D-Wave shows off its new entry in quantum computing race

Company noted for building quantum annealers now also making gate-based hardware.

D-Wave is a bit of an oddity in the quantum computing space, having been founded back in the last century. And its initial offering wasn’t a quantum computer like those being developed by IBM or Google. Instead, the company built what’s now called a quantum annealer, a machine that isn’t general-purpose but can solve a large class of optimization problems. While the hardware shares some similarities with the qubits used in gate-based quantum computers, it operates in a fundamentally different way.

But a few years back, D-Wave started working on gate-based hardware, apparently choosing a somewhat unusual qubit technology called fluxonium. And this year, the company acquired a startup called Quantum Circuits that spun out of Yale University and has been developing what’s called a dual-rail qubit (the same technology used by Amazon), which promises to make most errors very easy to detect, simplifying error correction.

On Wednesday, the company is publishing a paper in Nature that describes a key step in validating this dual-rail technology, showing that two of the qubits can be entangled without altering their best feature: Most are a single type that is easy to detect.

Resonating

The basic structure of a dual-rail qubit depends on making two linked resonators, which we’ll call left and right. If you place a single photon in the system and measure it, it will always be in either the left or right resonator. But it’s possible to place that single photon in a superposition of both left and right. Those are all the features you need to make a qubit.

The nice thing about the dual-rail setup is that the most common error is simply the photon escaping the hardware. (This is sometimes referred to as an “erasure qubit,” as the loss of the photon erases the information it contains.) The next most common error is flipping the phase of the qubit, with bit flips being a very distant third. Crucially, photon loss can be easily detected with the right hardware, without needing additional qubits required to run an error-correction code.

You’ll still need an error code to handle those other errors. But since they should be less frequent and phase flips will be far more common, it should be possible to use a simpler, more compact error code—one that can potentially be tailored to the relative frequency of the remaining two error types. That means you need to devote fewer hardware qubits to each logical, error-corrected qubit. D-Wave is betting that this will allow it to get useful quantum computation without having to build as much hardware.

But dual-rail qubits aren’t as widely studied as many other approaches, which is why the company is publishing a paper today. The relative frequencies of the three types of errors have primarily been studied in single qubits that are simply sitting still, trying to maintain their initial value. Performing computations will require a lot of operations on qubits, and it’s possible that those operations could have a different hierarchy of errors.

“Often when you start producing two qubit or entangling gates, you can distort your error hierarchy,” D-Wave’s Trevor Lanting told Ars. “So for dominant errors, things other than erasures can start cropping in. One of the fundamental things that we’re very excited about with this two-qubit gate [that] we’re publishing in Nature is that it preserves the error hierarchy.”

(For the curious, Amazon hasn’t needed to do something similar because it chose a system where any connections between dual-rail qubits are mediated by an intervening qubit based on the transmon. As a result, it never has to directly entangle dual-rail qubits.)

Validation

D-Wave set out to do three things in the paper: demonstrate entanglement of two dual-rail qubits, show that the operation was fast enough for practical quantum computing, and show that any errors occurred in roughly the same proportion as they do when using the qubit as memory.

The company set up a pair of dual-rail qubits, each containing two coupled resonators. Between them was a tunable coupler that could be switched on or off. Depending on the state of this coupler, the two qubits would either operate independently or interact. Practically, this involves setting the coupler so that one qubit (termed the control qubit) partially occupies the coupler, allowing it to interact with the second.

Running an entangling gate turned out to be extremely simple: Activate the connection, wait, and then shut it off again. The “wait” portion would not even give you time to check your watch; it’s about 200 nanoseconds, with the entire entanglement taking 500 ns. “We can do these operations in a few hundred nanoseconds, so these are fast operations,” Lanting said. “So not only do you have kind of the high fidelity of the dual rail devices, but you are producing fast entangling operations.”

Critically, it had the same hierarchy of error types. Photon loss rates were about 0.5 percent per entanglement, and they were five times more common than any other type of error. “Bit-flips are practically non-existent at the 10⁻⁶ level,” the paper says. Most of the losses affected the source qubit, which is also allowed to partially enter the coupler.

“The addition of a fast and high-fidelity entangling operation completes the toolbox of gates and operations for dual-rail cavity qubits,” Lanting said.

Work to be done

That doesn’t mean D-Wave’s work is done, of course. To begin with, the devices showed rising error rates as more operations were done. “As the number of gates increases, both the fidelity and purity of the final state exhibit an unexpected approximately quadratic decrease,” the paper acknowledges. “We propose the cause to be drift in calibration parameters or fluctuations in the frequency of the coupling transmon on timescales of experimental shot averaging.”

So there’s more work to do before the system is ready for more complex operations.

Error correction also requires that the hardware support measurements while a calculation is in progress, which a number of other technologies have already demonstrated. That’s still a work in progress for photon-loss detection. “We are working on techniques for doing mid-circuit erasure detection,” Lanting said. “We don’t have this perfected yet, but we’ve got ideas and a plan and a roadmap for detecting these, not just at the end but as you’re running your circuit.”

The other area that needs significant work is the theory of error detection when there are such large differences between classes of errors. The classical compute needed to interpret the error data—called the syndrome—of a logical qubit is substantial. But Lanting said he expects it can be reduced by the dual-rail qubit, providing what he termed a “richer set of information in the data stream that’s flowing off the [quantum processor].”

There’s time to work that out. D-Wave plans to reach 181 dual-rail qubits by 2028, which should be enough time to test several forms of the surface code error-detection scheme. Should that hardware perform as expected, the challenge will then be to make and link enough of it to host a hundred or more logical qubits.

Nature, 2026. DOI: 10.1038/s41586-026-10822-y (About DOIs).