A Delft team reports 99.18% average accuracy for two qubit logic operations on a small register of nuclear spins coupled to a single atom scale flaw in a diamond, just clearing a floor the field treats as the minimum for useful distributed quantum computing.
A small diamond-based device has run two-qubit logic operations with an average accuracy of 99.18% and a best of 99.61%, clearing a long-cited error threshold that has constrained multi-qubit experiments on this hardware family. The result, reported in a preprint from the Taminiau group at Delft, puts the platform just above the floor the field treats as the minimum for useful distributed quantum computing.
The team, led by Tim H. Taminiau, frames the number as a prerequisite rather than a destination. In the paper's own words, the work addresses one of the key prerequisites for scalable quantum networks based on solid-state spins, with the thresholds having remained out of reach for multi-qubit registers. That phrasing matters: a prerequisite cleared is not the same as a fault-tolerant machine built, and the paper stops short of claiming a network demonstration.
The hardware is a nitrogen-vacancy center, an atom-scale flaw in a diamond crystal that traps a single quantum bit, with seven surrounding nuclear spins acting as additional qubits. Together they form a register, a small, tightly coupled set of qubits that work as one unit. The authors measured two-qubit gate performance with gate set tomography, a procedure that characterizes gate errors and measurement errors together rather than benchmarking gates in isolation. That choice is significant: gate set tomography is more demanding than randomized benchmarking, and the 99.18% average fidelity the team reports is the result under that stricter lens.
The paper also runs a small chemistry simulation as a worked example. Using a variational quantum eigensolver, a hybrid quantum-classical algorithm that estimates a molecule's ground-state energy by iterating a parameterized quantum circuit, the team computed the ground-state energy of H2 and LiH. The point is not the chemistry itself; a seven-qubit register is too small to compete with classical methods on molecules of practical interest. The point is that the register can run a multi-qubit algorithm end to end, with gate errors low enough that the result is meaningful.
The limitations are visible. Seven qubits is a network node only in the early-stage sense. The 99.18% average fidelity is just above the often-cited surface-code threshold, not deep inside the fault-tolerant regime; logical error rates for real distributed applications will require significantly better gates. And the result is a single group's demonstration on one device, with independent reproduction on other nitrogen-vacancy systems still the next step the field will look for.
What changes now is what this specific hardware can demonstrably try. The Taminiau group's paper includes an explicit analysis of crosstalk, the unwanted coupling between qubits that becomes harder to suppress as registers grow, along with an efficient gate-optimization procedure that the authors argue generalizes. With the fidelity bar cleared, the platform can host multi-qubit experiments that were out of reach on this device yesterday, including small network-node demonstrations and more demanding algorithmic tests, while the harder problem of scaling past a handful of qubits remains open.