Loading...
Loading...
Individual atoms held by light, entangled through Rydberg interactions, and moved in programmable arrays offer a scalable, high-connectivity path to quantum computing.
Figures are representative order-of-magnitude values for leading neutral-atom systems.
From laser cooling and magnetic traps to optical tweezers and Rydberg gates, neutral atoms have evolved into a leading platform for scalable quantum computing.
Neutral-atom research began with magneto-optical traps and Bose-Einstein condensates, tools developed for precision spectroscopy and atomic clocks.
The invention of optical tweezers made it possible to trap individual atoms at specific sites, turning a dilute atomic cloud into a programmable register.
1980s–1990s
Laser cooling
Development of MOT, molasses, and sub-Doppler cooling enabled control of single atoms at microkelvin temperatures.
2000s
Tweezers & Rydberg
Optical tweezers isolated individual atoms, and Rydberg blockade was demonstrated as a coherent two-atom interaction.
2010s–today
Programmable arrays
Hundreds of atoms in reconfigurable arrays, with native multi-qubit gates and error-corrected logical qubits emerging.
Rydberg blockade, first observed in the 2000s, provided a fast, strong interaction between distant atoms and became the basis of two-qubit gates.
Today, neutral-atom systems combine hundreds of qubits, reconfigurable geometries, and mid-circuit measurement, placing them among the most scalable near-term platforms.
Key experimental and engineering milestones mark the transition from single-atom control to programmable neutral-atom processors.
Early experiments demonstrated trapping and cooling of individual atoms, paving the way for deterministic array loading.
Click a milestone to see details.
Each milestone increased qubit number, gate fidelity, or measurement capability, bringing the platform closer to fault-tolerant quantum computing.
Optical tweezers are focused laser beams that trap individual neutral atoms at their intensity maxima or minima, depending on the wavelength.
A tightly focused laser beam creates a dipole potential proportional to the atomic polarizability and the local intensity.
Atoms are loaded from a magneto-optical trap and then imaged to identify which tweezers are occupied, allowing deterministic rearrangement into defect-free arrays.
Dipole potential
Tweezer spacing is typically a few micrometers, large enough for individual addressing but small enough for strong Rydberg interactions.
Moving tweezers by steering acoustic-optic or spatial-light-modulator beams can shuttle atoms between zones or reshape the array geometry.
Red-detuned tweezers trap atoms at intensity maxima, while blue-detuned tweezers trap them at intensity minima, suppressing photon scattering.
Tweezer depths are usually tens to hundreds of microkelvin, balancing tight confinement against heating from photon scattering and Raman scattering.
High-numerical-aperture optics are required to focus each tweezer to a diffraction-limited spot only about one micrometer across.
Advanced systems use planar arrays of microlenses to generate hundreds or thousands of tweezers simultaneously.
Neutral-atom qubits are encoded in long-lived ground hyperfine states, with transitions driven by microwave, Raman, or Rydberg lasers.
The most common choice is two hyperfine sublevels of the ground electronic state, such as |F=1, m_F=0⟩ and |F=2, m_F=0⟩ in alkali atoms.
These clock states are insensitive to magnetic field fluctuations to first order, giving coherence times of seconds or longer.
Zeeman energy
Single-qubit rotations are performed with microwave pulses or two-photon Raman transitions, controlled by phase and amplitude.
Rydberg states with large principal quantum number n are used for two-qubit gates because they have strong van der Waals interactions.
The qubit readout is performed by fluorescence imaging: one state scatters photons while the other remains dark.
Rotation operator
Different atomic species trade off wavelength, nuclear spin, and available Rydberg blockade radius.
State preparation is done by optical pumping into a chosen hyperfine sublevel before computation.
Raman Rabi frequency
Because the qubit states are electronic ground states, they experience minimal decoherence from spontaneous emission.
Neutral atoms are slowed and cooled by laser light, then loaded into tweezers before quantum operations begin.
A magneto-optical trap combines counterpropagating red-detuned lasers with a magnetic field gradient to cool and trap atoms from room-temperature vapor.
The Doppler cooling limit sets a minimum temperature for a given atomic transition, typically around one hundred microkelvin for alkali atoms.
Doppler limit
Optical molasses removes the magnetic field and uses six laser beams to further cool atoms below the Doppler limit.
Sub-Doppler techniques such as Sisyphus cooling exploit the ac Stark shift of degenerate sublevels to reach temperatures below ten microkelvin.
Each absorbed photon transfers a recoil momentum that sets the ultimate quantum limit on localization.
Recoil energy
Low temperature and tight confinement are essential because hotter atoms can escape shallow tweezers and collide with neighbors.
After cooling, atoms are transferred from the MOT into the optical tweezer array by reducing the trap depth gradually.
Some platforms use a two-dimensional or three-dimensional optical lattice to increase atom density before array loading.
Neutral-atom systems can run as digital gate-based processors or as analog quantum simulators, each suited to different problems.
In digital mode, atoms are encoded as qubits and a universal gate set is applied sequentially, similar to superconducting or ion-trap computers.
In analog mode, the entire array evolves under a tunable Hamiltonian, which is ideal for studying quantum phases and optimization landscapes.
The Rydberg Hamiltonian is controlled by the global Rabi frequency Ω(t) and detuning δ(t), which define the local field and interaction terms.
Ising-type Hamiltonians are naturally implemented with pairwise van der Waals interactions between adjacent atoms.
Ising Hamiltonian
XY and other spin models can be engineered by choosing different Rydberg states, laser polarizations, or lattice geometries.
Analog simulation is well suited to finding ground states of combinatorial optimization problems through adiabatic evolution.
Hybrid algorithms combine analog evolution with digital gates, using the best features of both approaches.
Overhead for converting analog dynamics into digital gates is significant, so native analog execution can outperform gate decomposition for specific problems.
By tuning global parameters, the atom array becomes a programmable quantum simulator whose phase diagram can be explored interactively.
Analog quantum simulators use the natural Hamiltonian of the system rather than compiled gate sequences.
Changing the Rabi frequency and detuning drives the array across phase boundaries such as ordered, disordered, and crystalline states.
Rydberg Hamiltonian
Promoting atoms to highly excited Rydberg states creates strong interactions that enable fast, high-fidelity entangling gates.
A Rydberg atom has an electron in a state with large principal quantum number n and a huge polarizability.
Two nearby Rydberg atoms experience a van der Waals interaction that scales with the inverse sixth power of their separation.
van der Waals interaction
When the interaction exceeds the Rabi frequency, the doubly excited state is shifted out of resonance, an effect called Rydberg blockade.
Blockade ensures that only one atom in a pair can be excited, creating a controlled phase conditional on the state of its neighbor.
Blockade condition
The blockade radius is typically a few micrometers, matching the spacing of optical tweezers and allowing nearest-neighbor gates.
Multi-qubit gates and native Toffoli gates can be built by exploiting simultaneous blockade among several atoms.
Residual interactions with more distant atoms, known as spectator errors, are a major source of gate infidelity.
Trapping atoms while preserving qubit coherence requires finding wavelengths where ground and Rydberg states experience the same trapping potential.
An optical tweezer perturbs both the ground and Rydberg states, causing differential light shifts that dephase the qubit.
Magic condition
At a magic wavelength, the polarizabilities of the two relevant states are equal, so the transition frequency is unchanged by the trap.
Magic trapping is essential for performing high-fidelity gates inside tweezers without letting the trap light corrupt the qubit phase.
Effective Rabi frequency
The effective Rabi frequency for a two-photon Rydberg transition depends on the individual laser couplings and their detuning from an intermediate state.
Doppler shifts from finite atomic temperature also limit gate fidelity, especially during longer pulse sequences.
Doppler infidelity
Careful pulse shaping and optimal control can compensate for residual light shifts and motional excitation.
Different atomic species and Rydberg states have different magic wavelengths, so the trap design must match the target transition.
Trapping at a magic wavelength is one of the key reasons modern neutral-atom gates have reached fidelities above 99.5%.
Neutral-atom processors implement rotations with microwaves or Raman beams, and entangling gates through Rydberg pulses.
Single-qubit gates are fast, often below one microsecond, and can be addressed globally or with individual laser beams.
Local addressing uses tightly focused beams to illuminate only one atom while leaving its neighbors unaffected.
The two-qubit CZ gate is the most common entangling gate, using a sequence of Rydberg pulses to accumulate a conditional phase.
Geometric pulse shapes, such as optimal pulses derived from reverse engineering, can suppress leakage and motional errors.
CZ unitary
Native multi-qubit gates such as CCZ and Toffoli are implemented in a single pulse sequence, saving gate count compared to digital decomposition.
Intrinsic error bound
Intrinsic gate errors are bounded by the adiabatic speed limit, which relates the interaction strength and pulse duration to the minimum possible error.
Gate benchmarking with randomized benchmarking and cross-entropy measures helps separate coherent errors from stochastic ones.
Native gates can be used directly in quantum algorithms or decomposed into a universal set of one- and two-qubit gates.
A controlled-Z gate accumulates a conditional phase by exciting the control atom to a Rydberg state and performing a 2π rotation on the target atom.
The CZ gate applies a π pulse on the control atom, a 2π pulse on the target atom, and a final π pulse on the control atom.
CZ matrix
When the control atom is in |1⟩, the Rydberg blockade shifts the target atom's 2π pulse, preventing a full return to the ground state and accumulating a π phase.
When the control atom is in |0⟩, the target atom completes a full 2π rotation and returns unchanged, leaving no phase.
The geometric phase is robust because it depends on the area enclosed in the Bloch sphere rather than the exact pulse amplitude.
Optimal pulse shapes reduce the effect of finite blockade strength, motional excitation, and laser noise.
Square pulses are simple but cause more leakage; smooth optimal pulses keep the population closer to the desired subspace.
The same sequence can be extended to multi-qubit gates by involving several control atoms simultaneously.
Achieving fault-tolerant gates requires identifying and suppressing every error source, from atomic motion to laser noise and environmental fields.
The intrinsic error of a Rydberg gate is set by the ratio of the interaction strength to the pulse duration, which obeys a quantum speed limit.
Minimum intrinsic error
Thermal motion of the atoms causes Doppler shifts and time-varying laser coupling during the gate, converting motional energy into phase errors.
Blackbody radiation from the room-temperature environment can ionize or dephase Rydberg atoms, especially for high n states.
Laser intensity and phase noise directly affect the Rabi frequency and must be stabilized below the part-per-thousand level.
Spectator atoms near the target qubit experience weak Rydberg shifts and can pick up unwanted phases.
Stray electric and magnetic fields shift Rydberg levels and must be shielded or compensated to maintain gate calibration.
Neutral atoms can implement multi-controlled gates directly, reducing the number of pulses and improving algorithm efficiency.
A native three-qubit gate, such as CCZ or Toffoli, can be performed with one Rydberg pulse sequence instead of decomposing into many CNOTs.
The resource savings are large: a Toffoli may need fewer than ten pulses natively versus six CNOTs and eight single-qubit gates in digital form.
Native multi-qubit gates are especially useful for arithmetic, reversible logic, and quantum error-correction protocols.
Pulse labels in the chart compare the native pulse count to the decomposed count for a representative Toffoli-like circuit.
The geometry of the atom array is not fixed; it can be reconfigured to match the connectivity required by an algorithm or Hamiltonian.
Static 2D arrays can be generated with spatial light modulators or microlens arrays, producing hundreds of traps with regular spacing.
Acousto-optic deflectors can create moving tweezers that transport atoms between zones, enabling a QCCD-like architecture.
Reconfigurable arrays allow all-to-all connectivity for problems that are not local on a fixed grid, such as MaxCut on random graphs.
The QCCD architecture separates storage, entangling, readout, and reservoir zones, so each step is optimized in its own region.
Atoms can be rearranged during computation to reduce the number of shuttling steps or to place interacting pairs next to each other.
As arrays scale to thousands of atoms, fast imaging and feedback loops must keep track of defect positions and trigger reloads.
QCCD architectures move atoms between specialized zones, combining long storage coherence with high-fidelity entangling regions.
In a QCCD-like neutral-atom processor, atoms are stored in a low-noise storage zone and only shuttled to an entangling zone when a gate is needed.
This separation protects qubits from the strong Rydberg lasers used during gates and from the heating caused by readout illumination.
Atoms move between zones on a fast, repeated cycle.
After measurement, spent ancilla atoms can be moved to a reservoir zone and replaced with fresh atoms from a loading zone.
Move atom pairs between storage and entangling zones.
CZ or CCZ gate in the entangling zone.
Image, decide, and refill ancilla in the readout zone.
Rearrange array geometry for the next algorithm step.
A neutral-atom processor is built around a vacuum cell, high-NA optics, laser systems, and fast electronics for real-time control.
The vacuum cell maintains an ultra-high-vacuum environment to reduce collisions between trapped atoms and background gas molecules.
High-numerical-aperture objectives sit inside or outside the cell to focus tweezer beams and collect fluorescence photons.
Spatial light modulators and acousto-optic deflectors create and move the tweezer array under software control.
Rydberg excitation lasers are locked to stable references to maintain narrow linewidth and long coherence during gates.
Magnetic field coils provide quantization fields and allow fast switching for state preparation and addressing.
Atoms are trapped in a vacuum cell, but residual background gas causes loss. Cryogenic environments can extend the lifetime dramatically.
In a room-temperature vacuum cell, background gas collisions limit the trap lifetime to a few tens of seconds at typical pressures.
Cryogenic environments reduce the partial pressure of all background gases, extending trap lifetimes to hundreds or thousands of seconds.
A cryostat also reduces blackbody radiation on Rydberg atoms, improving gate coherence and lowering ionization rates.
The trade-off between trap depth and lifetime is governed by the scattering rate of tweezer photons, which heats the atoms.
Trap lifetime scaling
Deep traps increase photon scattering and shorten lifetime; shallow traps require colder atoms but live longer.
Real-time control, waveform generation, and image analysis coordinate every step of the experiment from circuit to measurement.
The user circuit is compiled into a sequence of microwave, Raman, and Rydberg pulses with precise timing and phase.
A scheduler reorders operations to minimize shuttling and maximize parallel gate execution across the array.
FPGAs and GPUs generate waveforms and analyze fluorescence images in real time, enabling fast feedback and mid-circuit correction.
Camera frames are processed to identify which atoms are present, which have been lost, and which must be replaced from a reservoir.
The feedback loop is essential for fault tolerance, allowing erasure errors to be detected and corrected before they propagate.
Fluorescence imaging reads out qubit states and can detect atom loss, enabling mid-circuit measurement and conditional reloading.
A near-resonant laser scatters photons from atoms in one qubit state while the other state remains dark.
Collected photons are imaged onto a high-efficiency camera with single-atom resolution, allowing each site to be read independently.
Photon recoil energy
The recoil energy from scattered photons can heat atoms out of shallow traps, so readout pulses must be brief and carefully tuned.
Fluorescence images reveal whether an atom is present or lost, making atom loss a detectable erasure error rather than a silent bit-flip.
Mid-circuit readout can measure ancilla qubits during computation and condition future operations on the outcome.
After readout, lost or measured atoms can be replaced from a reservoir and recooled, restoring the array for the next step.
This cycle of image, decide, and refill is what makes neutral-atom QCCD architectures possible.
Measuring ancilla qubits during computation, then replacing them with fresh atoms, converts loss into a detectable erasure error.
Mid-circuit readout begins by moving an atom to a readout zone or splitting it into two separate traps based on its qubit state.
A fluorescence image reveals the state, while the same process can confirm whether the atom has survived.
If the atom is lost, a fresh atom is moved from a reservoir and prepared in the desired initial state.
Because loss is detected and corrected, the effective error rate for error correction is lower than the raw physical error rate.
Mid-circuit measurement is central to syndrome extraction in surface-code and color-code implementations.
Neutral-atom loss is a detectable erasure error; loss-aware decoders can turn it into a significant advantage for surface-code error correction.
Surface codes detect errors by measuring stabilizers on a lattice of data and ancilla qubits.
Conventional decoders assume only Pauli errors and may misinterpret an atom loss as an X or Z error.
Loss-aware decoders use the fact that the missing qubit position is known, matching erasures directly to edges in the decoding graph.
Including loss information improves the effective error threshold and reduces the logical error rate per round.
Neutral-atom platforms can implement quantum error correction using surface codes, color codes, and transversal logical gates.
The long coherence time and high connectivity of neutral atoms make them well suited for stabilizer codes.
Surface codes require only nearest-neighbor gates and have a high error threshold, but logical operations can be complex.
Color codes and other LDPC codes may need long-range interactions, which neutral atoms can provide through reconfigurable shuttling.
Mid-circuit measurement enables syndrome extraction and ancilla replacement without destroying the logical state.
Native multi-qubit gates can simplify logical circuits, especially for non-Clifford operations and magic-state distillation.
The first neutral-atom logical qubits are being demonstrated by integrating shuttling, readout, and high-fidelity gates.
Building a useful quantum computer requires stacking many error-corrected qubits and running deep logical circuits efficiently.
Deep circuits consist of thousands of logical gates, each requiring a physical implementation of CNOTs, Hadamards, and T gates.
Logical qubits are encoded in many physical qubits so that errors stay below the threshold throughout the computation.
Transversal gates are especially valuable because they do not spread errors between physical qubits within a code block.
Teleportation and lattice surgery are used to move logical information between code blocks and to perform multi-qubit logical operations.
The Steane code demonstrates many concepts of fault tolerance, including transversal Clifford gates and syndrome extraction.
Logical teleportation
T gates are usually implemented through magic-state distillation, consuming many ancilla qubits to produce high-fidelity magic states.
Concatenated codes and surface codes with different distances are chosen based on the logical error budget of the target algorithm.
Native multi-qubit gates in neutral atoms can reduce the depth of logical circuits compared to standard two-qubit decompositions.
A high logical clock rate is essential for practical algorithms; it depends on the physical gate speed and the decoding latency.
[[16,6,4]] code
The [[16,6,4]] code and other small block codes are studied as testbeds for transversal gates and decoding strategies.
Scaling beyond a few logical qubits requires careful management of ancilla consumption, routing, and classical feedforward.
Neutral-atom systems are aiming to demonstrate tens of logical qubits and deep logical circuits in the near term.
Curated videos explaining neutral-atom quantum computing from first principles to industrial roadmaps.
An introduction to how neutral atoms are trapped, manipulated, and used for quantum computing and simulation.
A clear explanation of Rydberg blockade and how it enables fast entangling gates between neutral atoms.
A technical overview of QuEra's neutral-atom platform and its analog-digital programming model.
A research talk on shuttling, mid-circuit readout, and error correction with neutral-atom arrays.
Companies and research labs around the world are advancing neutral-atom hardware, software stacks, and cloud access.
Analog and digital Rydberg arrays
Building Aquila and future logical qubits with a hybrid analog-digital stack.
Tweezer arrays and analog simulation
Scaling programmable neutral-atom arrays for optimization and simulation workloads.
Strontium nuclear-spin qubits
Using alkaline-earth atoms with long coherence and nuclear-qubit readout.
Rydberg science and quantum simulation
Academic leader in Rydberg blockade, quantum many-body physics, and error correction.
Laser systems for atom cooling
Supplying integrated laser and control systems for neutral-atom experiments.
Cold-atom and neutral-atom platforms
Developing quantum sensing and computing systems with neutral atom technology.
German neutral-atom research
Advancing Rydberg-based quantum computing and simulation at the national level.
Carbon nanotube integration
Exploring atom-scale qubits and integration paths for neutral-atom architectures.
Selected papers and reviews covering neutral-atom quantum computing, Rydberg gates, and error correction.
M. Saffman, T. G. Walker & K. Mølmer, Quantum information with Rydberg atoms, Rev. Mod. Phys. 82, 2313–2363 (2010).
H. Labuhn et al., Tunable two-dimensional arrays of single Rydberg atoms for realizing quantum Ising models, Nature 534, 667–670 (2016).
H. Weimer et al., Rydberg atoms: a natural quantum simulator, Nat. Phys. 6, 382–388 (2010).
M. Endres et al., Atom-by-atom assembly of defect-free one-dimensional cold atom arrays, Science 354, 1024–1027 (2016).
D. Barredo et al., Atom-by-atom assembly of defect-free one-dimensional cold atom arrays, Science 354, 1021–1023 (2016).
S. Ebadi et al., Quantum phases of matter on a 256-atom programmable quantum simulator, Nature 595, 227–232 (2021).
S. de Léséleuc et al., Single-atom addressing in tightly packed two-dimensional arrays of optically trapped neutral atoms, Phys. Rev. Lett. 120, 113603 (2018).
M. Morgado & D. Porras, Universal quantum computing with atomic bosonic qubits, New J. Phys. 23, 043041 (2021).
P. Scholl et al., Quantum simulation and computing with Rydberg-interacting qubits, Appl. Phys. Rev. 9, 021303 (2022).
D. Bluvstein et al., Logical quantum processor based on reconfigurable atom arrays, Nature 626, 58–65 (2023).
T. M. Graham et al., Demonstration of multi-qubit entanglement and algorithms on a programmable neutral atom quantum computer, Nature 604, 457–462 (2022).
S. J. Evered et al., High-fidelity parallel entangling gates on a neutral-atom quantum computer, Nature 622, 268–272 (2023).
I. Cong, H. Levine & N. P. de Leon, Hardware-efficient error-corrected quantum memories with neutral atoms, arXiv:2308.06348 (2023).
A. D. Challis, M. A. Perlin & D. Bluvstein, Computational architecture for fault-tolerant photonic quantum computing, Nat. Phys. 20, 85–92 (2024).
QuEra Computing, QuEra Aquila: a 256-qubit neutral-atom quantum computer, QuEra Technical Documentation (2024).
Pasqal, Pasqal QPU roadmap and neutral-atom analog quantum computing, Pasqal Technical Whitepaper (2024).
L. Henriet et al., Quantum computing with neutral atoms, Quantum 4, 327 (2020).
M. Saffman, Quantum computing with atomic qubits and Rydberg interactions: Progress and challenges, J. Phys. B: At. Mol. Opt. Phys. 49, 202001 (2016).
D. Bluvstein et al., Architectural mechanisms of a universal fault-tolerant quantum computer, arXiv:2506.20661 (2025).
H. Levine et al., High-fidelity control and entanglement of Rydberg-atom qubits, Phys. Rev. Lett. 121, 123603 (2018).
D. Jaksch et al., Fast quantum gates for neutral atoms, Phys. Rev. Lett. 85, 2208–2211 (2000).
J. Preskill, Quantum computing in the NISQ era and beyond, Quantum 2, 79 (2018).
D. Isenhower et al., Demonstration of a neutral atom controlled-NOT quantum gate, Phys. Rev. Lett. 104, 010503 (2010).
K. M. Maller et al., Rydberg-blockade controlled-not gate and entanglement in a two-atom ensemble, Phys. Rev. A 92, 022336 (2015).
E. Dennis, A. Kitaev, A. Landahl & J. Preskill, Topological quantum memory, J. Math. Phys. 43, 4452–4505 (2002).
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).
S. Bravyi & A. Kitaev, Universal quantum computation with ideal Clifford gates and noisy ancillas, Phys. Rev. A 71, 022316 (2005).
M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, Cambridge University Press (2000).
Each hardware platform makes different tradeoffs. See how neutral-atom qubits stack up against superconducting qubits, trapped ions, and photonic qubits.
Browse All Platforms