Skip to main content
Quantum computing

Quantum computing

Neutral-atom quantum computers are having a moment

09 Nov 2023
Optics bathed in green and blue light
All systems go: A beam of laser light in Jeff Thompson's lab at Princeton University, where he and his colleagues recently demonstrated a new way of erasing errors in a neutral-atom quantum computer. (Courtesy: Frank Wojciechowski)

In the race for the quantum computing platform of the future, neutral atoms have been a bit of an underdog. While quantum bits (qubits) based on neutral atoms have several attractive characteristics, including the ease of scaling up qubit numbers and performing operations on them in parallel, most attention has focused on rival platforms. Many of the largest machines are built with superconducting qubits, including those developed at IBM, Google, Amazon, and Microsoft. Other companies have opted for ions, like Honeywell and IonQ, or photons, like Xanadu.

In the past few weeks, though, several eye-catching developments have pushed neutral atoms towards the front of the pack. One of them came from a start-up called Atom Computing, which announced in late October that it will soon have a 1000-qubit neutral-atom machine ready for customers – the first commercial quantum device to pass this milestone. The others came from three teams of researchers who published separate studies in Nature describing neutral-atom platforms with low noise, new error mitigation capacities and strong potential for scaling up to even larger numbers of qubits.

For any qubit platform, the biggest barriers to robust quantum operations are noise and the errors it causes. “Error correction is really the frontier of quantum computing,” says Jeff Thompson, a physicist at Princeton University, US who led one of the three studies together with Shruti Puri of Yale University, US. “It’s the thing that’s standing in between us and actually doing useful calculations.”

The reason error correction is so important is that it makes computations possible even if the underlying hardware is prone to noise. Classical computers use a simple error correction strategy called a repetition code: store the same information multiple times so that if there’s an error in one bit, the “majority vote” of the remaining bits will still point to the correct value. Quantum error correction algorithms are essentially more complex versions of this, but before a platform can benefit from them, their hardware must meet some minimal fidelity requirements. For traditional quantum algorithms, the rule of thumb is that the error rate for the minimum unit of quantum computation – a quantum gate – should be below 1%.

Bringing down the noise

Researchers led by Mikhail Lukin of Harvard University, US, are now reporting that their neutral-atom quantum computer has met that threshold, achieving an error rate of 0.5%. They reached this milestone by implementing two-qubit gates in a way pioneered by teams in Germany and France, and their machine, which they developed with colleagues at the neighbouring Massachusetts Institute of Technology (MIT) and QuEra Computing, works as follows.

First, a vapour of rubidium atoms is cooled to just above absolute zero. Then, individual atoms are captured and held by tightly focused laser beams in a technique known as optical tweezing. Each atom represents a single qubit, and hundreds are arranged in a two-dimensional array. The quantum information in these qubits – a zero or one or a quantum superposition of the two – is stored in two different energy levels of the rubidium atoms.

To perform a two-qubit gate, two atoms are brought near each other and simultaneously illuminated by a laser. The illumination promotes one of the atom’s electrons to a high energy level known as a Rydberg state. Once in this state, atoms easily interact with their near neighbours, making the gate operation possible.

To improve the fidelity of the operation, the team used a recently developed optimized pulse sequence for exciting the two atoms to the Rydberg state and bringing them back down. This pulse sequence is faster than previous versions, giving the atoms less chance to decay into the wrong state, which would break the calculation. Combining this with other technical improvements allowed the team to reach 99.5% fidelity for two-qubit gates.

Although other platforms have achieved comparable fidelities, neutral-atom quantum computers can do more computations in parallel. In their experiment, Lukin and his team applied their two-qubit gate to 60 qubits at once simply by illuminating them with the same laser pulse. “This makes it very, very special,” Lukin says, “because we can have high fidelities and we can do it in parallel with just a single global control. No other platform can actually do that.”

Erasing errors

An artist's drawing of five spheres in a line. The spheres represent atoms; four of the atoms are yellow, while one of them glows pink

While Lukin’s team optimized their experiment to meet the fidelity threshold for applying error correction schemes, Thompson and Puri, together with colleagues at the University of Strasbourg, France, found a way to convert certain kinds of errors to erasures, removing them from the system altogether. This makes these errors much easier to correct, lowering the threshold for error-correction schemes to work.

Thompson and Puri’s setup is similar to that of the Harvard-MIT team, with individual ultracold atoms held in optical tweezers. The main difference is that they used ytterbium atoms instead of rubidium. Ytterbium has a more complicated energy-level structure than rubidium, which makes it more difficult to work with but also provides more options for encoding quantum states. In this case, the researchers encoded the “zero” and “one” of their qubits in two metastable states, rather than the traditional lowest two energy levels. Although these metastable states have shorter lifetimes, many of the possible error mechanisms would bump the atoms out of these states and into the ground state, where they can be detected.

Being able to delete errors is a big boon. Classically, if more than half the bits in a repetition code have errors, the wrong information will be transmitted. “But with the erasure model, it’s much more powerful because now I know which bits have had an error, so I can exclude them from the majority vote,” Thompson explains. “So all I need is for there to be one good bit left.”

Thanks to their erasure conversion technique, Thompson and colleagues were able to detect about a third of the errors in real time. Though their two-qubit gate fidelity of 98% is less than that of the Harvard-MIT team’s machine, Thompson notes that they used almost 10 000 times less laser power to drive their gate, and increasing the power will boost the performance while also allowing a larger fraction of errors to be detected. The error erasure technique also lowers the threshold for error correction to below 99%; in a scenario where almost all errors are converted to erasures, which Thompson says should be possible, the threshold could be as low as 90%.

Multiplexing error erasure

In a related result, researchers at the California Institute of Technology, US (Caltech) also converted errors to erasures. Their strontium-based neutral atom machine is a more restricted kind of quantum computer known as a quantum simulator: while they can excite atoms up to the Rydberg state and create entangled superpositions between the ground and Rydberg states, their system has only one ground state, which means they cannot store quantum information long-term.

However, they created these entangled superpositions with unprecedented fidelity: 99.9%. They also made a huge superposition consisting of not just two atoms, but 26, and improved the fidelity of doing so by erasing some of the errors. “We basically show that you could meaningfully bring this technique into the realm of the many-body,” says Adam Shaw, a PhD student in Manuel Endres’ group at Caltech.

Together, the three advances show off the capabilities of neutral-atom quantum computers, and the researchers say their ideas can be combined into a machine that works even better than the ones demonstrated thus far. “The fact that all these works came out together, it’s a little bit of a sign that something special is about to come,” Lukin concludes.

Copyright © 2024 by IOP Publishing Ltd and individual contributors