D-Wave reports 99.9 percent fidelity two-qubit gate for dual-rail superconducting qubits
A gate whose qubits flag their own photon loss shows why error detection built into hardware can shrink the qubit count needed for fault tolerance.
D-Wave has unveiled a hardware two-qubit entangling gate for its dual-rail superconducting qubit architecture, with results published in the journal Nature, The Next Platform reported on 7 August 2026. According to the report, the gate reached about 99.9 percent fidelity with a gate time of roughly 500 nanoseconds.
In the dual-rail design, a unit of quantum information is carried by a single microwave photon shared between two superconducting cavities. Alongside the usual zero, one and superposition states, the encoding has a third detectable state that signals when the photon has been lost. This means the hardware can flag the most common error type, an erasure, as it happens rather than leaving it to be inferred later. Phase flips are less common and bit flips rarer still, the article says.
Simulations cited in the piece indicate the architecture could reduce logical error rates by about ten times with each increment of error-correction strength, which would cut the number of physical qubits needed for each logical qubit. Rob Schoelkopf, co-founder of Quantum Circuits and now D-Wave's chief scientist, said fault-tolerant, error-corrected quantum computing is within reach. D-Wave bought Quantum Circuits earlier in 2026 for US$550 million.
The roadmap described runs from a 49-physical-qubit system in 2026 with a 20-fold error reduction, to a 181-physical-qubit system in 2027 with a 2,000-fold reduction, a 10-logical-qubit machine in 2030 for the first fault-tolerant algorithms, and a 100-logical-qubit computer in 2032 aimed at quantum chemistry and AI. The article also notes that bookings for D-Wave's annealing business reached US$35.5 million in the first half of 2026, up from US$2.9 million a year earlier.
Why it matters
Most quantum roadmaps stall on the overhead of error correction, which can demand thousands of physical qubits for each usable one. Encodings that convert the dominant error into a detectable erasure change that arithmetic in a way fidelity gains alone cannot. The announcement also marks D-Wave as a serious gate-model entrant.
Most quantum roadmaps stall on the overhead of error correction, which can demand thousands of physical qubits for each usable one.
Encode one photon across two cavities
Photon in cavity A or cavity B represents the two basis states; superpositions span both.
Apply the two-qubit gate
About 500 nanoseconds at roughly 99.9 percent fidelity, per the report.
Detect photon loss
If both cavities are empty, the hardware sees a third state and flags an erasure.
Correct the located error
Flagged erasures are cheaper for a code to fix than silent errors.
Scale with fewer physical qubits
Simulations suggest a tenfold logical error cut per correction increment.
What you can learn from this
- Logical qubits are built from many physical ones. A single physical qubit loses its state quickly, so error-correcting codes spread one logical qubit across a group of physical qubits and repeatedly measure check values that reveal errors without disturbing the data. The code's distance, roughly the size of that group, sets how many simultaneous errors it can tolerate. Increasing distance suppresses logical error exponentially but multiplies the hardware needed, which is why the overhead per logical qubit is the central engineering number. In common codes the physical qubit count grows with the square of the distance, so doubling protection can quadruple the hardware.
- Erasures are easier to correct than unlocated errors. An ordinary error corrupts a qubit silently, so the decoder has to work out both where it happened and what it did. An erasure comes with a flag saying which qubit failed, leaving only the correction to determine. Codes can tolerate a substantially higher rate of flagged erasures than of silent errors, so hardware that turns its most common failure into an erasure gets more protection from the same number of qubits.
- Gate fidelity has to clear a threshold before scaling helps. Every quantum code has a threshold error rate; below it, adding qubits makes the logical qubit better, and above it, adding qubits makes things worse. A two-qubit gate around 99.9 percent means roughly one error in a thousand operations, which is in the regime where common codes start to pay off. Fidelity and erasure conversion work together: one lowers the raw error rate, the other makes the remaining errors cheaper to fix. Reported fidelities are averages over many runs, so the worst-performing gates on a chip matter as much as the headline figure when estimating real overhead.
- Dual-rail encoding uses presence and position of a photon. The photon being in cavity A represents one state and in cavity B the other, with superpositions across both. If the photon leaks away, both cavities are empty, a condition that matches neither valid state and can be detected by measurement. This is how a physical loss mechanism becomes a clean, flagged erasure rather than a hidden corruption.
- Annealers and gate-model machines are different tools. A quantum annealer relaxes a physical system toward a low-energy configuration and is suited to certain optimisation problems, but it does not run arbitrary quantum circuits. Gate-model machines apply a programmable sequence of operations and are what general algorithms, including quantum chemistry simulations, are written for. A company offering both is covering two distinct markets with different maturity curves. The two also scale differently: annealers already field thousands of qubits, while gate-model counts remain small because each qubit must meet far stricter fidelity requirements.
We teach this
How to use this in practice
- Simulate a three-qubit repetition code. Using a free circuit simulator such as Qiskit or Cirq, encode one qubit into three, deliberately flip one of them, measure the two parity checks and apply the correction. Done looks like a notebook where the recovered qubit matches the original for every single-qubit flip you inject. Then flip two qubits at once and observe that the code decodes to the wrong answer, which demonstrates the limit of a distance-three code. Keep the notebook, since it becomes the starting point for trying a phase-flip code next.
- Tabulate error suppression by distance. Assume a starting logical error rate and a tenfold reduction per increment of distance, as the report's simulations suggest, and compute the rate for three successive increments. Then estimate how many physical qubits a code of each distance would need using the simple rule that qubit count grows with the square of distance. Done looks like a small table with two columns that shows why overhead, not just error rate, is the story. Plot the two columns against each other if you can, since the shape of that curve is the whole argument for erasure conversion. Record the assumptions you made so the table can be revisited if the real numbers differ.
- Draw the dual-rail encoding. Sketch two cavities side by side, show the photon in the left one, in the right one, and absent from both, and label which of those three pictures are valid states and which is the flagged error. Done looks like a diagram a colleague could use to explain why this architecture detects loss. Add a caption explaining why a silent phase flip would not show up in the same picture, and number the three pictures so the caption can refer to them.
- Read the abstract of the Nature paper and translate it. Find the paper through the journal's website, read only the abstract, and write five plain-English sentences covering what was built, what was measured, and what the authors claim it enables. Done looks like a short paragraph with no jargon that a non-physicist could follow. Compare your summary with the numbers in the news report and note any claim that appears in one but not the other.
Sources
- D-Wave Intros Two-Qubit, Error Correcting Gate For Its Dual-Rail Quantum Architecture — The Next Platform
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report