A team linked three individual atoms, used as long lived quantum "memories" (i.e., qubits), with single photons into a shared quantum state, with measured fidelity — a 0 to 1 quality score — between 0.84 and 0.88.
Three small ion traps, each holding a single atom, have been wired together with individual particles of light into a single shared quantum state. The authors report the result is the first fully-distributed three-node version of that state, called a Greenberger-Horne-Zeilinger (GHZ) state, produced across individual atomic "memories" rather than solid-state chips or bulk atomic clouds.
The work, described in arXiv preprint 2606.17173 and led by a group at Duke University's Quantum Center that includes Christopher Monroe, sits at the heart of a long-running program to build a quantum internet out of many small, addressable nodes rather than one large, monolithic machine. A GHZ state is the multipartite cousin of the better-known two-particle entanglement: instead of two atoms sharing one joint fate, all three share one. That kind of joint state, the authors argue, is what modular quantum computers, distributed sensors, and multi-party secure communication protocols will eventually need to function at scale.
What makes the new result different is the platform. Three-node GHZ states have already been produced between solid-state qubits and between atomic ensembles, but those systems trade off reproducibility, addressability, or both. Single atoms held in individual traps — in this case, three separated barium ion qubits, each roughly 22 meters apart — can in principle be replicated, detected, and controlled one by one, which is what a network of many small nodes would demand. The team's approach uses photonic interconnects: each node emits a single photon that carries a small piece of quantum information, and a measurement on those photons is what entangles the atoms. The paper is explicit that this is a "first" only within the individual-atom platform; the broader milestone of three-way entanglement across separated nodes is older.
The cost of that rigor is visible in the numbers. The authors bound the state fidelity, a 0-to-1 score for how close the lab state is to an ideal GHZ state, between 0.841 and 0.881, well short of a perfect 1. The entanglement generation rate is 0.095 per second — slow enough that assembling anything network-scale will require both faster links and longer memory coherence. The team also reports that the result closes the detection loophole, a worry that the apparent agreement between nodes might just reflect lost or dropped photons rather than real entanglement, by using a fair-sample assumption that the analysis defends. Each ion is measured with 99.7% detection efficiency per node, which the team argues makes the fairness assumption tenable in a way it has not been for prior photonic demonstrations.
Whether the loophole is "fully closed" in a multipartite setting is more nuanced than in a two-particle Bell test, where the same claim has been harder to make than to assert. The preprint makes a fairness argument the team can defend in print; outside experts have not yet weighed in on the analysis, and the work is not peer reviewed.
What comes next is engineering, not physics. The headline number a quantum internet will be measured against is the rate at which remote entanglement can be generated and held, and 0.095 per second is closer to a proof of concept than a backbone. The preprint's authors frame the work as a step toward modular quantum computers, where each node is small and the network stitches them together, and toward clock networks and secure multi-party protocols that need shared quantum resources across distance. The team estimates the rate could climb to 0.25 per second with improved cooling to eliminate ion de-crystallization events, but that engineering has not yet been demonstrated.
The honest read: the platform, single atoms with photonic interconnects, is now a peer of solid-state and ensemble systems in the three-node GHZ category, with a measured but imperfect state, a slow rate, and a loophole-closing argument that still has to clear peer review. The next round of work will say whether the rate can climb without giving the fidelity back.