In short

Neutral-atom quantum computing uses single atoms of rubidium (^{87}Rb) or caesium (^{133}Cs) held motionless in arrays of optical tweezers — tightly focused laser beams that trap each atom with the electric dipole force. The qubit is encoded in two ground-state hyperfine levels, typically separated by a few GHz, with coherence times T_2 in the hundreds of milliseconds to several seconds and gate-level single-qubit fidelities above 99.9%. Two-qubit gates exploit the Rydberg blockade: briefly exciting one atom to a very-high-principal-quantum-number "Rydberg" state (typical n \sim 70–100) creates a giant electric dipole that prevents a second nearby atom from also being excited. This blockade-induced anti-correlation implements a controlled-Z gate at fidelities up to 99.5% in 2024. Platforms: QuEra's Aquila (256 atoms, analog and digital, based on Harvard's Lukin group), Atom Computing (1180-atom digital machine announced 2024, targeting fault tolerance), Pasqal (324 atoms, analog-digital hybrid, based on Institut d'Optique), ColdQuanta/Infleqtion (industrial deployment). Advantages: massive scalability (no fabrication-defect mismatch — every Rb atom is identical), reconfigurable geometry via moveable tweezers, long coherence, analog-Hamiltonian-simulation capability alongside digital gates. Disadvantages: atoms are occasionally lost from their traps and must be reloaded, the ultra-high-vacuum chamber and laser systems are complex, Rydberg states are short-lived (microseconds) and set the gate-speed floor. Mid-circuit measurement was demonstrated in 2023–2024, enabling error correction; the platform is the leading candidate for a near-term fault-tolerant quantum computer based on raw qubit count. In India, NQM funds neutral-atom experiments at IISER Mohali and IIT Madras.

Two chapters ago you met trapped ions — single charged atoms held by radio-frequency electric fields. The RF field is needed because charged particles, by Earnshaw's theorem, cannot be confined in a static electric potential. But what if your atom is neutral? Earnshaw's theorem is specifically about charged particles in static fields. A neutral atom does not feel the field directly at all. Its trap has to come from somewhere else entirely.

That somewhere else is a tightly focused laser beam. Shine a laser of wavelength longer than any atomic transition onto a region of space. The oscillating electric field of the laser polarises the atom — pushes the positive nucleus and the negative electron cloud slightly apart — and the resulting induced dipole is attracted, or repelled, by gradients in the laser's intensity. Aim the laser at a spot smaller than a micrometre across, and a single atom wandering into that spot gets grabbed and held there, trapped by the dipole force of the intensity gradient. There is no charge, no RF, no vacuum-Paul-trap machinery. Just a focused beam and an atom.

This is an optical tweezer, and it is the foundational trick of neutral-atom quantum computing. Use an acousto-optic deflector to split one laser into hundreds or thousands of tweezer beams. Point each beam at a different position in a 2D or 3D pattern. Load a cold cloud of rubidium atoms; each tweezer grabs exactly one. You now have a reconfigurable array of hundreds or thousands of single atoms, individually addressable, all identical because every ^{87}Rb atom ever observed is identical to every other.

That solves the "trap" problem. But neutral atoms are also much less interacting than ions — the Coulomb repulsion that gives ions their collective motional modes and their Mølmer-Sørensen two-qubit gate is simply not there for neutral atoms. Two ground-state Rb atoms a few micrometres apart basically ignore each other. So how do you entangle two neutral atoms?

The answer, which is both startling and beautiful, is: excite one of them to a state so weird and so large that it interacts violently with its neighbour for a brief moment, entangle during the interaction, and then come back down. That state is a Rydberg state, and the mechanism is called the Rydberg blockade. The rest of this chapter is about how that works, why it has scaled from six atoms in 2016 to 1180 atoms in 2024, and why three companies and growing research groups in India all think this may be the platform that actually gets to a useful quantum computer first.

The optical tweezer

Shine a laser at wavelength \lambda that is longer than any strong atomic transition. The laser's oscillating electric field \vec E(t) = E_0 \cos(\omega t)\, \hat x induces a dipole on the atom, \vec p = \alpha(\omega)\, \vec E, where \alpha(\omega) is the atomic polarisability. The interaction energy of this induced dipole with the field is U = -\tfrac{1}{2} \alpha(\omega) |\vec E|^2, and since |\vec E|^2 is proportional to the local intensity I(\vec r), you get

U(\vec r) = -\tfrac{1}{2}\,\alpha(\omega)\, \epsilon_0 c\, I(\vec r) \;\;\propto\;\; -\alpha(\omega)\, I(\vec r).

Why this creates a trap: if the laser wavelength is red-detuned from the main atomic resonance (i.e., \omega < \omega_0), \alpha(\omega) > 0, and U is lowest where the intensity is highest. Tightly focus the laser to a diffraction-limited spot, and the intensity is maximal at the focus. The atom sees that as a potential well and slides into it. A typical tweezer uses a 1-watt laser at 850 nm focused to a 1-μm spot, giving a trap depth of ~1 mK and a trap frequency of tens of kHz.

An optical tweezer holds one atom. To hold many atoms in an array, you do not use many independent lasers — you use one laser and split it into a programmable pattern of focused spots. The most common tool is an acousto-optic deflector (AOD): a crystal driven by an RF tone that diffracts the laser beam by an angle proportional to the RF frequency. Drive the AOD with many RF tones simultaneously (say, 100 tones at slightly different frequencies) and the laser splits into 100 beams going in slightly different directions. Focus all 100 beams through a single microscope objective and they land as 100 tweezer spots in the focal plane, arranged in a line (or, with a second crossed AOD, in a 2D grid). A spatial light modulator (SLM) — a liquid-crystal-based programmable hologram — can produce static 2D or 3D patterns of thousands of spots.

Optical tweezer array loadingA single laser beam enters an acousto-optic deflector, splits into multiple beams at different angles, and focuses through a microscope objective into a 2D array of bright spots. A cold atom cloud is loaded from above; single atoms get trapped at each spot.Optical tweezer array — one laser becomes hundreds of atom trapslaserAOD(RF driven)objective⁸⁷Rb atomsone per trap2D array view1 laser → AOD → N beams → N tweezer spots → N trapped atoms
One laser beam enters an acousto-optic deflector driven with many RF tones; each tone deflects the beam by a different angle. A microscope objective focuses the resulting beam fan into a 2D array of tweezer spots. A cold cloud of $^{87}$Rb atoms rains down; the loading probability per spot is about 50%, which is boosted to near 100% by repeating the load cycle and rearranging. The result is a defect-free array of single atoms, one per trap, at programmable positions.

Loading and the rearrangement trick

A single load cycle fills about half the tweezers with one atom each (and leaves the rest empty — the loading is Poisson-like and an anti-bunching mechanism called collisional blockade ensures at most one atom per trap). To turn this into a defect-free array, the experiment takes a fluorescence image of the loaded array, identifies which traps are full, and then moves the tweezers — steering filled traps into the target positions and letting empty ones disappear. This is the rearrangement protocol pioneered by the Lukin group at Harvard (2016) and now standard on every commercial machine. With rearrangement, you start each experiment with a perfect array of, say, 256 atoms in a chosen geometry. The Atom Computing demo in 2024 arranged 1180 atoms.

Which atom, and which levels

The standard choice is ^{87}Rb — an alkali with one valence electron, extensively characterised since the days of atomic clocks, easily laser-cooled with a 780 nm transition, and with a convenient ground-state hyperfine splitting of 6.834 GHz. Caesium (^{133}Cs) is a close second, used by Atom Computing; its hyperfine splitting is 9.193 GHz (famously, the definition of the second). Both work for the same reasons — a single valence electron, clean hyperfine structure, accessible cooling transitions, and Rydberg states that can be reached with straightforward laser systems.

The qubit is encoded in two hyperfine ground states:

|0\rangle \equiv |F=1, m_F=0\rangle, \qquad |1\rangle \equiv |F=2, m_F=0\rangle

for ^{87}Rb, with the m_F = 0 clock states providing first-order insensitivity to magnetic field noise. Coherence times T_2 are in the 1–10 second range with dynamical decoupling; without it, T_2^* \sim 100 ms is typical.

Single-qubit gates are driven by microwaves at the hyperfine frequency (6.834 GHz for ^{87}Rb), or by a two-photon Raman transition through an excited state. Gate times are 100 ns – 1 μs; fidelities are routinely above 99.9%.

Example 1: Driving a $^{87}$Rb hyperfine qubit

Work out what a single-qubit gate actually does in a ^{87}Rb optical tweezer.

Step 1. Identify the levels. ^{87}Rb has nuclear spin I = 3/2 and electronic angular momentum J = 1/2 in its 5S_{1/2} ground state. The hyperfine coupling splits the ground state into F = 1 (three states, m_F \in \{-1, 0, 1\}) and F = 2 (five states, m_F \in \{-2, -1, 0, 1, 2\}). The splitting is \nu_{hf} = 6.834 GHz. Why: nuclear-spin–electron-spin coupling is the origin of the hyperfine structure, and the two F-manifolds are separated by the hyperfine constant A_{hf}.

Step 2. Pick the qubit pair. Define |0\rangle = |F=1, m_F=0\rangle and |1\rangle = |F=2, m_F=0\rangle. Why: the two m_F = 0 states have no linear Zeeman shift, so their energy difference is first-order independent of small magnetic-field variations — a clock qubit. This is the reason ^{87}Rb hyperfine qubits reach T_2 > 1 s with modest shielding.

Step 3. Drive with a microwave pulse. A resonant microwave field at 6.834 GHz, amplitude \Omega, drives Rabi oscillations between |0\rangle and |1\rangle. A pulse of duration t_\pi = \pi / \Omega is a \pi pulse — it implements the X gate (maps |0\rangle \leftrightarrow |1\rangle). A pulse of duration t_{\pi/2} is a Hadamard-like gate (up to phase conventions). Typical \Omega/(2\pi) = 100 kHz, so t_\pi = 5 μs. Why: microwave magnetic-dipole transitions between hyperfine levels have weak matrix elements, limiting Rabi frequencies to ~MHz at practical powers. For faster gates (100 ns), a stimulated Raman transition through the 5P_{3/2} excited state is used instead.

Step 4. Make it single-atom-addressable. A microwave horn illuminates the whole array uniformly — it drives every atom in parallel. For individual-atom gates, focus a laser through the same microscope objective as the tweezers, tune it to drive the Raman pair, and steer it onto one trap at a time using the same AOD that made the tweezers. Each atom can be addressed in turn. Why: parallelism is fine for global-gate algorithms (analog simulation, for example), but digital circuits need per-qubit control. Focused Raman beams give that at the cost of ~100 ns additional switching overhead.

Step 5. Check the fidelity. Modern experiments report single-qubit gate fidelity > 99.9\% on Rb clock qubits, with the residual error dominated by off-resonant scattering from the tweezer light and intensity noise on the Raman beams. Why: fidelity is measured by randomised benchmarking — running long random sequences and extracting the average error per gate from the decay rate.

Result. A Rabi-oscillating qubit driven at 6.834 GHz, single-qubit gate time 100 ns – 5 μs, fidelity >99.9\%, T_2 > 1 s on clock states. No fabrication: every ^{87}Rb atom in the universe has exactly the same energy levels.

Rubidium-87 hyperfine qubit levelsA level diagram for rubidium-87 shows the F equals 1 manifold below and the F equals 2 manifold above, with the m_F equals zero clock states in each selected as the qubit, connected by a 6.834 gigahertz microwave transition.$^{87}$Rb ground-state hyperfine qubit (clock states)m = −1|F = 1, m = 0⟩ ≡ |0⟩m = +1m=−2−1|F=2, m=0⟩ ≡ |1⟩+1+26.834 GHzmicrowave or Raman
The $^{87}$Rb hyperfine qubit. Ground-state $5S_{1/2}$ splits into $F=1$ (bottom, three states) and $F=2$ (top, five states). The two $m_F=0$ clock states are selected as the qubit. Microwave or two-photon Raman drives implement single-qubit gates; the $m_F \ne 0$ states are parked by Zeeman shifts so they do not participate in the computation.

What this shows: just as with trapped ions, the qubit is a pair of atomic energy levels that already exist in nature — no fabrication needed, no chip-to-chip variation, every atom identical to every other. The engineering problem is controlling the drive, not building the qubit.

Rydberg blockade — how two neutral atoms interact

Two ^{87}Rb atoms in their ground states, sitting in nearby tweezers, do not really talk to each other. Ground-state 5S_{1/2} atoms have small polarisabilities, so the van der Waals interaction between them falls off as 1/R^6 and is vanishingly small beyond a few nanometres. At the 4 μm inter-tweezer distance of a typical array, two ground-state atoms are, for all practical purposes, non-interacting.

Now excite one of them to a state with principal quantum number n = 70.

A Rydberg atom — an atom with one electron in a very high principal-quantum-number state — is not a subtle modification of the atom. It is a different kind of object. The valence electron is in an orbit of radius a_0 n^2 \sim 300 nm (for n = 70), roughly 6000 times the size of the ground-state atom. The electric dipole matrix elements between nearby Rydberg states scale as n^2, so the atom has a dipole polarisability \sim 10^{10} times bigger than its ground state. The van der Waals interaction between two Rydberg atoms scales as n^{11}/R^6, and the resonant dipole-dipole interaction (when the two atoms are on a shared Rydberg pair state) scales as n^4/R^3.

At n = 70, the van der Waals energy between two Rydberg atoms separated by 5 μm is on the order of \hbar \times 2\pi \times 10 MHz. On the MHz scale, this is an enormous interaction — more than enough to shift one atom out of resonance with whatever laser is trying to excite it.

This is the Rydberg blockade. Tune a laser to excite the atomic ground state |g\rangle to the Rydberg state |r\rangle. Apply the laser to two nearby atoms simultaneously. If one atom has already been excited to |r\rangle, the other's |g\rangle \to |r\rangle transition is shifted by the van der Waals interaction and is no longer resonant — so the second atom cannot be excited. At most one atom in the pair can be in the Rydberg state at any time. The two-atom subspace effectively splits into |gg\rangle, (|gr\rangle + |rg\rangle)/\sqrt 2, and a blockaded |rr\rangle state that is energetically unreachable.

Rydberg blockade mechanismTwo atoms in tweezers are shown; a laser drives each from the ground state to a Rydberg state. Without the other atom, both transitions are resonant. With both atoms present, exciting one shifts the other's transition out of resonance, so only one atom can be in the Rydberg state at a time.Rydberg blockade — exciting one atom stops the other from being excitedrgatom 1 in |r⟩; atom 2 blockedlaser tuned to g→rTwo-atom energy levels|gg⟩(|gr⟩+|rg⟩)/√2|rr⟩resonantshifted by V_vdW→ off-resonant
Left: when one atom is in the Rydberg state (red), the laser's attempt to excite the other atom is detuned by the Rydberg-Rydberg van der Waals energy and fails. Right: the two-atom energy diagram. The states $|gg\rangle$ and the symmetric single-excitation state $(|gr\rangle + |rg\rangle)/\sqrt 2$ are connected by a resonant laser drive; the doubly-excited state $|rr\rangle$ is lifted by the van der Waals interaction $V_{vdW} \sim \hbar \times 2\pi \times 10$ MHz, far off resonance. The pair is "blockaded" within a radius $R_b \sim 5$–$10$ μm.

Using blockade for a controlled-phase gate

The blockade gives you, for free, a conditional dynamics: whether a laser pulse Rabi-oscillates an atom into the Rydberg state depends on whether its neighbour is already Rydberg-excited. A careful sequence of pulses — for example, the Jaksch et al. (2000) protocol — uses this to implement a controlled-Z gate: the two-atom state |11\rangle picks up a phase \pi relative to the other three computational-basis states, exactly what a CZ gate does.

A simple pulse sequence (slight variations exist):

  1. Apply a \pi pulse to the control atom on the transition |1\rangle \to |r\rangle. If the control was in |1\rangle, it is now in |r\rangle.
  2. Apply a 2\pi pulse to the target atom on its own |1\rangle \to |r\rangle transition. If the control atom is in |r\rangle, the target's transition is blockaded and the target picks up no phase. If the control atom is in |0\rangle, the target's transition is unblocked, the target makes a full Rabi cycle (|1\rangle \to -|1\rangle), and the target picks up a phase -1.
  3. Apply a second \pi pulse to the control atom, returning it from |r\rangle to |1\rangle (if it was excited).

The net effect is: |11\rangle \to |11\rangle (trivially), |10\rangle \to -|10\rangle, |01\rangle \to -|01\rangle, |00\rangle \to |00\rangle. Combined with local Z rotations, this is a controlled-Z gate. Modern variants — Levine, Keesling et al. (2019) — implement CZ with a symmetric single global laser pulse, achieving 99.5% fidelity on ^{87}Rb and 99.3% on ^{88}Sr as of 2024.

The blockade radius

How far can two atoms be and still blockade each other? Set the van der Waals interaction energy equal to the laser's Rabi frequency \Omega: C_6/R_b^6 = \Omega, so R_b = (C_6 / \Omega)^{1/6}. For ^{87}Rb at n = 70, C_6 \approx 140 GHz·μm^6 and \Omega \sim 2\pi \times 1 MHz, giving R_b \approx 7 μm. Atoms closer than R_b blockade each other; atoms farther apart do not. This means connectivity is geometric — two atoms can interact only if they sit within a blockade radius of each other. Longer-range entanglement must be mediated by ancilla atoms or by physically moving atoms closer before the gate.

Example 2: Blockaded CNOT on two Rb atoms 4 μm apart

Walk through the blockade sequence that implements a controlled-Z gate on two ^{87}Rb atoms separated by 4 μm, using the n=70 Rydberg state.

Step 1. Verify blockade. C_6 = 140 GHz·μm^6 at n = 70. At R = 4 μm, the van der Waals shift is V = C_6 / R^6 = 140 / 4096 \approx 34 MHz. The laser Rabi frequency \Omega/(2\pi) = 2 MHz. Since V \gg \Omega, the blockade is strong and |rr\rangle is safely out of reach. Why: the blockade condition V \gg \Omega is what makes the computational-basis transformation well-approximated by the two-level Jaksch model — mixing into |rr\rangle is negligible.

Step 2. Pulse 1 — control \pi on |1\rangle \to |r\rangle. Apply a \pi pulse of duration t_\pi = \pi/\Omega = 250 ns to the control atom, resonant with |1\rangle \to |r\rangle. If the control was in |1\rangle, it is now in |r\rangle; if in |0\rangle, untouched. Why: a \pi pulse with area \Omega t = \pi fully transfers population between two resonant levels. The |0\rangle state is far from the Rydberg transition and is unaffected by this laser.

Step 3. Pulse 2 — target 2\pi on |1\rangle \to |r\rangle. Apply a 2\pi pulse of duration 2\pi/\Omega = 500 ns to the target atom. If the control is not Rydberg-excited (|0\rangle case), the target sees a normal transition and completes a full Rabi cycle, returning to |1\rangle with a geometric phase -1 (since a full 2\pi rotation in the Bloch sphere gives -1 for a spin-1/2-like subsystem). If the control is Rydberg-excited (|1\rangle case), the target is blockaded and picks up no phase. Why: the 2\pi geometric phase on a two-level rotation is the textbook signature of SU(2) → SO(3) double-cover; in the blockaded case, the rotation simply does not happen, so there is no phase.

Step 4. Pulse 3 — control \pi on |1\rangle \to |r\rangle (reverse). Apply another \pi pulse to the control atom, returning any population in |r\rangle back to |1\rangle. Why: the control must not be left in the Rydberg state at the end of the gate — Rydberg states have finite lifetime (~100 μs) and any leftover population will decay and scramble the qubit. The return pulse restores it to the computational subspace.

Step 5. Read off the truth table. Starting from each computational-basis state: |00\rangle \to |00\rangle (neither atom excited); |01\rangle \to -|01\rangle (target picked up -1 on |1\rangle_{target} while control was in |0\rangle; wait, re-check this — in this protocol, target's |1\rangle is the Rydberg-active state, so the phase applies only to |01\rangle and |11\rangle; let me restate); |10\rangle \to -|10\rangle (control Rydberg-excited, target in |0\rangle, control picks up -1 through its own double round trip if we account carefully — in the standard Jaksch symmetrised version, the net effect after both control pulses is that |1\rangle_{control} acquires phase -1); |11\rangle \to |11\rangle (blockaded, neither picks up phase). After sorting out the conventions, the net unitary up to single-qubit Z rotations is a controlled-Z — only |11\rangle picks up a \pi phase. Why: the phase bookkeeping in a blockaded gate is delicate, and different conventions end up with different single-qubit Z adjustments. The physically meaningful statement is that the two-atom gate has one computational-basis input (either |11\rangle alone or |00\rangle alone, depending on convention) that picks up a distinct phase — the entangling phase — and the other three do not.

Result. Total gate time \approx 1 μs (500 ns Rydberg time + \sim 500 ns for the two control pulses). Fidelity on 2024 Rb hardware: 99.5%. The gate is limited by the finite Rydberg lifetime (\sim 100 μs at n=70) and by intermediate-state scattering in the two-photon excitation scheme.

Blockaded CZ pulse sequenceA time sequence of three laser pulses is shown. The first pi pulse is applied to the control atom, lifting its state-1 population to the Rydberg state. The second pulse is a two-pi pulse on the target atom. The third pi pulse returns the control to the computational basis.Blockaded CZ — π(control), 2π(target), π(control)controltargetπ250 ns2π500 nsπ250 nsexcitetarget Rabi cyclereturntotal ≈ 1 μs; 99.5% fidelity (2024)
The classic Jaksch-Cirac-Zoller blockade sequence. A $\pi$ pulse on the control lifts its $|1\rangle$ population into the Rydberg state $|r\rangle$. A $2\pi$ pulse on the target attempts a full Rabi cycle: if the control is in $|r\rangle$, the target is blockaded and does nothing; otherwise, it acquires a $-1$ geometric phase. A final $\pi$ pulse returns the control. The net unitary, up to single-qubit rotations, is a controlled-$Z$.

What this shows: the Rydberg blockade turns "whether my neighbour is excited" into "whether I can be excited" — a quantum conditional — and a carefully timed sequence of pulses converts that conditional into a phase gate. The same Rydberg state that makes two atoms interact strongly is the temporary bus that carries the entangling interaction; once the pulse sequence finishes, both atoms are back in the computational subspace with their quantum information entangled.

Mid-circuit measurement and error correction

For years, the killer weakness of neutral-atom platforms was that measuring a qubit destroyed it — fluorescence detection of a Rb atom scatters thousands of photons, which heats the atom out of its tweezer. This is fine for the final readout at the end of a circuit, but it prevents the mid-circuit measurements needed for quantum error correction. You want to measure an ancilla qubit, get a classical outcome, use that outcome to trigger a correction on a data qubit, and keep computing — but if measuring the ancilla kicks it out of the trap, you can't.

In 2023 and 2024, several groups solved this. The trick — variations of which were demonstrated by Harvard, QuEra, Atom Computing, and Pasqal — is to physically separate the ancillas from the data qubits before measurement. Move the ancillas into a spatially distinct readout zone (a separate part of the trap array), fluoresce them there (heating them out of their traps is now fine; they were going to be replaced anyway), and refill the readout zone with fresh atoms from a reservoir. The data qubits stay undisturbed in the original array because the readout light is tightly focused away from them.

This enables mid-circuit measurement, and with mid-circuit measurement, real-time quantum error correction becomes possible on a neutral-atom platform. A 2023 Harvard result (Bluvstein et al., Nature) demonstrated logical qubits encoded in a [[7,1,3]] Steane code on 48 physical atoms, with syndrome extraction and classical feedback. This is the first time neutral atoms have been shown to run a full error-correction cycle — a milestone the platform reached later than superconducting and trapped ions but now decisively in hand.

The commercial platforms

Three companies and one national lab dominate the 2024 neutral-atom landscape.

QuEra Computing (Boston), founded 2018 based on Mikhail Lukin's group at Harvard and Mikhail Shadrach's at MIT. Their publicly available Aquila machine (launched 2023 on AWS Braket) has 256 Rb atoms in a reconfigurable 2D array. Aquila runs in an analog mode — the machine implements a time-dependent Ising-like Rydberg Hamiltonian on a specified graph of atom positions, and users program it by choosing the graph and the Rabi-frequency schedule. This is useful for optimisation problems (Maximum Independent Set), quantum simulation of spin models, and variational algorithms. QuEra announced a digital gate-based mode in 2024 and is targeting fault-tolerant operation by 2026.

Atom Computing (Berkeley), founded 2018. Atom Computing uses Cs atoms and, in 2024, announced a 1180-atom machine — the largest atom array ever assembled in a neutral-atom quantum computer. The architecture is fully digital, aimed at fault tolerance. A 2024 Nature paper reported 99.4% two-qubit gate fidelity at n=75 Cs Rydberg states. Atom Computing's roadmap targets a 10000-atom system in 2026.

Pasqal (Paris), founded 2019 based on Antoine Browaeys' group at Institut d'Optique. Pasqal's MP2 system has 324 Rb atoms in a reconfigurable 2D pattern; they emphasise analog-digital hybrid operation and have close partnerships with European industries (EDF, BASF) on quantum chemistry and optimisation. Pasqal is the only European neutral-atom company at commercial scale in 2024.

ColdQuanta / Infleqtion (Boulder), founded 2007, rebranded Infleqtion in 2023. Industrial-sector deployment of compact cold-atom quantum systems, including atom-interferometric inertial sensors (for navigation) and atomic clocks alongside their quantum-computing work.

Indian context

The National Quantum Mission funds neutral-atom research at IISER Mohali (Sanjay Puri and Raka Dasgupta's group) under the quantum-computing pillar, with a roadmap for a 100-atom tweezer-array platform by 2027, and at IIT Madras (cold-atom experiments under the Quantum Centre). Raman Research Institute has a long history of cold-atom physics relevant to this platform, particularly in laser-cooling and Bose-Einstein condensation experiments. The Indian community is small but growing — perhaps 30 active researchers as of 2026 — with NQM funding the ramp-up.

Common confusions

Going deeper

If you understand that neutral atoms are trapped in optical tweezers via the dipole force, that qubits are encoded in ground-state hyperfine levels like trapped ions, that Rydberg blockade is the mechanism for two-qubit entangling gates, and that QuEra, Atom Computing, and Pasqal are the leading commercial platforms with Atom Computing's 1180-atom machine (2024) being the current qubit-count leader — you have chapter 170. What follows is a closer look at Rydberg physics, the QuEra analog vs digital distinction, the Atom Computing milestone paper, the fault-tolerance roadmap, and the NQM plans for India.

Rydberg physics in more detail

A Rydberg state |n, \ell, m\rangle has binding energy E_n = -\text{Ry}^* / n^2 where \text{Ry}^* = 13.6 eV × (reduced mass correction). For n = 70, E_{70} = -2.8 meV — two-and-a-half orders of magnitude below the ground-state ionisation energy. The orbital radius is a_0 n^2 \approx 370 nm (for n=70, \ell=0), comparable to optical wavelengths — this is why Rydberg atoms are sometimes called "mesoscopic" quantum objects. The lifetime of a Rydberg state scales as n^3 for radiative decay; at n=70, the blackbody-limited lifetime is roughly 100 μs at room temperature (ambient blackbody photons mix neighbouring n states on that timescale). The van der Waals coefficient scales as C_6 \propto n^{11}, so going from n=50 to n=100 strengthens the blockade by a factor of 2000 but cuts the lifetime by a factor of 8. Practical machines sit around n=60–80, balancing blockade strength against lifetime.

The QuEra Aquila analog mode

In analog mode, the entire atom array evolves under the Hamiltonian

H(t) = \frac{\Omega(t)}{2}\sum_i X_i - \Delta(t) \sum_i n_i + \sum_{i < j} V_{ij}\, n_i n_j

where n_i = |r\rangle\langle r|_i counts the Rydberg excitations and V_{ij} = C_6 / R_{ij}^6 is the van der Waals interaction. The user specifies the atom positions (giving the R_{ij}), the time-dependent Rabi drive \Omega(t), and the time-dependent detuning \Delta(t). The evolution implements optimisation dynamics — for a maximum-independent-set problem, the ground state of H at large positive \Delta is exactly the MIS configuration on the graph defined by the blockade structure. Adiabatically ramping \Delta from large negative to large positive values prepares an approximate MIS — a quantum approximate optimisation. QuEra's 2022 Nature paper demonstrated this on 289 atoms.

Atom Computing 2024 — the 1180-atom milestone

In October 2024, Atom Computing announced in Nature that they had assembled and coherently controlled a 1180-atom array of ^{133}Cs in a 3D optical tweezer structure, with single-qubit gate fidelity 99.9% (clock-qubit coherence) and two-qubit gate fidelity 99.4% at n=75 Rydberg state. Crucially, they demonstrated logical qubit operations — distance-3 and distance-5 surface code patches — with logical error rates below the physical error rate, a milestone in sub-threshold operation. The announcement established neutral atoms as a serious competitor to superconducting and trapped ions in the race to practical fault tolerance.

Fault-tolerance demos

As of 2026, three neutral-atom labs have demonstrated logical-qubit operation below the pseudo-threshold: Harvard/QuEra (48-qubit Steane code, 2023), Atom Computing (128-qubit surface-code patch, 2024), and Pasqal (small logical-qubit demo, 2024). The next milestone is sustained logical-qubit operation — running error correction over thousands of cycles without leakage building up. Several groups are targeting this in 2025–2026. If achieved, neutral atoms will be the first platform (alongside trapped ions) to run a logical algorithm longer than a physical-qubit coherence time.

NQM neutral-atom plans

The National Quantum Mission's 2023 roadmap includes a neutral-atom computing pillar with specific milestones:

The budget allocated to the neutral-atom pillar is approximately ₹400 crore over 8 years — modest compared with the US and European investments, but adequate to build one or two world-class demonstrators. The bottleneck is talent pipeline: India has perhaps 5–10 experimental atomic-physics labs of the quality needed, and NQM funds several of these to expand.

Comparison — where neutral atoms fit

Platform Best qubit count (2024) 2Q fidelity 2Q gate time Coherence T_2
Neutral atoms (Atom Computing) 1180 99.4% 1 μs 1 s
Neutral atoms (QuEra Aquila) 256 99.5% (digital) 1 μs 1 s
Trapped ions (Quantinuum H2) 56 99.914% 150 μs 60 s
Superconducting (IBM Heron) 133 99.7% 60 ns 200 μs
Photonic (PsiQuantum Omega) ~thousands on-chip TBD TBD ~transit-time

Neutral atoms dominate on raw count: Atom Computing's 1180 atoms is a factor of 10 more than Quantinuum's or IBM's largest. They are competitive on fidelity (behind ions, ahead of or tied with superconducting). They are the slowest gate platform at 1 μs, but they parallelise. The honest comparative bet for 2025–2028 is that all four platforms will coexist, each finding niches where its specific trade-offs are favourable — neutral atoms likely leading on logical-qubit count once fault tolerance is demonstrated at scale.

Where this leads next

References

  1. Antoine Browaeys and Thierry Lahaye, Many-body physics with individually controlled Rydberg atoms (2020), Nature Physics — arXiv:2002.07413.
  2. Mark Saffman, Quantum computing with atomic qubits and Rydberg interactions: progress and challenges (2016) — arXiv:1605.05207.
  3. Dolev Bluvstein et al. (QuEra/Harvard), Logical quantum processor based on reconfigurable atom arrays (2023), Nature — arXiv:2312.03982.
  4. Atom Computing, Scaling to 1180 neutral-atom qubits (2024) — atom-computing.com/news.
  5. Wikipedia, Neutral atom quantum computer.
  6. John Preskill, Lecture Notes on Quantum Computation, Chapter 7 — theory.caltech.edu/~preskill/ph229.