Loss-tolerant distributed lattice surgery using fusion networks

Building a quantum computer that can solve real-world problems requires more than just a few thousand tiny particles acting as information carriers; it requires millions. The sheer number of physical components needed to protect a single piece of information from the constant noise of the universe means that no single machine will ever be large enough to hold them all. Instead, scientists are looking toward a future where many smaller quantum processors are linked together, forming a distributed network. The challenge lies in the connection itself. While the processors inside the machines are made of solid matter, the links between them rely on light. This creates a fundamental mismatch: the solid parts suffer from slow, steady errors, while the light-based links are prone to sudden, catastrophic failures where photons simply vanish.

In this new study, researchers at Imperial College London have developed a way to bridge this gap. They propose a hybrid method that treats the loss of light not as a disaster to be avoided, but as a known event that can be managed. By combining the standard circuit-based operations used inside the processors with a measurement-based approach designed for light, they have created a protocol for merging two separate quantum memories. Their work suggests that by arranging the connection in a specific, elongated pattern, they can double the distance over which information is protected during the merge, making the entire system significantly more robust against the inevitable loss of photons.

The core of the problem is how to perform a joint operation, such as merging two quantum memories, across a network where the connection is unreliable. In a standard setup, two quantum processors might be linked by a direct line of communication. If a photon is lost in that line, the information is corrupted. Traditional methods try to fix these errors after they happen, but when the loss rate is high, the system becomes overwhelmed. The researchers turned to a different strategy used in photonic computing, where the loss of a particle is treated as a "heralded" event. This means the system knows exactly when a photon is missing, allowing the error correction software to adjust its strategy in real time rather than guessing what went wrong.

To test this, the team used a mathematical language called ZX calculus, which allows them to visualize and manipulate quantum circuits as diagrams. They started with a standard procedure for merging two quantum memories, known as lattice surgery, and transformed the diagram to see how it could be built using a mix of solid-state components and light-based links. They discovered that by splitting the connection into a series of smaller, linear chains of resources, they could create a "hybrid" interface. In this setup, the solid processors handle the local operations, while the light-based links handle the connection between them. Crucially, they found that this hybrid arrangement could be designed to be much wider than a simple straight line.

The researchers simulated this new architecture under various conditions to see how well it held up. They found that when the local processors were nearly perfect, the new hybrid method could tolerate a photon loss rate of up to 50 percent. This is a theoretical limit for this type of error correction, meaning the system can survive even if half of the connecting photons disappear. More importantly, they measured the "distance" of the protection. In quantum error correction, distance refers to how many errors must occur simultaneously to break the system. Their simulations showed that the hybrid linear-chain design doubled the protection distance for the most critical part of the operation compared to a standard straight-line connection. While a standard connection offered a protection distance of roughly half the code size, the new method extended this to nearly the full size, and then some.

However, the advantage of this new design depends heavily on the quality of the local processors. The team ran simulations with different levels of noise to see where the new method shines. When the local processors had a moderate error rate, the benefits of the longer protection distance were largely hidden by the local noise. In these conditions, the system performed similarly to older methods. But as the researchers lowered the local error rate to a very low level, the advantage of the hybrid design became clear. The error rate for the new method dropped much faster than for the standard methods as the connection quality improved. This suggests that the new architecture is specifically designed for a future where local processors are highly reliable, but the network links remain noisy.

The study also explored what happens when the resource states used for the connection are smaller or "truncated." In a real-world scenario, generating massive chains of entangled particles might be too difficult, so the researchers tested shorter chains. They found that while shortening the chains lowered the maximum tolerable loss rate, it did not reduce the protection distance. This means that even with smaller, more manageable resources, the system retains its structural advantage of a wider protection zone, provided the loss rate stays below a certain threshold. For example, with chains containing eight particles, the system could still tolerate a loss rate of about 38 percent, and with five-particle chains, about 32 percent.

To make these results relevant to actual hardware, the team converted their theoretical loss rates into physical photon-loss thresholds. They accounted for the fact that in real optical systems, a failed fusion measurement can sometimes be boosted by adding extra photons to increase the chance of success. Their calculations showed that with the right level of boosting, the new hybrid protocols could tolerate physical photon loss rates of nearly 18 percent for the best-case scenarios, and around 9 percent for the truncated versions. This is a significant improvement over previous estimates for distributed quantum computing, which often struggled with much lower tolerance for loss.

The researchers also examined how the system protects different types of information. Quantum information has two main aspects, often called the merge observable and the perpendicular observable. The new hybrid design significantly improved the protection for the merge observable, effectively doubling the safety margin. However, for the perpendicular observable, the standard "zig-zag" connection pattern used in older designs still offered the best protection. This indicates that while the new hybrid method is a major step forward, it does not yet solve every aspect of the problem perfectly. The team suggests that future work could focus on tailoring the failure modes of the light-based links to create a more balanced protection for both types of information.

Ultimately, this work provides a blueprint for how to connect the quantum computers of the future. It demonstrates that by mixing the strengths of solid-state processors with the loss-tolerant nature of photonic networks, it is possible to build a distributed system that is far more resilient than the sum of its parts. The findings suggest that as local quantum processors become more reliable, the bottleneck will shift to the network links, and the hybrid protocols described here offer a way to overcome that bottleneck. The simulations confirm that with the right design, a quantum network can survive the loss of half its connecting photons, opening a realistic path toward scaling quantum computers to the massive sizes required for practical applications.