Physics guide ×
How a neutral-atom quantum computer holds, arranges, entangles and measures single atoms, in the order the machine does it, and how that becomes a computation. Each section has a button that sets up the scene to match.
Light that holds atoms
A laser tuned slightly to the red of an atom's resonance makes the atom's electron cloud oscillate with the light. That induced dipole has lower energy where the light is more intense, so the atom is pulled towards the brightest point. A high-numerical-aperture (NA) objective focuses the beam to a spot about a micrometre wide, and each focus becomes a trap: an optical tweezer.
U(r) ≈ −(3πc² / 2ω₀³)·(Γ/Δ)·I(r) far-detuned two-level atom; ω₀, Γ: the atom's resonance frequency and linewidth, Δ = ω − ω₀ < 0 (red detuning) I(r,z) = I₀ (w₀/w(z))² exp(−2r²/w(z)²) w₀ ≈ λ/(π·NA), w(z) = w₀√(1 + (z/z_R)²), z_R = πw₀²/λ
Near the focus the trap is harmonic: tight across the beam (ω_r = √(4U/mw₀²)) and looser along it (ω_z = √(2U/mz_R²)). A thermal atom therefore spreads σ_r = (w₀/2)√(T/U) sideways and σ_z = (z_R/√2)√(T/U) along the beam, which is why the atoms here look like small cigars.
Waist w₀ 0.5–1 µm Trap depth ~1 mK from ~1 mW per tweezer (Rb, ~800 nm) Atom size at 10 µK ~30 nm × ~110 nm (1σ)
One atom, or none
Tweezers are filled from a cloud of laser-cooled atoms. The trap is so small that the cooling light drives collisions that throw atoms out in pairs, so whatever number entered, the trap ends up with one atom or none, each about half the time. This is called collisional blockade. Tailoring the cooling light so collisions eject only one atom at a time raises the odds to 80–90%.
Loading ≈50% standard · 80–90% enhanced Temperature ~10 µK Trap lifetime tens of seconds to 23 minutes (2025), set by the vacuum Continuous reloading 3,000+ qubits kept filled for 2+ hours (2025)
Making a perfect grid
A camera images every atom's fluorescence to find which traps are full. A second set of tweezers, steered by acousto-optic deflectors (AODs), then carries atoms into the empty sites. An AOD is a crystal driven by radio-frequency sound; each frequency deflects the beam by a slightly different angle, so several tones make several movable tweezers at once.
θ ≈ λ·f / v_s deflection for RF tone f, sound speed v_s
In a two-dimensional AOD the tweezers sit on a grid of row tones and column tones, so whole rows and columns move together and can never pass through one another. The simplest assemblers move one atom at a time; faster ones move many at once, row by row or with several independent beams. The grid view here moves a few atoms at a time along straight lines, for clarity.
Assembly time tens of milliseconds for ~2,000 atoms Array sizes 6,100 atoms trapped (2025); ~1,000–3,000 typical in processors
Atoms as qubits
Each atom stores a qubit in two long-lived internal states, for example two hyperfine levels of rubidium or the two nuclear-spin states of ytterbium-171. Microwaves or focused lasers rotate between them to perform single-qubit gates. Every atom of an isotope is identical, so there is no fabrication spread between qubits; what still needs calibrating is the light around them, such as small differences in trap depth from site to site.
Coherence seconds; 12.6 s shown in a 6,100-atom array (2025) Single-qubit gates 99.9–99.98% fidelity
Rydberg blockade and two-qubit gates
To entangle two atoms, a laser excites them towards a Rydberg state: an electron orbit so large (principal quantum number n ≈ 50–100) that the atom becomes roughly a thousand times bigger. Two Rydberg atoms interact strongly. Closer than the blockade radius R_b, the doubly excited state is shifted out of resonance, so at most one of the pair can be excited. That conditional behaviour becomes a controlled-Z (CZ) gate.
V(r) = C₆ / r⁶, R_b = (C₆ / ħΩ)^(1/6) Ω: Rabi frequency of the drive
Blockade radius ~5–10 µm CZ fidelity 99.5% in parallel (2023); 99.7–99.85% in 2025–26 benchmarks Rydberg atom size grows as n²
The violet bubble in the scene has radius R_b. Change it under Physics settings.
A processor with zones
Zoned machines keep qubits in a storage zone and move them on demand. AOD tweezers carry whole rows into an entangling zone and set them down in close pairs: closer than R_b within a pair, farther than R_b from the next pair. A Rydberg pulse then performs many CZ gates in parallel, and the atoms return to storage or move on. Separating the zones protects idle qubits from gate and measurement light, and moving atoms lets any qubit be paired with any other.
Pair spacing ~2 µm (< R_b) Gap to the next pair ~10 µm (> R_b) Transport order-preserving row and column moves
Try the zoned-layout sliders: when R_b reaches the next pair, the label warns of crosstalk.
Reading out a qubit
To measure, atoms are lit with resonant light and the objective collects their fluorescence onto a camera. The measurement is made state-selective, for example by removing or shelving one of the two qubit states, so a bright atom reads as one value and a dark site as the other. Error correction needs some qubits measured mid-computation, which means keeping that scattered light away from the rest: a separate readout zone, or shelving the spectators in protected states.
Some machines, and the circuit mode here, read out three outcomes, not two. Atoms in |0⟩ are first shelved into a dark state and the camera images the |1⟩ atoms; then all atoms are made bright for a second image. Bright in the first image means 1, dark then bright means 0, and dark in both means the atom has gone. The camera inset in the panel combines the two images into one colour picture: gold for 1, blue for 0, an empty dashed circle for a lost atom. A lost atom is an "erasure": an error whose location is known, which error-correcting codes can handle far more cheaply than an invisible error.
Photons scattered thousands per atom Collected a few percent Imaging time milliseconds Detection fidelity 99.9% and above
From moving atoms to an answer
Six qubits are described by 2⁶ = 64 numbers, the amplitudes of every bitstring from 000000 to 111111. Each laser pulse changes those numbers; moving atoms only decides which qubits a pulse can link. Measurement then picks one bitstring b with probability P(b) = |⟨b|ψ⟩|², so a quantum program is run many times ("shots") and the answer is read from the histogram.
Hold red tweezers: no change to the state Move orange tweezers: choose who meets Prepare optical pumping: every qubit → |0⟩ Rotate green gate laser: single-qubit gates (H, Z) Entangle violet Rydberg pulse: CZ on close pairs Read blue imaging light: 1, 0, or – if the atom was lost
CNOT(c→t) = H_t · CZ(c,t) · H_t GHZ: (|000000⟩ + |111111⟩)/√2 → only two outcomes, all qubits agree
The circuit mode runs five short programs. Coin flips puts every qubit in superposition, so all 64 strings are equally likely. H twice applies the same gate again and every qubit returns to 0: randomness that interference undoes. Phase flip adds a Z gate in between, which changes only signs, yet every shot now reads 111111. Bell pairs entangles neighbours, so each pair always agrees. GHZ links all six, so every shot reads 000000 or 111111. By default the machine is ideal; switch on hardware errors to see how a fraction of a percent per gate spoils results, which is why large computations need error correction.
How it is computed: six qubits are held as 64 amplitudes. H, Z and CZ act on them exactly, and measurement samples a bitstring by the Born rule. Each CZ applies to the atoms that end up closer than R_b, so changing R_b can switch a gate off or add crosstalk. An optional error model adds random Pauli errors and atom loss at today's typical rates. Lost atoms read out as –, and the panel reports how many shots the ideal circuit would allow, with and without discarding shots that lost an atom. The README that comes with this tool explains each step.
Correcting errors
Every gate, wait and readout goes wrong now and then, so large computations need error correction: the information is spread over several atoms and checked continually, without ever reading the information itself. The simplest example, shown in the Error correction mode, is a repetition code. Three data atoms hold one bit as 000 or 111. Two check atoms measure the parities D1·D2 and D2·D3: each check is entangled with its two neighbours by CZ gates and then imaged in a separate readout zone. The parities reveal which atom flipped, but not the stored value.
S_A = Z₁Z₂, S_B = Z₂Z₃ the checks (stabilisers); stored states 000 and 111 p_L ≈ 3p² one round, perfect checks: two flips fool a distance-3 code
Checks can be misread too, so they are repeated (three rounds here) and the decoder looks at changes between rounds, called detection events. The decoder in this tool is exact maximum likelihood: for every run it weighs all 2¹⁸ combinations of the 18 possible error events. Because checking adds operations, the code helps only when errors are rare enough: in this tool's model the protected bit beats a lone atom below about 2.9% error per operation. Erasures (atom loss that is detected and repaired) are easier, because the decoder knows where they happened: a distance-3 code can fix two of them, against one hidden bit flip.
A repetition code catches only bit flips; with the "+" input you can watch phase flips get through. Codes such as the surface code check for both kinds with a two-dimensional pattern of checks, and below a threshold error rate (about 1% per operation for the surface code) their logical error rate falls exponentially as the code grows. Neutral-atom machines have already run circuits on dozens of logical qubits, using atom moves to connect checks with any data atom.
Code here repetition, distance 3, 3 rounds Fixes any 1 bit flip, or 2 erasures Surface-code threshold ~1% per operation Below threshold shown experimentally (2024–25) Neutral-atom logical qubits 48 in one processor; 24 entangled (2024)
Model used in the mode: each wait, gate or readout fails with probability p; a data atom sees three operations per round and a check atom five, and every error of a round is drawn at its start. Hook errors and leakage are not modelled.
The machine around the atoms
The qubits occupy a fraction of a millimetre, but the apparatus around them fills a room. The atoms sit inside a glass cell held at ultra-high vacuum, at the focus of a high-NA objective that both forms the tweezers and collects fluorescence for the camera. Before each run, a magneto-optical trap (coils plus crossing cooling beams) gathers a cloud of cold atoms from an atomic beam, and the tweezers are loaded from it. Around the vacuum system, an optical table carries the lasers, a spatial light modulator for the static tweezers, acousto-optic deflectors for the moving ones, and the camera; control electronics time every pulse.
Vacuum ~10⁻¹¹ mbar (ultra-high vacuum) Cell glass, a few cm across Objective NA ≈ 0.5–0.7, working distance ~1–3 cm Atom cloud ~1 mm, millions of atoms Footprint one or more optical tables, plus racks
The zoom-out (button under the title) has six stops, from the array to the whole lab, including one that follows the tweezer light from laser to atoms. Tap any label there for a short explanation. It is a schematic of a typical laboratory, not a specific machine: some designs use two cells, a second cooling stage or an objective inside the vacuum.
What is to scale here
Computed from your settings: beam waists and Rayleigh ranges, spacings, zone dimensions, and each atom's 3σ thermal size.
Real-world reference: the human hair, about 70 µm wide.
Not to scale: the glow around each atom, the swelling of an excited atom, the soft edge of the blockade bubble.
Circuit mode: the quantum state is computed exactly; errors and atom loss appear only when you switch them on. Atom glows are enlarged there so their colours can be read.
Illustrative: zone dimensions are typical rather than taken from one machine; transport paths are straight segments, and in the grid view several atoms move independently at once; infrared light is drawn red; motion is slowed by a factor of thousands (atom moves) to millions (Rydberg pulses) so it can be followed.
References
R. Grimm, M. Weidemüller, Yu. B. Ovchinnikov, Optical dipole traps for neutral atoms , Adv. At. Mol. Opt. Phys. 42, 95 (2000). arXiv:physics/9902072
A. M. Kaufman, K.-K. Ni, Quantum science with optical tweezer arrays of ultracold atoms and molecules , Nat. Phys. 17, 1324 (2021). doi
N. Schlosser et al., Sub-poissonian loading of single atoms in a microscopic dipole trap , Nature 411, 1024 (2001). doi
T. Grünzweig et al., Near-deterministic preparation of a single atom in an optical microtrap , Nat. Phys. 6, 951 (2010). doi
M. O. Brown et al., Gray-molasses optical-tweezer loading: controlling collisions for scaling atom-array assembly , Phys. Rev. X 9, 011057 (2019). doi
D. Barredo et al., An atom-by-atom assembler of defect-free arbitrary two-dimensional atomic arrays , Science 354, 1021 (2016). doi
M. Endres et al., Atom-by-atom assembly of defect-free one-dimensional cold atom arrays , Science 354, 1024 (2016). doi
D. Jaksch et al., Fast quantum gates for neutral atoms , Phys. Rev. Lett. 85, 2208 (2000). doi
M. Saffman, T. G. Walker, K. Mølmer, Quantum information with Rydberg atoms , Rev. Mod. Phys. 82, 2313 (2010). doi
S. J. Evered et al., High-fidelity parallel entangling gates on a neutral-atom quantum computer , Nature 622, 268 (2023). doi
D. Bluvstein et al., Logical quantum processor based on reconfigurable atom arrays , Nature 626, 58 (2024). doi
M. A. Norcia et al., Midcircuit qubit measurement and rearrangement in a ¹⁷¹Yb atomic array , Phys. Rev. X 13, 041034 (2023). doi
J. W. Lis et al., Midcircuit operations using the omg architecture in neutral atom arrays , Phys. Rev. X 13, 041035 (2023). doi
A. E. Siegman, Lasers , University Science Books (1986), chapters 16–17 on Gaussian beams.
H. J. Manetsch et al., A tweezer array with 6100 highly coherent atomic qubits , Nature (2025). doi
N.-C. Chiu et al., Continuous operation of a coherent 3000-qubit system , Nature (2025). arXiv:2506.20660
S. J. Evered, M. Xu, S. H. Li et al., High-fidelity entangling gates and nonlocal circuits with neutral atoms , preprint (2026). arXiv:2604.25987
M. A. Nielsen, I. L. Chuang, Quantum Computation and Quantum Information , Cambridge University Press (2010).
B. W. Reichardt et al., Fault-tolerant quantum computation with a neutral atom processor (Microsoft and Atom Computing), preprint (2024). arXiv:2411.11822
Y. Wu, S. Kolkowitz, S. Puri, J. D. Thompson, Erasure conversion for fault-tolerant quantum computing in alkaline earth Rydberg atom arrays , Nat. Commun. 13, 4657 (2022). doi
E. L. Raab, M. Prentiss, A. Cable, S. Chu, D. E. Pritchard, Trapping of neutral sodium atoms with radiation pressure , Phys. Rev. Lett. 59, 2631 (1987). doi
S. Chu, L. Hollberg, J. E. Bjorkholm, A. Cable, A. Ashkin, Three-dimensional viscous confinement and cooling of atoms by resonance radiation pressure , Phys. Rev. Lett. 55, 48 (1985). doi
W. D. Phillips, H. Metcalf, Laser deceleration of an atomic beam , Phys. Rev. Lett. 48, 596 (1982). doi
H. J. Metcalf, P. van der Straten, Laser Cooling and Trapping , Springer (1999).
F. Nogrette et al., Single-atom trapping in holographic 2D arrays of microtraps with arbitrary geometries , Phys. Rev. X 4, 021034 (2014). doi
H. Bernien et al., Probing many-body dynamics on a 51-atom quantum simulator , Nature 551, 579 (2017). doi
S. Ebadi et al., Quantum phases of matter on a 256-atom programmable quantum simulator , Nature 595, 227 (2021). doi
L. Henriet et al., Quantum computing with neutral atoms , Quantum 4, 327 (2020). doi
K. Barnes et al. (Atom Computing), Assembly and coherent control of a register of nuclear spin qubits , Nat. Commun. 13, 2779 (2022). doi
J. F. O'Hanlon, A User's Guide to Vacuum Technology , 3rd ed., Wiley (2003).
P. W. Shor, Scheme for reducing decoherence in quantum computer memory , Phys. Rev. A 52, R2493 (1995). doi
E. Dennis, A. Kitaev, A. Landahl, J. Preskill, Topological quantum memory , J. Math. Phys. 43, 4452 (2002). doi
A. G. Fowler, M. Mariantoni, J. M. Martinis, A. N. Cleland, Surface codes: towards practical large-scale quantum computation , Phys. Rev. A 86, 032324 (2012). doi
Google Quantum AI, Exponential suppression of bit or phase errors with cyclic error correction , Nature 595, 383 (2021). doi
Google Quantum AI and collaborators, Quantum error correction below the surface code threshold , Nature 638, 920 (2025). doi
Q. Xu et al., Constant-overhead fault-tolerant quantum computation with reconfigurable atom arrays , Nat. Phys. 20, 1084 (2024). doi
Using and citing this explorer
You are welcome to use, share and adapt this explorer, in classes, talks, websites or your own projects, under the Creative Commons Attribution 4.0 licence (CC BY 4.0). The one condition is credit: name the author, link to the original and say if you changed it.
Suggested citation: G. Valentí-Rojas, Atoms held by light: an interactive explorer of neutral-atom quantum computers (2026), https://valenti-rojas.com/tweezers/ . Built with Claude (Anthropic). The three.js library (MIT licence) and the fonts (SIL Open Font Licence) keep their own licences.