Fault-tolerant distributed quantum computing with a single nucleus per node

Distributed quantum computing architectures typically suffer from an asymmetry: nodes have high-fidelity data qubits, but inter-node communication relies on noisy, low-coherence qubits and faulty photonics. Traditionally, this requires multiple ancilla qubits per node to perform complex distillation of Bell pairs or GHZ states. This paper asks whether this hardware overhead can be eliminated by exploiting the specific noise characteristics of the communication link.

The authors propose a protocol that avoids distillation entirely by engineering a noise bias in the photonic Bell pairs, where phase errors (Z) are frequent but bit-flip errors (X) are rare. By designing syndrome-extraction circuits such that phase noise manifests only as measurement errors rather than data-qubit errors, the system can suppress the noise through simple, repeated syndrome measurements. This approach allows the error-correcting code to perform the purification intrinsically. The authors analyze this architecture for both Floquet codes (requiring only one data qubit per node) and standard stabilizer codes (requiring one additional ancilla qubit).

The study demonstrates that this bias-exploiting architecture significantly reduces hardware requirements while maintaining high error-correction thresholds. For the honeycomb code, the authors report a threshold of approximately 11% for phase errors on individual Bell pairs. For surface code memory, the bulk-measurement threshold exceeds 30%. Furthermore, the authors show that lattice surgery is inherently robust in this setting, enabling logical operations at thresholds comparable to those of quantum memory. These results suggest that the performance of a distributed quantum computer can be limited by the high-quality data qubits rather than the noisy communication links.