Trapped-ion quantum computers explained

A quantum computer stores information in qubits, which can hold combinations of states that ordinary bits cannot. Trapped-ion quantum computers make each qubit from a single charged atom, called an ion. Electromagnetic fields hold the ions in place; in Quantinuum's Helios machine they hover above a chip (Quantinuum blog). Laser pulses (or, in IonQ's newer prototypes, precision electronics) change the internal state of each ion. Two ions can then be made to interact, which links qubits into a program. The approach matters because its developers report some of the most accurate qubits so far (company-reported figures, covered below), and accuracy decides how much of a calculation survives before errors take over.

How the approach works

An ion is an atom with one or more electrons removed, so it carries an electric charge. Its internal energy levels can hold the 0 and 1 states of a qubit. Because the ions are held in place, they can be addressed one at a time and, in some designs, moved between zones. Quantinuum's Helios system uses the quantum charge-coupled device (QCCD) design, in which ions are moved between separate operation zones, and it uses barium-137 ions as its qubits (Helios preprint; arXiv abstract). A 2019 peer-reviewed review by Bruzewicz and colleagues says the basic requirements for universal quantum computing have all been demonstrated with ions (Applied Physics Reviews).

What the Helios preprint reports

The most detailed public description of the current generation that we found is a preprint about Quantinuum's 98-qubit Helios system, submitted to arXiv on 7 November 2025 (arXiv). It is company-authored. When we checked on 23 October 2026, arXiv listed only version 1 and we found no peer-reviewed journal version, so these are preprint results. The company describes the 98 qubits as fully connected (Quantinuum blog).

The preprint reports error rates averaged over all operational zones of the machine:

• single-qubit operations: 2.5 × 10⁻⁵, roughly one failure in 40,000;
• two-qubit operations: 7.9 × 10⁻⁴, roughly one failure in 1,270;
• state preparation and measurement: 4.8 × 10⁻⁴.

A two-qubit operation takes about 70 microseconds (arXiv HTML, Section II.3). These are averages, not a guarantee for every qubit. The preprint gives per-pair values in its appendix tables, which we did not check.

What is reported but not independently checked

The figures in this section come from the companies involved. The sources we reviewed do not show an independent reproduction of any of them.

Quantinuum's launch announcement, dated 5 November 2025, gives 99.921% two-qubit and 99.9975% single-qubit gate fidelity (Quantinuum press release). Its companion blog post says the two-qubit figure applies across all qubit pairs (Quantinuum blog). The percentages match the preprint's averages when written the other way round (1 minus 7.9 × 10⁻⁴ is 99.921%), but neither company source says the two sets come from the same measurements. The "all qubit pairs" wording is stronger than an average, which cannot show that every pair performs equally.

Quantinuum also reports logical qubits, which group several physical qubits so that errors can be detected or corrected.

• 94 error-detected logical qubits in a globally entangled state, and 50 error-detected logical qubits in a magnetism simulation. The company says both perform better than the physical qubits they are built from. Error detection flags that an error has happened but does not fix it.
• 48 error-corrected logical qubits at a 2:1 encoding rate, using code concatenation, which the company also says perform better than physical qubits. It quotes 99.99% state preparation and measurement fidelity for this result (Quantinuum press release; Quantinuum blog). The sources we reviewed do not say how that figure was measured, so it should not be compared directly with the preprint's physical-qubit average of 4.8 × 10⁻⁴.

Quantinuum's blog also calls Helios the most accurate commercial quantum computer in the world. That is a company claim, and nothing in the sources we reviewed tests it independently.

IonQ announced on 21 October 2025, in a company press release republished by The Quantum Insider, a two-qubit gate fidelity above 99.99% using Electronic Qubit Control, which uses precision electronics rather than lasers to control its qubits (The Quantum Insider). The result came from prototype systems in IonQ's research labs, not from a commercial service. The release refers to technical papers, which we did not check; the release itself does not give the number of qubits used or how the fidelity was measured.

Limits and open problems

Speed. A two-qubit operation takes about 70 microseconds, and the preprint lists increasing clock speed as one scaling challenge for the QCCD design (arXiv HTML, Section IV).

Scaling the traps and optics. The 2019 review discusses what may be required to increase the scale of trapped-ion computers while mitigating decoherence and control errors (Applied Physics Reviews). The Helios preprint reports much larger crosstalk errors for ions sitting next to the 493 nm laser light used to measure or reset a neighbouring ion mid-circuit (arXiv HTML, Section III.2.5). The preprint reports these on the 98-qubit machine, and the sources we reviewed do not show how they would be handled in much larger machines.

Error sources. The preprint's error budget predicts a two-qubit error of 3.5 × 10⁻⁴. That is a prediction, not a measurement, and it is set against a measured average of 7.9 × 10⁻⁴. The preprint lists several possible reasons for the gap, including higher leakage error in the operational zones, crosstalk and other unaccounted-for effects (arXiv HTML, Section III.2.3).

Cost and control. None of the sources we checked gives a cost for a trapped-ion system, so we do not estimate one. Control electronics come up in two places: the clock speed limit above, and IonQ's approach, which uses electronics rather than lasers to control its qubits. That approach is still a prototype result.

Verification. The Helios figures come from a company-authored preprint (arXiv version 1). If a peer-reviewed version appears, its figures may differ.

Speculative uses

The following have not been shown in the sources we reviewed:

• error-corrected trapped-ion machines running chemistry or materials simulations that classical computers cannot match;
• trapped-ion modules linked by photonic connections into larger machines;
• faster control that cuts the time per logical operation enough for long error-corrected circuits.

What to watch next

• Whether IonQ demonstrates the 256-qubit systems it said, in October 2025, it would demonstrate in 2026 (a company target), and with what fidelity.
• Whether the Helios results appear in a peer-reviewed journal with matching figures, and whether outside groups reproduce the averages.
• Whether two-qubit error rates hold as qubit numbers rise.
• Whether the time per two-qubit operation falls well below 70 microseconds.
• Whether independent cost figures for trapped-ion systems become available.