Much like classical supercomputers, scaling up quantum computers requires an optical interconnect. However, signal attenuation leads to irreversible qubit loss, making quantum interconnect design guidelines and metrics different from conventional computing. Inspired by the classical Dragonfly topology, we propose a multi-group structure where the group switch routes photons emitted by computational end nodes to the group's shared pool of Bell state analyzers (which conduct the entanglement swapping that creates end-to-end entanglement) or across a low-diameter path to another group. We present a full-stack analysis of system performance, a combination of distributed and centralized protocols, and a resource scheduler that plans qubit placement and communications for large-scale, fault-tolerant systems. We implement a prototype three-node switched interconnect to justify hardware-side scalability and to expose low-level architectural challenges. We create two-hop entanglement with fidelities of 0.6-0.76. Our design emphasizes reducing network hops and optical components to simplify system stabilization while flexibly adjusting optical path lengths. Based on evaluated loss and infidelity budgets, we find that moderate-radix switches enable systems meeting expected near-term needs, and large systems are feasible. Our design is expected to be effective for a variety of quantum computing technologies, including ion traps and neutral atoms.
翻译:与经典超级计算机类似,量子计算机的规模化扩展需要光学互连。然而,信号衰减会导致不可逆的量子比特损耗,这使得量子互连的设计准则与性能指标不同于传统计算体系。受经典Dragonfly拓扑结构启发,我们提出一种多群组架构:群组交换机将计算终端节点发射的光子路由至该群组的贝尔态分析器共享池(通过纠缠交换实现端到端纠缠建立),或通过低直径路径转发至其他群组。我们开展了系统性能的全栈分析,提出分布式与集中式相结合的控制协议,并设计了面向大规模容错系统的资源调度器,用于规划量子比特布局与通信任务。通过实现三节点交换互连原型系统,我们验证了硬件侧的可扩展性并揭示了底层架构挑战。实验实现了保真度0.6-0.76的双跳纠缠。本设计通过减少网络跳数与光学元件来简化系统稳定性控制,同时支持光学路径长度的灵活调节。基于对损耗与保真度误差的量化评估,我们发现中等端口数的交换机即可满足近期预期需求,且大规模系统具备可行性。该设计预期可适用于离子阱与中性原子等多种量子计算技术路线。