Optical links for scalable quantum computers: how microwave-to-optical converters pave the way for distributed computing ✦
Context
To build a quantum computer capable of solving practically important problems, millions of qubits are needed. However, leading platforms like superconducting circuits face a fundamental limitation: heat removal in cryostats. Scaling is only possible by linking multiple modules through quantum channels. Optical photons are ideal carriers of quantum information thanks to low loss and noise resilience even at room temperature. The key element is a converter from microwave signals (on which qubits operate) to optical ones. The concept of quantum computing laid down by Richard Feynman demands scaling, and this is exactly the approach explored by the authors.
Methods
The authors consider two types of converters: electro-optic and electro-optomechanical, based on the Pockels effect and optomechanical interaction. These devices, built on lithium niobate or silicon chips, parametrically enhance the coupling between microwave and optical fields using a laser pump. Despite current efficiencies on the order of 10⁻³–10⁻² and added noise, protocols for on-demand entanglement generation have been proposed. For instance, using two-mode squeezing or up-conversion followed by Bell measurement (John Stewart Bell) on a beam splitter setup. To compensate for losses, repeated attempts with a fixed delivery time are used, and quantum decoherence is mitigated by storing states in long-lived memory qubits. The entanglement purification procedure was advanced, in particular, by Charles Bennett.
Results
Simulations show that even with current converter parameters (microwave loading efficiency 0.8, noise 0.1 photons per attempt, repetition rate 1 MHz) one can obtain entangled pairs with 91% fidelity in 15 microseconds using 20 parallel channels. After four rounds of quantum distillation, the fidelity rises to 99%. This exceeds the threshold for error correction in the surface code (90%). With more optimistic parameters (noise 0.01, optical detection efficiency 0.5), a single line delivers 91% fidelity in 400 microseconds. Thus, even imperfect converters can create high-quality links.
Implications
The work shows that implementing optical links does not require perfect converters. This changes the development strategy for quantum computers: instead of chasing maximum efficiency, one can focus on integration and scaling up the number of channels. The achieved fidelities are sufficient for algorithmic advantages, for instance, in generating GHZ states or performing the Bernstein–Vazirani test.
Future development
In the coming years, experimental demonstration of optical entanglement between two cryostats is expected. Further material improvements (e.g., quasi-two-dimensional optomechanical crystals) and reduced heating from the optical pump will allow efficiencies above 50% and noise below 0.01. This will pave the way for full-scale lattice surgery—stitching logical qubits through optical channels at every error correction cycle.
Impact
Such distributed systems will impact the development of quantum algorithms, cryptography, and materials simulation. Quantum data centers will be able to pool resources of thousands of qubits, solving problems beyond the reach of classical supercomputers.
Next steps
Next steps include implementing non-destructive state transfer between a qubit and the converter, integration with optical memory, and scaling to hundreds of channels per cryostat.
Key open problems
This work is directly linked to the fundamental problem of scaling superconducting quantum systems and overcoming decoherence in extended channels. The proposed methods are also relevant for building the quantum internet.
🎯 Optical fibers used for classical telecommunications can transmit single photons over hundreds of kilometers without significant attenuation—this is the very basis of the quantum network idea.
Key numbers
- conversion efficiency: 0.01–0.5
- added noise: less than 0.1 photons per attempt
- repetition rate: 1 MHz
- entanglement delivery time: 15–400 µs
- required number of channels per cryostat: 100–10,000