Jülich benchmark tests the building blocks of error correction on Quantinuum, IBM and IQM hardware

[Sinead_47]

Researchers at Germany's Jülich Supercomputing Centre have tested the basic operations that quantum error correction depends on, Quantum Computing Report says. J. A. Montañez-Barrera and Kristel Michielsen looked at mid circuit measurement, reset and feed forward control across ten processors from Quantinuum, IBM and IQM. The paper is on arXiv as 2610.05928. Their benchmark is based on a quantum optimisation algorithm called LR-QAOA and used up to 2,950 mid circuit measurements per circuit

Quantinuum came out on top. Its Helios-1 and H2-1 machines had effective error rates of about 0.22% to 0.49%. They were tested on colour codes up to 91 data qubits, surface codes up to 81 and a 48 qubit qLDPC code. The report says Quantinuum's trapped ion systems demonstrated the strongest overall performance across the tested primitives

IBM's results were mixed. On its Heron processors, mid circuit measurement error remains about ten times the two qubit gate error, with effective error rates of 1.1% to 3.2%. That gap matters, because error correction relies on measuring qubits constantly in the middle of a computation. If measurement is your weak link, it drags everything else down with it

IQM had a tougher time on this particular test. According to the report, signal decay sped up beyond two data qubit chains because of conditional control and routing limits. That does not mean IQM's hardware is bad overall. It means this specific set of operations is not where its chips shine yet

One detail I liked is that the benchmark was checked against reality. On IBM's ibm_phoenix chip, the benchmark's results correlated with real logical memory results at 0.83 to 0.88. That suggests a cheap, low sampling test can predict how well a machine will do at proper error correction. Independent benchmarks like this are worth far more than vendor claims, because nobody is marking their own homework

Are you surprised Quantinuum came out ahead? And should independent benchmarks like this become standard before anyone claims error correction progress?

Jülich Supercomputing Centre Benchmark Evaluates Quantum Error Correction Primitives Across Quantinuum, IBM, and IQM Hardware - Quantum Computing Reportquantumcomputingreport.com

[TheLegendJohn32]

IBM's measurement gap is the key finding for me. Ten times worse than the two qubit gates is a big problem for surface codes. IBM knows this, which is partly why it has moved towards qLDPC codes with different requirements. It will be interesting to see if their next chips close the gap. Measurement on superconducting qubits is hard because it disturbs nearby qubits

[Jaffa12]

Correlation of 0.83 to 0.88 with real logical memory results is pretty good for a cheap benchmark. If it holds on other hardware, labs could use it as a quick health check before running expensive experiments

That would save a lot of time. It might even become a standard part of calibration

[Drifter]

Hardly surprising. Trapped ions have had the best fidelities for years, and Quantinuum has invested heavily in mid circuit measurement. The trade off is speed, which this kind of benchmark does not capture. A fast machine with worse errors can still win in some cases

Gate speeds on ions are still much slower than superconducting chips. That gap matters for long computations

[HHH86]

Helios-1 holding up on a 91 qubit colour code is impressive. Colour codes are harder to run than surface codes in some ways, but they allow certain gates to be done more easily. Seeing them tested on real hardware at that size is a nice step

A year ago this would have been a simulation paper. Quantinuum seems to be pulling away on fidelity