The hardest problem in quantum computing isn’t building more qubits. It’s stopping the ones you have from making mistakes. A new peer-reviewed paper in Nature shows that D-Wave’s dual-rail qubits can now entangle with each other cleanly. That’s been a critical missing puzzle piece for solving quantum error correction efficiently. D-Wave says its new qubit design could reduce logical error rates 10X at each step, and clears a path to a commercially useful 100-logical-qubit system by 2032.
The catch: we still need to see if it will scale.
As anyone with even a passing interest in quantum computing knows, qubits are fantastically delicate. A stray photon, a flicker of heat, a whisper of electromagnetic noise, and the fragile quantum state that promises to create the most powerful computing platform we can currently imagine collapses. The industry’s answer has been quantum error correction: to bundle hundreds or even thousands of physical qubits together into a single, more reliable “logical” qubit that can catch and fix its own errors. It works in principle, but the nasty problem is the overhead. Depending on design, a useful machine could need a thousand physical qubits to build one logical qubit you can actually trust … which makes it incredibly hard to build a commercially useful quantum computer.
D-Wave, the Canadian-American company best known for its quantum annealers, thinks it has found a shortcut. And a new paper published this week in the peer-reviewed journal Nature is the first hard evidence the bet might pay off.
The new trick: make errors easy to catch
D-Wave’s gate-model machines use what’s called a dual-rail erasure qubit: a design that comes largely from Yale physicist Rob Schoelkopf, a co-inventor of the transmon qubit – a tiny superconducting electrical circuit cooled to near absolute zero – that most of the industry runs on.
D-Wave acquired his startup, Quantum Circuits Inc., in a $550 million deal that closed in January, and it now anchors the company’s push into gate-model computing.
Here’s the idea: dual-rail qubits.
Essentially, you encode a single bit of quantum information across two superconducting microwave cavities using exactly one shared photon. If that photon leaks away (the most common failure mode) the qubit ends up empty, and you can simply check for that and flag it. In conventional transmon qubits, it doesn’t show up as empty: it bit-flips from a 1 to a 0, and computation continues erroneously. The benefit of the dual-rail qubit is that it takes that typical hard-to-fix quantum computing error and makes it a known, located, understood issue. That’s converting a loss into an erasure, and erasures are much cheaper to correct than silent errors.
The numbers D-Wave is reporting
The performance is legitimately good.
The gate runs in about 500 nanoseconds; it leaks a photon (resulting in an erasure) roughly half a percent of the time; the kind of subtle phase error that’s harder to detect shows up about 0.1% of the time. And the nastiest errors of all, the bit-flips, are suppressed to around one in a million operations. That last number is the crown jewel; the paper notes bit-flips are “practically non-existent.”
Overall two-qubit fidelity lands near 99.9%, which the authors say is “among the best reported for superconducting qubits.”
Also important: the paper shows the qubit’s favorable “error hierarchy” — the property that makes the easy-to-catch errors far more common than the hard ones — survives the entangling gate.
That’s huge, and it’s already operational:
“The entangling gate demonstrated through this research is already integrated into our gate-model systems, where it is delivering comparable performance,” said Rob Schoelkopf, D-Wave’s chief scientist.
But there are some caveats (of course)
This looks like a genuine milestone, but there are a few caveats.
What D-Wave demonstrated is a high-quality gate between two physical qubits. It is not, however, a working logical qubit, or a full error-correction cycle, or running at scale. The biggest headline in D-Wave’s announcement is that the design can cut logical error rates “by as much as a factor of 10 for each increment in error correction.” D-Wave calls this a lambda of 10, and it’s impressive, but it is derived from scaled simulations of the currently demonstrated tech.
When I asked Trevor Lanting, D-Wave’s chief development officer, about that, he was straightforward.
“When we leverage the current gate performance demonstrated in the Nature paper, our simulations indicate a Lambda of 10,” he said. “Experimentally determining Lambda is part of D-Wave’s roadmap to fault-tolerance.”
In addition, the headline fidelity is “post-selected:” the researchers throw out the runs where an error was flagged and report the fidelity of what’s left. That’s a standard thing to do with erasure qubits, but it’s probably not how a finished error-corrected computer would work. The point of the architecture is that eventually you correct those flagged errors instead of just discarding them.
Lanting says that’s exactly the next demonstration.
“Near-term, we plan to demonstrate a distance-3 surface code with full repeated [error correction] cycles, on 17 dual-rail qubits, and with no post-selection, since erasures will be corrected,” he said. He also noted the paper’s supplementary data reproduces the result across seven more qubit pairs in an eight-qubit chip — a sign the gate isn’t a one-off fluke.
That coming 17-qubit demonstration will be the real scaling test.
There’s a business behind the physics
None of this is happening in a vacuum, and D-Wave isn’t alone.
Making error correction cheaper is one of the hottest ideas in the field: Amazon and the Paris-based startup Alice & Bob are chasing a related concept with “cat” qubits, and several groups are pursuing erasure qubits in neutral atoms. D-Wave’s distinction is that it’s now running two horses at once – annealing and gate-model — and it’s the only company doing both.
All of this comes just as D-Wave reported second-quarter results.
D-Wave is still spending aggressively against a startup-scale revenue base, with flat quarterly revenue at $3.1 million but a massive 1,120% year over year jump in first-half bookings to $35.5 million. (Thanks to a $20 million system sale to Florida Atlantic University.) Also, the company’s backlog of contracted work rose 668% to $40.7 million. D-Wave posted a $48 million net loss for the quarter, and operating expenses more than doubled in the first half, likely due to integrating the Quantum Circuits acquisition and staffing up a new gate-model research center in New Haven.
The product roadmap is not short: D-Wave expects to have a 49-qubit system in 2027, 181 qubits in 2028, a 10-logical-qubit machine in 2030, and a 100-logical-qubit system in 2032. That machine, D-Wave says, could run more than a million operations and tackle massive quantum-chemistry and AI problems.
A journey of a thousand miles starts with a single step, however, and the next big step is the 17-qubit chip test that corrects its own errors.
