Multiresonator Quantum Memory

• Multiresonator quantum memory is a quantum storage architecture that maps electromagnetic fields into a discrete network of coupled resonant modes.
• It leverages spectral engineering by tuning resonator frequencies, couplings, and linewidths to optimize impedance matching and maximize storage efficiency.
• Recent implementations in superconducting circuits and hybrid resonator–spin modules demonstrate broadband, on-demand, and multimode storage capabilities.

Searching arXiv for recent and foundational papers on multiresonator quantum memory. Multiresonator quantum memory is a class of quantum memory architectures in which a propagating electromagnetic field is reversibly mapped to a discrete network of coupled resonant modes rather than to a single cavity or to a spatially continuous medium. In the literature considered here, the memory medium is realized by arrays of high-$Q$ microresonators, ring resonators, coplanar superconducting resonators, or hybrid resonator–spin modules coupled to a common waveguide or common resonator. The central design principle is spectral engineering: resonator frequencies, couplings, and linewidths are chosen to form a controllable frequency comb or spatial-frequency comb whose transfer function approximates an impedance-matched broadband absorber and, under suitable conditions, an echo-based or delay-based quantum memory (1705.01536, Perminov et al., 2018, Perminov et al., 2017, Bao et al., 2021, Matanin et al., 2022).

1. Architectures and defining features

Multiresonator quantum memory has been developed in several closely related architectures. A foundational waveguide-coupled formulation employs an array of high-$Q$ single-mode microresonators with resonance frequencies

$\omega_n=\omega_0+n\Delta$

placed along a broadband one-dimensional waveguide at positions $z_n=nz$, with the paper focusing on $z=\lambda/2=\pi/k_0$, so that the resonators form a spatial-frequency comb (1705.01536). In this setting the resonators are point-like relative to the wavelength and collectively act as a distributed resonator with bandwidth $\sim N\Delta$ (1705.01536).

A second architecture uses a compact controllable frequency-comb block made of three ring microresonators coupled to a common waveguide. Resonators 1 and 3 couple to the waveguide, resonator 2 couples only to resonators 1 and 3, and the resonance detunings are

$$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$

This three-line comb is treated as an elementary broadband memory unit whose spectral response can be optimized and then “glued” with other blocks to expand total bandwidth (Perminov et al., 2018).

A third family introduces a common resonator as an interface between the waveguide and several miniresonators. In the experimentally demonstrated microwave quantum memory-interface, five cylindrical mini-resonators are coupled to one common broadband resonator, which in turn is coupled to an external waveguide through a tunable slit; the mini-resonators are tuned into a periodic comb with spacing $\Delta$ and echo time $T_1=1/\Delta$ (Moiseev et al., 2017). Closely related superconducting on-chip realizations use one common coplanar waveguide resonator coupled capacitively to eight internal $\lambda/4$ resonators; these eight resonators are divided into two four-resonator groups that act as two spectral-mode memory cells (Matanin et al., 2022).

A fourth architecture hybridizes resonators with long-lived matter excitations. In the long-lived and cascade schemes, each miniresonator or ring resonator contains a spin ensemble with inhomogeneous broadening, and all resonators couple to a common waveguide or common resonator. The four-resonator cascade scheme uses ring resonators containing long-lived spin ensembles and achieves a transfer function with a nearly vanishing reflection plateau over $Q$0, yielding ideal storage efficiency $Q$1 across that band in the optimized lossless model (Perminov et al., 2018). The long-lived hybrid multiresonator memory with four miniresonators and spin ensembles likewise reports a broadband absorption plateau with $Q$2 over $Q$3 in units of the comb spacing (Perminov et al., 2017).

A more recent integrated proposal combines three interacting resonators with a common resonator and a switchable coupler to the waveguide. In that design the waveguide coupling is dynamically switched between $Q$4 during loading or retrieval and $Q$5 during storage, so that the resonator network becomes a closed high-$Q$6 storage subsystem supporting on-demand retrieval at selected times (Perminov et al., 2023).

These implementations share three defining traits. First, the memory is discrete in frequency: the relevant spectral structure is built from a finite set of resonator lines rather than from a continuous cavity linewidth. Second, the spectral response is engineered by couplings and detunings, so multiresonator quantum memory is fundamentally a transfer-function design problem. Third, multimode operation follows from the composite bandwidth created by the resonator comb, not from a single broad cavity alone (1705.01536, Perminov et al., 2018, Moiseev et al., 2017, Matanin et al., 2022).

2. Dynamical models and transfer functions

Theoretical treatments are predominantly based on input–output theory, Heisenberg–Langevin equations, or equivalent coupled-mode equations. In the waveguide-coupled spatial-frequency-comb model, the Hamiltonian is

$Q$7

with $Q$8 waveguide modes and $Q$9 resonator modes (1705.01536). In the single-excitation sector this yields coupled equations for the waveguide amplitude $\omega_n=\omega_0+n\Delta$0 and resonator amplitudes $\omega_n=\omega_0+n\Delta$1, from which one derives a collective retrieval efficiency that factorizes as

$\omega_n=\omega_0+n\Delta$2

where $\omega_n=\omega_0+n\Delta$3 depends only on the coupling strength and comb spacing, and $\omega_n=\omega_0+n\Delta$4 encodes rephasing dynamics and the initial distribution across resonators (1705.01536).

For the three-ring controllable-frequency-comb block, the intracavity fields $\omega_n=\omega_0+n\Delta$5 and waveguide fields $\omega_n=\omega_0+n\Delta$6 obey the coupled-mode system

$\omega_n=\omega_0+n\Delta$7

supplemented by the waveguide input–output relations (Perminov et al., 2018). The transfer function of one block is

$\omega_n=\omega_0+n\Delta$8

with spectral delay

$\omega_n=\omega_0+n\Delta$9

In the idealized lossless model, $z_n=nz$0 for all $z_n=nz$1; the memory behavior is therefore encoded entirely in the phase or group-delay structure, and optimization targets flatness of $z_n=nz$2 in the working band (Perminov et al., 2018).

Common-resonator architectures lead to rational transfer functions in which the internal resonators enter through an effective susceptibility. In the on-chip eight-resonator superconducting device, the Heisenberg–Langevin equations are

$z_n=nz$3

$z_n=nz$4

with input–output relation

$z_n=nz$5

The output field satisfies

$z_n=nz$6

where

$z_n=nz$7

and $z_n=nz$8 is the resonator-comb susceptibility (Matanin et al., 2022). This form makes explicit that the multiresonator memory is spectrally a structured reflector/absorber whose design reduces to shaping $z_n=nz$9.

Hybrid resonator–spin memories add an additional layer of effective susceptibility from inhomogeneously broadened spin ensembles. In the four-resonator cascade scheme, the total transfer function is

$z=\lambda/2=\pi/k_0$0

$z=\lambda/2=\pi/k_0$1

and ideal storage efficiency is

$z=\lambda/2=\pi/k_0$2

when cavity losses are neglected (Perminov et al., 2018). The hybrid multiresonator memory with atomic ensembles generalizes this construction to a common resonator plus many atomic-ensemble-loaded miniresonators and derives a common-resonator transfer function

$z=\lambda/2=\pi/k_0$3

along with a reflection coefficient $z=\lambda/2=\pi/k_0$4 obtained from the input–output relation $z=\lambda/2=\pi/k_0$5 (Moiseev, 7 Jul 2025). This suggests that the distinction between “purely photonic” and “hybrid” multiresonator memory is mainly a distinction in internal susceptibility: both are optimized through the same spectral-matching logic.

3. Impedance matching, frequency combs, and optimization

A recurring concept in multiresonator quantum memory is impedance matching generalized from single-cavity memory to structured resonator networks. In the spatial-frequency-comb waveguide model, retrieval efficiency reaches its maximum when the Bragg-type impedance matching condition

$z=\lambda/2=\pi/k_0$6

is satisfied (1705.01536). Equivalently, the dimensionless coupling $z=\lambda/2=\pi/k_0$7 should equal 1, at which point $z=\lambda/2=\pi/k_0$8 in the factorized efficiency formula (1705.01536). The physical interpretation given there is that the periodic spatial arrangement at $z=\lambda/2=\pi/k_0$9 creates a Bragg grating, and the matching condition balances comb spacing against resonator linewidth so the distributed resonator absorbs broadband input without destructive leakage (1705.01536).

The three-ring controllable-frequency-comb block uses a different optimization target because its idealized transfer function is lossless. There the objective is to flatten the normalized delay

$\sim N\Delta$0

across the band, and the authors use an objective function

$\sim N\Delta$1

to optimize the waveguide coupling $\sim N\Delta$2 and inter-resonator coupling $\sim N\Delta$3 (Perminov et al., 2018). Partial optimization with $\sim N\Delta$4 gives $\sim N\Delta$5, whereas full optimization gives $\sim N\Delta$6, reducing the spread in normalized delay over $\sim N\Delta$7 from $\sim N\Delta$8 to $\sim N\Delta$9 and thereby improving spectral flatness by a factor of 1.5 (Perminov et al., 2018). This suggests that “impedance matching” in purely resonator-based memories often takes the form of phase-delay flattening rather than simple amplitude matching.

In the common-resonator microwave interface, the first-echo efficiency is approximated by

$$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$0

under the optimal coupling condition

$$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$1

and the reflection coefficient at the comb center is

$$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$2

When $$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$3, the numerator vanishes and central reflection is suppressed (Moiseev et al., 2017). This architecture implements impedance matching directly at the bus resonator and makes explicit the tradeoff between common-resonator loss $$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$4, miniresonator loss $$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$5, comb spacing $$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$6, and coupling $$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$7 (Moiseev et al., 2017).

Hybrid schemes with spin ensembles introduce “extended matching conditions” that flatten reflection over a finite band rather than only at $$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$8. In the four-resonator cascade memory, the optimization is formulated as

$$\Delta_1=-\Delta,\quad \Delta_2=0,\quad \Delta_3=\Delta.$$9

with $\Delta$0, and the free parameters are $\Delta$1 (Perminov et al., 2018). For $\Delta$2 (four resonators), the optimized symmetric parameter set is

$\Delta$3

$\Delta$4

$\Delta$5

$\Delta$6

and this yields $\Delta$7 across $\Delta$8 (Perminov et al., 2018). The long-lived hybrid scheme with spin ensembles uses a parallel optimization philosophy and derives both a basic impedance-matching condition

$\Delta$9

and a spectral matching condition

$T_1=1/\Delta$0

to flatten the low-reflection band (Moiseev, 7 Jul 2025).

In the integrated switchable-coupler proposal, optimization is performed step by step. First the closed system with $T_1=1/\Delta$1 is designed to have a multiple spectrum, then the open system with $T_1=1/\Delta$2 is tuned for flat group delay via

$T_1=1/\Delta$3

leading to an explicit algebraic relation

$T_1=1/\Delta$4

for the waveguide coupling in a three-mini-resonator plus common-resonator system (Perminov et al., 2023). This is a particularly clear example of a “spectral-topological” design rule: closed-system eigenfrequency structure and open-system matching are optimized separately but consistently.

4. Echo dynamics, delay lines, and on-demand retrieval

Multiresonator quantum memory is closely related to photon-echo and atomic-frequency-comb logic. In the waveguide spatial-frequency-comb scheme, without additional control the field re-emits at the natural echo time

$T_1=1/\Delta$5

and the bandwidth is $T_1=1/\Delta$6, with minimum signal duration

$T_1=1/\Delta$7

(1705.01536). The explicit analytical solution for resonator amplitudes

$T_1=1/\Delta$8

exhibits coherent rephasing and interference among the resonators (1705.01536). This framework also shows how the comb can be “frozen” dynamically by setting $T_1=1/\Delta$9, placing the system in a nonradiative dark state so that emission resumes only when the comb spacing is restored (1705.01536).

The three-resonator controllable-frequency-comb block is, in its static formulation, a short-delay memory or engineered delay line rather than a full on-demand memory. The output obeys

$\lambda/4$0

so if $\lambda/4$1 across the pulse spectrum, the pulse is delayed by $\lambda/4$2 with little distortion (Perminov et al., 2018). The authors note that longer storage would require dynamic disconnection and reconnection of resonators to the waveguide, citing relevant switchable-coupler technologies (Perminov et al., 2018). This suggests that static multiresonator blocks are naturally suited to fixed-delay buffering, while true on-demand retrieval requires fast control over couplers or detunings.

Microwave experimental demonstrations realize both fixed-delay echo retrieval and dynamic on-demand release. In the room-temperature five-miniresonator interface, the miniresonators are tuned to a periodic comb with

$\lambda/4$3

and the first echo efficiency is maximized by adjusting the waveguide–common-resonator coupling to the matching condition above (Moiseev et al., 2017). In the superconducting four-resonator on-demand memory, the resonator frequencies are dynamically tuned: the comb is used for write and read, while during storage all resonators are aligned to the same frequency to stop rephasing and create a non-radiative dark state. When the comb is restored, the echo process resumes, and the total storage time becomes

$\lambda/4$4

where $\lambda/4$5 is the controlled “comb-closed” interval (Bao et al., 2021). This experiment demonstrated tunable memory bandwidth from 10 MHz to 55 MHz and on-demand storage and retrieval of weak coherent microwave photon pulses at the single-photon level, with overall storage efficiency up to 12% and preserved phase coherence (Bao et al., 2021).

The integrated switchable-coupler proposal generalizes this logic by separating the loading stage, where the memory is impedance matched to the waveguide, from a closed storage stage with $\lambda/4$6, and then a retrieval stage where coupling is restored at a chosen moment (Perminov et al., 2023). In that design the internal eigenfrequencies are arranged so that the resonator subsystem exhibits periodic revivals at

$\lambda/4$7

and on-demand release is accomplished by re-opening the switchable coupler when the internal state is favorable for emission (Perminov et al., 2023).

A distinct but related realization of reversible multiresonator dynamics appears in the three-resonator “programmable quantum motherboard” coupled to atoms. There the spectrum is engineered to be equidistant, yielding exactly periodic single-excitation dynamics; this enables reversible quantum state transfer between distributed atoms and the generation of logical qubits and qutrits (Perminov et al., 2019). While that work is not a waveguide memory interface, it demonstrates how spectrum engineering in a small multiresonator system can create perfect revivals, which is a memory-like property.

5. Performance benchmarks across implementations

The literature reports performance across a wide range of physical platforms and assumptions. The following table summarizes concrete figures explicitly stated in the cited works.

System	Key result	Notes
Five mini-resonators + common resonator, room temperature	$\lambda/4$8 storage efficiency	Gaussian pulses, X-band microwave, echo at $\lambda/4$9 ns (Moiseev et al., 2017)
Four tunable superconducting CPW resonators	up to $Q$00 overall storage efficiency	Single-photon-level weak coherent pulses; tunable bandwidth 10–55 MHz (Bao et al., 2021)
Eight on-chip superconducting resonators in two 4-resonator groups	$Q$01 single-photon efficiency, $Q$02 at higher intensity	Two spectral modes; noiseless storage confirmed by coherent-state process tomography (Matanin et al., 2022)
Four-resonator cascade with spin ensembles	$Q$03 over $Q$04	Idealized lossless optimized model (Perminov et al., 2018)
Four-resonator hybrid long-lived memory with spin ensembles	$Q$05 over $Q$06	Idealized optimized model (Perminov et al., 2017)

The room-temperature microwave interface demonstrated storage of microwave pulses with an efficiency of 16.3% using five mini-resonators coupled through a common resonator, with Gaussian pulses and strong suppression of input reflection when the tunable slit was adjusted close to the impedance-matching condition (Moiseev et al., 2017). The same work argued that for high-$Q$07 resonators at low temperature, taking $Q$08 MHz with $Q$09 MHz, the efficiency estimate becomes

$Q$10

which suggests near-unit efficiency is feasible in cryogenic implementations (Moiseev et al., 2017).

The first on-demand superconducting multiresonator memory stored weak coherent microwave pulses at the single-photon level using four frequency-tunable CPW resonators. It reported tunable memory bandwidth from 10 MHz to 55 MHz and overall storage efficiency up to 12%, along with on-demand storage and retrieval of a time-bin flying qubit and preserved phase coherence (Bao et al., 2021). That work identified resonator internal loss and imperfect impedance matching as the dominant limitations, with the storage decay time extracted as approximately $Q$11 (Bao et al., 2021).

The eight-resonator superconducting memory significantly improved efficiency. The device comprised one common $Q$12 coplanar resonator coupled to eight internal $Q$13 resonators designed from 5.9895 to 6.0105 GHz with spacing $Q$14 MHz and couplings $Q$15 MHz (Matanin et al., 2022). Dividing the eight resonators into two four-resonator combs yielded two independent spectral-mode memories with echo times $Q$16 ns and $Q$17 ns (Matanin et al., 2022). The measured power efficiency reached $Q$18 for one memory cell and $Q$19 for the other at high intensity, while coherent-state process tomography gave a single-photon efficiency of $Q$20 and showed that the probability of generating an extra noise photon was below 1% (Matanin et al., 2022).

Several proposals and optimized hybrid models report much higher theoretical values. The cascade multiresonator quantum memory with four ring resonators and spin ensembles optimized a symmetric parameter set with $Q$21, $Q$22, $Q$23, $Q$24, $Q$25, $Q$26, and $Q$27, obtaining a reflection plateau corresponding to $Q$28 for $Q$29 (Perminov et al., 2018). The long-lived hybrid multiresonator memory with four miniresonators and spin ensembles reported an optimized lossless absorption plateau $Q$30 over $Q$31 with $Q$32, $Q$33, $Q$34, $Q$35, $Q$36, $Q$37, and $Q$38 (Perminov et al., 2017).

A plausible implication is that current experimental superconducting memories are already in the regime where the core spectral-engineering ideas have been validated, but losses, control errors, and incomplete matching still separate them from the $Q$39 regime predicted in optimized hybrid models.

6. Applications, constraints, and open directions

Multiresonator quantum memory is motivated primarily by the need for broadband, multimode, and integrable memory modules in photonic and superconducting quantum information processing. The microwave works explicitly connect the memory to superconducting quantum computers, where an efficient microwave memory can store weak coherent states or photonic qubits alongside circuit-QED processors (Moiseev et al., 2017, Bao et al., 2021, Matanin et al., 2022). The experimental demonstration of a time-bin flying qubit in the four-resonator device shows direct compatibility with temporal-mode encodings used in communication protocols (Bao et al., 2021). The eight-resonator superconducting device further suggests use as a broadband two-spectral-mode memory with low added noise, an attractive capability for multiplexed superconducting architectures (Matanin et al., 2022).

Hybrid resonator–spin schemes extend the application space by importing long coherence times from spin ensembles. The long-lived and cascade schemes are aimed at multiqubit microwave memory for superconducting quantum computers and broadband quantum interfaces that could connect several different memory elements into one frequency-comb memory block (Perminov et al., 2017, Perminov et al., 2018). The integrated optical proposal with atomic ensembles explicitly targets integrated photonic implementations, arguing that ring resonators, photonic crystal cavities, and rare-earth-ion ensembles can provide high-$Q$40, small-volume, and waveguide-compatible memory modules (Moiseev, 7 Jul 2025). This suggests a route to telecom-band or integrated optical multiresonator memories, although the detailed experimental benchmarks in that regime are still developing.

Several constraints recur across the literature. The first is parameter sensitivity. High performance depends on precise control of resonator frequencies, coupling rates, and linewidths. The integrated switchable-coupler proposal emphasizes high-precision parameter matching and spectral-topological optimization, and the on-chip superconducting experiment notes that effective inter-resonator coupling makes precise comb alignment difficult at small $Q$41, preventing optimal impedance matching (Perminov et al., 2023, Bao et al., 2021). The second is internal loss. Both the room-temperature microwave interface and the superconducting on-demand memory identify resonator loss as a primary limitation (Moiseev et al., 2017, Bao et al., 2021). The eight-resonator superconducting experiment attributes the drop from $Q$42 high-power efficiency to $Q$43 single-photon efficiency to unsaturated two-level-system defects that increase $Q$44 at low photon number (Matanin et al., 2022). The third is scaling complexity. As the number of resonators grows, optimization becomes higher-dimensional, fabrication tolerances tighten, and parasitic couplings become harder to suppress (Perminov et al., 2018, Matanin et al., 2022).

Common misconceptions are clarified by the literature. Multiresonator quantum memory is not a single protocol: some realizations are static delay-line or echo-line devices, while others are true on-demand memories with switchable couplers or dynamic comb control (Perminov et al., 2018, Bao et al., 2021, Perminov et al., 2023). It is also not restricted to purely photonic resonator networks: spin ensembles can be embedded into each resonator, producing hybrid schemes where the resonators shape the spectrum and the spins provide long-lived storage (Perminov et al., 2017, Perminov et al., 2018, Moiseev, 7 Jul 2025). Nor is a large number of resonators always required. Several papers show that three- or four-resonator structures already support substantial spectral control, with three-resonator blocks serving as optimized elementary units and four-resonator structures achieving broad and flat absorption bands in optimized models (Perminov et al., 2018, Perminov et al., 2023, Perminov et al., 2018).

Open directions are explicit in the cited works. One is modular composition: the three-resonator controllable-frequency-comb block can be “glued” into broader composite memories, with two optimized blocks already outperforming an earlier four-direct-resonator comb in spectral flatness over $Q$45 (Perminov et al., 2018). Another is improved materials and architectures: the eight-resonator superconducting memory suggests replacing aluminum with lower-TLS materials such as Nb, Ta, or TiN to reduce internal losses and approach near-unity efficiency (Matanin et al., 2022). A third is hybridization with long-lived media in integrated photonics, which could reconcile high bandwidth, long storage time, and telecom compatibility (Moiseev, 7 Jul 2025). A fourth is dynamic control sophistication: fast switchable couplers, tunable detunings, and time-dependent couplings are central to turning echo-based multiresonator devices into full on-demand memories (1705.01536, Bao et al., 2021, Perminov et al., 2023).

Taken together, the literature portrays multiresonator quantum memory as a spectrally engineered family of quantum interfaces in which a finite resonator network replaces or supplements a continuous memory medium. Its theoretical unification lies in transfer-function design and collective rephasing; its experimental trajectory has moved from room-temperature proof-of-principle interfaces through on-demand superconducting chip memories to higher-efficiency on-chip multimode devices; and its long-term significance lies in providing broadband, multimode, and integrable memory primitives for both microwave and optical quantum technologies (1705.01536, Moiseev et al., 2017, Bao et al., 2021, Matanin et al., 2022, Moiseev, 7 Jul 2025).