Gate Teleportation vs Circuit Cutting in Distributed Quantum Computing

TL;DR

This paper tackles the challenge of distributing quantum circuits across modules by comparing entanglement-based remote gates (gate teleportation) with circuit cutting (classical links). It models noisy microwave-to-optical transducers and Bell-pair generation to assess GHZ-state fidelity via the Hellinger metric, revealing a break-even regime where remote gates match gate-cutting performance under realistic noise ($N_{add}$) and shot budgets ($N_{shot}$). A key finding is that a roughly 10× reduction in transducer-noise can render remote gates competitive, with larger GHZ states favoring remote gates due to the exponential sampling overhead of circuit cutting. The work also proposes a network-aware hybrid distributed approach that dynamically selects between remote gates and circuit cuts based on Bell-pair availability to minimize quantum runtime, informing near-term hardware metrics and compiler strategies for scalable modular quantum computing.

Abstract

Distributing circuits across quantum processor modules will enable the execution of circuits larger than the qubit count limitations of monolithic processors. While distributed quantum computation has primarily utilized circuit cutting, it incurs an exponential growth of sub-circuit sampling and classical post-processing overhead with an increasing number of cuts. The entanglement-based gate teleportation approach does not inherently incur exponential sampling overhead, provided that quantum interconnects of requisite performance are available for generating high-fidelity Bell pairs. Recent advances in photonic entanglement of qubits have motivated discussion on optical link metrics required to achieve remote gate performance approaching circuit-cutting techniques. We model noisy remote (teleported) gates between superconducting qubits entangled via noisy microwave-to-optical (M2O) transducers over optical links. We incorporate the effect of the transducer noise added ($N_{add}$) on the Bell pair fidelity and inject noisy Bell pairs into remote CNOT gates. We perform a comparative simulation of Greenberger-Horne-Zeilinger (GHZ) states generated between processor modules using remote gates and gate cuts by studying the dependence of the Hellinger fidelity on the primary source of error for the two approaches. We identify break-even points where noisy remote gates achieve parity with gate-cuts. Our work suggests that a 10-fold reduction in the present M2O transducer noise added figures would favor generating multipartite entangled states with remote gates over circuit cutting due to an exponential sampling overhead for the latter. Our work informs near-term quantum interconnect hardware metrics and motivates a network-aware hybrid quantum-classical distributed computation approach, where both quantum links and circuit cuts are employed to minimize quantum runtime.

Figures (9)

• Figure 1: (a) Distributed processor architecture, quantum links are established through photonic Bell state measurements over optical fiber, while classical links are established through circuit-cutting techniques. (b) Intermodule GHZ N circuit generated using non-local CNOT gates using remote gates or gate-cuts.
• Figure 2: Remote CNOT telegate between qubits $\ket{q_1}$ and $\ket{q_2}$ implemented via Bell pairs shared between local communication qubits $\ket{q_{1c}}$ and $\ket{q_{2c}}$.
• Figure 3: Quantum link between qubits $\ket{q_1}$ and $\ket{q_2}$ in two different modules is established using communication qubits $\ket{q_{1c}}$ and $\ket{q_{2c}}$ and microwave-to-optical (M2O) transducers over an optical link. The qubits $\ket{q_{1c}}$ and $\ket{q_{2c}}$ share a noisy Bell pair, with the primary source of infidelity being the noise added by transducer $N_{add}$. A remote 2-qubit gate is executed between data qubits $\ket{q_1}$ and $\ket{q_2}$ through a combination of local gates and a Bell pair between the communication qubits.
• Figure 4: Illustration of the quasiprobability-based circuit cutting procedure applied to the CNOT gate connecting two modules. The measurement of the observable $X \otimes X$ is performed on data qubits 1 and 2. The ancilla qubits, $a_1$ and $a_2$, mediate the measurement of state overlaps for cases where $i \neq j$.
• Figure 5: GHZ Hellinger Fidelity with remote CNOT gates (solid lines) vs transducer noise added (${N_{add}}$) on the bottom x-axis and CNOT gate-cuts (dashes) vs number of shots (${N_{shot}}$) on the top x-axis. Remote gate fidelity shows a strong dependence on the transducer noise added, and the gate-cut fidelity depends on the number of shots. In the regime of low ${N_{add}}$ (remote) and high ${N_{shots}}$ (cut), the local gate errors dominate the GHZ fidelity. Red arrow indicates state-of-the-art transducer ${N_{add}}$ figure Meesala2024-mkWarner2025-sm.