-level truncation may fail;
• gauge choice and material sum rules matter; and
• switching or modulating the coupling can become part of the observable.

A large $\eta$ says nothing directly about $\kappa$, $\gamma$, addressability, or gate fidelity. Moreover, stationary virtual photons in a dressed ground state are not ordinary output photons. Detectable emission requires an operational protocol such as modulation, a quench, or another coupling that converts correlations into propagating excitations.

Multimode coupling is more than several anticrossings

A single-mode model fails when several field modes contribute materially over the emitter bandwidth or coupling scale. A generic multimode Hamiltonian is

$$\begin{aligned} H &= H_{\rm matter} +\sum_m\hbar\omega_m a_m^\dagger a_m \\ &\quad +\sum_m\hbar g_m (a_m+a_m^\dagger)X +H_{\rm sp}, \end{aligned}$$

where $X$ is a matter operator. The associated discrete spectral density can be written

$$J(\omega) = 2\pi\sum_m |g_m|^2 \delta(\omega-\omega_m).$$

When the free spectral range $\Delta_{\rm FSR}$ is comparable to $g$, an emitter linewidth, or a drive bandwidth, several modes may participate. The mode profiles and polarizations matter as much as the frequencies. At still larger mode density, a Green-tensor or continuum description is usually more natural.

Multimode models raise issues hidden by a single oscillator:

• ultraviolet cutoffs and parameter renormalization;
• the completeness of the electromagnetic basis;
• causality and propagation delay;
• interference among ports or coupling points;
• bound states near band edges;
• memory kernels and nonexponential decay; and
• whether a fitted mode count corresponds to independently controllable channels.

Waveguide QED redistributes spontaneous emission

For an emitter coupled to a guided continuum, a central metric is the beta factor

$$\beta = \frac{\Gamma_{\rm 1D}}{\Gamma_{\rm tot}},$$

where $\Gamma_{\rm 1D}=\Gamma_R+\Gamma_L$ is emission into the selected right- and left-propagating guided modes. Directionality can be summarized by

$$\mathcal D = \frac{\Gamma_R-\Gamma_L} {\Gamma_R+\Gamma_L}.$$

High $\beta$ and high $|\mathcal D|$ answer different questions. A device can emit almost entirely into a waveguide but equally in both directions, or emit directionally while losing substantial population elsewhere.

For two coupling locations separated by $L$, the propagation delay is

$$\tau=\frac{L}{v_g}.$$

If $\Gamma\tau\ll1$ and relevant drive envelopes vary slowly over $\tau$, the delay may be eliminated in a Markov approximation. If $\Gamma\tau$ is order unity, delayed self-interaction, frequency-dependent coupling, and nonexponential dynamics can become observable. “Giant atoms” realize this regime by coupling one artificial atom at multiple separated points.

Scattering amplitudes are not automatically gates

A coherent probe can reveal an emitter-induced complex transmission coefficient

$$t(\omega)=|t(\omega)|e^{i\phi(\omega)}.$$

A large phase $\phi$ is valuable evidence of strong interaction. A two-photon gate requires more. One must characterize the complete few-photon scattering map, including:

• photon-number dependence;
• loss and reflection;
• temporal- and spectral-mode distortion;
• emitter dephasing;
• residual emitter–photon entanglement;
• distinguishability; and
• unconditional process fidelity.

The phase of a weak coherent field averages over vacuum, one-photon, and multiphoton components. It cannot alone establish a controlled phase on two photonic qubits.

Circuit QED adds levels, control lines, and engineered reservoirs

A transmon is not exactly a two-level atom. Its weak anharmonicity permits fast control but creates leakage. Dispersive couplings permit quantum nondemolition readout and bosonic control but also produce measurement-induced dephasing, Purcell decay, residual couplings, and spectator shifts. Pumps activate useful interactions while creating Stark shifts, quasiparticles, heating, and mixing products.

The frontier is not merely to increase the number of fabricated qubits. It is to preserve a calibrated effective Hamiltonian while scaling:

$$\text{control} +\text{measurement} +\text{reset} +\text{routing} +\text{error correction}.$$

This page treats circuit QED mainly through bosonic memories, waveguide interfaces, and hybrid links. The canonical hardware account remains in Circuit QED Overview.

A transducer must report efficiency and noise together

For two pumped resonant modes coupled through an intermediate degree of freedom, a common idealized on-resonance internal conversion efficiency is

$$\eta_{\rm int} \simeq \frac{4C_1C_2} {(1+C_1+C_2)^2},$$

where $C_1$ and $C_2$ are cooperativities for the two conversion legs. External coupling and propagation multiply this value:

$$\eta_{\rm ext} = \eta_1\eta_2\eta_{\rm int}\eta_{\rm path}.$$

The exact expression is architecture- and convention-dependent, but the lesson is general: high internal efficiency can coexist with poor end-to-end efficiency.

For a phase-insensitive channel, an input-referred added-noise number $n_{\rm add}$ is also essential. The label quantum enabled is often used when $n_{\rm add}<1$ photon in a declared operating mode. That is an important noise boundary, not a complete channel certification. Low efficiency can replace the input by vacuum, and frequency-dependent noise can be hidden by an overly broad scalar average.

A credible transducer report includes

$$\left( \eta_{\rm ext}, n_{\rm add}, B, \text{temporal mode}, \text{pump power}, \text{duty cycle} \right),$$

plus uncertainty and calibration references.

A memory benchmark needs a declared reference

For a logical memory, a common gain is

$$G=\frac{\tau_L}{\tau_{\rm ref}},$$

where $\tau_L$ is a logical lifetime and $\tau_{\rm ref}$ is the lifetime of the best relevant unencoded physical alternative. The result is meaningful only if the state ensemble, error metric, encoding and decoding overhead, and reference hardware are specified.

Memory break-even is narrower than fault-tolerant computation. A complete processor also needs:

• fault-tolerant logical state preparation;
• a logical gate set;
• repeated syndrome extraction;
• measurement and reset;
• leakage management;
• scalable decoding; and
• logical error suppression as resources increase.

Bosonic encodings are especially rich because oscillator loss, dephasing, ancilla errors, and control nonlinearities transform differently under cat, binomial, and Gottesman–Kitaev–Preskill encodings.

An evidence ladder prevents category errors

The following ladder is useful across all six frontier themes:

1. Parameter evidence: independently calibrated frequencies, coupling rates, linewidths, temperatures, and port efficiencies.
2. Spectral evidence: anticrossings, linewidth changes, sidebands, and complex scattering amplitudes.
3. Dynamical evidence: swaps, revivals, delayed feedback, conditional trajectories, and time-resolved correlations.
4. State evidence: tomography, entanglement witnesses, nonclassical correlations, or logical-state survival.
5. Task evidence: a memory, transfer, gate, readout, or sensing benchmark against a declared reference.
6. Scaling evidence: improvement with mode count, code distance, link distance, or module count while the full metric vector remains controlled.

Each rung supports claims at that rung and below. It does not automatically support the next one.

What Is Known

Coherent single-emitter strong coupling is mature

Established. Optical atoms, quantum dots, microwave Rydberg atoms, and superconducting artificial atoms have all exhibited normal-mode splitting, vacuum Rabi exchange, and cavity-modified emission. The underlying single-mode theory is quantitatively predictive when:

• one transition and one mode dominate;
• linewidth conventions are explicit;
• drive power is low enough for the intended model;
• emitter motion or spectral diffusion is included when relevant; and
• useful-port coupling is separated from parasitic loss.

The mature question is no longer whether strong coupling exists. It is how to make it uniform, multiplexed, stable, and useful in a complete protocol.

Ultrastrong coupling is experimentally established

Established. Ultrastrong and, in several artificial systems, deep-strong coupling have been observed through spectra and dynamics that require counter-rotating interactions. Platforms include superconducting circuits, intersubband and cyclotron polaritons, molecular and phonon polaritons, and other collective solid-state excitations.

Three conclusions are firm:

1. The rotating-wave approximation can fail quantitatively.
2. The identity of a “photon” or “matter excitation” becomes basis- and protocol-dependent in a dressed system.
3. Gauge consistency, self-polarization terms, and level truncation can affect predictions.

What is not firm is that every large normalized splitting provides a useful quantum-information resource. Many record coupling ratios involve collective material excitations at finite temperature or with limited state-level control.

Multimode ultrastrong coupling now has direct three-dimensional evidence

Established spectral result; active operational interpretation. In 2025, Tay and collaborators reported multimode ultrastrong coupling between terahertz modes of a three-dimensional photonic-crystal cavity and the cyclotron resonance of a Landau-quantized gallium-arsenide two-dimensional electron gas. Polarization-dependent spectra required spatial mode profiles and were described by an extended multimode Hopfield model.

This result establishes controlled multimode ultrastrong spectroscopy in that collective solid-state platform. It does not establish a single-emitter multimode qubit, arbitrary multimode quantum processing, or a realized superradiant phase transition. The paper discusses possible ground-state correlations; extracting those correlations as an operational resource remains an active problem.

Waveguide QED supports collective bright and dark states

Established. Emitters coupled to a common one-dimensional continuum can exchange excitations and acquire collective decay rates. Constructive interference produces superradiant bright states; destructive interference produces subradiant or dark states.

In 2022, a four-transmon waveguide-QED experiment coherently controlled a collective dark state using local drives. Its decay time exceeded those of the waveguide-limited single qubits by more than two orders of magnitude. That qualifier matters: suppression of one radiative channel does not remove material loss, dephasing, control error, or finite-size imperfections.

Giant artificial atoms create frequency-selective coupling

Established. A superconducting artificial atom coupled at multiple well-separated positions can interfere with its own emission. Experiments have shown large tunable on–off ratios, engineered frequency dependence, and waveguide-mediated interactions that persist while selected radiative channels are suppressed.

The 2020 multi-point superconducting experiment demonstrated decoherence-free interactions and entangling dynamics in engineered geometries. “Decoherence free” here names protection from the designed waveguide channel in the relevant subspace. It is not a claim of zero total decoherence.

Few-photon waveguide nonlinearities are directly measurable

Established measurement; active gate problem. In 2024, a quantum dot in a planar nanophotonic waveguide produced a directly measured optical phase shift of $0.19\pi\pm0.03\pi$, about $34^\circ$, for a weak coherent probe. The response saturated at the single-photon scale.

This is strong evidence for a coherent, power-dependent emitter response in an integrated waveguide. It is not a demonstrated deterministic two-photon controlled-phase gate. Such a gate would require a specified photonic encoding, two-photon input states, full modal characterization, loss accounting, and an unconditional process benchmark.

Bosonic memories have crossed several break-even boundaries

Established for declared memory tasks. Superconducting cavities controlled by transmon ancillas have supported repeated error correction of cat, binomial, and Gottesman–Kitaev–Preskill states. The field has progressed from extending one encoded qubit lifetime to operating beyond the best relevant physical memory and to encoding higher-dimensional logical systems.

In 2025, an oscillator GKP experiment demonstrated an error-corrected logical qutrit and ququart with memory gains

$$G_3=1.82\pm0.03, \qquad G_4=1.87\pm0.03$$

relative to the best physical qutrit and ququart in the same system. Reinforcement learning optimized the control policy. Reported dominant limitations included transmon bit flips, cavity photon loss, and cavity dephasing; the latter was associated primarily with a measured transmon thermal population of $2.2\%\pm0.1\%$.

Also in 2025, a superconducting device combined stabilized cat qubits with an outer repetition code. A distance-five memory corrected phase flips below the measured repetition-code threshold while retaining physical suppression of bit flips. Its best average logical error per cycle, $1.65\%\pm0.03\%$, was comparable to the average $1.75\%\pm0.02\%$ of two distance-three subsections. The experiment showed the expected stronger phase-error scaling with distance, but not an unambiguous decrease of the total optimum error from distance three to five.

Both are memory milestones. Neither experiment by itself demonstrates a universal fault-tolerant logical gate set.

Quantum-enabled hybrid conversion has been demonstrated

Established component results; active end-to-end problem. Several architectures have entered regimes where useful conversion and low added noise coexist:

• A 2023 cold-rubidium hybrid device coupled atoms to an optically accessible superconducting resonator and an optical cavity at $5\ {\rm K}$. It reported $58\%\pm11\%$ internal millimetre-wave-to-optical conversion efficiency, a $360\pm20\ {\rm kHz}$ bandwidth, and $0.6$ photons of added thermal noise.
• A 2025 silicon electro-optomechanical transducer reported continuous-wave input-referred added noise $n_{\rm add}=0.58$ and an upconversion rate of $0.47$–$1.9\ {\rm kHz}$.
• A 2025 ytterbium-171-doped yttrium orthovanadate device achieved percent-level microwave–optical efficiency without an engineered optical cavity and input-referred added noise as low as $1.24\pm0.09$ photons. It also showed interference between light from two simultaneously operated transducers.
• A 2025 electro-optic device reached up to $1.18\%$ conversion efficiency with low added microwave noise and used the converted field to drive Rabi oscillations of a superconducting qubit.

These results establish low-noise conversion, coherent control, or interference in the stated operating modes. They do not yet establish high-fidelity transfer of arbitrary qubit states through a low-loss optical network.

What Is Actively Developing

Parallel cavity arrays

Active. A single global cavity mode can measure or couple an atom array, but it limits spatial addressability. In 2026, a free-space cavity-array microscope assigned individual atoms to individual cavities across a two-dimensional array of more than 40 modes. It achieved above-unity peak cooperativity, homogeneous coupling, and fast nondestructive parallel readout on millisecond timescales. Fibre-array readout was demonstrated as a networking proof of principle.

The same report realized a next-generation optical structure with more than 500 cavities and nearly ten times higher finesse as an outlook. That fabricated mode count is not equivalent to a demonstrated 500-atom, simultaneously calibrated, strongly coupled network. The active tasks are:

• loading one controlled emitter into each intended mode;
• calibrating the full cooperativity distribution;
• suppressing mode cross-talk and frequency disorder;
• parallelizing control and readout electronics;
• connecting modes to low-loss external channels; and
• showing a task whose performance improves with array size.

Multimode quantum dynamics

Active. Multimode experiments are moving from fitting several spectral branches toward controlling mode–mode correlations and nonlocal matter interactions. Promising directions include:

• cavity-mediated interactions with programmable spatial kernels;
• synthetic dimensions built from frequency modes;
• multimode memories and routers;
• band-edge bound states;
• dynamical extraction of ultrastrong-coupling correlations; and
• cavity-modified materials and chemistry.

The hard part is validation. A model with many adjustable modes can fit a spectrum while misidentifying the microscopic coupling. Spatially resolved fields, polarization dependence, sum rules, and out-of-sample dynamical tests are increasingly important.

Directional and topological waveguide interfaces

Active. Chiral coupling, interference between separated ports, and photonic band engineering can route emission asymmetrically. In 2026, a superconducting Rice–Mele waveguide with an Xmon qubit realized qubit-frequency-controlled directional edge states. The authors estimated $99.996\%$ directionality fidelity, constrained by the measurement noise floor, and observed the states through transmission and qubit emission.

This is a strong finite-device directionality result. “Theoretically zero opposite population” is a model property, while the reported fidelity is noise-floor limited. A scalable bus still must demonstrate:

• low insertion loss;
• robustness to fabrication disorder and frequency crowding;
• compatible state preparation and release;
• negligible unwanted dynamical phases;
• operation with multiple simultaneous emitters; and
• an end-to-end entangling or transfer benchmark.

Delayed and non-Markovian waveguide networks

Active. Longer links, slow-light sections, giant atoms, and multiple coupling points make propagation time a dynamical variable. This opens feedback without measurement, structured reservoirs, and photon-bound-state physics. It also invalidates familiar local master equations.

Current theory uses delay-differential equations, matrix-product states, time-bin tensor networks, pseudomodes, and scattering methods. The experimental frontier is to tune $\Gamma\tau$ across unity while independently measuring loss, dispersion, and dephasing. A revival caused by a reflected pulse should not be relabelled intrinsic information backflow without a complete channel model.

Bosonic logical operations

Active. Memory protection is becoming reproducible enough that attention is shifting to logical operations and architectures. Central goals include:

• bias-preserving entangling gates;
• fault-tolerant preparation of non-Gaussian resource states;
• repeated error correction during gates;
• autonomous stabilization with low pump-induced error;
• leakage and ancilla-fault detection;
• modular coupling between bosonic cavities; and
• logical benchmarking that includes encoding, decoding, and reset.

Different bosonic codes make different compromises. Cat codes exploit noise bias. GKP codes turn small phase-space displacements into correctable shifts but require finite-energy grid states and precise control. Binomial codes use finite Fock support and tailored error detection. No code dominates independently of hardware noise and task.

Quantum transduction as a complete channel

Active. The component frontier is moving toward simultaneous optimization of

$$\eta_{\rm ext}, \quad n_{\rm add}, \quad B, \quad \text{pump heating}, \quad \text{stability}.$$

In 2026, two frequency-matched aluminium-nitride electro-optic transducers in separate dilution refrigerators carried a coherent signal over $1\ {\rm km}$ of telecom fibre. The experiment reported an overall $80\ {\rm dB}$ transduction-efficiency improvement over commercial electro-optic modulators. The demonstrated result was coherent signal transfer; the authors described it as paving the way toward a fully quantum-enabled link. Arbitrary quantum-state transfer or remote entanglement was not demonstrated.

The next decisive experiments will combine low noise and efficiency with a state-level benchmark: transfer of nonclassical microwave states, entanglement distribution between independently controlled nodes, or a measured quantum channel fidelity above the best classical strategy for the declared ensemble.

Hybrid systems with useful division of labour

Active. Hybrid designs are credible when each subsystem supplies a specific resource:

• superconducting circuits provide fast nonlinear control;
• optical cavities and waveguides provide collection and routing;
• spins or rare-earth ions provide narrow transitions or storage;
• mechanical modes bridge frequency scales;
• Rydberg atoms provide strong microwave dipoles and optical access; and
• telecom photons provide long-distance propagation.

Hybridization also imports every subsystem’s loss and calibration problem. The frontier is therefore not to couple the largest number of materials. It is to show that the combined task outperforms the best single-platform alternative after interface overhead.

What Is Debated or Uncertain

Which ultrastrong-coupling effects are operational resources?

Active and partly controversial. Dressed ground-state correlations, modified vacuum fluctuations, and non-RWA selection rules are well-defined within a consistent microscopic model. Their usefulness depends on a protocol that prepares, changes, and measures the system faster than decoherence while properly defining input and output excitations.

Claims should distinguish:

• a fitted ground-state correlation;
• a measurable response caused by that correlation;
• extraction of real propagating excitations;
• entanglement verified in accessible subsystems; and
• an advantage in sensing, simulation, or information processing.

The no-go conditions for equilibrium superradiant phase transitions also depend on microscopic degrees of freedom, self-polarization terms, spatial structure, and truncation. Removing one assumption from a theorem does not demonstrate the proposed phase.

How should gauge choice and truncation be handled?

Established problem; active best practice. Gauge-equivalent microscopic theories can produce inequivalent few-level models after premature truncation. In ultrastrong and multimode regimes, a defensible calculation should state:

1. the microscopic variables and gauge;
2. the material-level truncation;
3. the electromagnetic mode basis and cutoff;
4. the self-polarization or diamagnetic contribution;
5. which parameters are bare, renormalized, or fitted; and
6. convergence under increasing levels and modes.

Agreement with one spectrum is necessary but may not identify a unique effective model.

What counts as a quantum memory?

Active. Lifetime extension, average state fidelity, entanglement fidelity, logical error per cycle, and break-even gain are not interchangeable. Postselected storage can answer a different question from unconditional storage. A code optimized for six cardinal states can behave differently on a larger state ensemble.

A strong memory claim declares:

$$(\text{state ensemble}, \text{metric}, \text{reference}, \text{overhead}, \text{accepted trials}).$$

Whether a protected bosonic mode is best called a memory, qubit, qudit, or small processor should follow the demonstrated operation set, not the dimension of its underlying oscillator.

Can high directionality survive network scaling?

Conjectural at large scale. Directionality in a calibrated finite device can be excellent. Scaling introduces disorder, propagation loss, reflections, frequency crowding, and unwanted loops. Topological language does not make state preparation, coupling, or output collection immune to all errors.

A scalable claim needs distributions, not only a best device:

• directionality across sites and frequencies;
• insertion loss;
• cross-talk matrix;
• sensitivity to disorder;
• switching speed;
• residual population after release; and
• network-level process fidelity.

Which transducer benchmark predicts entanglement distribution?

Active. The pair $(\eta,n_{\rm add})$ is indispensable but may still omit modal mismatch, phase noise, nonstationarity, and pump correlations. Added noise must be referred to a named input, bandwidth, and temporal mode. Efficiency must specify internal, on-chip, fibre-to-fibre, or end-to-end.

For a specific state ensemble, the strongest test is a reconstructed channel or entanglement benchmark. Until then, phrases such as “quantum enabled” and “single-photon level” should retain their authors’ operational definitions rather than be treated as universal certification.

Will bosonic hardware reduce total fault-tolerance overhead?

Conjectural. Bosonic codes can encode protection into one oscillator and can exploit biased noise. The oscillator still needs nonlinear control, ancillas, pumps, couplers, measurement, reset, and decoding. Hardware savings at the data-code level may be offset by calibration or gate overhead.

The decisive comparison is a complete logical resource estimate under measured noise, including control hardware and throughput. Comparing one bosonic mode with one bare two-level qubit is rarely sufficient.

Key Experimental Platforms

Platform	Native advantage	Frontier observable	Dominant caveat
Free-space optical cavities with atoms	clean natural transitions, optical ports, long atom–surface distance	uniform single-atom cooperativity, parallel readout, network collection	mechanical stability, mode matching, loading
Nanophotonic cavities and waveguides	small mode volume, high beta factor, chip integration	indistinguishable emission, phase response, directional scattering	spectral diffusion, surface noise, fabrication disorder
Three-dimensional microwave cavities	long bosonic lifetimes, high-purity modes	logical memory gain, repeated QEC, multimode control	ancilla-induced loss and thermal population
Planar superconducting circuits	strong tunable coupling and fast control	gates, shaped emission, chiral or giant-atom coupling	dielectric loss, leakage, wiring, pump errors
Terahertz polaritonic cavities	large collective normalized coupling	multimode polariton spectra and ground-state correlations	limited single-excitation control, material loss
Photonic-crystal and metamaterial waveguides	tailored dispersion and band edges	bound states, directionality, delayed exchange	disorder, finite size, outcoupling
Electro-optomechanical converters	microwave–optical frequency bridge	external efficiency, input-referred noise, bandwidth	pump heating, mechanical loss, duty cycle
Electro-optic converters	direct coherent conversion, no mechanical intermediary	frequency matching and long-fibre coherent links	small single-photon coupling, optical heating
Rare-earth spin ensembles	narrow atomic resonances and reproducible optical frequencies	low-noise conversion and multi-device interference	inhomogeneity, modest efficiency, initialization
Rydberg-atom hybrids	strong millimetre-wave dipoles plus optical access	internal conversion and hybrid entanglement	cryogenic optical access, ensemble complexity

A platform comparison should preserve denominators. “Efficiency” without port definitions, “coherence” without a state ensemble, and “coupling” without a linewidth convention are not comparable quantities.

A practical reporting matrix

For cavity or waveguide interfaces, report:

$$\left( g,\kappa_{\rm useful},\kappa_{\rm loss}, \gamma,\gamma_\phi,\beta,\mathcal D \right).$$

For converters, report:

$$\left( \eta_{\rm int},\eta_{\rm ext},n_{\rm add}, B,P_{\rm pump},\text{duty cycle} \right).$$

For memories, report:

$$\left( \tau_L,\tau_{\rm ref},G, \epsilon_L,\text{cycle time}, \text{overhead} \right).$$

The uncertainty, calibration method, accepted-event rule, and date belong beside every vector.

Key Theoretical Tools

Open quantum systems and input–output theory

Lindblad master equations, quantum Langevin equations, and input–output relations remain the workhorses for few-mode Markovian devices. They connect internal operators to measurable reflection, transmission, fluorescence, and homodyne records.

Use them when the reservoir correlation time and propagation delay are short relative to system dynamics. Check complete positivity, port normalization, and linewidth conventions. The relevant canonical pages are Input–Output Theory and What Non-Markovian Means.

Few-photon scattering theory

Single- and multiphoton $S$ matrices expose transmission, reflection, bound components, spectral entanglement, and mode distortion. They are essential when a classical transfer function is being promoted to a photonic gate claim.

A one-photon amplitude constrains a linear channel. A two-photon connected amplitude is needed to identify genuine photon–photon interaction:

$$S^{(2)} = S^{(1)}S^{(1)} +S_{\rm connected}^{(2)}.$$

Gate usefulness depends on whether the connected term produces the intended conditional phase without unacceptable loss or wavepacket distortion.

Green tensors and quasinormal modes

Structured electromagnetic environments are naturally described through a dyadic Green tensor. It encodes the local density of states, coherent exchange, collective decay, and propagation. Quasinormal modes provide a useful reduced representation for open resonators, provided normalization and background continua are handled carefully.

These tools are preferable to manually adding modes until a spectrum fits. They also clarify when a “cavity mode” is leaky, overlapping, or inseparable from a continuum.

Hopfield, quantum Rabi, and gauge-consistent circuit models

Quadratic Hopfield models describe collective bosonic matter excitations coupled to several field modes. Quantum Rabi and multilevel circuit models describe nonlinear few-level matter. Microscopic circuit quantization and material sum rules determine self-polarization terms and parameter renormalization.

Convergence tests should increase both matter levels and field modes. A model that converges in one truncation while omitting the compensating self-polarization term is not automatically physical.

Delay equations and tensor networks

Time-delayed feedback can be represented with delay-differential equations, time bins, matrix-product states, or enlarged Markovian embeddings. Tensor networks are particularly useful when only a bounded number of excitations occupies a long one-dimensional field.

The computational cost is controlled by temporal entanglement, not simply link length. Strong drives, many emitters, and repeated feedback can increase the required bond dimension sharply.

Floquet and driven-dissipative methods

Pumps activate sidebands, longitudinal couplings, beam-splitter interactions, two-mode squeezing, and reservoir engineering. Floquet theory organizes the periodic Hamiltonian; driven master equations or Keldysh methods treat dissipation and nonequilibrium occupation.

Every pump also creates a calibration problem. Stark shifts, heating, pump phase noise, spurious mixing products, and quasiparticle generation should be included in the effective model or bounded experimentally.

Bosonic-code simulation and decoding

Bosonic QEC requires phase-space methods, Fock-basis simulation, stochastic trajectories, and code-specific decoders. Useful validation includes:

• truncation convergence in photon number;
• measured ancilla and cavity error channels;
• finite-duration control pulses;
• correlated faults;
• decoder latency;
• training–test separation for learned control; and
• comparison with the best physical reference under the same metric.

Reinforcement learning can discover effective feedback policies. It does not remove the need for held-out validation, uncertainty estimates, and interpretability of failure modes.

Quantum-channel and systems analysis

Component parameters should ultimately feed a channel model. For Gaussian conversion, transmissivity and noise determine covariance propagation. For non-Gaussian states, process tensors, tomography, entanglement fidelity, or task-specific witnesses may be necessary.

A system model should compose:

$$\mathcal E_{\rm total} = \mathcal E_{\rm receive} \circ \mathcal E_{\rm link} \circ \mathcal E_{\rm convert} \circ \mathcal E_{\rm emit}.$$

High fidelity of one factor cannot compensate for vanishing success probability of another unless the protocol explicitly heralds and budgets the resulting rate.

Representative Papers and Reviews

Broad reviews

1. A. Blais, A. L. Grimsmo, S. M. Girvin, and A. Wallraff, “Circuit quantum electrodynamics,” Reviews of Modern Physics 93, 025005 (2021).
2. A. F. Kockum, A. Miranowicz, S. De Liberato, S. Savasta, and F. Nori, “Ultrastrong coupling between light and matter,” Nature Reviews Physics 1, 19–40 (2019).
3. A. S. Sheremet, M. I. Petrov, I. V. Iorsh, A. V. Poshakinskiy, and A. N. Poddubny, “Waveguide quantum electrodynamics: Collective radiance and photon–photon correlations,” Reviews of Modern Physics 95, 015002 (2023).
4. D. Roy, C. M. Wilson, and O. Firstenberg, “Colloquium: Strongly interacting photons in one-dimensional continuum,” Reviews of Modern Physics 89, 021001 (2017).
5. A. Reiserer, “Colloquium: Cavity-enhanced quantum network nodes,” Reviews of Modern Physics 94, 041003 (2022).
6. X. Gu, A. F. Kockum, A. Miranowicz, Y.-x. Liu, and F. Nori, “Microwave photonics with superconducting quantum circuits,” Physics Reports 718–719, 1–102 (2017).
7. Z.-L. Xiang, S. Ashhab, J. Q. You, and F. Nori, “Hybrid quantum circuits: Superconducting circuits interacting with other quantum systems,” Reviews of Modern Physics 85, 623–653 (2013).
8. G. Kurizki et al., “Quantum technologies with hybrid systems,” Proceedings of the National Academy of Sciences 112, 3866–3873 (2015).
9. E. Zeuthen, A. Schliesser, A. S. Sørensen, and J. M. Taylor, “Figures of merit for quantum transducers,” Quantum Science and Technology 5, 034009 (2020).

Strong, ultrastrong, and multimode coupling

10. R. J. Thompson, G. Rempe, and H. J. Kimble, “Observation of normal-mode splitting for an atom in an optical cavity,” Physical Review Letters 68, 1132–1135 (1992).
11. A. Wallraff et al., “Strong coupling of a single photon to a superconducting qubit using circuit quantum electrodynamics,” Nature 431, 162–167 (2004).
12. P. Forn-Díaz et al., “Ultrastrong coupling of a single artificial atom to an electromagnetic continuum in the nonperturbative regime,” Nature Physics 13, 39–43 (2017).
13. F. Tay et al., “Multimode ultrastrong coupling in three-dimensional photonic-crystal cavities,” Nature Communications 16, 3603 (2025).
14. A. L. Shaw et al., “A cavity-array microscope for parallel single-atom interfacing,” Nature 650, 320–326 (2026).

Waveguide QED

15. B. Kannan et al., “Waveguide quantum electrodynamics with superconducting artificial giant atoms,” Nature 583, 775–779 (2020).
16. M. Zanner et al., “Coherent control of a multi-qubit dark state in waveguide quantum electrodynamics,” Nature Physics 18, 538–543 (2022).
17. M. J. R. Staunstrup et al., “Direct observation of a few-photon phase shift induced by a single quantum emitter in a waveguide,” Nature Communications 15, 7583 (2024).
18. P. Pakkiam et al., “Experimental realization of qubit-state-controlled directional edge states in waveguide QED,” npj Quantum Information (2026).

Bosonic memories

19. N. Ofek et al., “Extending the lifetime of a quantum bit with error correction in superconducting circuits,” Nature 536, 441–445 (2016).
20. P. Campagne-Ibarcq et al., “Quantum error correction of a qubit encoded in grid states of an oscillator,” Nature 584, 368–372 (2020).
21. V. V. Sivak et al., “Real-time quantum error correction beyond break-even,” Nature 616, 50–55 (2023).
22. H. Putterman et al., “Hardware-efficient quantum error correction via concatenated bosonic qubits,” Nature 638, 927–934 (2025).
23. B. L. Brock et al., “Quantum error correction of qudits beyond break-even,” Nature 641, 612–618 (2025).

Hybrid conversion and links

24. A. Kumar et al., “Quantum-enabled millimetre wave to optical transduction using neutral atoms,” Nature 615, 614–619 (2023).
25. H. Zhao et al., “Quantum-enabled microwave-to-optical transduction via silicon nanomechanics,” Nature Nanotechnology 20, 602–608 (2025).
26. T. Xie, R. Fukumori, J. Li, and A. Faraon, “Scalable microwave-to-optical transducers at the single-photon level with spins,” Nature Physics 21, 931–937 (2025).
27. H. K. Warner et al., “Coherent control of a superconducting qubit using light,” Nature Physics 21, 831–837 (2025).
28. Y. Zhou et al., “A 1-km photonic link connecting superconducting circuits in two dilution refrigerators,” Nature Photonics 20, 579–585 (2026).

Review articles establish vocabulary and broad context. Numerical records and “this year” statements should be checked against the primary papers and their port, bandwidth, uncertainty, and postselection definitions.

Common Misconceptions

“Cooperativity above one means coherent strong coupling”

Not necessarily. Cooperativity, resolvable splitting, and underdamped swaps use different combinations of $g$, linewidths, and dephasing. State the criterion and convention actually tested.

“Ultrastrong coupling means low loss”

No. Ultrastrong coupling compares $g$ with a bare frequency. Conventional strong coupling compares $g$ with decay. A device can be ultrastrong and highly dissipative, or coherently strong while far from ultrastrong.

“Virtual photons leak out of a stationary ground state”

Not as ordinary stationary output flux. Bare-excitation occupation in a dressed ground state is basis dependent. Real emission requires a protocol that changes parameters or otherwise couples the correlations to output modes.

“More cavity modes mean more usable channels”

No. Modes may overlap, share loss, couple nonuniformly, or be impossible to address independently. Useful channel count requires calibrated preparation, routing, discrimination, and cross-talk.

“A high beta factor makes an interface directional”

No. $\beta$ describes guided versus nonguided emission. $\mathcal D$ describes right versus left emission within the guided channel. Report both.

“A single-emitter phase shift is already a photonic controlled-phase gate”

No. A weak-coherent-state phase measurement does not determine the complete two-photon scattering channel. Gate claims require encoded inputs, modal fidelity, loss, and an unconditional process benchmark.

“Quantum-enabled transduction means high-fidelity state transfer”

No. Sub-one-photon input-referred added noise is an important condition. Low efficiency, mode mismatch, or phase instability can still destroy an unknown input state.

“Internal efficiency is the network efficiency”

No. Escape efficiencies, fibre coupling, propagation, filtering, receiver acceptance, and detection multiply the internal conversion probability.

“Beyond break-even memory means fault tolerance”

No. It establishes a declared storage benchmark against a declared physical reference. Fault-tolerant processing additionally requires logical operations, state preparation, measurement, and scalable error suppression.

“Topological directionality means zero cross-talk in hardware”

No. Ideal-model localization or chirality can be robust to selected perturbations. Finite devices still have disorder, loss, ports, preparation error, and a measurement noise floor.

Exercises

Exercise 1: Compare strong-coupling metrics

An emitter–cavity system has

$$\frac{g}{2\pi}=50\ {\rm MHz}, \qquad \frac{\kappa}{2\pi}=10\ {\rm MHz}, \qquad \frac{\gamma}{2\pi}=5\ {\rm MHz}.$$

Using $C=4g^2/(\kappa\gamma)$, calculate the cooperativity. Does this number alone prove high-fidelity vacuum Rabi swaps?

Solution

The factors of $2\pi$ cancel:

$$C = \frac{4(50)^2}{(10)(5)} = 200.$$

This is a large cooperativity in the stated convention. It does not by itself prove high-fidelity swaps. One must also know pure dephasing, detuning, preparation and readout errors, cavity-port partition, emitter motion or spectral diffusion, and the exact damping convention. A time-domain exchange measurement would test the intended dynamical claim more directly.

Exercise 2: Classify an ultrastrong device

A circuit has $g/2\pi=1.2\ {\rm GHz}$ and $\omega_c/2\pi=6.0\ {\rm GHz}$.

1. Compute $\eta=g/\omega_c$.
2. Is the rotating-wave approximation expected to be quantitatively safe?
3. Can you infer coherent operation from these data?

Solution

The normalized coupling is

$$\eta=\frac{1.2}{6.0}=0.20.$$

This lies in the conventional ultrastrong regime, so counter-rotating terms and associated renormalizations should not be discarded without a controlled error estimate. Nothing was specified about $\kappa$, $\gamma$, dephasing, temperature, or control fidelity. Therefore coherent operation cannot be inferred from $\eta$ alone.

Exercise 3: Estimate a multimode window

Equally spaced cavity modes have spacing $\Delta_{\rm FSR}/2\pi=100\ {\rm MHz}$. A material resonance couples over an approximate half-width $g/2\pi=250\ {\rm MHz}$. How many mode centres lie within $\pm g$ of a resonant central mode?

Solution

The mode offsets in the interval are

$$-200,\ -100,\ 0,\ 100,\ 200\ {\rm MHz}.$$

Thus five mode centres lie within $\pm250\ {\rm MHz}$. This count is only a first diagnostic. Mode-dependent coupling, polarization, linewidth, and self-polarization terms determine whether all five materially participate. Modes outside this window may still renormalize fitted parameters.

Exercise 4: Separate beta factor and directionality

An emitter has $\beta=0.92$. Within its guided emission, $\Gamma_R:\Gamma_L=9:1$.

1. Compute $\mathcal D$.
2. Express $\Gamma_R$, $\Gamma_L$, and nonguided decay as fractions of $\Gamma_{\rm tot}$.

Solution

The directionality is

$$\mathcal D = \frac{9-1}{9+1} = 0.8.$$

The guided fraction is $0.92$, split in the ratio $9:1$:

$$\frac{\Gamma_R}{\Gamma_{\rm tot}}=0.92(0.9)=0.828,$$ $$\frac{\Gamma_L}{\Gamma_{\rm tot}}=0.92(0.1)=0.092.$$

The nonguided fraction is $1-\beta=0.08$. High directionality therefore does not mean unit total collection into the desired port.

Exercise 5: Test a Markov approximation

A giant atom has coupling points separated by $L=0.50\ {\rm m}$ along a waveguide with group velocity $v_g=0.40c$. Its waveguide decay rate is $\Gamma/2\pi=50\ {\rm MHz}$. Calculate $\tau$ and $\Gamma\tau$. Is delay negligible?

Solution

Using $c\simeq3.0\times10^8\ {\rm m\,s^{-1}}$,

$$\tau = \frac{0.50}{0.40(3.0\times10^8)} \simeq 4.17\ {\rm ns}.$$

The angular decay rate is

$$\Gamma = 2\pi(50\times10^6)\ {\rm s^{-1}},$$

so

$$\Gamma\tau \simeq 1.31.$$

The delay is dynamically significant. A memoryless local master equation is not justified merely by calling the waveguide a reservoir. Dispersion, loss, and the phases at all coupling points must also be included.

Exercise 6: Convert internal to external efficiency

A two-leg converter has $C_1=C_2=10$. Use

$$\eta_{\rm int} = \frac{4C_1C_2}{(1+C_1+C_2)^2}$$

and external coupling efficiencies $\eta_1=0.80$ and $\eta_2=0.70$. Neglect path loss.

1. Find $\eta_{\rm int}$ and $\eta_{\rm ext}$.
2. If $n_{\rm add}=0.60$, is high-fidelity arbitrary-state transfer established?

Solution

The internal efficiency is

$$\eta_{\rm int} = \frac{4(10)(10)}{(1+10+10)^2} = \frac{400}{441} \simeq 0.907.$$

The external efficiency is

$$\eta_{\rm ext} = (0.80)(0.70)(0.907) \simeq 0.508.$$

The noise is below one input-referred photon under the stated definition, but roughly half the input is still lost before any omitted path and receiver losses. High-fidelity arbitrary-state transfer is therefore not established. A channel or state-level benchmark is needed.

Exercise 7: Interpret a memory gain

A logical qutrit memory has $\tau_L=1.82\ {\rm ms}$, while the best physical qutrit under the same storage metric has $\tau_{\rm ref}=1.00\ {\rm ms}$.

1. Compute $G$.
2. Name four facts still needed before calling the device a fault-tolerant qutrit processor.

Solution

The memory gain is

$$G=\frac{1.82}{1.00}=1.82.$$

This establishes beyond-break-even storage if the state ensemble, metric, and overhead are matched. A fault-tolerant processor still needs, for example:

1. fault-tolerant logical state preparation;
2. a logical gate set with error characterization;
3. repeated syndrome extraction during operations;
4. logical measurement and reset;
5. leakage and ancilla-fault handling; and
6. improving logical error with increasing resources.

Any four suffice. Underlying Hilbert-space dimension alone supplies none of these demonstrations.

Exercise 8: Rewrite an overclaim

Rewrite the statement

A 99.996% directional topological waveguide has eliminated cross-talk for scalable superconducting quantum networks.

so that it states the demonstrated result, uncertainty boundary, and missing network evidence.

Solution

A defensible version is:

In a finite superconducting Rice–Mele waveguide coupled to an Xmon qubit, transmission and emission measurements identified frequency-controlled directional edge states. The inferred directionality fidelity was $99.996\%$, constrained by the measurement noise floor. Demonstrating a scalable low-cross-talk quantum interconnect still requires insertion-loss, disorder, multi-emitter, switching, and end-to-end state-transfer or entanglement benchmarks.

This separates the measured finite-device result from the architecture-level extrapolation.

Connections to Other Volumes

Quantum Optics Frontiers owns dated questions about nonclassical optical sources, squeezing, integrated photonic processing, and long-distance optical networks. The present page owns matter–field interfaces, superconducting resonators, microwave waveguides, conversion, and bosonic memories.

Composite Systems and Entanglement supplies Schmidt structure, continuous variables, and multipartite-state tools used to certify emitted, stored, and converted states.

Measurement and Open Quantum Systems supplies master equations, quantum trajectories, non-Markovianity, channels, feedback, and receiver models. Its Strong Coupling and Memory Effects page explains why a large interaction rate is not itself a non-Markovianity witness.

Bosonic Qubits owns complete oscillator modules, logical channels, nonlinear control, correction architectures, and processor-level resource accounting. Bosonic memory results belong here only as dated cavity and circuit-QED evidence; detailed code theorems remain in the Quantum Information and Computation error-correction chapter.

Interconnects and Transduction owns hardware-level link semantics, end-to-end channel composition, added-noise cascades, mode acceptance, service rate, and cross-platform converter comparison. The present page retains the physical frontier audit for cavity, circuit, waveguide, and hybrid light–matter devices.

Many-Body and Statistical Physics provides collective-mode, phase-transition, and driven-dissipative tools needed for cavity arrays, polaritons, and waveguide-mediated many-body systems.

The planned Relativistic Quantum Mechanics and the QFT Bridge volume will provide the field-theoretic perspective behind gauge invariance, mode truncation, and renormalized light–matter descriptions.

What Changed This Year

Parallel cavity QED moved from one global mode to a cavity array

New in 2026. The cavity-array microscope demonstrated individual atom–cavity interfacing across more than 40 free-space modes, with above-unity peak cooperativity and millisecond-scale nondestructive parallel readout. The reported next-generation structure has more than 500 cavities and higher finesse, but full many-atom operation at that scale remains future work.

A kilometre-scale cryogenic photonic link preserved coherent signals

New in 2026. Frequency-matched aluminium-nitride transducers connected superconducting circuits in separate dilution refrigerators through one kilometre of telecom fibre. The result is coherent signal transfer and a major transduction-efficiency improvement over commercial modulators, not yet arbitrary quantum-state transfer.

Directional edge states became tunable in situ

New in 2026. A superconducting Rice–Mele waveguide demonstrated qubit-frequency-controlled directional edge states with an inferred $99.996\%$ directionality fidelity limited by the measurement noise floor. Multi-emitter, lossy-network, and state-transfer validation remain open.

The previous year set the current bosonic-memory benchmark

Current 2025 evidence. Error-corrected GKP qutrit and ququart memories operated beyond their physical references, and a distance-five concatenated cat memory demonstrated below-threshold phase-flip correction. These are important scaling and memory results, but neither closes the logical-gate and full fault-tolerance problem.

Multimode ultrastrong coupling gained a three-dimensional platform

Current 2025 evidence. Terahertz three-dimensional photonic-crystal cavities exhibited polarization- and mode-profile-dependent multimode ultrastrong coupling to a collective cyclotron resonance. Operational access to the predicted multimode ground-state correlations remains active.

Transducers improved along different, noninterchangeable axes

Current 2025 evidence. Silicon nanomechanics crossed the sub-one-photon continuous-wave noise boundary; rare-earth spins combined percent-level efficiency with multi-device optical interference; and electro-optic conversion drove coherent superconducting-qubit rotations. None of these individual component milestones yet supplies a low-loss, high-fidelity, end-to-end optical quantum link between processors.

The next review should update the full metric vectors rather than only record values. In particular, it should ask whether any experiment has simultaneously demonstrated:

$$\text{useful efficiency} +\text{low added noise} +\text{state-level verification} +\text{network distance}.$$