How neutral-atom qubits work
How single atoms held in laser light become qubits, how Rydberg gates and atom moves work, why arrays reach thousands of atoms, and what limits their speed.
A neutral-atom quantum computer stores each qubit in a single uncharged atom held in a tightly focused laser beam, an optical tweezer. The processor is not a fabricated chip but a pattern of light inside a vacuum cell, and every atom is identical. Neutral atoms lead on raw scale: Caltech held more than 6,100 atoms as qubits in 2025, and Harvard and MIT kept more than 3,000 running for over two hours. The costs are speed, because atoms must be moved and imaged, and atom loss. QuEra, Atom Computing, Pasqal, Infleqtion and planqc build them.
This article is part of a series on the main ways to build a qubit. The overview compares all six side by side.
How the qubit forms
Two families of atoms are used: alkali atoms with one outer electron, such as rubidium-87 and cesium-133, and atoms with two, such as strontium and ytterbium-171. In alkali atoms the qubit is a pair of hyperfine levels of the ground state, split by about 6.8 GHz in rubidium and 9.2 GHz in cesium. In ytterbium-171 it is the spin of the nucleus, which barely couples to stray fields. Coherence is long: typically around a second for rubidium, about 7 seconds for ytterbium nuclear spins (Atom Computing), and up to 12.6 seconds in Caltech's cesium array.
Loading starts from a vapor. Lasers slow and cool the atoms in a magneto-optical trap to tens of microkelvin, and an array of tweezers, each about a micrometer wide and a few micrometers apart, picks single atoms out of the cloud. Each tweezer ends up holding one atom about half the time, so a camera images the array and movable tweezers rearrange the atoms into a pattern with no gaps.
Entanglement uses Rydberg states. A laser lifts an atom's outer electron into a very large orbit, hundreds of nanometers across, and two such Rydberg atoms interact strongly over several micrometers. Once one atom is excited, the interaction shifts its neighbor's Rydberg level so the neighbor cannot follow: the Rydberg blockade. Driving a close pair with the right pulses gives a controlled-Z gate in about 0.2 to 0.5 microseconds (µs).
How the machine is built
The core is a glass cell or small steel chamber at ultra-high vacuum. Collisions with the remaining gas set how long an atom survives: Caltech measured a vacuum-limited lifetime of 22.9 minutes at room temperature.
Spatial light modulators shape one laser into a large static grid of tweezers. Acousto-optic deflectors make movable rows and columns of tweezers that carry atoms around, and metasurfaces are a newer route to tens of thousands of sites. Fast cameras read the atoms out by their fluorescence: Caltech reports 99.994% imaging fidelity.
Several stabilized lasers, for cooling, trapping, gates and Rydberg excitation, and their electronics dominate cost and reliability. There is no dilution refrigerator. Apart from the microkelvin atoms, the apparatus runs at room temperature, though it needs a clean, vibration-controlled room.
How it is controlled and read out
Single-qubit gates are usually global: microwaves or laser pairs rotate every atom at once, and Caltech reports 99.98% across its array. Focused addressing beams act on chosen atoms, more slowly and less accurately.
Two-qubit gates are also driven globally, over an entangling zone. Which pairs interact is set by where the atoms sit, not by which beams fire, so no qubit needs its own control wiring. Typical two-qubit fidelities are about 99.5 to 99.6%: Harvard and QuEra reported 99.5% in parallel on 60 atoms in 2023, Atom Computing about 99.6% (with 0.96% of atoms lost per gate), and Princeton and USTC about 99.6% in 2026. The record is 99.854%, in an April 2026 preprint from Harvard and MIT; Caltech held the earlier record of 99.71%.
Moving atoms gives any-to-any connectivity. Acousto-optic deflectors carry whole rows of atoms while they keep their quantum state, and Caltech moved atoms 610 µm at 99.95% fidelity. Most machines now use zones: storage, an entangling zone under the Rydberg laser, and a readout zone where imaging light cannot disturb the data qubits. The move dominates the gate: about 0.1 to 1 millisecond (ms) per two-qubit gate including transport, against 0.2 to 0.5 µs for the laser pulse.
Readout takes milliseconds, so one round of error correction with imaging in the middle of the circuit takes about 1 to 10 ms, roughly 1,000 to 10,000 times longer than on superconducting qubits.
What makes it hard to scale
Atom loss. Atoms escape through gas collisions, heating, Rydberg decay and imaging, and a lost atom is an error that standard codes handle poorly. Atom Computing's loss per gate (0.96%) is larger than its gate error (0.43%), and discarding runs with lost atoms lifts Harvard and MIT's 99.854% to 99.941%.
Slow cycles. Moves of hundreds of microseconds and imaging of milliseconds set the clock for error correction, and the time budget for decoding.
Optical scaling. Trap laser power grows with the number of sites, and tweezer optics have topped out around 10,000 sites. Rydberg lasers need high power and low noise to drive many atoms fast and accurately.
Replenishment. Even with a 20-minute atom lifetime, a 10,000-atom array loses an atom about every tenth of a second, so long computations need atoms reloaded continuously without disturbing the qubits.
Crosstalk and calibration. Stray Rydberg interactions, scattered imaging light and drifting lasers need constant calibration.
What could solve it
Continuous reloading. Harvard and MIT kept more than 3,000 qubits running for over two hours by feeding atoms from a reservoir on optical conveyor belts, replacing more than 30,000 qubits a second (Nature, 2025).
Erasure conversion. Atoms such as ytterbium and strontium can be operated so that most errors leave the atom in a detectable state. The location of each error is then known, which makes it cheaper to correct.
Zones and reuse. Measuring and reusing helper atoms in the middle of a circuit raised Harvard's cycle rates about a hundredfold.
Logic by moving atoms. Because atoms move freely, a logical gate can act on whole code blocks in one parallel step, and codes that need long-range connections become practical. Harvard showed error suppression of 2.14 times below threshold in 2025, and with QuEra and MIT it distilled logical magic states (Nature, 2025).
New optics and modules. Metasurfaces for tens of thousands of sites, cryogenic vacuum for longer atom lifetimes, and photonic links between vacuum modules.
Strengths and limits
Strengths. Identical qubits; the largest arrays, with thousands of atoms; coherence of seconds; any-to-any connectivity by moving atoms; global control with few wires; no dilution refrigerator; and rapid progress in error correction.
Limits. A slow clock, with error-correction rounds of milliseconds; atom loss as a leading source of error; two-qubit fidelity typically 99.5 to 99.6%, with a 99.854% record in a 2026 preprint, still behind trapped ions; complex lasers that are hard to keep stable; and immature links between modules.
Key numbers
Two-qubit gates: typically 99.5 to 99.6%; record 99.854% (Harvard and MIT, April 2026 preprint).
Gate time: 0.2 to 0.5 µs for the Rydberg pulse; about 0.1 to 1 ms including the move.
Coherence: typically around 1 s (rubidium); about 7 s (ytterbium, Atom Computing); up to 12.6 s (cesium, Caltech).
Largest arrays: 6,100 atoms held as qubits (Caltech, 2025); more than 3,000 run continuously for over two hours (Harvard and MIT, 2025).
Readout: 99.994% imaging fidelity (Caltech), in milliseconds.
Error-correction round: about 1 to 10 ms.
Operating environment: a room-temperature vacuum apparatus with atoms cooled to microkelvin.
Quick answers
What is an optical tweezer? A tightly focused laser beam that holds a single atom at its focus. Thousands of tweezers, made by shaping one laser, form the array of qubits.
What is the Rydberg blockade? When one atom is excited to a Rydberg state, its strong interaction shifts a nearby atom's energy so that atom cannot be excited too. This conditional behavior produces a two-qubit gate.
Why are neutral-atom quantum computers slow? The laser pulse for a gate is fast, but atoms must be moved together first and imaged to be measured, which takes hundreds of microseconds to milliseconds.
How many qubits do neutral-atom computers have? Arrays of thousands of atoms have been held as qubits, 6,100 at Caltech, but not all of them are linked by gates in one computation. The number that can be entangled at high fidelity in one run is smaller.
Sources
Caltech, a tweezer array with 6,100 highly coherent atomic qubits, Nature (2025)
PostQuantum, summary of Harvard and MIT's continuous 3,000-qubit operation (Nature, 2025)
Lukin group, high-fidelity entangling gates with neutral atoms, 99.854% (arXiv 2604.25987, preprint)
Tsai et al. (Caltech), 99.71% Rydberg gates, PRX Quantum 6, 010331 (2025)
Atom Computing, toric-code error correction with reloading (arXiv 2606.04079)
About Anyon Computing
Anyon Computing Inc. (also known as Anyon Technologies) is a quantum computing company that provides superconducting modular quantum supercomputers, which it tightly couples with NVIDIA GPUs via NVQLink for data center deployment at scale. Anyon pioneered deploying NVIDIA NVQLink inside a commercial production data center and is among the first superconducting QPU backends integrated with NVIDIA CUDA-Q. Founded by pioneers from Caltech and UC Berkeley. https://www.anyoncomputing.com/
NVIDIA, CUDA-Q and NVQLink are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and product names may be trademarks of the respective companies with which they are associated.
Media contactAnyon Computing · info@anyoncomputing.com