Summary

• Quantinuum reports 99.9995% logical memory fidelity per error-correction cycle on Helios, equivalent to an estimated error probability of five in one million.
• The colour-code implementation produced a tenfold reduction against the company's September memory-error result, using up to 37 physical qubits per logical qubit.
• Improved error suppression supports Quantinuum's planned Apollo system for 2029, but the experiment does not demonstrate one million error-free logical operations in a complete computation.

Quantinuum has reported a tenfold reduction in logical quantum memory errors using its Helios processor, demonstrating a method of protecting quantum information that could help reduce the hardware requirements of future fault-tolerant computers.

The company achieved 99.9995% logical memory fidelity per error-correction cycle using a colour-code implementation, equivalent to an estimated failure probability of five in one million cycles. The result improves on its September demonstration, which used a different error-correction code on the same hardware.

Quantinuum says the measurement brings the accuracy of its error-correction architecture into the target range for Apollo, its planned universal fault-tolerant quantum computer expected by 2029. The experiment does not establish that the company can already execute a million useful quantum operations without error.

The distinction is important because preserving information in a protected quantum memory is not the same as performing a long, reliable computation. A practical machine must combine memory protection with logical operations, state preparation, decoding, and enough physical hardware to support the intended calculation.

Quantum information requires protection from errors

Conventional digital computers use established techniques to detect and correct errors in stored and processed information. Quantum computers face a more difficult problem because qubits are sensitive to environmental noise and imperfect control operations.

A qubit can carry quantum information through a physical system such as a trapped ion, but the information can become corrupted as operations are performed or the state is maintained over time.

Errors accumulate as computations grow longer, limiting the complexity of work that can be completed reliably. Increasing the number of physical qubits alone does not solve the problem if those qubits introduce errors faster than the system can correct them.

Quantum error correction addresses this limitation by distributing protected information across groups of physical qubits. Together, they represent logical qubits whose errors can be detected through specially designed measurements.

Those measurements must reveal the presence of errors without directly exposing and destroying the quantum information being protected. Software and control systems then interpret the results and determine the corrections needed.

This introduces additional hardware and computational overhead. A fault-tolerant quantum computer may require many physical qubits for each logical qubit used in the intended algorithm.

The engineering objective is consequently to improve reliability without requiring an impractical amount of hardware.

A colour code changes the reliability trade-off

Quantinuum’s latest experiment used a colour code, a form of quantum error-correcting code that organises physical qubits into structures designed to detect and suppress logical errors.

The company tested code distances of three, five, and seven on Helios. Code distance is related to the extent of protection provided, with larger distances generally requiring more physical qubits.

At distance seven, Quantinuum reports a logical memory failure probability of approximately five times ten to the power of minus six per syndrome-extraction cycle. That corresponds to the headline fidelity of 99.9995%.

The measured result is an estimate derived from experimental data. Quantinuum reports observing three failures across 580,000 test rounds, which provides the statistical basis for its published figure.

This is more precise than describing the result as five actual failures recorded in a completed one-million-operation computation. The experiment measured logical memory protection, not the successful execution of a million-step general-purpose algorithm.

The company also reports that increasing the tested code distance reduced logical memory errors by an average factor of 12 at each step.

Such error suppression is important because adding protective qubits should make logical information more reliable. If the additional error-correction operations themselves introduce too much noise, increasing the code distance may fail to deliver the expected improvement.

Quantinuum’s measurements provide experimental evidence of improving reliability under the tested conditions, although further work is needed to establish performance as larger systems and more complicated computations are introduced.

Different codes impose different hardware costs

The October demonstration follows Quantinuum’s September work using its Helix error-correction code.

Both studies were conducted on the Helios trapped-ion computer, allowing the company to explore different approaches without relying on a new generation of physical hardware between results.

The Helix implementation used ten physical qubits per logical qubit, compared with up to 37 for the colour-code experiments. The colour code achieved lower logical error rates but required more physical hardware for the tested configuration.

That trade-off is relevant because stronger protection is not necessarily the most efficient option for every stage of a computation.

An algorithm may involve operations with different reliability requirements, and the ability to use several error-correction codes could allow a future system to balance accuracy against resource consumption.

Quantinuum’s architecture includes techniques for adapting the extraction of error information and connecting different codes. Such flexibility could help reduce the physical resources required for particular tasks, but the commercial benefits remain dependent on demonstrating those arrangements in larger working systems.

The company compares the October result with a previously published superconducting-hardware benchmark, claiming a logical memory error rate more than 100 times lower.

Cross-platform comparisons require care because quantum technologies can use different physical arrangements, error-correction codes, and experimental measurements. A record in one form of logical memory protection does not establish superiority across all operations or applications.

Logical gates introduce further requirements

Quantinuum also tested logical operations using protected qubits, including two-qubit Clifford gates.

The company reports an average error-suppression factor of nine when increasing the tested code distance for those gates. This is separate from the factor of 12 reported for logical memory.

Reliable logical operations are necessary because a quantum computer must do more than preserve information. It must manipulate that information through sequences of gates while keeping the resulting errors sufficiently low.

Some operations are more difficult to protect than others, particularly those needed to support universal computation.

Quantinuum’s technical discussion identifies additional work required to combine the relevant components, including fault-tolerant state preparation and the integration of larger computations.

The company has previously demonstrated techniques used for preparing specialised quantum states, but its October results do not establish that every component of a universal fault-tolerant calculation has been shown to improve in reliability as the code size increases.

Its decoding technology is another part of the programme. Error-related measurements must be interpreted quickly enough for corrective action to keep pace with the hardware.

Quantinuum reports testing several approaches, including machine-learning decoders, with processing latency below one millisecond under its experimental conditions.

The speed of decoding matters because a highly accurate code could still be unsuitable for practical use if the classical processing needed to identify corrections became a bottleneck.

Apollo remains a development target

Helios is a trapped-ion system using a quantum charge-coupled device architecture, which moves ions between operating regions to support quantum operations and control.

The architecture enables flexible interactions between qubits, while placing demanding requirements on precision control, ion transport, optical systems, and the coordination of error-correction measurements.

Quantinuum is using Helios to investigate how these components perform together and to guide the development of larger processors.

Its longer-term objective is Apollo, a universal fault-tolerant quantum computer targeted for availability by 2029. The company believes its recent error-correction results support that roadmap, although the date and proposed capabilities remain forward-looking.

Greater reliability could eventually make it possible to execute quantum algorithms requiring substantially more operations than current systems can support. Potential applications include selected problems in chemistry, materials science, and other areas where conventional calculations may be difficult.

However, the commercial case requires more than a favourable memory-error measurement. Businesses considering quantum computing need evidence that a complete system can execute useful workloads reliably, within acceptable time and cost constraints.

Quantinuum’s latest experiment demonstrates important progress in protecting logical quantum memory and provides additional evidence about the trade-offs between code selection, accuracy, and hardware requirements.

Whether those results can be extended to large fault-tolerant calculations will depend on integrating reliable logical gates, state preparation, decoding, and more extensive hardware. Those remain the engineering milestones separating the October demonstration from the universal computer described in the company’s 2029 roadmap.