Quantum Computing Modalities: Neutral Atom (Rydberg)
Table of Contents
Updated May 2026
(For other quantum computing modalities and architectures, see Taxonomy of Quantum Computing: Modalities & Architectures)
What It Is
Neutral-atom quantum computing uses uncharged atoms, typically rubidium-87 (⁸⁷Rb) or strontium-87 (87Sr), trapped individually in arrays of focused laser beams called optical tweezers. Each atom’s internal quantum state (two hyperfine ground-state energy levels) encodes |0⟩ and |1⟩. To entangle two atoms, a precisely timed laser pulse promotes them to a Rydberg state, a highly excited electronic configuration with an enormous electric dipole moment. At that energy level, neighboring atoms interact so strongly that only one can be excited at a time, an effect called Rydberg blockade. That conditional behavior is the basis for two-qubit gates.
The combination of properties is unusual among quantum computing modalities. Neutral atoms operate at room temperature (the vacuum chamber, optics, and control systems are all at ambient conditions, with no dilution refrigerator or cryogenic infrastructure required). They scale to large qubit counts (over 1,000 atoms trapped in a single array has been demonstrated, and Caltech achieved 6,100 atoms in an optical tweezer array in September 2025). Their connectivity is reconfigurable: optical tweezers can rearrange atom positions between circuit layers, enabling non-planar code layouts without fixed hardware routing. And they have produced the most verified logical qubits of any platform: QuEra / Harvard / MIT demonstrated 96 error-corrected logical qubits on a 448-atom processor in November 2025, published in Nature.
From a CRQC perspective, neutral atoms present a distinctive risk profile. Their gate speeds sit between trapped ions (microseconds to milliseconds) and superconducting qubits (nanoseconds). Their QEC efficiency rivals or exceeds trapped ions. And their room-temperature operation means they can be deployed in standard data center racks without the cryogenic infrastructure that constrains superconducting scaling. If neutral-atom systems can close the remaining fidelity gap with trapped ions, they become a serious contender for the first modality to reach fault-tolerant scale.
How It Works
Trapping and Array Assembly
Neutral atoms are first collected and cooled in a magneto-optical trap (MOT), which uses a combination of magnetic fields and counter-propagating laser beams to slow the atoms from thermal velocities (hundreds of meters per second) to microkelvin temperatures. From the MOT, atoms are loaded into optical tweezers: tightly focused laser beams that create microscopic potential wells through the optical dipole force. Each tweezer holds at most one atom, enforced by light-assisted collisions that eject atom pairs until a single atom remains.
Laser systems for trapping. The trapping laser must be far-detuned from atomic resonances (to avoid scattering photons that would heat the atoms) and tightly focusable (to create deep, sub-micrometer potential wells). For ⁸⁷Rb, the standard choice is 1064 nm (Nd:YAG), which is far red-detuned from the 780 nm D2 line and available at high power (10–100 W) from commercial fiber lasers. For 87Sr, 515 nm light is commonly used (green, from frequency-doubled 1030 nm fiber lasers), chosen to work with strontium’s different level structure. The trap depth is proportional to laser intensity and inversely proportional to detuning, so high-power, well-focused beams are essential for reliable atom retention during gate operations.
SLM vs. AOD for array generation. Two technologies compete for generating the tweezer array:
A spatial light modulator (SLM) is a liquid crystal device that shapes the phase of a single laser beam, producing an arbitrary 2D (or 3D) pattern of focal spots via a computed hologram. SLMs can generate hundreds to thousands of tweezer sites in arbitrary geometries (grids, triangular lattices, rings, application-specific topologies), which is valuable for mapping specific QEC code layouts onto the atom array. The tradeoff: SLMs update slowly (30–60 Hz typical refresh rate), so they cannot dynamically rearrange atoms at the speeds needed for mid-circuit qubit shuttling.
Acousto-optic deflectors (AODs) use acoustic waves in a crystal to steer a laser beam’s direction at MHz rates. By driving the AOD with multiple radio frequencies simultaneously, multiple tweezer spots can be generated along one axis. Crossed AODs (one for each axis) produce a 2D array. AODs are fast enough to shuttle atoms between positions in microseconds, making them essential for the Dynamic Qubit Array architecture and for the sort-and-fill procedure during array loading. The tradeoff: AODs produce rectangular grids with limited geometric flexibility compared to SLMs, and the number of simultaneous spots is limited by the AOD’s bandwidth.
In practice, most systems use both: an SLM to define the static array geometry, and AODs to perform dynamic rearrangement and qubit shuttling within that geometry. The laser suppliers for these systems are primarily Toptica Photonics (Germany) and M Squared Lasers (UK) for the Rydberg excitation lasers, and IPG Photonics or NKT Photonics for high-power trapping sources.
The initial loading is probabilistic: each tweezer site fills with roughly 50% probability. The critical engineering innovation that made large-scale arrays practical was developed by the Harvard group (Endres et al., 2016): real-time rearrangement. After loading, a camera image identifies which sites are occupied. AOD-driven tweezers pick up atoms from random positions and move them into a target defect-free configuration. This “sort-and-fill” procedure produces arrays of 100–1,000+ atoms in arbitrary 2D (or 3D) geometries with near-unity filling.
Atom Computing demonstrated 1,180 atoms trapped in a 1,225-site array (October 2023), and Caltech achieved a 6,100-atom array in September 2025. Harvard and MIT demonstrated continuous operation of a 3,000-atom system for over two hours by replenishing lost atoms mid-computation (September 2025), solving the atom-loss problem that had been the modality’s most cited scaling bottleneck.
Rydberg Excitation Paths
Two approaches exist for promoting atoms from the ground state to a Rydberg state, and the choice has practical consequences:
Direct UV excitation. A single laser at ~297 nm (for ⁸⁷Rb) drives the ground-to-Rydberg transition directly. This is conceptually simple but technically demanding: 297 nm light requires frequency-quadrupled lasers, UV-compatible optics, and careful management of photoionization (UV photons can ionize the atom if the laser intensity is too high or the atom is in the wrong state).
Two-photon excitation. Two lasers, typically at 420 nm (blue) and 1013 nm (infrared) for rubidium, combine to drive the Rydberg transition via a virtual intermediate state. This avoids the need for deep-UV light and uses laser wavelengths where commercial sources are more mature and powerful. The two-photon approach also allows more flexible pulse shaping and better suppression of off-resonant excitation. Most current commercial systems (QuEra, Pasqal) use two-photon Rydberg excitation. The suppliers for the excitation lasers are primarily Toptica (tunable diode lasers) and M Squared (Ti:Sapphire and OPO systems for high-power, narrow-linewidth output).
Qubit Encoding and Single-Qubit Gates
Qubits are encoded in two hyperfine ground states of the atom. For ⁸⁷Rb, these are typically the “clock states” |F=1, mF=0⟩ and |F=2, mF=0⟩, separated by 6.8 GHz. These are the same transitions used in atomic clocks, which is why they are extraordinarily stable and long-lived. Coherence times of 40 seconds have been demonstrated on Atom Computing‘s strontium-87 platform, and seconds-scale coherence is routine on rubidium.
Single-qubit gates are performed by driving the hyperfine transition with microwave pulses (global addressing, same rotation on all qubits) or stimulated Raman transitions using two laser beams (individual addressing, different rotations on different qubits). Single-qubit fidelities exceed 99.9% and are not the limiting factor for computation.
Two-Qubit Gates via Rydberg Blockade
The Rydberg gate is the mechanism that makes neutral atoms a viable universal quantum computer, and it works through one of the strongest interactions in atomic physics.
When a ground-state atom absorbs a carefully tuned laser pulse (either a direct UV excitation at ~297 nm for rubidium, or a two-photon path combining 420 nm and 1013 nm), it is promoted to a Rydberg state: an electronic configuration where the outermost electron orbits at enormous distance from the nucleus (hundreds of Bohr radii). At these radii, the atom’s electric dipole moment is proportional to n², where n is the principal quantum number (typically n = 60–100). Two Rydberg atoms separated by a few micrometers experience a van der Waals interaction that shifts each other’s energy levels by many MHz, far more than the laser’s linewidth.
This energy shift is the blockade: if atom A is in a Rydberg state, the laser cannot excite atom B to the same Rydberg state because the transition is no longer on resonance. The controlled-Z gate exploits this by applying a Rydberg excitation pulse to two neighboring atoms simultaneously. If both atoms are in |1⟩, only one can be excited, and the system accumulates a conditional phase of π that entangles them. If either atom is in |0⟩, the Rydberg pulse has no effect, and no phase is accumulated. The result is a CZ gate.
The blockade radius (the distance within which blockade is effective) is typically 5–10 µm for n ~ 70 Rydberg states, which matches the spacing of atoms in typical tweezer arrays. Gate times are on the order of hundreds of nanoseconds to a few microseconds, including the Rydberg excitation, interaction, and de-excitation steps. This is faster than trapped-ion gates (tens of microseconds) but slower than superconducting CZ gates (20–100 ns).
As of May 2026, the best published two-qubit Rydberg gate fidelities are around 99.5% (QuEra‘s Gemini: >99.2% two-qubit fidelity). The gap between this and trapped-ion fidelities (99.921% on Quantinuum Helios) is the primary technical challenge for the modality.
Reconfigurable Connectivity: The Dynamic Qubit Array
The feature that most distinguishes neutral atoms from other modalities is reconfigurable connectivity. Between circuit layers, the optical tweezers can physically move atoms to new positions, bringing distant qubits into proximity for Rydberg interaction without SWAP overhead. QuEra‘s Dynamic Qubit Array (DQA) architecture formalizes this by dividing the system into two zones: a storage zone (where atoms wait with long coherence times) and an entanglement zone (where Rydberg gates are performed on selected pairs). Atoms are shuttled between zones by AODs.
This reconfigurability has a direct impact on quantum error correction. Non-local stabilizer measurements, which superconducting planar chips cannot perform without SWAP chains, are native operations on a reconfigurable atom array. That is why neutral atoms can implement high-rate qLDPC codes that require long-range connectivity, achieving encoding efficiencies (like the [[16,6,4]] code that produced 96 logical qubits from 448 physical) that surface-code architectures cannot match.
Readout
Measurement uses fluorescence imaging. A resonant laser drives a cycling transition on one hyperfine state (say |1⟩), causing it to scatter thousands of photons and appear “bright” on an EMCCD or sCMOS camera. The other state (|0⟩) is off-resonant and appears “dark.” Each atom’s position is known from the tweezer geometry, so the camera image directly reads out the qubit register.
Readout fidelity is currently ~98–99%, limited by off-resonant scattering that occasionally flips the “dark” state to “bright” or causes atom loss. Mid-circuit measurement (measuring some qubits while preserving others) is more challenging than in trapped ions because the fluorescence photons from measured atoms can scatter onto neighboring unmeasured atoms, disturbing their states. Strategies include spatial separation (moving measured atoms away from the computation zone) and shelving (hiding unmeasured atoms in a dark state that doesn’t scatter). This is an active area of engineering improvement.
Atom Loss and Erasure Conversion
Atoms occasionally escape their tweezers during computation, due to background gas collisions, off-resonant scattering from the trapping laser, or failed Rydberg excitations that kick the atom out of the trap. This “atom loss” reduces the qubit count mid-circuit, a problem that neither superconducting nor trapped-ion systems face at comparable rates.
Two approaches address this. First, continuous atom replenishment: the Harvard/MIT group demonstrated that new atoms can be loaded into empty sites during computation, sustaining a 3,000-atom array for over two hours. Second, erasure conversion: atom loss can be detected (an empty tweezer site is visible on camera), converting the error from an unknown quantum error into a known “erasure” error. Erasure errors are cheaper to correct than unknown errors, because the QEC decoder knows which qubit failed. Codes designed for erasure-dominant noise can tolerate much higher loss rates than codes designed for depolarizing noise. This turns what appears to be a weakness of neutral atoms into a structural advantage for error correction.
Key Academic Papers
Jaksch et al. (2000). “Fast Quantum Gates for Neutral Atoms.” The theoretical proposal for Rydberg-blockade-based two-qubit gates using nanosecond laser pulses. Established the physical basis for the entire modality. Published in Physical Review Letters.
Lukin et al. (2001). Proposed dipole blockade for deterministic entanglement of neutral atoms via Rydberg states. Together with Jaksch’s work, this defined the theoretical framework that labs would spend the next decade verifying experimentally.
Endres et al. / Harvard (2016). Demonstrated defect-free assembly of 2D atom arrays by dynamically rearranging atoms with optical tweezers. Solved the loading-efficiency problem and made large-scale neutral-atom arrays practical. Published in Science.
QuEra / Harvard / MIT (November 2025, Nature). 96 verified logical qubits on a 448-physical-atom processor using a [[16,6,4]] high-rate code. Magic-state distillation demonstrated within the logical layer. Simulations indicated 580 logical qubits on 1,152 physical atoms, and 1,156 on 2,304. The most efficient fault-tolerant demonstration on any hardware platform at the time. My analysis.
Harvard / MIT continuous operation (September 2025). 3,000-atom array operating continuously for over two hours with mid-computation atom replenishment. Solved the atom-loss scaling problem. Published in Nature.
Caltech 6,100-atom array (September 2025). Demonstrated the largest optical tweezer array, trapping 6,100 strontium atoms with high filling fraction. Published in academic proceedings. My analysis.
QuEra 2:1 qLDPC encoding (April 2026). Achieved a 2:1 physical-to-logical qubit ratio using ultra-high-rate qLDPC codes, matching Quantinuum‘s trapped-ion efficiency on a neutral-atom platform. My analysis.
The Vendor Landscape (May 2026)
Major Vendors
QuEra (Boston, USA). The performance and QEC leader in neutral atoms. Spun out of Harvard and MIT research (Mikhail Lukin, Markus Greiner, Vladan Vuletić). QuEra’s product line uses ⁸⁷Rb atoms in a Dynamic Qubit Array (DQA) architecture. Current system: Gemini (260 physical qubits, >99% single-qubit fidelity, >99.2% two-qubit fidelity, commercially available). The Gemini system has been installed on-premises at AIST in Japan, paired with an NVIDIA-powered supercomputer. QuEra’s Aquila (256 atoms, analog mode) is available on AWS Braket.
The QuEra / Harvard / MIT collaboration has produced four landmark Nature papers in 2025: 96 logical qubits on 448 atoms, continuous 3,000-atom operation, below-threshold error correction with error rates decreasing as system size grows, and magic-state distillation within the logical layer. In April 2026, QuEra demonstrated a 2:1 physical-to-logical ratio using qLDPC codes. QuEra raised over $230 million in 2025 and is doubling its workforce. Roadmap: 10,000 physical atoms and 100 logical qubits targeted by 2026.
Pasqal (Paris, France). The deployment leader, with the most neutral-atom systems installed at customer sites worldwide. Pasqal uses ⁸⁷Rb atoms and offers both analog (quantum simulation) and digital (gate-based) modes on the same hardware. The Orion product line ships in rack-mountable form at room temperature, consuming roughly 3 kW, compatible with standard data center infrastructure.
Deployments completed: Orion Beta “Ruby” at GENCI France, Orion at Forschungszentrum Jülich (Germany), Orion at CINECA Italy (February 2026), Orion at OVHcloud (November 2025). Cloud access via Azure Quantum. Pasqal has trapped over 1,000 atoms in a single processor and is developing a 250-qubit QPU optimized for quantum advantage demonstration in early 2026. Pasqal acquired Aeponyx (Canadian photonic integrated circuit company) to enable chip-scale parallel gate operations in next-generation processors. Roadmap: Orion Gamma (140+ qubits, late 2025), Vela (200+ qubits, PIC-enabled, 2027), Centaurus (early FTQC, 2028), Lyra (impactful FTQC, 2029). Logical qubit targets: 2 by 2025, 20 by 2027, 100 by 2029, 200 by 2030. Pasqal demonstrated QRMI integration with NVIDIA CUDA-Q in March 2026, making neutral-atom QPUs Slurm-schedulable alongside CPUs and GPUs.
Atom Computing (Boulder, USA). Uses 87Sr (strontium) atoms and focuses on long coherence and high qubit count. Demonstrated 1,180 atoms trapped in a 1,225-site array (October 2023), with 40-second coherence time on the Phoenix platform. Partnered with Microsoft for logical qubit development. The upcoming Magne system (in collaboration with Microsoft) will deliver 50 logical qubits from ~1,200 physical qubits, targeted for operational deployment to the Export and Investment Fund of Denmark and the Novo Nordisk Foundation by early 2027.
Other Vendors
Infleqtion / ColdQuanta (Boulder, USA). Builds the Hilbert neutral-atom system. Also active in quantum sensing (atomic clocks, magnetometers) and cold-atom technology. Infleqtion’s quantum computing hardware is less commercially advanced than QuEra’s or Pasqal’s, but its broader cold-atom technology base positions it for sensing and networking applications.
planqc (Munich, Germany). Uses strontium atoms in optical lattices (periodic potentials formed by interfering laser beams) rather than individual tweezers. The Pioneer system is under development. Munich-based, with support from the Bavarian quantum ecosystem.
China 100-qubit neutral-atom deployment (November 2025). China deployed a 100-qubit neutral-atom quantum computer, marking its entry into this modality at scale. Details remain sparse but confirm that the global neutral-atom ecosystem extends beyond the US and Europe.
Google Quantum AI. In early 2026, Google announced a parallel neutral-atom quantum computing program at a new laboratory in Boulder, Colorado, operated in collaboration with JILA (led by Professor Adam Kaufman). Google became the only company pursuing both superconducting (Willow roadmap) and neutral-atom approaches simultaneously. This is a significant validation of the modality from the company that leads superconducting QEC.
Two Business Models: Academic Collaboration vs. Commercial Deployment
QuEra and Pasqal represent fundamentally different approaches to building a neutral-atom business, and the contrast is worth understanding.
QuEra operates through deep academic partnerships. Its landmark results (96 logical qubits, 3,000-atom continuous operation, magic-state distillation, 2:1 qLDPC ratio) are all published collaborations with Harvard and MIT, often with the academic groups doing the experimental work on QuEra-built hardware. The company’s Gemini system (260 qubits, commercially available) is the product of this research pipeline: breakthroughs demonstrated in the lab are productized into shipping hardware. QuEra’s first on-premises installation at AIST Japan pairs the Gemini system with NVIDIA HPC infrastructure. The Aquila analog simulator is on AWS Braket. The model’s strength is that QuEra consistently produces Nature-grade science while simultaneously building commercial products. The risk is dependence on a small number of academic groups for the most advanced demonstrations, and the question of whether the transition from lab results to production hardware happens fast enough to maintain competitive advantage.
Pasqal prioritizes rapid deployment to customers. Its Orion line is installed at five major HPC centers across Europe (GENCI France, Jülich Germany, CINECA Italy, OVHcloud, plus Azure Quantum access), and additional deployments are planned for Canada, the Middle East, and further European sites. Pasqal offers both analog (quantum simulation) and digital (gate-based) modes on the same hardware, which broadens the customer base to include near-term simulation use cases that don’t require fault tolerance. The company demonstrated QRMI integration with NVIDIA CUDA-Q in March 2026, making neutral-atom QPUs Slurm-schedulable. Pasqal’s Aeponyx acquisition positions it for PIC-enabled parallel gates in the Vela generation (2027). The model’s strength is commercial traction and hardware deployed in real HPC environments. The risk is that Pasqal’s published two-qubit fidelities lag behind QuEra’s, and the company’s fault-tolerance demonstrations are less advanced.
The Data Center Advantage
One detail worth lingering on: Pasqal‘s Orion fits in a standard server rack. Room temperature. 3 kW power draw. No dilution refrigerator, no helium-3 supply chain, no vibration isolation pads, no chilled-water plant, no 3-meter ceiling clearance. Compare that to the facility requirements for a superconducting system: 750 kg cryostat, 30 kW peak power during cooldown, dedicated cooling water, EMI shielding, vibration exclusion zones, and $100,000 of He-3 per cryostat. The infrastructure burden per qubit is lower for neutral atoms by an order of magnitude. I cover both sides of this comparison in my deep dives on building a neutral-atom system and cryogenic infrastructure.
Quantum Error Correction: The Logical Qubit Record
As of May 2026, neutral atoms hold the record for verified logical qubits on real hardware, though Quantinuum‘s 94-logical-qubit result from the same-sized 98-qubit Helios chip is close in absolute count. The comparison illuminates a deeper point about how different modalities approach error correction.
QuEra’s 96 logical qubits used a [[16,6,4]] high-rate code, encoding 6 logical qubits per 16-atom block. This is possible because the Dynamic Qubit Array can rearrange atoms between circuit layers, implementing the non-local stabilizer measurements that high-rate codes demand. The 2:1 physical-to-logical ratio demonstrated in April 2026 used qLDPC codes that push encoding efficiency even further.
Magic-state distillation, the resource-intensive process of generating the non-Clifford states needed for universal computation, was demonstrated within the logical layer in the November 2025 Nature paper, achieving higher output fidelity than input, a “beyond-breakeven” result for magic states. This is one of the CRQC capabilities I track in my framework.
The erasure-conversion advantage is structurally important here. Because atom loss is detectable (the camera sees an empty site), neutral-atom QEC codes can treat loss as a known error. This allows higher tolerable loss rates: where a standard depolarizing-noise code might require <1% error per gate, an erasure-optimized code can tolerate 3–5% erasure rates. This makes neutral atoms more resilient to their primary error channel than the raw fidelity numbers would suggest.
Comparison to Other Modalities
Neutral Atom vs. Superconducting
Superconducting qubits are faster (nanosecond gates vs. microsecond Rydberg gates) and have a more mature fabrication ecosystem (lithographic, chip-based manufacturing with an established QOA component supply chain). Google’s Willow demonstrated below-threshold surface code scaling. IBM’s qLDPC Gross code reached the teraquop regime.
Neutral atoms counter with room-temperature operation (no cryogenics), reconfigurable connectivity (no SWAP overhead), larger demonstrated qubit arrays (6,100 atoms vs. 1,121 transmons), and superior QEC encoding efficiency (2:1 ratio on qLDPC codes vs. ~100:1 on surface code). The erasure-conversion advantage further favors neutral atoms for error correction.
The honest comparison for the CRQC path: superconducting processors execute more error-correction cycles per second (MHz vs. kHz), but neutral atoms encode more logical information per physical qubit. Which approach reaches a useful number of high-quality logical qubits first depends on whether raw clock speed or encoding efficiency matters more for the specific codes and algorithms being run. As I discussed in the superconducting modality article, the metric that resolves this is error-corrected operations per second, and neither modality has a decisive advantage yet.
Neutral Atom vs. Trapped Ion
The closest competitor. Both modalities use atomic qubits with long coherence, both have demonstrated high-rate QEC codes with ~2:1 efficiency, and both benefit from reconfigurable or all-to-all connectivity. Trapped ions have higher two-qubit fidelity (99.921% on Quantinuum Helios vs. ~99.2% on QuEra Gemini), more mature mid-circuit measurement, and longer coherence per qubit. Neutral atoms have larger qubit arrays (6,100 vs. 98), room-temperature operation, faster gates (µs vs. tens of µs), and the erasure-conversion advantage.
The scaling paths differ. Trapped-ion scaling requires either complex QCCD trap chips with hundreds of zones or photonic interconnects between modules (both carry engineering overhead). Neutral-atom scaling requires larger tweezer arrays and higher-power laser systems, which leverage existing photonics technology. Pasqal‘s acquisition of Aeponyx for photonic integrated circuits is a bet that PIC technology can parallelize Rydberg gate operations, which would address one of the remaining bottlenecks (current systems apply Rydberg gates to one zone at a time).
Neutral Atom vs. Photonic
Photonic systems share the room-temperature advantage but use probabilistic entangling gates, requiring measurement-based or fusion-based architectures with high photonic resource overhead. Neutral atoms perform deterministic Rydberg gates, producing entanglement with near-unity success probability per attempt. This gives neutral atoms a significant gate-efficiency advantage for algorithmic computation. Photonic systems may excel at long-distance quantum networking (photons travel natively through fiber), while neutral atoms are stronger for local computation. PsiQuantum‘s bet on fusion-based quantum computing at GlobalFoundries fab scale is architecturally distinct from anything in the neutral-atom world, and the two modalities may ultimately serve different roles in a heterogeneous quantum ecosystem.
Neutral Atom vs. Silicon Spin
Silicon spin qubits and neutral atoms occupy opposite ends of the maturity-versus-scalability spectrum. Silicon spin has demonstrated only ~12 qubits but benefits from CMOS-compatible fabrication on 300 mm industrial wafers (Diraq‘s >99% fidelity on foundry devices). Neutral atoms have demonstrated 6,100 trapped qubits but rely on precision optics rather than lithographic mass production. Silicon operates at ~1 K (still cryogenic, but far warmer than superconducting 10 mK); neutral atoms operate at room temperature. Silicon spin’s long-term cost-per-qubit advantage (if CMOS yield curves hold) could be transformative, but the modality is years behind neutral atoms in qubit count, QEC demonstrations, and logical qubit production. The first logical operations on silicon were demonstrated in March 2026, while neutral atoms had 96 logical qubits four months earlier. For the foreseeable future, neutral atoms are the more advanced platform.
Neutral Atom vs. Topological
Microsoft‘s topological approach (Majorana-based qubits) promises hardware-level error protection that would eliminate most error-correction overhead. If it works, it would leapfrog all existing modalities. But Majorana 1 has not demonstrated a topological qubit, and Microsoft’s own partnership with Atom Computing (the Magne system for Denmark) suggests Microsoft itself is hedging with neutral atoms while topological development continues. Neutral atoms are shipping hardware with demonstrated logical qubits; topological qubits remain a research program. The comparison is not competitive today, but worth tracking for the decade ahead.
Advantages
Room-temperature operation. No dilution refrigerator, no helium-3, no cryogenic wiring, no vibration isolation beyond standard optical-table practice. Pasqal’s Orion runs in a standard server rack at 3 kW. This eliminates the most expensive, longest-lead, and physically largest component in a superconducting quantum computer, and the entire helium-3 supply chain risk that I discuss in my cryogenic infrastructure deep dive.
Scalable qubit count. Optical tweezers generated by SLMs and AODs can produce arrays of thousands of trapping sites using a single laser source and standard photonics components. The scaling path is increasing laser power, improving SLM resolution, and engineering better atom-loading efficiency. Caltech’s 6,100-atom array demonstrates that the physics supports arrays far larger than any current computation requires.
Reconfigurable connectivity. Atoms can be physically moved between circuit layers, enabling non-local qubit interactions without SWAP gates. This is uniquely valuable for high-rate QEC codes (qLDPC, color codes) that require non-local stabilizers. No other modality offers this flexibility natively.
QEC encoding efficiency. The 96-logical-qubit result on 448 physical atoms and the 2:1 qLDPC ratio represent the best demonstrated encoding efficiencies on any hardware. Neutral atoms’ reconfigurable connectivity makes them natural hosts for the high-rate codes that minimize the physical-qubit overhead for fault tolerance.
Erasure-conversion advantage. Atom loss is detectable, which converts the primary error channel into a known erasure. Erasure-optimized codes tolerate higher raw error rates than depolarizing-noise codes, making neutral atoms more fault-tolerant per unit of raw fidelity than the headline gate-fidelity numbers alone suggest.
Identical qubits. Every ⁸⁷Rb atom is identical to every other ⁸⁷Rb atom. No fabrication variation, no frequency targeting challenges, no chip-to-chip variability.
Disadvantages
Two-qubit gate fidelity gap. At ~99.2% on QuEra Gemini, neutral-atom Rydberg gates lag behind trapped ions (99.921% on Quantinuum Helios) and superconducting qubits (99.67% on Google Willow). The fidelity is limited by Rydberg state decay during the gate (spontaneous emission from the excited state), laser intensity noise, and Doppler shifts from residual atomic motion. Closing this gap to >99.5% is the single most important engineering challenge for the modality.
Mid-circuit measurement challenges. Fluorescence readout scatters photons that can disturb neighboring qubits. Measuring a subset of qubits while preserving others requires spatial separation or shelving protocols that add overhead. This is less mature than in trapped ions, where individual-ion fluorescence detection is well-established with minimal cross-talk.
Atom loss during computation. Despite the erasure-conversion advantage, atom loss still represents a resource drain: each lost atom must either be replenished (adding time overhead) or tolerated by the QEC code (consuming error-correction budget). The continuous-replenishment technique demonstrated by Harvard/MIT is a mitigation but adds operational complexity.
Laser and optics infrastructure. While neutral-atom systems avoid cryogenics, they require extensive laser systems: a trapping laser (10+ W at 1064 nm), Rydberg excitation lasers (UV or two-photon), cooling and repumping lasers, and SLM/AOD systems for array generation and rearrangement. These are more compact than the laser systems for trapped-ion gates (because neutral atoms use broader-bandwidth excitations), but they are still substantial. Pasqal’s Aeponyx acquisition aims to shrink this infrastructure via photonic integrated circuits.
Gate parallelism. Current systems apply Rydberg gates zone-by-zone rather than globally in parallel, because the Rydberg blockade radius means that exciting one atom pair prevents nearby pairs from being gated simultaneously. PIC-enabled parallel gate operations (Pasqal’s Vela roadmap, 2027) and advanced addressing schemes aim to increase parallelism, but this remains an engineering challenge.
Vertically integrated supply chain. Like trapped ions, the neutral-atom market is vertically integrated. You buy a system from QuEra, Pasqal, or Atom Computing; there is no equivalent of the superconducting QOA component ecosystem where an integrator assembles a system from independently sourced QPUs, cryostats, and control electronics. The optics, vacuum, and laser subsystems are commercially sourced but not standardized into interchangeable modules. I discuss this in my deep dive on building a neutral-atom quantum computer.
Impact on Cybersecurity
Neutral atoms are one of three modalities (alongside superconducting and trapped ions) with a plausible path to a CRQC. Their QEC efficiency means they could reach the ~1,400 logical qubits required for RSA-2048 (per the Gidney 2025 estimate) with fewer physical qubits than superconducting surface-code approaches. At a 2:1 ratio, ~2,800 physical atoms would suffice, compared to ~1.4 million physical superconducting qubits at a 1,000:1 surface-code ratio.
The wall-clock time for the computation depends on gate speed. Rydberg gates at ~1 µs per two-qubit operation are faster than trapped-ion gates (~10–100 µs) but slower than superconducting CZ gates (~20–100 ns). For the Gidney estimate’s 6.5 billion Toffoli gates, a neutral-atom processor at 1 µs per gate would take roughly 1.8 hours, versus seconds on a comparable superconducting machine and days on a trapped-ion system.
The practical question is whether any neutral-atom vendor reaches the required qubit count and fidelity before 2030. QuEra’s roadmap targets 10,000 physical atoms by 2026 and continues scaling thereafter. If qLDPC codes maintain their demonstrated 2:1 efficiency at scale, a 10,000-atom system would encode ~5,000 logical qubits, more than enough for RSA-2048. Whether this is achieved depends on maintaining gate fidelity at scale (>99.5% on 10,000+ atoms) and demonstrating sustained fault-tolerant operation over the hours required.
My CRQC Quantum Capability Framework tracks these capabilities across modalities. The response for practitioners is the same regardless of which modality arrives first: begin PQC migration now. Regulators, insurers, investors, and clients are already setting deadlines that arrive before any quantum computer breaks a key.
Future Outlook
2026. QuEra targets 10,000 physical atoms and 100 logical qubits, which would make it the first platform to break three digits of verified logical qubits. Pasqal demonstrates quantum advantage on an industry-relevant problem with a 250-qubit QPU. Google’s neutral-atom lab at JILA/Boulder begins producing results, marking the first time a company pursues both superconducting and neutral-atom modalities simultaneously. China continues to scale its neutral-atom program. Atom Computing and Microsoft deliver the Magne system (50 logical qubits from ~1,200 physical) to Denmark.
2027–2028. Pasqal’s Vela (200+ qubits, PIC-enabled parallel gates) and Centaurus (early FTQC) arrive. QuEra scales toward the regime where CRQC resource estimates start to become relevant. The fidelity question becomes decisive: if Rydberg gates reach 99.5–99.7%, neutral atoms become competitive with trapped ions on a per-logical-gate basis while maintaining their qubit-count and infrastructure advantages.
2029–2030. Pasqal targets Lyra (impactful FTQC, 200 logical qubits). QuEra targets fault-tolerant computation at scale. If either delivers, neutral atoms become the first room-temperature modality to achieve practical fault tolerance, a milestone with significant implications for both quantum computing deployment and CRQC timelines.
The trajectory of this modality over the past three years has been remarkable. In September 2023, when the original version of this article was published, the largest controlled neutral-atom entanglement was around 50 atoms and the idea of 96 logical qubits on 448 physical atoms would have seemed aspirational at best. The speed of progress in neutral atoms should make anyone tracking Q-Day pay close attention.
Quantum Upside & Quantum Risk - Handled
My company - Applied Quantum - helps governments, enterprises, and investors prepare for both the upside and the risk of quantum technologies. We deliver concise board and investor briefings; demystify quantum computing, sensing, and communications; craft national and corporate strategies to capture advantage; and turn plans into delivery. We help you mitigate the quantum risk by executing crypto‑inventory, crypto‑agility implementation, PQC migration, and broader defenses against the quantum threat. We run vendor due diligence, proof‑of‑value pilots, standards and policy alignment, workforce training, and procurement support, then oversee implementation across your organization. Contact me if you want help.