Researchers report that more than 17,000 pairs of atoms can carry out a quantum swap with 99.91% accuracy across a single system.
That result shows a path toward quantum computers that keep working even when real-world noise and small errors build up.
Atomic grid in action
In the experiment, scientists arranged potassium atoms in a grid of light, placing pairs on neighboring sites before triggering the swap.
As those pairs were pushed into overlap, Yann Kiefer at ETH Zurich showed that the swap followed from the shared geometry of the motion rather than from exquisitely tuned timing.
Only one spin arrangement traveled that protected route, while three closely related arrangements remained fixed and served as stable reference points.
That separation kept the phase difference clean, but it also raised the deeper question of why a usually troublesome shared-state detour became the source of the gate’s protection.
Why doublons help
Their trick relied on a doublon, two qubits briefly occupying one site, which earlier gate schemes usually treated as leakage.
Here the shared stop widened the system’s available states, letting the pair trace a controlled loop instead of a fragile collision.
Because the atoms belonged to a type that cannot share the exact same quantum state, only certain combinations could mix during the loop.
Using that once-unwanted stop as part of the route changed the engineering problem from fine tuning to staying on track.
What geometry protects
Protection came from a geometric phase, a state change set by the path taken rather than the exact motion along it.
During the sweep, the active state kept zero ordinary energy, so timing errors had far less chance to affect the result.
Two symmetries in the governing equations kept that route pinned to the same loop even when control settings varied.
Only faster, stronger jolts could throw the atoms off the intended route, which is why the noise tests mattered so much.
The system-wide test
To test the idea at scale, the team arranged tens of thousands of atoms and formed more than 17,000 working pairs.
Each operation finished in less than a thousandth of a second as the atoms moved together and then separated again.
They checked the result by tracking a simple back-and-forth signal between two states that revealed whether the swap worked.
Across the whole array, corrected fidelity reached 99.91%, while raw performance held at 99.5% before survival adjustments.
Noise hits limits
Robustness became clearer when the team deliberately introduced noise into the lattice controls, with white noise spread across a 2 kilohertz bandwidth.
Gate fidelity held on a broad plateau until the added tunneling noise reached about 5%, an unusually forgiving margin.
Once the noise carried faster components, it could kick atoms off the protected route and across the nearby energy gap.
That window matters because real processors never stay perfectly stable, and this gate tolerated noise engineers face.
Beyond simple swaps
Adding interactions let the same setup do more than swap positions; it also produced an entangling gate, one that links two qubits.
In that mode, the shared-site route added a controllable phase from atom-atom interactions, so the gate could stop halfway instead of fully swapping.
Corrected fidelities stayed near 99% for two half-swap versions, outperforming the older indirect exchange route the team compared.
That advantage appeared because the key phase now depended mainly on interaction strength, not on a squared tunneling effect.
Why routing matters
Swap gates matter because crowded processors need constant rerouting, moving information between neighbors without leaving large gaps between qubits.
Neutral atoms help here because they carry no charge and fit into an optical lattice, a grid of atom traps made by light.
“A few years ago, researchers managed to realise such gates using neutral atoms in their lowest energy state, albeit by exploiting dynamical phases due to tunnelling and collisions,” said Kiefer.
His point was simple: earlier exchange gates worked, but they leaned on a dynamical phase, a timing-based quantum phase, from motion and collisions.
A path to scale
Scale also depends on moving qubits without scrambling them, and this group already showed coherent transport over long distances.
That earlier transport moved paired atoms over 50 positions in the grid with 99.57% accuracy at each step.
Combined with the new gate, this motion can bring distant pairs together, let them interact, and then separate them again.
Such a setup could create long links inside packed arrays without reserving wide empty lanes for constant shuttling.
What still limits
Even so, the remaining mistakes still came mostly from technical noise, especially drifting interaction strength and small laser-power changes.
Current preparation also leaves only 60–70% of atoms in the desired paired states, enough for tests but not full programs.
Better magnetic-field stability, steadier control of the light-based trap, and atoms less sensitive to field drift could push the error lower.
Just as important, a quantum gas microscope, a tool for seeing and controlling single atoms, would let engineers target chosen pairs.
What comes next
This experiment argues that better quantum gates do not always come from tighter control; sometimes they come from leaning into symmetry.
If cleaner preparation and targeted readout follow, neutral-atom processors could become both denser and much harder to knock off course.
The study is published in Nature.
—–
Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.
Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.
—–