Experimental demonstration of entanglement delivery using a quantum network stack

The work addresses how to transition from physics-level demonstrations of entanglement generation to robust, scalable quantum communication systems. The research question is how to implement and demonstrate a platform-independent link layer that abstracts entanglement generation into a reliable, on-demand service between adjacent quantum nodes, enabling higher-layer quantum networking applications. The study situates this within the vision of a Quantum Internet that enables secure communication, distributed/blind quantum computation, and metrology via entanglement distribution across diverse hardware (ions, neutral atoms, NV centers, quantum dots). Drawing from classical networking (TCP/IP, OSI), the authors emphasize layered abstractions where the link layer ensures reliable entanglement between neighboring nodes while higher layers remain agnostic of physical details. They also note coexistence and interdependence with classical networks for control and application-level classical communication, and the potential of SDN controllers for scheduling and QoS in quantum networks. The paper proposes and experimentally validates a quantum link layer protocol that provides robust, platform-independent, on-demand entanglement delivery and evaluates fidelity and latency metrics compared to non-abstracted implementations.

Prior works have demonstrated fundamental entanglement primitives across multiple platforms: trapped ions, neutral atoms, NV centers in diamond, and quantum dots. Building scalable quantum networks motivates layered architectural proposals inspired by classical stacks (TCP/IP, OSI), including specific quantum stacks where a link layer provides reliable adjacent-node entanglement and a network layer extends connectivity. Dahlberg et al. proposed the Quantum Entanglement Generation Protocol (QEGP) for the link layer and a supporting physical-layer Midpoint Heralding Protocol (MHP) for heralded links, relying on precise timing and mismatch checks. Additional literature explores SDN-style centralized control for scheduling and QoS, as well as routing techniques and resource scheduling policies for larger-scale, multi-user quantum networks. The study builds on these by revising QEGP/MHP to address challenges (scaling distributed queues, strict real-time constraints for triggering, and mismatch verification at the heralding station) via centralized scheduling, entrusting fine-grained synchronization to the physical layer, and performing mismatch checks at the nodes.

Architecture and protocols: The system implements a quantum link layer (QEGP) atop a physical layer (MHP) between two 2 m-separated NV-center-based quantum nodes. The link layer exposes a platform-independent entanglement delivery interface with parameters: remote node ID, number of pairs, minimum fidelity, delivery type (K keep, M immediate measure, R remote state preparation), measurement basis, and request timeout. It returns produced Bell state, measurement outcomes (for M/R), and unique entanglement IDs. The revised protocol replaces DQP with centralized scheduling, delegates precise entanglement attempt triggering to MHP, and moves request mismatch verification to the nodes. Centralized scheduling and TDMA: A logically centralized scheduler produces static TDMA schedules (manually installed for this single-link study), assigning time bins per request class. Time bins in experiments are 20 ms, accommodating up to four batches of 1000 attempts per bin. MHP synchronization and batching: After QEGP hands an entanglement command to MHP, nodes exchange readiness signals, synchronize using a shared clock/DIO, then trigger attempts at precise times. If peer not ready within timeout, control returns to QEGP avoiding wasted attempts. Attempts are batched so synchronization/reporting occurs once per batch to mitigate overhead. Interface and commands: A defined HAL/device interface supports commands: INI (qubit init), MSR (measurement), SQG (single-qubit gate), ENT (entanglement attempt; ENM for immediate measure types M/R), and PMG (premeasurement basis selection). Outcomes report execution status, measurement results, and produced Bell state. QEGP delivers |Φ+⟩ by default, applying a local Pauli correction if needed. Software/hardware stack: The network controller (QEGP, instruction processor, HAL) runs as a C/C++ runtime on FreeRTOS on an Avnet MicroZed (Zynq-7000 SoC, ARM Cortex-A9 @667 MHz). Inter-node communication uses TCP over Gigabit Ethernet; HAL-device controller uses 12.5 MHz SPI. Applications are written in Python using the NetQASM SDK and compiled to NetQASM low-level instructions. The physical layer device controller is a time-deterministic microcontroller (Jäger ADwin Pro II; Zynq-7000 SoC, dual-core ARM Cortex-A9 @1 GHz). Each node uses a Zurich Instruments HDAWG8 (2.4 GSa/s, 300 MHz sequencer) for nanosecond tasks. Qubit control and protection: Before operation, a charge and resonance (CR) check prepares the NV center. INI (100 μs) and MSR (10 μs) are deterministic. SQG pulses (~100 ns) are chosen from 64 precompiled waveforms via 32-bit DIO; X and Y rotations in π/16 steps are implemented (Z planned). To counter short idle coherence (~5 μs), a real-time qubit protection module on the AWG interleaves dynamical decoupling with requested gates until measurement. Timing and entanglement attempts: Device controllers share a 1 MHz common clock and DIO for synchronization; internal cycle counters are aligned and shared with the network controller. ENT execution uses a three-way handshake; if peer unavailable, a 0.5 ms ENT_SYNC_FAIL timeout returns; otherwise, optical phase stabilization runs, then both AWGs are triggered (server’s controller triggers both to achieve sub-ns jitter). Each attempt lasts 3.8 μs (fast init, communication-qubit–photon entanglement, probabilistic entanglement swapping). Up to 1000 attempts per batch are made before ENT_FAIL; longer batches are avoided due to charge-state risks (no charge-stabilization implemented). Success probability per attempt is ~5×10^-5. Success is heralded by a CPLD signal; AWGs enter protection and device controllers report the Bell state to the link layer. For M/R types, ENM executes immediate measurement based on prior PMG. Deployment specifics: QEGP peers communicate over TCP/GbE; the device controller–network controller coupling is via SPI. The simple, single-link schedule is static; mismatch verification at MHP is specified but not implemented due to reliable classical links in this setup. Evaluation applications and measurements: Three applications run back-to-back without hardware/software changes: (1) Full state tomography at requested min fidelity 0.80, measuring all 9 two-qubit correlators with ± basis variants (36 combinations; 125 shots each; total 4500 states). Bayesian tomography (QInfer, 10^5 particles) reconstructs the density matrix. Post-processing corrects known tomography errors and discards events where devices were in wrong charge state. (2) Fidelity/latency vs requested fidelity: requested fidelities {0.50,0.55,0.60,0.65,0.70,0.75,0.80}; measure XX, YY, ZZ (and ± variants; 12 correlators), 1500 states per fidelity (10,500 total). Physical layer is calibrated to target ~0.03 above requested min. Latency is measured from CREATE issuance to successful physical-layer outcome, decomposed into link layer protocol overhead, inter-controller interface, physical-layer CR check, and entanglement generation; requests >10 s latency are excluded (plateau periods). (3) Remote state preparation (R-type): client measures immediately in one of six cardinal bases to prepare ±x, ±y, |0⟩, |1⟩ on the server; server performs tomography across six bases; requested fidelity 0.80; 6×6×125 = 4500 preparations. Corrections analogous to tomography are applied.

- The implemented link (QEGP) and physical (MHP) layers deliver a robust, platform-independent entanglement service usable by applications in real-time on solid-state NV nodes separated by 2 m. - Full-state tomography of delivered states yields fidelity F = 0.783(7) to |Φ+⟩. This is ≈3% lower than prior non-abstracted experiments, attributed to added decoupling (≈300 μs) and an extra Pauli correction gate. - Fidelity versus requested minimum fidelity: measured fidelities meet or exceed requests (physical layer targeted ~+0.03 above requested). Without charge-state event correction, fidelities decrease by a few percent. - Latency analysis: total latency dominated by physical-layer entanglement attempts and CR check, consistent with expected attempts from single-photon protocol. The link layer adds ~10 ms overhead (largely TDMA bin waiting and peer synchronization) largely independent of requested fidelity; inter-controller interface contributes negligible latency. Batching and multi-pair CREATEs can amortize overhead. - Stable entanglement delivery sustained over ~30 minutes; delivery pauses and rate changes correlate with resonance conditions and CR checks. - Remote state preparation achieves average preparation fidelity F = 0.853(8). Client measurement fidelities F0 = 0.928(3), F1 = 0.997(1) influence prepared-state identification. Asymmetry between |0⟩ and |1⟩ fidelities traces to population imbalance (|01⟩ vs |10⟩) due to double-occupancy error in the single-photon entanglement protocol. - Entanglement attempt success probability ≈ 5×10^-5; up to 1000 attempts per batch; each attempt 3.8 μs.

The results demonstrate that a quantum link layer can abstract the complexities of physical entanglement generation into a reliable, on-demand, platform-independent service suitable for higher-layer applications. The implemented QEGP revisions address scalability and synchronization challenges by adopting centralized TDMA scheduling, delegating precise timing to the physical layer, and preparing for node-side mismatch verification. Experiments show that delivered entanglement meets requested fidelity targets across a range, with only marginal overhead relative to non-abstracted setups. Latency is predominantly governed by physics (attempt rates and CR checks), while link-layer overhead (~10 ms) is modest and can be further mitigated by optimized scheduling (shorter/matched time bins, multi-pair requests) and tighter controller integration. Successful remote state preparation validates application-level utility (a step toward blind quantum computation) and highlights the importance of readout fidelity and protocol-specific error modes. The work bridges physics experiments and quantum communication systems, providing a blueprint for scaling to multi-node networks and integrating higher layers (network/transport) while remaining agnostic to hardware specifics. Security considerations (e.g., authentication of classical control) and scheduling/routing remain active areas that will become more critical as networks scale.

This study experimentally realizes a quantum link layer and supporting physical layer that deliver entanglement as a robust, platform-independent service between NV-based network nodes. It validates the stack by running full-state tomography, fidelity/latency benchmarking across requested fidelities, and remote state preparation, showing that requested fidelities are met with modest additional latency and that applications can operate without awareness of physical details. Key contributions include (i) a revised QEGP/MHP with centralized TDMA scheduling, physical-layer synchronization with batching, and a defined HAL interface; (ii) real-time qubit protection enabling on-demand gates; and (iii) quantitative fidelity/latency characterization under realistic control. Future work includes implementing node-side mismatch verification, dynamic/optimized centralized schedulers, shortening and adapting TDMA bins, integrating device and network controllers to reduce interface overhead, enabling user-selectable Bell states and avoiding unnecessary Pauli corrections, adding charge-state stabilization for longer attempt batches, authenticating classical control messages, and extending to multi-node networks with network/transport layers for end-to-end entanglement and qubit transmission.

- Single-link, two-node setup (2 m separation); results may differ over longer distances and in multi-link networks. - Centralized scheduling is static and manually configured; no dynamic scheduler or contention scenarios were evaluated. - Mismatch verification at the physical layer (MHP) was specified but not implemented; reliance on a reliable classical channel. - Charge-state stabilization not implemented, limiting batch size to 1000 attempts and contributing to pauses/latency. - Post-processing removes wrong charge-state events and corrects tomography errors; such corrections are not yet integrated into the link-layer service for arbitrary applications. - Additional Pauli correction (to always deliver |Φ+⟩) and extended decoupling sequences slightly reduce fidelity relative to prior non-abstracted experiments. - Link-layer overhead (~10 ms) arises from TDMA binning and synchronization; time-bin optimization and multi-pair requests were not employed in measurements. - Z-axis single-qubit rotations not yet implemented (planned); device and network controllers are separate, adding minor interface latency. - Security aspects (e.g., authentication of classical control messages) not implemented and could affect robustness in open networks.