Quantum error correction well below the color code threshold

To build a computer that can solve problems beyond the reach of today's machines, scientists are turning to the strange rules of quantum physics. These quantum computers use tiny particles called qubits to hold information, but they are incredibly fragile. A whisper of heat, a stray magnetic field, or even a tiny glitch in the machine's own control system can cause a qubit to lose its data. This fragility has long been the biggest obstacle to building a useful quantum computer. To overcome it, researchers use a strategy called quantum error correction. Instead of relying on a single, fragile qubit, they spread the information across many physical qubits, creating a redundant safety net. If one part of the net fails, the others can detect the mistake and fix it without destroying the information. However, this safety net comes with a cost: adding more qubits to fix errors also adds more places where new errors can happen. The goal is to find a point where adding more redundancy actually makes the system more reliable, rather than less. This tipping point is known as the "threshold." If the physical components of the computer are good enough to stay below this threshold, then making the computer larger and more complex will cause the errors to disappear exponentially, paving the way for machines that can run for hours without failing.

A team of researchers at Quantinuum has now demonstrated that they can operate a quantum processor well below this critical threshold, a major step toward building a machine that can actually solve real-world problems. Using a processor called Helios, which traps individual atoms of barium in a vacuum and holds them in place with electric fields, the team tested a specific type of error-correcting code known as the triangular color code. They ran experiments using three different sizes of these codes, ranging from small blocks of thirteen qubits to larger blocks of seventy-three qubits. The results were striking: as they increased the size of the code, the rate at which the computer made logical mistakes dropped dramatically. For the largest code they tested, the error rate fell to roughly five errors in a million attempts. This represents a suppression of errors by more than ten times with each step up in size, proving that the system is indeed operating in the regime where scaling up leads to greater stability.

The researchers achieved this success by tackling two specific problems that often plague quantum computers: leakage and circuit complexity. In a perfect world, a qubit would stay in its designated state forever, but in reality, it can sometimes "leak" out of its working space into other energy levels where it no longer behaves correctly. If a leaked qubit is used in a calculation, it can spread errors to its neighbors, causing a cascade of failures. To stop this, the team employed a two-pronged approach. First, they used a hardware-level technique that acts like a reset button, using carefully timed pulses of light to pump any leaked atoms back into their correct working state. Second, they designed a circuit-level method that swaps out data qubits for fresh ones before they can cause trouble, effectively removing the leaked particles from the calculation. By combining these methods, they ensured that the errors remained manageable and could be corrected by the code.

The team also tested how well different methods could read the error signals and fix the mistakes in real time. They compared a traditional, rule-based computer program with two new approaches based on artificial intelligence: a convolutional neural network and a transformer-based model. These neural networks were trained on simulated data from the Helios processor and then applied directly to the real hardware without further adjustment. Remarkably, both the traditional program and the two AI models performed nearly identically, successfully decoding the error signals and correcting the logical errors. The neural networks were particularly impressive because they could process the data in a few hundred microseconds, fast enough to keep up with the speed of the quantum processor. This suggests that in the future, these fast, intelligent decoders could be used to fix errors as they happen, rather than waiting until the end of a calculation.

Beyond just fixing errors, the experiment provided a deep look into the inner workings of the quantum processor. By analyzing the patterns of errors, the researchers could identify exactly where the machine was struggling. They found that the most common source of mistakes was the interaction between two qubits, followed by errors related to moving the ions around and the noise introduced when measuring them. This kind of detailed diagnosis is crucial because it tells engineers exactly which parts of the machine need the most improvement. The study also showed that the error-correcting code worked effectively even when the physical components were not perfect, provided that the leakage was managed. The logical error rates they achieved are approaching the levels required for what scientists call "megaquop" scale computation, a scale where quantum computers could tackle problems in chemistry, materials science, and cryptography that are currently impossible.

While this work proves that the error correction is working, the researchers note that the journey is not yet complete. The next major challenge is to demonstrate that the entire process, including the preparation of special "magic states" needed for the most powerful calculations, can also be kept below the error threshold. The current study focused on the memory of the computer and the basic logic gates, but a fully universal quantum computer will need to perform a wider variety of operations. The team suggests that the next logical step is to run a complete calculation that includes these more complex operations, using the fast, real-time decoding methods they have developed. If successful, this would mark the transition from a machine that can correct errors in a controlled experiment to one that can run a long, complex program without failing. The results indicate that trapped-ion processors, like the one used in this study, have the potential to reach application-relevant performance with smaller blocks of code than previously thought possible, bringing the era of practical quantum computing closer to reality.