Quantum technologies
Quantum computing approaches compared
How do neutral-atom and superconducting quantum computers compare on scale, error correction and their main limits, and how far is the encryption-breaking goal?
This comparison sets neutral-atom and superconducting quantum computers against the goal of breaking today's public-key encryption, using the largest reported system, a reported error-correction result and the main limit for each. Every cell is a lab or theory result, so counts and error rates are not like-for-like across platforms.
| Option | Largest system reportedLargest physical qubit or atom count in a lab or theory result, with year. Units: physical qubits or atoms held. | Logical qubit or error-correction resultReported logical-qubit or error-correction figure, with code size or error rate. Units: logical qubits, error rate per correction cycle, or error suppression factor. | How qubits are heldThe physical form the qubits take and how they are held or controlled. | Main limitationA main limit on scale or reliability stated in the cited sources. |
|---|---|---|---|---|
| Neutral atoms | Over 3,000 atoms held for more than two hours in a lab array (2025). Reconfigurable arrays of up to 448 neutral atoms are reported in a 2025 preprint.Demonstrated[1][2] | 24 logical qubits encoded in 48 atoms (2024 preprint). A 2.14-fold below-threshold suppression in a four-round characterisation circuit (2025 preprint).Reported, unverified[3][2] | Neutral atoms in reconfigurable arrays. The 2024 preprint uses ytterbium atoms. The 2025 paper does not name the atom species.Demonstrated[3][1][2] | Atoms are lost during operation. The 2024 preprint corrected 1.8 lost atoms on average, and the 2025 paper says losses typically limit these systems to pulsed operation, with continuous reloading used to work around them.Demonstrated[3][1] |
| Superconducting qubits with error correction | 101 physical qubits in a distance-7 surface-code memory on superconducting processors (lab, 2024).Demonstrated[4] | Logical error rate fell 2.14-fold as code distance rose from 5 to 7, reaching 0.143% per correction cycle at distance 7 (lab, 2024).Demonstrated[4] | Superconducting transmon qubits in a square grid, with a real-time classical decoder setting corrections (lab, 2024).Demonstrated[4] | Rare correlated error events, about once an hour, limit logical performance in the repetition-code experiment. The paper does not show scaling to useful machines.Demonstrated[4] |
| Breaking encryption goal | No machine of this size is reported in the sources reviewed. A 2025 theory estimate gives fewer than one million noisy qubits for RSA-2048.Proposed[5] | Stable logical memories are shown in the 2024 distance-7 experiment. The sources reviewed do not report logical gates on error-corrected qubits at useful scale.Demonstrated[4] | Hardware not specified. The 2025 estimate assumes a surface-code architecture, a 0.1% gate error rate and a 1 microsecond cycle.Proposed[5] | The estimate depends on its stated error and cycle assumptions, and no machine of that size is reported in the sources reviewed.Proposed[5] |
What the table shows
- Reported physical counts range from 101 qubits in a 2024 superconducting memory to over 3,000 atoms held in a 2025 neutral-atom array.
- The superconducting error-correction result is one distance-7 memory, whose logical error rate fell 2.14-fold when the code grew from distance 5 to 7.
- The neutral-atom logical-qubit results are preprints, with 24 logical qubits in 48 atoms and 2.14-fold suppression over four rounds.
- The encryption-breaking row has no reported machine, and its 2025 estimate gives fewer than one million noisy qubits for RSA-2048.
Read with care
- Atoms held in an array and qubits on a processor measure different things, and an array held for operation is not a working computer.
- Error-correction results come from single experiments on small codes, so logical figures are not directly comparable across platforms.
- The encryption estimate is a theory preprint whose qubit count depends on the error rate and cycle time it assumes.
- Several neutral-atom figures come from arXiv preprints, which were checked on arXiv rather than against the full journal record.
Sources
- preprint
Continuous operation of a coherent 3,000-qubit system
Neng-Chun Chiu, et al.. arXiv, 25 Jun 2025. Accessed 21 Oct 2026. - preprint
Architectural mechanisms of a universal fault-tolerant quantum computer
Dolev Bluvstein, Alexandra A. Geim, Sophie H. Li et al.. arXiv, 25 Jun 2025. Accessed 21 Oct 2026. - preprint
Fault-tolerant quantum computation with a neutral atom processor
Ben W. Reichardt, et al.. arXiv, 18 Nov 2024. Accessed 21 Oct 2026. - preprint
Quantum error correction below the surface code threshold
Rajeev Acharya, et al.. arXiv, 24 Aug 2024. Accessed 21 Oct 2026. - preprint
How to factor 2048 bit RSA integers with less than a million noisy qubits
Craig Gidney. arXiv, 21 May 2025. Accessed 21 Oct 2026.