The future Quantum Internet promises advancements in secure communication, distributed quantum computation, and metrology by sharing entangled states over long distances. Various physical platforms have demonstrated fundamental entanglement primitives, but scaling these to large networks requires robust control and abstraction of underlying physical implementations. Classical networking utilizes layered architectures (like TCP/IP) to manage complexity. This paper adopts a similar approach, proposing a quantum network stack with layers for application, transport, network, link, and physical functionalities. The physical layer handles entanglement generation, while higher layers provide abstraction and reliability. Crucially, the Quantum Internet is expected to coexist with the classical internet, using classical communication to manage network resources and facilitate quantum applications. This work experimentally demonstrates a link layer protocol for entanglement-based quantum networks, focusing on robust, platform-independent, and on-demand entanglement generation between two nodes. This link layer, following the TCP/IP model, sits above the physical layer and aims to abstract away the complexities of entanglement generation from applications.

Several network stacks have been previously proposed for quantum network nodes, drawing inspiration from classical architectures such as TCP/IP and OSI model. These proposals often include a link layer responsible for reliable quantum communication between directly connected nodes, with higher layers managing end-to-end connectivity and qubit transmission. The existing literature highlights the importance of abstraction to improve scalability and platform independence in quantum networks. Challenges in implementing such stacks, including time synchronization and mismatch verification, have also been discussed in previous studies. This work builds upon the QEGP (Quantum Entanglement Generation Protocol) and MHP (Midpoint Heralding Protocol) proposed in earlier research, addressing limitations identified within those previous protocols.

The researchers implemented a revised version of the QEGP, addressing three key challenges: centralized request scheduling (replacing the original link-local protocol), delegating entanglement attempt triggering to the physical layer to reduce real-time constraints, and performing request mismatch verification at the nodes instead of a central heralding station. A centralized request scheduler uses a TDMA (Time-Division Multiple Access) schedule to manage entanglement requests. Entanglement attempts are batched at the physical layer to reduce synchronization overhead. The link layer can deliver |Φ⁺⟩ Bell states, applying Pauli corrections as needed. The physical layer, implemented using diamond NV centers separated by 2 meters, uses a state-machine-based algorithm on a microcontroller (device controller) and an AWG (Arbitrary Waveform Generator) for real-time control. The device controller performs qubit initialization, measurement, single-qubit gates, and entanglement attempts. The AWG generates precisely timed pulses for qubit manipulation. A qubit protection module interleaves decoupling sequences with gate operations to maintain coherence. Entanglement generation involves a three-way handshake between device controllers to synchronize attempts. The system uses a 1 MHz shared clock and a DIO (Digital Input/Output) connection for synchronization. The software stack includes a network controller (with QEGP, instruction processor, and HAL—Hardware Abstraction Layer) implemented in C/C++ on an Avnet MicroZed platform. User applications are written in Python using a NetQASM SDK, which translates them into low-level instructions. Three applications—full quantum state tomography, fidelity-latency trade-off analysis, and remote state preparation—are used to evaluate the system.

The researchers performed full quantum state tomography on the delivered entangled states, achieving a fidelity of 0.783(7) with the |Φ⁺⟩ state. This is slightly lower than results without the network stack (≈3%), attributed to additional overhead from real-time operation and Pauli corrections. The experiment also investigated the fidelity-latency trade-off, demonstrating the ability to deliver states at various fidelities (0.50 to 0.80) with a consistent increase in latency at higher fidelities, primarily due to increased entanglement attempt requirements at the physical layer. The link layer adds approximately 10ms latency per request, although this can be mitigated by batching multiple entanglement requests. Finally, remote state preparation of a qubit on the server by the client was successfully demonstrated with an average fidelity of 0.853(8). The fidelity of the prepared states is affected by the measurement error of the client. The system demonstrated stable entanglement delivery over 30 minutes, with occasional pauses due to variations in resonance conditions of the NV centers.

The results demonstrate a functional link and physical layer for entanglement-based quantum networks, showcasing robust and platform-independent entanglement delivery. The quantified latency overhead introduced by the link layer is manageable and can be further reduced with optimizations. The success of full state tomography, fidelity-latency trade-off analysis, and remote state preparation validates the system's capabilities. While the current implementation uses a simple TDMA schedule, the link layer is not tied to a specific scheduling algorithm, enabling future adaptation to more advanced schemes. The researchers acknowledge the need for further research on security aspects, such as authentication of classical control messages to prevent unauthorized entanglement generation. The techniques presented are not specific to diamond NV centers and could be adapted to other quantum network platforms, promoting the development of heterogeneous quantum networks.

This work presents a significant step towards building large-scale quantum networks by experimentally demonstrating a functional link and physical layer. The platform-independent approach allows for the creation of quantum networking applications that are not tied to specific hardware implementations. Future research should focus on addressing the limitations outlined in the paper, such as optimizing resource scheduling for larger networks and enhancing the security of classical control channels.

The current implementation uses a simple static TDMA scheduling approach, which may not be optimal for larger, more complex networks. The security of classical control messages is an area needing further investigation to prevent unauthorized entanglement generation and potential impact on qubit quality. The study is limited to a two-node network; scalability to larger networks needs further experimental verification. The experimental setup uses a specific type of quantum hardware; the generalizability to other platforms needs to be explored.