Neutral-Atom Quantum Computers: How Laser Tweezers Work

Most quantum chips are built by people. A neutral-atom machine uses qubits made by nature. Every atom of one kind is exactly like every other. You do not have to make a thousand tiny circuits all the same. You just catch a thousand atoms.

That is why this approach holds some of the largest qubit counts in our data. It is also why companies like QuEra, Atom Computing, Pasqal and Infleqtion are building on it. This page explains how it works, in plain steps, and what each company reports.

First, one word. A neutral atom has no electric charge. Its electrons balance its protons. That sets it apart from a trapped-ion machine, where each atom has lost an electron and is held by electric fields.

A neutral atom cannot be grabbed with electric fields. So these machines use light.

A laser beam is focused down to a very small, bright spot. An atom near that spot is gently pulled toward the brightest part. It sits there, like a marble at the bottom of a bowl. This trap is called an optical tweezer.

Now make many tweezers at once, in a grid. Spray a cloud of cooled atoms over them. Some tweezers catch an atom and some stay empty. A camera checks which spots are full. Then movable tweezers slide atoms into the empty spots, until the grid is filled in. It is a bit like a parking lot where a valet moves every car into a neat row.

The atoms sit in a vacuum, so air does not knock them out. Lasers also cool them until they barely move.

An atom has many energy levels, like rungs on a ladder. Builders pick two rungs that are steady and easy to control. The lower one is called 0. The upper one is called 1.

Laser or microwave pulses act as one-qubit gates. A pulse of the right color and length moves the atom between the two rungs, or part way. A pulse can hit the whole grid at once, which is called a global gate. Or it can hit one atom, which is called a local gate.

Different companies use different atoms and different rungs:

• QuEra and Pasqal use rubidium.
• Atom Computing uses ytterbium-171. It stores the qubit in the atom's nucleus, which is well shielded.
• Infleqtion uses cesium, with what it calls clock-state qubits.

To read the qubits, a laser makes atoms in one state glow. A camera takes a picture. A bright spot means one result. A dark spot means the other.

Atoms in a grid sit a few millionths of a meter apart. At that distance, they normally ignore each other. So how do you make a two-qubit gate?

The trick is the Rydberg state. A laser kicks an atom's outer electron very far out. The atom swells to many times its normal size. A swollen atom pushes strongly on any close neighbor.

That push creates a rule called the Rydberg blockade. If one atom is swollen, a close neighbor cannot swell too. Think of two people on a narrow bench. If one spreads out, the other cannot. So what happens to one atom depends on the other. That dependence is exactly what a two-qubit gate needs.

The gate this builds is called a CZ (controlled-Z) gate. It flips the sign of the amplitude only when both qubits are 1. The blockade works only over a short distance. So the layout of the atoms decides which pairs can be linked. The chapter on neutral-atom systems shows how geometry becomes the wiring diagram.

Many textbook circuits use a CX gate, which flips one qubit when another is 1. A neutral-atom machine does not do CX directly. Its native gate is CZ.

But an H gate before and after the target qubit turns a CZ into a CX. That is what the circuit above does. The result is an entangled pair. Count the results: about 512 of 1,024 shots show 00 and about 512 show 11, because half of 1,024 is 512. You should see no 01 or 10.

Every machine has its own native gates. A compiler swaps your gates for the ones the machine can run. Read native gates and transpilation to see how.

This is the neat part. The tweezers can move. So a machine can carry atoms across the grid while a program runs.

On a normal chip, qubits are fixed in place. If two far-apart qubits need a gate, the chip must pass their states along a chain of neighbors. That adds extra gates, and each gate adds error. Our lesson on connectivity costs shows the price.

With moving atoms, you can instead carry one atom next to the other. QuEra describes Gemini-class as a dynamic array with movable tweezers and two zones, one to store atoms and one to link them. Our data lists its connectivity as all-to-all.

Moving is not free. It takes time, and atoms can heat up or be lost. So the machine trades extra gates for extra travel time.

Here are the figures from our records. All of them are vendor-reported. Where a figure is missing, the maker has not published it.

QuEra Gemini-class (gate-based):

• 260 physical qubits.
• Two-qubit gate fidelity 99.2%. The page does not say how it was measured.
• One-qubit gate fidelity 99.9% for global gates and 99.7% for local gates.
• 1 shot per second.
• Coherence times: not publicly disclosed.
• First on-site install at AIST in Japan. Not on a public cloud when we checked.

QuEra Aquila (analog):

• Up to 256 qubits.
• Coherence T2* of 5.8 microseconds and T2 echo of 11.4 microseconds, from QuEra's 2023 whitepaper.
• Two-qubit gate fidelity: not publicly disclosed. It is an analog machine. You set how the lasers change over time, instead of giving it a list of gates.
• Open on Amazon Braket.

Atom Computing second-generation system:

• 1,180 qubits on a 1,225-site array, announced 2023-10-24.
• Qubits can store information for 40 seconds, per the press release.
• Two-qubit CZ fidelity 99.72% with post-selection and 99.40% without. Post-selection means throwing away runs where something visibly went wrong, like a lost atom. So the lower number is closer to what every run gets. This comes from a paper written by Atom Computing's own team.
• Readout fidelity and cloud access: not publicly disclosed.

Pasqal Orion Beta (analog):

• Up to 100 qubits.
• At least 0.25 successful runs per second. Pasqal notes the real rate depends on the qubit count.
• Two-qubit gate fidelity and coherence times: not publicly disclosed.
• Machines delivered in 2024 to GENCI in France and to Forschungszentrum Jülich in Germany.

Infleqtion Sqale:

• 100+ physical qubits on the current system.
• Two-qubit CZ fidelity: best shown 99.73% (August 2024). The current system is listed at 99.48% with post-selection and 98.8% raw.
• Arrays of up to 1,600 sites shown. That is a test of array size, not the qubit count of the system you can use.
• Coherence times and readout fidelity: not publicly disclosed.

Notice how often the "best" figure and the "everyday" figure differ. Always check which one you are reading.

Good points:

• Large arrays. Atom Computing's 1,180 qubits is among the largest gate-based counts in our catalog.
• Identical qubits, made by nature.
• Flexible links, because atoms can move.
• Long memory for some designs, like Atom Computing's 40 seconds.

Hard points:

• Slow repeat rates, like Gemini-class at 1 shot per second. Programs that need many shots take a long time.
• Lost atoms, which is one reason post-selected and raw numbers differ.
• Many key figures are not yet published.

Many qubits is not the same as many good qubits. Our qubit count page explains why the number alone can mislead. To see these machines next to other types, read types of quantum computers, or open the compare page.

A logical qubit is one reliable qubit made by spreading information over many physical ones. It is the path to fixing errors. See logical vs physical qubits.

Neutral-atom teams have early results. Atom Computing's 2025 whitepaper cites work with Microsoft that entangled 24 logical qubits and ran an algorithm with 28 logical qubits. Infleqtion lists 8+ logical qubits with error detection on its current system. Detection spots an error but does not fix it, which is a weaker step than correction.

These are real but early steps. For the wider picture, see our page on fault tolerance.