A team at ETH Zurich has built a quantum gate that swaps the states of two atoms with 99.91 percent accuracy, operating simultaneously across more than 17,000 atom pairs in under a millisecond. Published in Nature in May 2026, the result represents the highest reported fidelity for a two-qubit gate on a neutral-atom platform and suggests a practical route toward quieter, more reliable quantum processors.
“The key idea is that our gate is protected by geometry,” said Tilman Esslinger, the ETH Zurich professor who leads the research group, in a university statement. Rather than relying on precisely timed energy pulses, the gate encodes its operation in the shape of a path traced by atom pairs through a shifting optical lattice. That geometric phase is naturally resistant to small fluctuations in laser intensity or timing, the kinds of imperfections that plague other gate designs and force engineers to spend extra qubits on error correction.
What the experiment achieved
The eight-member team, led by first author Philipp Kiefer, loaded pairs of cesium atoms into a dynamical optical lattice, a grid of standing laser waves whose geometry can be smoothly deformed. By steering each atom pair around a closed loop in the lattice, the researchers induced a quantum SWAP: the two atoms exchanged their internal states without any direct energy kick. Because the swap depends only on the loop’s shape, not on how fast or slowly the atoms traverse it, the operation tolerates a range of experimental imperfections that would degrade a conventional pulse-based gate.
The reported loss-corrected amplitude fidelity of 99.91(7) percent was measured across 17,000 atom pairs operating in parallel. That scale matters. Earlier neutral-atom experiments typically demonstrated high-fidelity gates on single pairs or small clusters, leaving open the question of whether performance would hold up when thousands of qubits operated at once. The ETH Zurich result shows that it can, at least for this particular operation.
For comparison, a separate result published in Nature around the same time reported collisional entangling gates on fermionic atoms at 99.75(6) percent fidelity. The gap of roughly 0.16 percentage points may sound trivial, but in quantum error correction, every fraction of a percent translates directly into hardware savings: higher gate fidelity means fewer physical qubits are needed to protect a single logical qubit from errors. (The original ETH Zurich paper cites this fermionic-atom work, but because the specific group, institution, and publication details have not been independently confirmed by this newsroom, readers should consult the Nature paper’s reference list for full attribution.)
Why it matters for quantum computing
Quantum computers built from neutral atoms have attracted growing investment because atoms are naturally identical, eliminating the manufacturing variation that complicates superconducting circuits. Companies such as Atom Computing, QuEra, and Pasqal are all pursuing neutral-atom architectures. But the approach has lagged behind superconducting and trapped-ion systems in one critical metric: two-qubit gate fidelity. The ETH Zurich result narrows that gap substantially.
The work builds on more than two decades of theoretical and experimental groundwork. A 1999 proposal by Dieter Jaksch and colleagues first outlined how cold, controlled collisions between neutral atoms could serve as the basis for quantum logic gates. Subsequent experiments demonstrated controlled exchange interactions and collision-based operations in optical traps. The 2026 Nature paper extends that lineage by wrapping the collision in a geometric phase, a design choice that trades precise calibration for topological protection.
Fault-tolerant quantum computing, the threshold at which a machine can correct its own errors faster than they accumulate, demands two-qubit gate fidelities above roughly 99 percent, with higher values dramatically reducing overhead. Crossing 99.9 percent on a massively parallel platform puts the geometric SWAP gate in the range where early fault-tolerant schemes become plausible, at least on paper.
Important caveats
The 99.91 percent figure is loss-corrected: atom pairs that vanished during the experiment were excluded from the final count. The Nature paper details the atom-loss rate, but no independent group has yet benchmarked it against competing platforms using the same correction methodology. That makes direct comparisons with superconducting or trapped-ion gates, which face different loss mechanisms, difficult to interpret at face value.
More critically, no data have been published on how this gate performs when chained into longer circuits. A single high-fidelity gate is necessary but not sufficient for practical computation; errors compound as circuits grow deeper. Superconducting processors already implement SWAP-equivalent operations by decomposing them into native gates such as CZ or iSWAP and running them in multi-layer circuits. Whether the ETH Zurich geometric approach can integrate into a similarly rich instruction set remains an open question.
The gate’s noise resistance, while genuine, is specific. Geometric phases naturally average out certain fluctuations, such as variations in lattice depth or small timing errors. Other imperfections, including spatial inhomogeneities across a large lattice or slow thermal drifts, could still degrade performance over the longer timescales a real computation would require. The team has not yet published data under varied noise conditions.
Cost and scalability are also unaddressed. The press release does not disclose the experimental budget, the infrastructure needed to maintain the optical lattice at operating conditions, or a timeline for scaling beyond the current setup. Tens of thousands of simultaneous atom pairs are impressive, but a programmable quantum computer also needs individual qubit addressing, mid-circuit measurement, error-correction protocols, and robust classical control electronics, none of which this experiment demonstrates.
What to watch next
The clearest next milestone is a demonstration of the geometric SWAP gate embedded in a multi-gate circuit, ideally combined with entangling operations and measurements to show it can function as part of a universal gate set. Independent replication by another laboratory would also clarify how much of the reported fidelity stems from the geometric concept itself and how much depends on the specific hardware and tuning at ETH Zurich.
An open-access preprint of the paper is available for researchers who want to examine the methods in detail. For the broader quantum computing field, the result is a concrete step forward: it shows that neutral-atom gates can reach the fidelity tier previously dominated by superconducting and trapped-ion systems, and it does so at a scale that no other platform has matched for this class of operation. The distance between a record-setting gate and a working, fault-tolerant quantum computer remains large, but it just got a little shorter.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
