⚛️ quantum physics

Scalable quantum error correction tailored for a heavy-hex qubit array

This paper introduces the dynamic compass code, a subsystem error-correcting code tailored for heavy-hex qubit arrays, and experimentally demonstrates that combining its distance-5 implementation with noise-informed decoding strategies—including context-dependent error characterization and measurement soft information—achieves a 38.3% reduction in logical error rates.

Original authors: Seok-Hyung Lee, Xanda C. Kolesnikow, Jun Zen, Evan T. Hockings, Campbell K. McLauchlan, Georgia M. Nixon, Thomas R. Scruby, Stephen D. Bartlett, Robin Harper, Benjamin J. Brown

Published 2026-04-17

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Seok-Hyung Lee, Xanda C. Kolesnikow, Jun Zen, Evan T. Hockings, Campbell K. McLauchlan, Georgia M. Nixon, Thomas R. Scruby, Stephen D. Bartlett, Robin Harper, Benjamin J. Brown

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to build a library of books that never get damaged, even though the librarians are clumsy, the shelves are wobbly, and the air is full of dust. This is the challenge of quantum computing. The "books" are pieces of information called qubits, and the "dust" is noise (errors) that constantly tries to ruin the data.

To fix this, scientists use Quantum Error Correction (QEC). Think of this as a team of librarians who constantly check the books. If a page is torn, they fix it before the story is lost. But here's the catch: the librarians themselves make mistakes, and checking the books takes time.

This paper introduces a new, smarter way to run this library, specifically for a type of quantum computer made by IBM (the "Heavy-Hex" layout). Here is the breakdown of their breakthrough:

1. The New Blueprint: The "Dynamic Compass"

Previous attempts used a static map (the "Heavy-Hex code") to check for errors. It was like having a security guard check every single door in a building at the exact same time every hour. The problem? As the building gets bigger, the guard gets overwhelmed, and the system stops working.

The authors invented the "Dynamic Compass Code."

The Analogy: Instead of checking every door at once, imagine a security team that changes its patrol route every few minutes. Sometimes they check the front doors, sometimes the back, and sometimes they skip a few to focus on the most vulnerable spots.
The Result: This "dynamic" schedule allows the system to scale up to massive sizes without getting overwhelmed. It's like a dance where the steps change to keep the rhythm perfect, no matter how many dancers join in.

2. The Super-Smart Decoder: "Knowing Your Librarian"

Even with a good patrol route, the system needs a "Decoder"—a computer program that looks at the reports from the librarians and figures out what actually went wrong.

Usually, these decoders are like generic GPS apps. They assume the road is always dry and traffic is average. But in the real world, it might be raining, or a specific road might be under construction. If the GPS doesn't know about the rain, it gives you a bad route.

The authors made the decoder context-aware:

The "ACES" Scan: Before running the experiment, they used a tool called ACES to map out the exact "personality" of every single qubit. It's like knowing that Librarian A is clumsy with heavy books, while Librarian B is great at them but slow.
The Benefit: The decoder now knows, "Oh, Librarian A is reporting a problem? That's probably just their clumsiness, not a torn page." This alone improved the system's accuracy by about 30%.

3. Reading the "Whispers": Soft Information

When a quantum computer measures a qubit, it doesn't just shout "Zero!" or "One!". It actually produces a fuzzy signal (called IQ data) that is somewhere in between.

The Old Way: The computer forces a decision: "It's a 0!" or "It's a 1!" immediately. This is like squinting at a blurry sign and guessing the letter.
The New Way (Soft Decoding): The authors let the decoder see the fuzziness. It's like saying, "The sign looks 60% like a 'B' and 40% like an '8'."
The Bonus: This fuzzy signal also reveals if a qubit has "leaked" out of its normal state (like a book falling off the shelf and getting lost in the basement). By spotting these leaks early, they can throw away the bad data runs (a process called post-selection) and only keep the good ones.

The Grand Result

By combining these three things:

A smarter patrol route (Dynamic Compass).
A decoder that knows the specific quirks of the hardware (ACES).
A decoder that listens to the "whispers" of the data rather than just shouting guesses (Soft Information).

The team achieved a 38% reduction in errors for their test system.

Why This Matters

Think of this as the difference between a novice driver and a race car driver.

The novice drives the same way every day, hoping the road is clear.
The race car driver knows exactly which tires are worn, which part of the track is slippery, and can feel the car's vibrations to adjust their speed instantly.

This paper shows that by treating quantum computers like race cars—customizing the strategy to the specific machine and listening to every tiny signal—we can build much more reliable quantum computers. This is a crucial step toward building the massive, error-free quantum supercomputers of the future that could solve problems we can't even imagine today.

1. Problem Statement

The realization of fault-tolerant quantum computing requires logical qubits with error rates significantly lower than physical qubits. However, current quantum hardware (specifically superconducting qubits) suffers from noise, leakage, and device-specific imperfections.

Hardware Constraints: Many quantum processors, such as IBM's Heron architecture, utilize a heavy-hex lattice connectivity. Standard surface codes are difficult to implement efficiently on this topology without increasing circuit depth or qubit count.
Code Limitations: The existing "heavy-hex code" (a subsystem code adapted to this lattice) lacks an error threshold for large distances because its X-type stabilizers grow linearly with code distance, making them ineffective for large-scale error correction.
Decoder Limitations: Standard decoding algorithms often rely on simplified, uniform noise models that do not account for the specific, context-dependent noise characteristics of the actual hardware, leading to suboptimal logical error rates.

2. Methodology

The authors propose a holistic approach combining a new quantum error correction (QEC) code, advanced noise characterization, and noise-informed decoding strategies.

A. The Dynamic Compass Code

Concept: A subsystem code built upon the heavy-hex lattice but utilizing a dynamic measurement schedule.
Mechanism: Unlike static codes, this code employs a sequence of weight-two and weight-four checks over four sub-steps. Crucially, it omits certain two-body X measurements in specific steps.
Advantage: This dynamic schedule ensures that the "detectors" (syndrome extraction circuits) maintain a constant space-time volume even as the code distance increases. This allows the code to possess a threshold, enabling logical error rates to be arbitrarily reduced as the device scales, overcoming the deficiencies of the static heavy-hex code.

B. Noise Characterization (ACES)

Technique: The team uses Averaged Circuit Eigenvalue Sampling (ACES) to characterize the device.
Application: ACES provides a rapid, comprehensive, and context-dependent error model for every gate and layer in the syndrome extraction circuit, including mid-circuit measurements.
Output: It generates a full Pauli error model for the specific hardware configuration, replacing generic calibration data with precise, shot-independent detector error models.

C. Noise-Informed Decoding Strategies

The decoding process integrates three layers of information to improve performance:

Hard Decoding with ACES: Using the ACES-derived noise model to inform a matching decoder (BeliefMatching), replacing uniform noise assumptions.
Soft Decoding: Utilizing raw IQ data (In-phase and Quadrature) from the qubit readout resonators. Instead of a binary 0/1 outcome, the system fits a Gaussian Mixture Model (GMM) to the IQ data to assign probabilities for the qubit being in state $|0\rangle$ , $|1\rangle$ , or the leaked state $|2\rangle$ . These probabilities are fed into the decoder as "soft" information.
Leakage Post-Selection: The soft information allows for the detection of leakage events (qubits falling into $|2\rangle$ ). Runs where the leakage probability exceeds a specific cutoff are rejected (post-selected) before calculating the logical error rate.

3. Key Contributions

Dynamic Compass Code: Introduction of a scalable code with a proven threshold tailored specifically for the heavy-hex lattice, solving the scalability issues of previous heavy-hex variants.
Experimental Demonstration: Implementation of a distance-5 instance of this code on the IBM Quantum Heron (ibm_pittsburgh) processor (156 qubits).
Integrated Noise-Informed Strategy: A novel pipeline combining ACES characterization, soft decoding from analog IQ data, and leakage post-selection.
Quantitative Gains: Demonstration that leveraging detailed device-specific data significantly outperforms standard calibration-based decoding.

4. Results

The experiments were conducted on a distance-5 code instance. The performance was measured by the logical error rate per round ( $\epsilon_L$ ) compared to baselines using global or per-qubit IBM calibration data.

Overall Improvement: The noise-informed approach yielded a 24.2% to 38.3% reduction in logical error rates compared to the global calibration baseline.
- Z-basis: 24.2% improvement.
- X-basis: 38.3% improvement.
Component Analysis:
- ACES Characterization: Using ACES data alone improved the X-basis error rate by 21.0% (statistically significant) but had a negligible effect on the Z-basis (3.0%).
- Soft Decoding & Post-Selection: The most significant gains came from combining soft decoding with leakage post-selection.
  - Z-basis: An additional 16.3% improvement in logical error rate.
  - X-basis: An additional 11.3% improvement.
- Leakage Impact: Post-selection rejected only ~3.5% of runs (abort rate) but significantly reduced logical errors, indicating that leakage is a dominant error source even at low rates.
Baseline Comparison: The "ACES soft + leakage PS" method consistently outperformed both global calibration and per-qubit calibration baselines across different code placements and the standard heavy-hex code.

5. Significance and Future Outlook

Scalability: The Dynamic Compass Code offers a viable path to large-scale fault tolerance on heavy-hex architectures, which are currently being deployed by major quantum hardware providers.
Practical QEC: The study proves that noise-informed decoding is not just a theoretical concept but a practical necessity. By utilizing analog data (IQ) and detailed characterization (ACES), current hardware can achieve much higher logical fidelities without requiring new physical qubits.
Leakage Management: The results highlight that leakage is a critical bottleneck. While post-selection works for near-term experiments, the authors note that scalable solutions will require active leakage reset mechanisms rather than just rejection.
Blueprint for Future Hardware: The techniques demonstrated (ACES, soft decoding, leakage detection) are scalable and provide a blueprint for optimizing QEC on future processors with hundreds of qubits, where device inhomogeneities and leakage will be even more pronounced.

In summary, this paper bridges the gap between theoretical code design and experimental reality by tailoring both the code structure and the decoding algorithm to the specific constraints and noise profiles of current superconducting quantum hardware.