⚛️ quantum physics

Reducing quantum error correction overhead using soft information

This paper demonstrates that leveraging soft information from imperfect measurements significantly enhances quantum error correction performance across various platforms, enabling up to 33% reductions in physical qubit requirements for achieving target logical error rates.

Original authors: Joonas Majaniemi, Elisha S. Matekole

Published 2026-03-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Joonas Majaniemi, Elisha S. Matekole

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 send a secret message across a very noisy room. You have a team of friends (the qubits) who repeat the message to each other to make sure it gets through. However, the room is full of static, and sometimes your friends mishear the message or shout the wrong thing back.

In the world of quantum computers, this "noise" causes errors. To fix this, scientists use Quantum Error Correction (QEC). Think of QEC as a super-smart referee who listens to all your friends, figures out who made a mistake, and corrects the message before it gets ruined.

The Problem: The "Hard" Decision

Traditionally, the referee (the decoder) listens to the friends and forces a binary choice: "Did they say 'Yes' or 'No'?"

If the signal is clear, the referee shouts "YES!" or "NO!" immediately.
But what if the signal is fuzzy? Maybe it sounds 60% like "Yes" and 40% like "No"?
In the old way, the referee just picks the most likely one (e.g., "Yes") and throws away the rest of the information. This is called Hard Decoding. It's like a teacher grading a test with only "Pass" or "Fail," ignoring the fact that a student got 49% right. That 49% of useful info is lost forever.

The Solution: The "Soft" Touch

This paper introduces Soft Information Decoding. Instead of forcing a "Yes" or "No" immediately, the referee says:

"I'm 60% sure they said 'Yes' and 40% sure they said 'No'. Let's keep that uncertainty in mind while we solve the puzzle."

This extra data—the "soft" probability—is like having a confidence meter for every piece of information. It tells the computer, "This part of the message is shaky, but that part is rock solid."

What the Paper Found

The researchers (from a company called Riverlane) tested this idea on two types of quantum computers:

Superconducting Qubits: Like tiny, super-cold electrical circuits (think of them as the "fast cars" of the quantum world).
Neutral Atom Qubits: Like floating atoms held by laser beams (think of them as the "heavy lifters" that can talk to many friends at once).

They simulated these computers using two different types of "referees" (decoders):

LCD (Local Clustering Decoder): A fast, local referee who looks at neighbors.
BP (Belief Propagation): A global referee who passes messages back and forth to solve the whole puzzle.

The Results: Saving Money and Space

The big news is that using "Soft Information" allows the computer to be much more efficient.

The Analogy of the Construction Site:
Imagine you are building a skyscraper (a large-scale quantum computer). To make it stable, you usually need a massive foundation (thousands of extra "physical" qubits) to protect the main structure.
- Without Soft Info: You need a huge, expensive foundation to keep the building from falling over.
- With Soft Info: Because the referee is smarter and uses the "confidence meters," the building is more stable. You can get away with a smaller, cheaper foundation.

The Numbers:

For Superconducting computers, they found they could reduce the number of required physical qubits by 13% to achieve the same reliability.
For Neutral Atom computers, the savings were even bigger: a 33% reduction in the number of qubits needed.

That might not sound like a lot, but in quantum computing, qubits are incredibly expensive and hard to build. Saving 33% is like cutting the cost of a skyscraper by a third.

Speeding Up the Process

Another cool finding is about time.
Usually, to hear a friend clearly in a noisy room, you have to wait a long time for them to finish speaking. This slows everything down.

Old Way: Wait until the signal is perfect, then decide.
New Way: Because the "Soft Decoder" is so good at handling fuzzy signals, you don't have to wait as long. You can make a decision faster without making more mistakes. This means the quantum computer can run faster cycles without crashing.

Why This Matters

We are currently in the "noisy" era of quantum computing. The machines are fragile and make mistakes. To build a useful quantum computer (one that can cure diseases or design new materials), we need to fix these mistakes.

Currently, fixing mistakes requires so many extra parts that the machines are too big and expensive to build. This paper shows a "software upgrade" (Soft Decoding) that acts like a force multiplier. It doesn't require building better hardware; it just requires the computer to be smarter about how it listens to the data it already has.

In a nutshell: By listening to the "shades of gray" instead of just black and white, quantum computers can become smaller, cheaper, and faster, bringing us closer to the day when they can solve real-world problems.

1. Problem Statement

Quantum Error Correction (QEC) is essential for achieving fault-tolerant quantum computing, but current hardware suffers from high physical error rates and imperfect measurements.

The Bottleneck: Standard QEC protocols typically convert analog measurement signals into binary outcomes (0 or 1) before decoding. This "hard decoding" discards valuable probabilistic information about the confidence of the measurement, leading to sub-optimal error correction.
The Challenge: While "soft information" (probabilistic measurement data) has been shown to improve thresholds in small-scale simulations, it remains unclear if it offers significant advantages for large-scale systems, different qubit platforms (superconducting vs. neutral atoms), and different code families (Surface codes vs. LDPC codes). Furthermore, there are concerns about whether soft decoding can be implemented in real-time with the low latency required for future quantum computers.

2. Methodology

The authors developed a comprehensive simulation framework to model soft information decoding across two leading qubit platforms and two distinct error-correcting codes.

A. Soft Measurement Model

Instead of binary outcomes, the model treats measurement signals as random variables ( $\mu$ $μ$ ) following specific probability density functions (PDFs):
- Superconducting Qubits: Modeled as Gaussian distributions based on dispersive readout.
- Neutral Atoms: Modeled as Poisson distributions based on fluorescence photon counts.
Soft Flip Probability: The model calculates the posterior probability $P(1|\mu)$ for each measurement. A "soft flip" occurs when the measurement is misclassified. The authors distinguish between Low-Flip (LF) regimes (where measurement errors are rare, $p_S \approx p$ ) and High-Flip (HF) regimes (where measurement errors are dominant, $p_S = 5p$ ).

B. Decoder Architectures
The study integrated soft information into two real-time compatible decoder architectures:

Local Clustering Decoder (LCD): Used for Surface Codes. The authors modified the graph weights dynamically based on the posterior probabilities of measurement errors.
Belief Propagation (BP) + Ordered Statistics Decoding (OSD): Used for Bivariate Bicycle (BB) codes (a type of LDPC code). Soft information was incorporated by updating the prior probabilities (log-likelihood ratios) of the variable nodes in the Tanner graph without adding virtual nodes, ensuring compatibility with circuit-level noise.

C. Simulation Setup

Platforms: Superconducting transmon qubits (using the Google Quantum AI SI1000 noise model) and Neutral Atom qubits (using a Z-biased noise model).
Codes: Rotated Planar Surface Codes and Bivariate Bicycle (BB) codes.
Metrics:
- $\Lambda$ -factor: The error suppression rate, quantifying how much the logical error rate decreases as code distance increases ( $\epsilon_d \propto \Lambda^{-(d+1)/2}$ ).
- Physical Qubit Footprint: The number of physical qubits ( $N$ ) required to achieve target logical error rates (KiloQuop: $10^{-3}$ and MegaQuop: $10^{-6}$ ).

3. Key Contributions

Platform-Agnostic Validation: Demonstrated that soft information provides consistent error suppression improvements across both superconducting and neutral atom platforms, not just one specific hardware type.
Code Family Generalization: Showed that the benefits of soft decoding extend beyond Surface Codes to high-rate LDPC codes (Bivariate Bicycle codes), reducing logical error probabilities by an order of magnitude.
Real-Time Feasibility: Proposed and validated methods to implement soft decoding in real-time, including:
- Quantizing posterior probabilities to 8-bit integers.
- Using pre-computed Look-Up Tables (LUTs) for edge weight updates to avoid complex runtime arithmetic.
Measurement Time Optimization: Demonstrated that soft decoding allows for shorter measurement times (faster QEC cycles) without sacrificing logical fidelity, a critical advantage for reducing idle errors.

4. Key Results

A. Error Suppression ( $\Lambda$ -factor) Improvements

Superconducting Qubits: In the High-Flip regime (realistic for current mid-circuit measurements), soft LCD decoding improved the $\Lambda$ -factor by 11% compared to hard decoding.
Neutral Atom Qubits: Soft decoding showed a 20% improvement in the $\Lambda$ -factor in the High-Flip regime.
Bivariate Bicycle Codes: Soft BP+OSD decoding reduced logical error probabilities by an order of magnitude (e.g., $30\times$ to $80\times$ reduction depending on code distance and error rate) compared to hard decoding.

B. Physical Qubit Footprint Reduction
The improvements in error suppression translate directly to hardware savings. To achieve a logical error rate of $10^{-6}$ (MegaQuop regime):

Superconducting: Soft decoding reduces the required physical qubit count by 13% (e.g., from 1,681 to 1,457 qubits).
Neutral Atoms: Soft decoding reduces the required physical qubit count by 33% (e.g., from 2,177 to 1,457 qubits).

C. Measurement Time Trade-offs

Soft decoders are more resilient to noisy, short-duration measurements.
For superconducting qubits, measurement times could be shortened by 55% while maintaining the same error suppression performance as hard decoders with optimal (longer) measurement times.
For neutral atoms, measurement times could be shortened by 40%.

5. Significance and Implications

Cost Reduction: The reduction in physical qubit footprint (up to 33%) is a critical step toward making large-scale, fault-tolerant quantum computers economically and physically feasible.
Hardware Efficiency: By allowing shorter measurement times, soft decoding enables faster QEC cycles, which reduces the accumulation of idle errors (decoherence) during syndrome extraction.
Scalability: The work bridges the gap between theoretical soft decoding and practical implementation, proving that high-throughput, low-latency soft decoding is achievable on current decoder architectures (LCD and BP).
Future Roadmap: The paper suggests that as hardware improves, the relative benefit of soft decoding may decrease if measurement fidelity becomes perfect. However, given current limitations, soft information offers a powerful "software" lever to trade measurement fidelity for logical fidelity, allowing manufacturers to build robust systems even with imperfect readout hardware.

In conclusion, this work establishes soft information decoding as a vital tool for the next generation of quantum error correction, offering substantial reductions in hardware overhead and enabling faster, more efficient quantum computing cycles across diverse physical platforms.