Microsoft is focused on building a utility-scale quantum computer, and that requires fault-tolerant logical qubits. The effort has been underway for more than a decade and approaches the problem holistically by designing hardware, software, and system architecture together as a single engineered stack.

Recently, Nature published Improved quantum processor logical error rates via correction and detection reporting that Microsoft-designed error-correction protocols delivered order-of-magnitude improvements in logical error rates on Quantinuum's state-of-the-art trapped-ion processors—the product of a rewarding technical collaboration between the two teams. Those results were a first step toward a more ambitious goal: a fully productized logical-qubit machine called Magne, built and deployed in partnership with QuNorth in Copenhagen, using Atom Computing's neutral-atom hardware and Microsoft's full-stack quantum software. Magne is expected to be the first general-purpose commercial machine to make logical qubits available to users on demand.

In parallel, we continue to innovate on the error-correcting codes and system architectures needed to scale fault-tolerant computation, as well as the toolchain that makes that work possible. We are also announcing deq, the newest open-source package in the Microsoft Quantum Development Kit (QDK) for error correction—presented at QEC 2026 and now on GitHub under an MIT license.

The Nature paper: foundational error-correction results

The Nature paper reports on two new quantum error-correction codes that Microsoft designed and optimized for an ion-trap architecture—a ⟦12, 2, 4⟧ carbon code (a substantially streamlined construction derived from Knill's C₄/C₆ concatenation) and a ⟦16, 6, 4⟧ tesseract color code on the 4-D hypercube (Figure 1)—together with the fault-tolerant protocols that run them. 

The main results included:

• Logical operations became dramatically more reliable than the physical operations they were built on—by up to roughly 800× on the strongest protocol tested.

• Error correction turned into a worthwhile investment. The logical error rate per round of correction fell below the idle error rate of an unencoded physical qubit—the cost of actively protecting a qubit dropped below the cost of leaving one alone.

• The gains held in multi-qubit, deep-circuit experiments. Error correction worked during computation, not only at the start and end. This was the first demonstration of repeated mid-computation syndrome extraction improving the result rather than degrading it—a prerequisite for running any long fault-tolerant algorithm.

Together, the carbon and tesseract codes cleared the break-even bar—encoded qubits performing better than the physical qubits that compose them—by between one and nearly three orders of magnitude, depending on the protocol (Figure 2). Specifics, including the circuit-level constructions and per-round error rates, are in the Nature paper.

From research to product: the QEC stack for Magne

The Nature experiments were hand-built. Every code choice and every syndrome circuit were tailored to the experimental setup. Turning the same capability into a platform is a different kind of engineering, and it is the work now in flight. Encoding, synthesis, and real-time decoding are becoming distinct, testable, performance-budgeted modules in Microsoft's quantum software stack, with defined interfaces and measurable error budgets. The objective has shifted from setting individual records to making reliable logical qubits available to users on demand.

That platform finds its first commercial home in Magne—the Level 2 Resilient quantum computer being co-designed and built by Microsoft and Atom Computing and operated by QuNorth in Copenhagen—a joint initiative of the Export and Investment Fund of Denmark (EIFO) and the Novo Nordisk Foundation. Magne pairs Atom Computing's neutral-atom hardware—more than 1,200 physical qubits—with Microsoft's full quantum software stack, and is expected to come online early in 2027.

“Quantum Error Correction (QEC) is a definite prerequisite to achieving meaningful quantum advantage with broad application across domains. With Magne, users will finally start to reap the rewards of seamless QEC in the form of more reliable computation, without having to worry about the mechanics of QEC. All they need is to bring their logical circuits.”

—Dr. Kasra Nowrouzi, CTO of QuNorth

In practical terms, Magne will be one of the first systems to enable quantum error correction so that users who want to benefit from logical qubits can do so without the need to implement their own encoding and decoding. Delivering that experience is less about adding more physical qubits than building an end-to-end software stack—synthesis, encoding, decoding, control, compilation, and scheduling—that holds together while more than 1,200 physical qubits do their work.

Microsoft's progress on fault tolerance extends beyond Magne. Our quantum error-correction team continues to design new code families, more efficient decoding strategies, and the system architectures needed to make those innovations practical at the scale a useful algorithm demands. This is all done with the mission of deploying these software innovations on a Majorana-based quantum machine in the near future.

QDK for error correction: deq and an open toolchain for the community

The toolchain that Microsoft Quantum's error-correction researchers use to develop, test, and deploy new error-correction techniques is being open-sourced in the QDK for error correction. These are the same packages that are being incorporated into Magne, and they are being made available to the wider community both for use on Magne and for error-correction research more broadly, to help accelerate the path to fault-tolerant quantum computing.

The QDK for error correction is a family of focused, composable packages. Those already available include high-performance Rust crates and Python packages for binary linear algebra (binar), Pauli and Clifford algebra (paulimer), and fast stabilizer simulation (pauliverse)—the building blocks needed to design, validate, and benchmark a custom code or fault-tolerant gadget before it touches hardware.

The newest open-sourced addition is deq—the first decoding system that makes fault-tolerant computation truly composable. On the physical side, composability is free: A sequence of logical gates is simply the physical gates running back-to-back. Decoding is the opposite. Each logical gate's decoding problem depends on the surrounding circuit, not the gate in isolation, and stitching those problems together across a real computation has, until now, required a human expert to hand label every piece. deq automates that stitching end-to-end, delivering a feature that all scalable proposals assume but none have realized. 

deq also features a modular design that lets the core decoding algorithm be chosen and revisited as an engineering decision. The same deq that targets Magne's classical control plane can be re-targeted to a surface-code experiment in a university lab or used offline to benchmark a custom code under design. Modularity matters here because decoding is the layer at which physics, code distance, classical-compute budget, and end-application accuracy interact; the best decoder for a given workload depends on all four.

Microsoft’s QDK offers capabilities that extend beyond error correction. Its quantum resource estimator allows researchers to test the functionality of quantum algorithms and accurately estimate the quantum resources that would be required to run them. The QDK for error correction, deq, and the rest of the currently available toolchain are on GitHub under an MIT license. Expect continued releases of new open-source QDK for error correction packages as the toolchain grows. 

“Magne will also enable QEC researchers to develop, test, and deploy novel QEC codes. The combination of Microsoft’s QEC tools with openness and modularity, and Atom Computing’s all-to-all connectivity at the scale of 1200+ physical qubits, enables Magne to be an excellent platform to advance QEC, paving the way to future Level 3 fault-tolerant quantum computers.”

—Dr. Kasra Nowrouzi, CTO of QuNorth

If you build quantum software—algorithms, codes, decoders, or the classical control plane—we welcome your participation. Clone the QDK for error correction, try deq, and tell us what you think.

Learn more about Microsoft’s research in quantum error correction

• Read the Nature paper on logical qubits
• Read Dr. Matthias Troyer’s blog post
• Watch the lecture series: Building the Modern Quantum Architecture