Quantum repeaters based on atomic ensembles and linear optics

Nicolas Sangouard; Christoph Simon; Hugues de Riedmatten; Nicolas Gisin

Source-linked AI summary

Quantum repeaters based on atomic ensembles and linear optics

Nicolas Sangouard, Christoph Simon, Hugues de Riedmatten, Nicolas Gisin

arXiv:0906.2699v2Jun 15, 2009quant-ph

TL;DR

Photon loss makes direct long-distance quantum-state distribution impractical, while quantum repeaters require heralded entanglement, quantum memories, and entanglement swapping. This review assesses atomic-ensemble repeaters using linear optics and photon counting, emphasizing quantitative comparisons and the prospects for surpassing direct transmission. It concludes that multiplexing may make a simple repeater outperform direct transmission, although practical systems still lack simultaneous high storage time, efficiency, and multimode capacity.

Problem
Photon loss limits direct quantum-state distribution over long distances, motivating repeater architectures that create entanglement from shorter links.
Method
The review compares atomic-ensemble quantum-repeater protocols that use linear optics and photon counting for heralded entanglement generation, storage, and swapping.
Results
Multiplexing appears especially promising for making a simple atomic-ensemble repeater outperform direct quantum-state transmission.
Takeaways & Limitations
Elementary links and key repeater parameters have been demonstrated, but storage time, memory efficiency, and multimode capacity have not yet been achieved simultaneously.
Takeaways & Limitations
DLCZ performance is constrained by a fidelity–rate trade-off from multiphoton errors and by phase-stability requirements over long timescales.

Abstract

from arXiv · show

The distribution of quantum states over long distances is limited by photon loss. Straightforward amplification as in classical telecommunications is not an option in quantum communication because of the no-cloning theorem. This problem could be overcome by implementing quantum repeater protocols, which create long-distance entanglement from shorter-distance entanglement via entanglement swapping. Such protocols require the capacity to create entanglement in a heralded fashion, to store it in quantum memories, and to swap it. One attractive general strategy for realizing quantum repeaters is based on the use of atomic ensembles as quantum memories, in combination with linear optical techniques and photon counting to perform all required operations. Here we review the theoretical and experimental status quo of this very active field. We compare the potential of different approaches quantitatively, with a focus on the most immediate goal of outperforming the direct transmission of photons.

I. INTRODUCTION

Quantum repeaters address photon-loss limits by building long-distance entanglement from shorter links through heralded entanglement creation, quantum memories, and entanglement swapping. The DLCZ approach combines atomic ensembles, linear optics, and photon counting, but faces rate, fidelity, stability, and memory-storage challenges.

Quantum repeater principle: For N = 2^n elementary links of length L0 = L/N, n hierarchical swapping levels can distribute entanglement over distance L with better long-distance scaling than direct transmission.The repeater creates elementary-link entanglement independently and connects neighboring links successively.
Quantum repeater principle: Heralded elementary-link entanglement is created by detecting a photon whose emission path is indistinguishable between remote systems, erasing which-way information.The detection event signals successful entanglement creation at a distance.
Required capabilities: Quantum memories must store elementary and higher-level entanglement while neighboring links are established for later swapping.Without memories, the protocol cannot wait for independently generated links before performing the required connections.
Required capabilities: Entanglement swapping requires local joint measurements on neighboring memories, but general quantum gates are difficult, motivating dedicated probabilistic measurements.The operation connects previously separate entangled pairs without requiring the distant endpoint systems to interact directly.
DLCZ approach: The DLCZ protocol uses atomic ensembles as memories and linear optics with photon counting to implement the repeater requirements.Spontaneous Raman emission correlates Stokes photons with collective spin excitations, which can later be reconverted into photons for swapping.
DLCZ approach: Multiple-pair emissions become more likely as the single-excitation probability increases, limiting repeater fidelity and forcing a rate–fidelity trade-off.The probability of two emitted photons scales as (χt)^4 when the single-pair probability is (χt)^2.

B. Protocol

The DLCZ protocol creates remote single-excitation entanglement through heralded Stokes-photon detection and extends it by readout, interference, and photon detection. Repeated swapping increases distance, while imperfect efficiencies introduce vacuum components and reduce performance.

Entanglement creation: DLCZ entanglement creation uses simultaneous excitation of two remote ensembles and central detection of a single Stokes photon to herald a delocalized atomic excitation.The detected photon could have originated at either ensemble, so which-way information is erased.
Entanglement creation: The elementary-link success probability is P0 = pηdηt, where ηt accounts for transmission through the link.For telecom fibers, ηt = exp(−L0/(2Latt)) and Latt = 22 km corresponds to 0.2 dB/km loss around 1550 nm.
Entanglement swapping: To connect neighboring links, stored excitations are read out as anti-Stokes photons, interfered on a beam splitter, and conditionally detected to project the outer ensembles into entanglement.Successive repetitions establish entanglement between increasingly distant ensembles.
Entanglement swapping: Nonunit detector and memory efficiencies allow false single-click events from two stored photons, adding vacuum components to the swapped state.The effective efficiency is η = ηdηm, and swapping success probabilities depend on the entangled-component weight at each level.
Protocol behavior: Single-photon swapping amplifies the vacuum component approximately twofold at every entanglement-swapping operation, unlike two-photon schemes that keep it constant.This behavior contributes to reduced performance as the nesting level increases.

3. Post-Selection of Two-Photon Entanglement

The protocol post-selects more useful two-photon entanglement from two independently distributed single-excitation chains. Its performance is constrained by multiphoton errors, transmission assumptions, and long memory-storage requirements at the direct-transmission crossover.

Two-photon post-selection: Two independently distributed chains are read out at corresponding locations, and the emitted anti-Stokes photons are interfered and counted to post-select two-photon entanglement.Changing beam-splitter transmission coefficients and phases enables measurements in arbitrary bases.
Two-photon post-selection: The post-selected two-photon state provides a more directly useful resource than single-excitation entanglement, whose measurements are largely limited to the Fock basis.The post-selection probability contributes to the total entanglement-distribution time.
Performance constraints: Multiphoton emissions occur with probability of order p^2 and reduce the fidelity of the distributed state, constraining the allowed pair-emission probability p.For tolerated fidelity reduction 1 − F = 0.1, the maximum p is determined from the error coefficients.
Performance constraints: The repeater comparison assumes 0.2 dB/km fiber attenuation, c = 2 × 10^8 m/s, ηm = ηd = 0.9, and a 10 GHz direct-transmission source.These assumptions define the reference used for distribution-time comparisons.
Performance comparison: 630 km is the DLCZ crossover distance, with Ttot = 340 seconds for n = 2 and p = 0.01; at 1000 km, Ttot = 4100 seconds versus 10^10 seconds for direct transmission.The repeater is faster at longer distances, but the 340-second distribution time implies a comparably long memory-storage requirement.

D. Discussion - Limitations

The DLCZ protocol is limited by multiphoton errors, phase stability, attempt rates, and telecom-wavelength constraints. Two-photon-detection and multiplexed-memory variants target these bottlenecks, though their implementation introduces additional optical and detection requirements.

D. Discussion - Limitations: DLCZ multiphoton errors grow quadratically with the number of elementary links, forcing low emission probability and creating a fidelity–rate trade-off.Improved swapping schemes can keep the vacuum component constant and reduce multiphoton-error growth to linear in the number of links.
D. Discussion - Limitations: For L0 = 125 km, the relevant phase must remain stable over an average 4.5 s interval, requiring active stabilization or self-compensating configurations.Two-photon-detection schemes greatly reduce the channel-stability requirement.
D. Discussion - Limitations: DLCZ permits at most one entanglement-generation attempt per elementary link per interval L0/c, a limitation that multimode memories can overcome.The proposed remedy stores multiple distinguishable temporal modes.
D. Discussion - Limitations: Long-distance operation requires Stokes photons near 1.5 µm, restricting atomic-species choices or requiring inefficient wavelength conversion because of coupling losses.Separating entanglement generation and storage is described as one route around this requirement.
D. Discussion - Limitations: A two-photon-detection swapping design uses polarization-resolved ensembles, reads out the intermediate memories, and projects the remaining modes through coincident detection.Only four of 16 Schmidt-decomposition terms contribute, reducing the swapping success probability to 1/8.

2. Higher-Level Entanglement Swapping

Higher-level swapping extends shorter-range spin-wave entanglement by converting selected memories into anti-Stokes photons and conditioning on two-photon detections. Two-photon-based protocols improve error scaling, but loss sensitivity and multiple-excitation errors still limit practical distribution times.

Higher-Level Entanglement Swapping: Two spin-wave entangled states distributed across ensembles A–D and E–H are connected toward A–H by combining anti-Stokes photons from D and E at a central station.A twofold coincidence projects the remaining memories into a longer-range entangled state, enabling successive swapping operations.
Higher-Level Entanglement Swapping: Multiple-pair errors grow linearly with the number of elementary links in the two-photon protocol, rather than quadratically for single-photon-based entanglement connection.The improved scaling is attributed to the stationary vacuum component and the interaction of vacuum and multiphoton components in the final state.
Higher-Level Entanglement Swapping: 610 km is the crossover distance for the Jiang et al. protocol, which achieves a 190-second entanglement distribution time and is about four times faster than DLCZ.The optimum number of links at this distance is four, although the distribution time remains long for a single entangled pair.
Higher-Level Entanglement Swapping: The Jiang et al. protocol’s improvement remains modest because multiple-excitation errors force a low emission probability, p = 0.037 for four links versus p = 0.010 for DLCZ.These excitations are difficult to detect because the corresponding Stokes photons propagate over long distances and are often lost.
Entanglement Generation via Two-Photon Detections: Two-photon detection in the elementary generation step removes the need for long-distance phase stability because entanglement depends on the photons’ internal degrees of freedom.This approach contrasts with single-photon detection, where propagation phases determine the delocalized excitation’s entangled character.
Entanglement Generation via Two-Photon Detections: 640 km is the distance at which the two-photon elementary-generation protocol begins to outperform direct transmission, with an entanglement distribution time of 610 seconds.Its rate is comparable to, but slightly worse than, the DLCZ protocol under the stated assumptions.

3. Other Protocol

These protocols separate entanglement generation from storage using photon-pair sources and absorptive memories, then improve rates through temporal multimode operation. Temporal multiplexing increases elementary-link attempts, but swapping requires mode matching and remains subject to phase-stability constraints.

Entanglement generation and storage: Photon-pair sources and absorptive memories emulate DLCZ-type entanglement generation while allowing greater wavelength flexibility for the memory.The stored modes need not share the telecom-wavelength requirement imposed on the photons sent to the central station.
Entanglement generation and storage: A single detected photon at a central beam splitter heralds entanglement between memories holding the partner modes.Each source emits a pair; one mode is stored locally and the other is sent to the central station.
Temporal multiplexing: Temporal multiplexing lets sources be triggered many times within one communication interval instead of only once after unsuccessful attempts.The memory stores a train of time-bin modes labeled i = 1, ..., Nm.
Temporal multiplexing: For NmP0 ≪ 1, storing Nm modes increases the elementary-link entanglement-generation probability by a factor Nm and directly increases the distribution rate by the same factor.The gain occurs at the lowest repeater level, while higher levels remain unchanged.
Temporal multiplexing: Successful swapping with temporal multimode memories requires recombining precisely the modes whose partners generated heralded entanglement.The memory bandwidth can limit the number of storable modes even when storage capacity is larger.
Performance and limitations: The multimode protocol starts to outperform direct transmission at 510 km, with a 1.4-second entanglement distribution time for four elementary links under ηm = ηd = F = 0.9 and Nm = 100.The assumed source operates at 10 GHz; the result is also more compatible with realistic memory times.
Performance and limitations: The protocol retains phase-stability issues because it relies on single-photon-detection entanglement generation, although its shorter distribution time somewhat reduces them.More robust protocols are harder to multiplex because local single- or two-photon entanglement preparation consumes repetitions within each communication interval.
Multiplexing outlook: Spatial multiplexing can greatly reduce required memory times, while combining temporal and spatial multiplexing is identified as a longer-term way to increase repeater rates.The rate advantage over parallel independent repeaters is described as moderate.

D. Spatially Multiplexed Memories

Spatial multiplexing offers a moderate rate advantage over independent parallel repeaters and a substantial reduction in memory-time requirements. The single-photon-source protocol separately improves rates by avoiding errors proportional to the entanglement-creation probability, outperforming direct transmission at 580 km under stated assumptions.

D. Spatially Multiplexed Memories: Strongly multiplexed repeaters provide a moderate rate advantage over Nr independent repeater architectures, with related work finding rate scaling as Nr^1.12.The comparison concerns spatial multiplexing against completely independent parallel architectures.
D. Spatially Multiplexed Memories: Spatial multiplexing greatly reduces required memory times compared with strictly parallel repeaters, whose memories must wait for each repeater individually.The paper identifies this memory-time reduction as a very significant advantage of the multiplexed architecture.
E. Single-Photon Source based Protocol: The single-photon-source protocol uses beam splitters to distribute each source photon between a local memory and a central detection mode.A single-photon detection heralds entanglement between the two remote memories.
E. Single-Photon Source based Protocol: Unlike DLCZ-type protocols, the single-photon-source scheme has no error term proportional to the entanglement-creation probability, improving distribution rates despite a larger vacuum component.The larger vacuum component lowers higher-level swapping success probabilities, partially offsetting the gain.
E. Single-Photon Source based Protocol: 580 km marks the distance where the protocol begins outperforming direct transmission, achieving a 44-second entanglement distribution time with four links and β2 = 0.16.The assumptions are ηm = ηd = p1 = 0.9 and a 10 GHz single-photon source; the time is about an order of magnitude faster than DLCZ.

3. Implementing the Single Photon Source Protocol with Atomic Ensembles

Atomic ensembles can approximate ideal single-photon or photon-pair sources, enabling repeater protocols based on partial storage and local entangled-pair preparation. Partial readout produces higher-fidelity local entangled pairs than emission followed by storage, yielding improved overall repeater rates.

Single-photon sources: A DLCZ-type ensemble can serve as a single-photon source because a stored atomic excitation can later be reconverted into a photon.The resulting source can be combined with partial readout to implement the single-photon-source protocol.
Single-photon sources: The single-photon-source implementation requires both the source and memory to operate at telecom wavelength, while appropriate multimode memories could still permit temporal multiplexing.A 150 km link has a communication time of about 750 µs, leaving considerable scope for multiplexing.
Local entangled-pair generation: Local entangled pairs of atomic excitations can be prepared from four ensemble-based single-photon sources, linear optics, and two quantum memories.The approach targets effective single-pair sources for two-photon-detection repeater protocols.
Local entangled-pair generation: Partial readout generates higher-fidelity entangled atomic pairs than emission followed by storage, improving the overall repeater rate for equal memory and detection efficiencies.The scheme avoids an emission-and-storage step.
Local entangled-pair generation: Four ensembles are independently charged through heralded Stokes-photon detections before simultaneous partial readout releases anti-Stokes photons for central-station detection.A twofold coincidence projects the remaining spin-wave modes nondestructively onto an entangled state.
Protocol operation: Stored excitations can be converted at different times, allowing one excitation to support entanglement generation and the other to be reserved for swapping or final use.Simultaneous conversion also permits the setup to function as a photon-pair source.
Local entangled-pair generation: The ideal local entangled-pair preparation succeeds with probability Ps = 2α4β4.Coincidences in the four detector-output combinations can contribute after appropriate one-qubit transformations.
Imperfections: Nonunit detector and memory efficiencies introduce additional detected-photon contributions that alter the conditionally prepared state's fidelity.The analysis explicitly considers events where three or four photons are emitted but only two are detected.

2. Repeater Protocol using Two-Photon Detections

This protocol creates heralded remote entanglement between atomic ensembles using two-photon detections and then extends it through repeated linear-optical entanglement swapping. Under stated assumptions, it outperforms direct transmission at 560 km, while remaining robust to channel phase fluctuations.

Entanglement creation: Remote ensembles generate heralded entanglement through a central linear-optical station and twofold photon detections.The resulting state retains vacuum and single-spin-wave components, with the remaining memories projected after detection.
Entanglement swapping: Two-photon detections enable n successive entanglement-swapping operations using the same linear-optical elements and detectors.The protocol propagates the distributed state across successive repeater levels while preserving the relevant source-state weights.
Success probabilities: The swapping success probability is Pi = 2η^2 (cs^2/2 + cs^1)^2 for the i-th local swapping operation.Because swapping is local, transmission losses do not enter this success probability.
Performance: At r = 10 MHz, the protocol begins outperforming direct transmission from a 10 GHz single-photon source at 560 km, distributing an entangled pair in 15 seconds.The stated repeater uses 8 links, p = 0.013, α2 = 0.26, and ηm = ηd = F = 0.9.
Performance: The protocol achieves the best performance among known non-multiplexed atomic-ensemble, linear-optics protocols and is robust to channel phase fluctuations because it uses two-photon detections.Its repetition rate is limited by source-preparation time, making multiplexing difficult in the stated implementation.

IV. COMPARISON AND DISCUSSION

The comparison evaluates repeater distribution times and robustness under realistic losses, efficiencies, memory times, phase stability, and purification choices. All protocols beat direct transmission around 500–650 km, but rates and technological requirements differ substantially.

Entanglement Distribution Time: The comparison targets final fidelity F = 0.9 and includes multi-photon emission errors, while other imperfections can affect protocols differently.The plotted quantity is the average time to distribute one entangled pair as a function of distance.
Entanglement Distribution Time: Two-photon protocols are more sensitive to photon loss, favoring larger numbers of elementary links; the comparison therefore limits protocols to at most 16 links.This limit has little effect on most shown protocols and changes the Chen et al. protocol by about a factor of 2.
Entanglement Distribution Time: All protocols begin outperforming direct transmission between 500 and 650 km, but the fastest protocol still requires very long distribution times.These long times imply low quantum-communication rates and demanding quantum-memory requirements.
Entanglement Distribution Time: Multiplexing gives the multi-mode-memory protocol a rate advantage, while multiplexing other protocols is more challenging because of source repetition-rate and memory-bandwidth demands.Temporal multiplexing is emphasized, although spatial and frequency multiplexing may provide additional benefits.
Technological imperfections: When τ ≪ L/c, repeater rates decline as exp(−L/cτ), making memory times comparable to total entanglement-creation times essential.Certain multiplexing strategies can reduce memory-time requirements, whereas simply running repeaters in parallel does not.
Technological imperfections: Single-photon protocols require long-distance phase stability, and residual phase errors are typically amplified by a factor of 2 at every swapping operation.Two-photon protocols are less severely affected because propagation and laser phases contribute only a global phase, although stabilization may still be required for photon overlap.
Technological imperfections: Entanglement purification adds complexity and slows protocols, so it is omitted from the comparisons focused on beating direct transmission.The authors favor minimizing errors and avoiding purification for this immediate goal.
Memory Efficiency: A memory-efficiency reduction from 90% to 89% increases entanglement-distribution time by 10–14%, depending on the protocol.Single-photon protocols are less sensitive than two-photon protocols because they require fewer memories.

C. Complexity

Comparing quantum repeater complexity requires more than counting components: practical performance also depends on stabilization, cooling or trapping, and demanding memory and detection specifications.

Resource comparisons: 510 km is the cross-over distance for a two-link protocol using four sources and four multi-mode memories per link, with a 2.8-second entanglement distribution time.A four-link repeater achieves 1.4 seconds under the cited comparison.
Practical complexity: Element counts alone are insufficient because protocols differ in phase-stabilization requirements and memory implementations.Temporal multimode memories may use cryogenic solid-state ensembles, whereas DLCZ experiments used optically cooled and trapped atomic gases.
Performance requirements: Practical repeaters likely need several-second storage, memory and detection efficiencies of at least 90 percent, stabilized fiber links, and minimal local coupling losses.The review concludes that outperforming direct transmission appears possible with moderately complex architectures, provided these components are excellent.
Experimental progress: DLCZ experiments established nonclassical Stokes–anti-Stokes correlations, with an early measurement of R = 1.84 ± 0.06 > 1.Residual fluorescence and excitation-pulse leakage limited the inequality violation, while later improvements produced substantially higher correlations.
Memory constraints: DLCZ experiments require managing atomic motion, which can cause diffusion and dephase collective spin excitations during storage.Motion-induced decoherence is particularly restrictive in non-collinear geometries and can be mitigated through larger beams, colder atoms, or optical lattices.
Memory constraints: For θ = 3° and T = 100 µK, the predicted storage lifetime is 25 µs, while a collinear configuration achieves a lifetime of about 1 ms.Reducing the angle increases the spin-wave wavelength and storage lifetime, but non-collinear operation supports spatial filtering of excitation beams.

2. Heralded Entanglement between two Atomic Ensembles

Heralded entanglement between spatially separated atomic ensembles was demonstrated using single-photon detection and quantum interference, then improved through higher retrieval efficiency and parallel ensemble chains. These experiments established elementary repeater segments and verified stored or effective entanglement.

Heralded entanglement: Heralded entanglement was first demonstrated between two cold cesium ensembles separated by 3 m using quantum interference in emitted light.A beam-splitter detection erased which-way information and heralded a delocalized single excitation.
Improved entanglement: At gS,AS = 60, the improved experiment measured output concurrence C = 0.35 ± 0.1 and inferred atomic concurrence C = 0.9±0.3, with entanglement persisting for at least 20µs.The earlier low concurrence was attributed mainly to limited retrieval efficiency of 10%.
Elementary repeater segment: Parallel chains enabled effective two-excitation entanglement by post-selection while relaxing phase-stability requirements and allowing independent elementary-link generation.The approach was realized with four ensembles in two nodes separated by about 3 m.
Elementary repeater segment: The first experimental elementary repeater segment retrieved excitations simultaneously, post-selected one detection at each node, and verified effective entanglement by violating a Bell inequality.The experiment’s independent entanglement generation was limited by a memory time of about 10µs.

B. Entanglement Creation and Swapping based on Two-Photon Detections

Two-photon interference supports entanglement generation, swapping, and final post-selection in ensemble-based repeater schemes. Experiments demonstrated interference and photon-source enhancement, while visibility, multiphoton components, and interferometric stability remained important constraints.

Two-photon interference: Two-photon interference is used for entanglement swapping, elementary-link generation, and post-selection in protocols based primarily on single-photon detections.The Hong-Ou-Mandel effect appears as a dip in coincidences when the photons are indistinguishable.
Experimental demonstrations: A first experiment with separate cold cesium memories demonstrated two-photon interference and a 28-fold increase in pair-generation probability through conditional control.The photons were heralded single photons generated independently in the two ensembles.
Experimental demonstrations: With Rb atoms, the measured photon coherence time was 25 ns and Hong-Ou-Mandel visibility was 80%, limited by two-photon contributions.The dip was measured by varying both relative read-pulse delay and frequency detuning.
Experimental demonstrations: Unconditional Stokes photons produced a Hong-Ou-Mandel visibility of 33%, reflecting their thermal character and equal probabilities for same-ensemble and separate-ensemble photon pairs.This configuration differs from interference between conditional anti-Stokes photons retrieved from stored excitations.
Repeater requirements: High Hong-Ou-Mandel visibility is essential because the dip visibility determines swapping fidelity, while accumulated swapping errors grow linearly with the number of links.The reviewed experiments had visibility too low for that repeater context and were mainly limited by two-photon components.
Light–matter entanglement: Encoding a logical qubit in collective excitations across spatial modes enables tunable relative excitation probabilities but requires interferometric stability.The technique supported mapping an atomic qubit to a photonic qubit and teleportation onto a matter qubit.
Light–matter entanglement: In a 300 m-fiber connection, the post-selected fidelity was F = 0.83 ± 0.02, while a 6 m connection supported Bell-inequality violation for storage times up to ∼4µs.These results used retrieved photons from the two ensembles.

4. Deterministic Local Generation of Entanglement

The review surveys atomic-ensemble sources for generating photons and entanglement compatible with quantum memories, including deterministic local entanglement, heralded single photons, and narrow-band or telecom-wavelength pairs.

Deterministic Local Generation of Entanglement: Deterministic local entanglement of nearby ensembles is useful for robust repeater architectures but is highly sensitive to loss for remote entanglement.These schemes are discussed as potential sources of high-fidelity number-state entanglement between nearby ensembles.
Photon-pair sources: 74% of Stokes photons were paired, with R = 11600 and 0.75 MHz linewidth in an improved two-dimensional magneto-optical-trap source.The generated photon pairs had coherence times up to 900 ns.
Narrow-band and telecom-wavelength sources: DLCZ-type sources lack direct flexibility at 1.5 µm, motivating wavelength conversion and atomic-cascade schemes for non-degenerate photon pairs.Telecommunication compatibility is a central requirement for repeater applications.
Single photon sources: DLCZ-like ensembles generate strongly correlated Stokes and anti-Stokes fields, enabling heralded single photons with programmable delay.A detected Stokes photon heralds a stored excitation that can later be converted into an anti-Stokes photon.
Quantum memories based on EIT: EIT maps light into collective atomic spin excitations by slowing and compressing the field, then retrieves it by restoring the control laser.The technique has also mapped single-photon polarization qubits into remote matter-qubit entanglement.
Quantum memories based on EIT: EIT multimode storage requires extremely high optical depth because efficient storage of many modes simultaneously demands slow group velocity and a large transparency window.These requirements create poor scaling with optical depth.

2. Photon Echo based Quantum Memories

Photon-echo memories reverse or engineer inhomogeneous dephasing to rephase atomic dipoles and retrieve stored light. CRIB and AFC offer complementary routes toward high-efficiency, multimode, and long-lived storage, with experimental progress but important constraints.

Photon-echo limitations and alternatives: Traditional optical rephasing pulses can produce fluorescence that blurs stored single-photon states.This motivates modified photon-echo approaches such as CRIB and AFC.
Controlled Reversible Inhomogeneous Broadening: CRIB reverses each emitter’s detuning after absorption so the atomic dipoles rephase and collectively emit the stored light at a later time.The detuning transformation is δ_i → −δ_i, producing collective emission at 2τ after absorption.
CRIB performance: CRIB storage and retrieval can reach 54% in one configuration, limited by reabsorption, while storage time is constrained by the initial absorption width.Transfer to a long-lived ground state can extend storage time beyond excited-state coherence limits.
CRIB performance: Theoretically, ground-state-transfer CRIB can reach η_B = 100% for sufficiently high optical depth, and longitudinal broadening can also enable 100% efficiency in a two-level system.These results are theoretical proposals or limits within the stated configurations.
Multimode storage: CRIB can store a number of high-efficiency temporal modes proportional to the initial optical depth, N_m ∼ d.This contrasts with AFC, whose mode capacity is independent of optical depth under the stated scaling.
Atomic Frequency Combs: AFC mode capacity is proportional to the number of comb peaks, N_m ∝ N_p, while the total bandwidth is limited by atomic level spacings.AFC experiments demonstrated 32 stored pulses for 1.5 µs and later full protocol operation including ground-state transfer.

E. Detectors

The review evaluates detector technologies and broader implementation constraints for ensemble-based quantum repeaters. Detector performance, passive losses, stabilized fiber links, and simultaneous system-level capabilities remain central practical boundaries.

Detector technologies: APDs are practical without cryogenic cooling, but present performance is insufficient for practical quantum repeater architectures.InGaAs APDs detect 1550 nm photons but have a worse efficiency-to-noise ratio, and standard APDs usually lack photon-number resolution.
Detector technologies: Superconducting nanowire detectors achieved 57% efficiency at 1550 nm and 67% at 1064 nm with low dark counts and excellent temporal resolution below 4 K.Their cryogenic operation remains a practical requirement.
Detector technologies: Transition-edge sensors provide 95% efficiency at 1556 nm with photon-number resolution, but respond in about 1 µs and require operation near 100 mK.These trade-offs matter for telecom-wavelength repeater operation.
Implementation constraints: A practical repeater that beats direct transmission requires considerable reduction of passive losses.The review identifies passive losses as a direct constraint on the target comparison.
Conclusions and outlook: Efficient multiplexing makes a simple atomic-ensemble repeater that outperforms direct transmission seem realistic in the not too distant future.The review reports theoretical and experimental progress, while noting that key capabilities have not yet been demonstrated simultaneously in one system.
Implementation constraints: The first useful repeater also requires stabilized long-distance fiber links and near-elimination of coupling losses between components.These requirements accompany entanglement generation, storage, and swapping capabilities.

APPENDIX A: Calculating the Entanglement Distribution Time

The appendix derives waiting times for creating and swapping entanglement across nested repeater links. Higher-level waiting-time distributions become more complicated, but numerical evidence supports a useful approximation for their scaling factor.

Creation of entanglement for an elementary link: Elementary-link entanglement creation has an exponential waiting-time distribution governed by success probability P0.The waiting time is expressed in units of L0.
Waiting for a success in two neighboring elementary links: The first swapping attempt requires successful entanglement creation in two neighboring elementary links, whose combined waiting time is the maximum of their individual waiting times.The combined distribution is denoted ˜p(n), with expectation value ⟨˜n⟩.
First entanglement swapping: A successful first-level swapping event occurs probabilistically across repeated attempts, with success probability P1 and a waiting time accumulated over the possible attempt numbers.The calculation includes immediate success, second-attempt success, and subsequent possibilities.
Higher levels in the repeater protocol: At the lowest repeater level, waiting for two neighboring-link successes takes 3/2 times the average waiting time for one link.This factor is stated as essentially exact for the lowest level.
Higher levels in the repeater protocol: Higher-level waiting-time distributions depend on P0 and all lower-level swapping probabilities Pi, preventing useful general analytical results so far.Numerical evidence suggests that 3/2 remains a good approximation for the factor f, with 1 ≤ f ≤ 2.
Second entanglement swapping and general formula: At each higher level, two lower-level copies are combined through a success probability and an f factor, including the final post-selection step.The general formula treats higher-level swapping and single-photon-detection post-selection analogously.

APPENDIX B: Multiphoton Errors in the DLCZ Protocol

The appendix analyzes multiphoton errors in the DLCZ protocol through the repeater's swapping and post-selection sequence. The resulting fidelity corrections grow with nesting level, while ideal number-resolving detection and perfect memories would eliminate the considered errors.

Error analysis setup: The DLCZ error analysis starts from the entanglement-generation state and explicitly retains O(p) terms with phases set to zero.The resulting state is then propagated through entanglement swapping and final post-selection.
Error analysis setup: Higher orders in p produce errors, whereas the O(√p) terms are the desired protocol contributions.This separates the intended state components from multiphoton-error contributions.
Entanglement swapping: The protocol reconverts atomic modes to photonic modes and combines them on a beam splitter during entanglement swapping, which can succeed with a single click.The analysis applies this operation to states established between non-neighboring stations.
Final post-selection and fidelity: Final post-selection projects onto one photon on each side, with imperfect detectors requiring probabilities for single detections in the final density matrix.The fidelity is obtained from the overlap of the properly renormalized final state with the ideal final state.
Final post-selection and fidelity: The fidelity F(n) is expanded to second order in p as a function of nesting level n, with coefficients An and Bn governing the correction terms.The lowest listed coefficients are A0 = 8, A1 = 18, A2 = 56, A3 = 204, A4 = 788 and B0 = 37, B1 = 250, B2 = 2966, B3 = 43206, B4 = 669702.
Scaling of multiphoton errors: From n = 3 onward, An scales approximately as N 2 and Bn approximately as N 4, where N = 2n is the number of links.The N 2 scaling is linked to combining a multiphoton component from one ensemble pair with the vacuum component from another.
Scaling of multiphoton errors: Ideal photon-number-resolving detectors and perfect memories would identify all undesirable multiphoton events and eliminate the considered errors.The dependence on (1−η) indicates the role of detector inefficiency in the fidelity reduction.