How trapped-ion qubits work

A trapped-ion qubit is a single atom with one electron removed, held by electric fields 50 to 100 micrometers (µm) above a chip. Every ion of a given species is identical, so the qubits themselves have no fabrication variation. Trapped ions hold the highest published gate fidelities of any approach and can entangle any pair of qubits directly. They are slow, though: once the movement and cooling that gates need are counted, a two-qubit gate takes about 1 to 10 milliseconds (ms), by our estimate from measured circuit times. Quantinuum, IonQ, AQT and others build them.

This article is part of a series on the main ways to build a qubit. The overview compares all six side by side.

How the qubit forms

Ions with one outer electron, such as ytterbium-171, barium-137 and calcium-40 or calcium-43, have simple energy levels, much like hydrogen. The most common qubit uses two hyperfine levels of the ion's ground state, split by a few gigahertz: about 12.6 GHz in ytterbium-171 and 8.0 GHz in barium-137. Neither level decays on its own, and the usual "clock" pair barely responds to magnetic fields. Coherence of seconds is routine in working processors, and the coherence of a single ytterbium ion has been estimated at more than an hour in the laboratory (Tsinghua University, 2021).

A Paul trap holds the ions. A static electric field alone cannot hold a charge in place, so the trap combines a radio-frequency field, which confines the ions across the trap's axis, with static voltages on segmented electrodes that confine them along it and move them. Laser cooling brings the ions close to their lowest state of motion, and several ions settle into a line a few micrometers apart.

The ions repel each other, so they vibrate together. These shared vibrations are the bus for entanglement: a gate briefly couples the internal states of two ions to a shared mode of motion. Any two ions in a chain or gate zone can therefore be entangled directly.

How the processor is made

Modern traps are microfabricated chips. All the electrodes lie in one plane of patterned metal, such as gold, aluminum or niobium, on silicon, sapphire or fused silica. Quantinuum's Helios trap has 1,228 electrodes and a four-way junction that links a ring-shaped storage region to its gate zones. Quantinuum is working with GlobalFoundries on 300 mm trap wafers, and IonQ is acquiring the SkyWater foundry.

The trap sits in ultra-high vacuum, around 10⁻¹¹ millibar, because a single collision with a gas molecule can knock an ion out. Some systems run at room temperature (AQT, eleQtron, IonQ Forte). Quantinuum's run cryogenically, which improves the vacuum and reduces electric-field noise from the electrodes.

Light does most of the work. A laser-based system needs five to eight wavelengths for loading, cooling, gates and readout, delivered through free-space optics, modulators and high-aperture lenses. That optical bench is a large share of a system's size and cost.

How it is controlled and read out

Single-qubit gates use a pair of laser beams whose frequency difference matches the qubit splitting, or a direct microwave drive. Helios reports 99.9975% single-qubit fidelity, and Oxford University measured an error of 1.5 in 10 million (99.99998%) with microwaves on calcium-43 in 2025, a record for any qubit.

Two-qubit gates of the Mølmer-Sørensen type push two ions with a spin-dependent force that briefly excites their shared motion and then returns it, leaving the two qubits entangled. The gate itself takes about 10 to 500 µs, about 70 µs on Helios. Quantinuum reports 99.921% averaged across all qubit pairs of the 98-qubit Helios, and IonQ reports 99.99% on one pair of ions using electronic (microwave) control. Both are company figures.

Moving ions. In the quantum charge-coupled device (QCCD) design that Quantinuum uses, ions are shuttled between storage, gate and readout zones by changing electrode voltages, and chains are split and merged. Moving and re-cooling dominate the time: Helios averages 55 ms per layer of 49 two-qubit gates across its 98 qubits. Counted this way, a two-qubit gate takes about 1 to 10 ms by our estimate, roughly 10,000 to 100,000 times longer than a superconducting gate.

Readout is by fluorescence. A laser makes the ion scatter thousands of photons if it is in one qubit state and none in the other, and a lens collects the light onto a camera or photon counter in about 0.1 to 1 ms. Helios reports 99.952% readout fidelity.

What makes it hard to scale

Speed. With transport and cooling, a full circuit layer takes tens of milliseconds, so deep fault-tolerant algorithms could run for days. It is the most common criticism of the approach, partly offset by high fidelity and a lower error-correction overhead.

Chain length. In a single chain the modes of motion crowd together as ions are added, and chains become hard to control beyond roughly 30 to 50 ions. Every vendor has moved to traps with many zones.

Heating. Electric-field noise from the electrode surfaces heats the ions, and it grows steeply as traps shrink.

Lasers and optics. Many stabilized wavelengths, steered to hundreds of sites with tight pointing and phase stability, are hard to manufacture, and ultraviolet light is hard to route on a chip.

Wiring. Thousands of electrodes each need a voltage channel that updates fast enough for transport.

Linking modules. Connecting traps with photons has reached about 200 entangled pairs a second, and fidelities of up to about 97% in separate experiments, far below what local gates achieve.

System-level fidelity. Records usually come from one or two ions. Keeping 99.9% across 100 or more qubits is much harder.

What could solve it

Zones and 2D grids. Splitting the trap into zones and moving ions keeps chains short while keeping any-to-any connectivity. Junctions let ions be routed in two dimensions, and Quantinuum's next system, Sol, planned for 2027, uses a 2D grid of 192 qubits.

Integrated photonics. Waveguides and grating couplers built into the trap chip can deliver every wavelength from below, with detectors on the chip, replacing the optical bench. Ions with visible-wavelength transitions, such as barium, make this easier.

Electronic control. Microwave conductors and magnetic-field gradients on the chip can replace the gate lasers, so gates are defined by electronics that fabs already know how to build. IonQ (which acquired Oxford Ionics), eleQtron and Universal Quantum follow this route.

Two ion species. One species stores the qubits while another is cooled or measured, so cooling light does not disturb the data. Helios uses barium qubits with ytterbium coolants.

Efficient codes. High fidelity and all-to-all connectivity allow codes that need fewer physical qubits per logical qubit. Quantinuum reports 48 error-corrected logical qubits at a ratio of two physical qubits per logical qubit on Helios, a company figure.

Strengths and limits

Strengths. Identical qubits with no fabrication spread; the highest published gate fidelities; coherence of seconds; all-to-all connectivity within a zone, and across the chip by moving ions; accurate readout and measurement in the middle of a circuit; and no dilution refrigerator.

Limits. Slow gates once movement and cooling are counted; single chains do not scale, so transport adds complexity; bulky lasers and optics; heating that worsens in smaller traps; slow photonic links between modules; and qubit counts in the low hundreds.

Key numbers

Two-qubit gates: 99.921% across all pairs of 98 qubits (Quantinuum Helios); 99.99% on one pair (IonQ). Both are company figures.

Single-qubit gates: 99.9975% (Helios); 99.99998% in the laboratory (Oxford University, 2025).

Gate time: about 10 to 500 µs for the gate itself (about 70 µs on Helios); about 1 to 10 ms including shuttling and re-cooling (our estimate).

Circuit layer: 55 ms across 98 qubits (Helios).

Readout: 99.952% (Helios), in about 0.1 to 1 ms.

Coherence: seconds in processors; more than an hour estimated for one ion in the laboratory.

Largest systems: Quantinuum Helios, 98 qubits; IonQ Tempo, 100 qubits. IonQ has announced 256-qubit chips, with customer deliveries planned for 2027.

Operating environment: ultra-high vacuum, at room temperature or cryogenic.

Quick answers

Why are trapped ions so accurate? Every ion of a species is identical, the qubit levels barely interact with their surroundings, and gates are driven by precisely controlled light or microwaves.

Why are trapped-ion quantum computers slow? The gates themselves take microseconds to hundreds of microseconds, and in large machines the ions must be moved and re-cooled between gates, which brings each two-qubit gate to about 1 to 10 ms by our estimate.

Do trapped-ion quantum computers need a dilution refrigerator? No. Some run at room temperature inside a vacuum chamber; others cool the trap cryogenically to improve the vacuum and reduce noise, but none needs millikelvin temperatures.

What is a QCCD? The quantum charge-coupled device, proposed in 2002, splits the trap into many small zones and moves ions between them as needed, so chains stay short while any qubit can still interact with any other.

Sources

Quantinuum Helios launch: 98 qubits, 99.921% two-qubit fidelity (November 2025)

Quantinuum, Helios: a 98-qubit trapped-ion quantum computer (arXiv 2511.05465)

IonQ, 99.99% two-qubit fidelity with electronic qubit control (October 2025)

Oxford University, single-qubit error of 1 in 6.7 million (June 2025)

Trapped-ion execution speed: time budget of gates, transport and cooling (arXiv 2405.11450)

Monroe group, remote entanglement of ions at 97% fidelity (arXiv 2406.01761)

Quantinuum, second-quarter 2026 results: Sol and Apollo roadmap

About Anyon Computing

Anyon Computing Inc. (also known as Anyon Technologies) is a quantum computing company that provides superconducting modular quantum supercomputers, which it tightly couples with NVIDIA GPUs via NVQLink for data center deployment at scale. Anyon pioneered deploying NVIDIA NVQLink inside a commercial production data center and is among the first superconducting QPU backends integrated with NVIDIA CUDA-Q. Founded by pioneers from Caltech and UC Berkeley. https://www.anyoncomputing.com/

NVIDIA, CUDA-Q and NVQLink are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and product names may be trademarks of the respective companies with which they are associated.