Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2611.12833

GARI++: A General-Purpose Decoder for Quantum Error Correction

Arshpreet Singh Maan, Francisco García-Herrero, Daniel Báscones, Valentin Savin, Alexandru Paler

Published Nov 9, 2026• 0 citations
arXivHTMLPDF

Abstract

We present GARI++, a two-stage decoding pipeline achieving accuracy competitive with the state of the art across surface, bivariate bicycle and color codes without relying on ensembles. The first stage uses GARI, a normalized min-sum decoder with random serial scheduling on the GARI transformed matrix. The second stage decoder is used only for the cases that the first stage does not converge, and the decoder is configurable based on the choice of the QEC code. The first stage pipelines towards the second stage both the residual syndrome and the soft information. GARI++ pipelining is predictive, meaning that GARI++ chooses the first stage iteration outputs which are expected to deliver the best accuracy in the second stage. GARI++ motivates a hybrid FPGA-CPU architecture for real-time decoding. For example, under uniform circuit-level depolarising noise with a physical error rate of 0.001, the FPGA implementation of GARI achieves average latencies per round ranging from $201 ns$ to $11.8μs$ for surface code distances between 7 and 25. At the same time, due to the very low weight residuals output by the first stage, the average residual decoding time per round remains near constant and below one microsecond on a single Apple M2 core for surface code distances even up to 51 when using Union-Find and accuracy on par with correlated MWPM.

DOI
10.48550/arxiv.2611.12833

Related papers

A Scalable FPGA Architecture for Real-Time Decoding of Quantum LDPC Codes Using GARI

Daniel Báscones, Arshpreet Singh Maan, Valentin Savin +1 more

In this work, we introduce a new hardware architecture for decoding correlated errors in quantum LDPC codes. The decoder is based on message passing and exploits the structure of the detector error model obtained through the recently introduced Graph Augmentation and Rewiring for Inference (GARI) method. The proposed architecture enables flexible scaling and can, in principle, adapt to any quantum LDPC codes using the GARI framework. It leverages resource reuse while maintaining a modest degree of parallelism, thereby reducing power consumption and area requirements, while preserving low decoding latency. As a case study, the architecture was implemented on a VCU19P FPGA as an ensemble of three decoder cores targeting the [[144,12,12]] bivariate bicycle code, achieving an average latency of 596 ns per decoding round. This implementation consumes six times fewer resources than the previous GARI-based proposal, being the first reported implementation of multiple decoder cores for correlated errors on a single FPGA device. This enables better energy-conscious scaling of the quantum error correction layer on the classical side, reducing overall power consumption while meeting real-time constraints without compromising decoding accuracy under correlated errors.

quant-pharXiv:2605.01035•May 5, 2026•9 citations

Low Latency GNN Accelerator for Quantum Error Correction

Alessio Cicero, Luigi Altamura, Moritz Lange +2 more

Quantum computers can solve selected problems more efficiently than classical computers, but current devices are limited by high physical error rates. Quantum Error Correction (QEC) mitigates this by encoding many physical qubits into a logical qubit, with the surface code among the most widely studied approaches. Since syndrome measurements are produced continuously, the decoder must process them fast enough to avoid becoming a system bottleneck, making real-time decoding essential for fault-tolerant quantum computing. While most state-of-the-art real-time decoders rely on Minimum-Weight Perfect Matching (MWPM), we instead use a high-accuracy Graph Neural Network (GNN) that trades higher computational cost for lower logical error rates. To make this GNN practical for real-time decoding, we apply algorithm-hardware co-design. We reduce complexity through hardware-guided pruning and retraining, producing two hardware-friendly models that reduce parameter count by $3.1\times$ and $6.5\times$. These target, respectively, an average decoding latency of one syndrome cycle and a worst-case latency within one syndrome cycle. We further reduce hardware cost using input-graph filtering and post-training quantization. Based on these optimized models, we propose an FPGA-based architecture for low-latency inference and real-time decoding. Evaluated on surface codes up to distance 7 under circuit-level noise at physical error rate $p=10^{-3}$, our decoder outperforms MWPM in decoding accuracy for both average-latency and max-latency settings. It reduces logical error rate by 40% at $1μs$ average latency, and by 13% under a strict $1μs$ deadline.

quant-pharXiv:2603.22149•Mar 24, 2026•4 citations

Zero-G: A Pre-Decoder-Aware Decoder for Quantum Error Correction

Peter Wegmann, Theofilos Augoustis, Aleksandra Świerkowska +2 more

Fault-tolerant quantum computing requires classical decoders that keep pace with the underlying hardware, translating syndrome measurements into corrections fast enough to avoid an exponential backlog. To meet this real-time constraint, pre-decoders have emerged as part of a hierarchical decoding approach to resolve simple, local errors before passing a sparser residual syndrome to a strong decoder. While pre-decoding should, in theory, speed up the strong decoder, in practice, the speedup is only marginal, since existing strong decoders are designed to decode dense syndromes and cannot exploit the sparsity provided by pre-decoders. To address this, we present Zero-G, a strong decoder designed for use alongside pre-decoders. As a stochastic approximate minimum-weight perfect matching (MWPM) decoder, Zero-G exploits sparse residual syndromes, dynamically trading latency for accuracy rather than relying on an all-or-nothing runtime-accuracy trade-off. By decoupling hardware control from the decoding core itself, we enable heterogeneous deployment across both FPGAs and CPUs without maintaining separate implementations. Zero-G achieves a $10\times$ latency improvement over existing strong decoders at matching accuracy, with worst-case sub-350ns decoding at code distances up to d=15, while scaling to 640 logical qubits on a single 128-core CPU and 32 logical qubits on a single AMD Versal V80 FPGA.

quant-pharXiv:2608.02030•Aug 4, 2026•1 citation

Accelerating A*-Based Algorithms for Decoding Quantum Low-Density Parity-Check Codes

Lamia Yous, Francisco Garcia Herrero, Mark F. Flanagan

Quantum low-density parity-check (QLDPC) codes represent a promising approach for error correction in quantum computing. The recently proposed Tesseract decoder uses the A* search algorithm that guarantees finding the most likely error pattern. However, practical implementations of Tesseract often involve an extremely large graph, and the inherently sequential nature of the search results in high computational overhead and long runtime. To improve decoding efficiency, we propose a two-stage decoding framework. First, a belief propagation (BP) decoder efficiently processes the syndrome. This step generates hard decisions (a binary error vector) and soft information (per-qubit confidence levels). In non-convergent BP cases, a gating mechanism examines the BP decoder's output to identify and filter out qubits with oscillating confidence values. In these cases, the refined output serves as input to Tesseract, which then attempts to identify and correct any residual errors. This hybrid approach leverages the high speed of BP decoding and delegates the more challenging decoding instances to Tesseract. Numerical results demonstrate a substantial reduction in overall decoding complexity while maintaining the logical error rate (LER) of the stand-alone Tesseract. Across all tested physical error rates, the proposed method achieves at least 5x reduction in the number of expanded nodes in the Tesseract, with a peak reduction of nearly 15x at a physical error rate of p = 0.05 for the [[126, 12, d < 11]] T1 code, and approximately 8.8x at p = 0.05 for the [[72, 12, 6]] bicycle bivariate (BB) code.

quant-pharXiv:2609.10056•IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC)•Sep 10, 2026•0 citations

Exploring the FPGA and ASIC design space of belief propagation and ordered statistics decoders for quantum error correction codes

Daniel Báscones, Francisco García-Herrero, Javier Valls

Belief propagation (BP) combined with ordered statistics decoding (OSD) provides a good balance between accuracy and complexity for many quantum error correction (QEC) codes, making it nearly universal. However, the complexity of OSD can limit real-time decoding, particularly for superconducting qubits, and the limits of classical hardware decoders have not been fully explored. Therefore, it is important to assess the architecture of OSD for different code families, such as surface codes and bicycle bivariate codes, under realistic assumptions like the detector error model. This paper introduces a BP + OSD parallel architecture implemented on FPGA and ASIC for surface codes (distances 3–21) and bicycle bivariate codes (distances 6–24). Results show that for surface codes up to distance 9, the OSD post-processor fits into a single VCU129 FPGA, achieving a frequency of 200 MHz with a worst-case latency of 134 μ s. For bicycle bivariate codes, the limit is distance 12, with a frequency of 244 MHz and a worst-case latency of 84 μ s. In ASIC, with 45 nm technology, latency improves by 31%, but area resources grow significantly, making parallel implementation beyond distance 12 impractical on a single chip. The designs were verified using a hardware emulator, ensuring that the decoder’s behavior matches software simulations and revealing interesting results like potential error floors at low logical error rates.

EPJ Quantum Technology•Nov 20, 2025•2 citations

Rethink the Role of Neural Decoders in Quantum Error Correction

Ge Yan, Shanchuan Li, Yuxuan Du

Quantum error correction (QEC) is essential for enabling quantum advantages, with decoding as a central algorithmic primitive. Owing to its importance and intrinsic difficulty, substantial effort has been made to QEC decoder design, among which neural decoders have recently emerged as a promising data-driven paradigm. Despite this progress, practical deployment remains hindered by a fundamental accuracy-latency tradeoff, often on the microsecond timescale. To address this challenge, here we revisit neural decoders for surface-code decoding under explicit accuracy-latency constraints, considering code distances up to d=9 (161 physical qubits). We unify and redesign representative neural decoders into five architectural paradigms and develop an end-to-end compression pipeline to evaluate their deployability and performance on FPGA hardware. Through systematic experiments, we reveal several previously underexplored insights: (i) near-term decoding performance is driven more by data scale than architectural complexity; (ii) appropriate inductive bias is essential for achieving high decoding accuracy; and (iii) INT4 quantization is a prerequisite for meeting microsecond-scale latency requirements on FPGAs. Together, these findings provide concrete guidance toward scalable and real-time neural QEC decoding.

quant-pharXiv:2605.12046•May 13, 2026•8 citations
We present GARI++, a two-stage decoding pipeline achieving accuracy competitive with the state of the art across surface, bivariate bicycle and color codes without relying on ensembles. The first stage uses GARI, a normalized min-sum decoder with random serial scheduling on the GARI transformed matrix. The second stage decoder is used only for the cases that the first stage does not converge, and the decoder is configurable based on the choice of the QEC code. The first stage pipelines towards the second stage both the residual syndrome and the soft information. GARI++ pipelining is predictive, meaning that GARI++ chooses the first stage iteration outputs which are expected to deliver the best accuracy in the second stage. GARI++ motivates a hybrid FPGA-CPU architecture for real-time decoding. For example, under uniform circuit-level depolarising noise with a physical error rate of 0.001, the FPGA implementation of GARI achieves average latencies per round ranging from 201ns201 ns201ns to 11.8μs11.8μs11.8μs for surface code distances between 7 and 25. At the same time, due to the very low weight residuals output by the first stage, the average residual decoding time per round remains near constant and below one microsecond on a single Apple M2 core for surface code distances even up to 51 when using Union-Find and accuracy on par with correlated MWPM.
In this work, we introduce a new hardware architecture for decoding correlated errors in quantum LDPC codes. The decoder is based on message passing and exploits the structure of the detector error model obtained through the recently introduced Graph Augmentation and Rewiring for Inference (GARI) method. The proposed architecture enables flexible scaling and can, in principle, adapt to any quantum LDPC codes using the GARI framework. It leverages resource reuse while maintaining a modest degree of parallelism, thereby reducing power consumption and area requirements, while preserving low decoding latency. As a case study, the architecture was implemented on a VCU19P FPGA as an ensemble of three decoder cores targeting the [[144,12,12]] bivariate bicycle code, achieving an average latency of 596 ns per decoding round. This implementation consumes six times fewer resources than the previous GARI-based proposal, being the first reported implementation of multiple decoder cores for correlated errors on a single FPGA device. This enables better energy-conscious scaling of the quantum error correction layer on the classical side, reducing overall power consumption while meeting real-time constraints without compromising decoding accuracy under correlated errors.
Quantum computers can solve selected problems more efficiently than classical computers, but current devices are limited by high physical error rates. Quantum Error Correction (QEC) mitigates this by encoding many physical qubits into a logical qubit, with the surface code among the most widely studied approaches. Since syndrome measurements are produced continuously, the decoder must process them fast enough to avoid becoming a system bottleneck, making real-time decoding essential for fault-tolerant quantum computing. While most state-of-the-art real-time decoders rely on Minimum-Weight Perfect Matching (MWPM), we instead use a high-accuracy Graph Neural Network (GNN) that trades higher computational cost for lower logical error rates. To make this GNN practical for real-time decoding, we apply algorithm-hardware co-design. We reduce complexity through hardware-guided pruning and retraining, producing two hardware-friendly models that reduce parameter count by 3.1×3.1\times3.1× and 6.5×6.5\times6.5×. These target, respectively, an average decoding latency of one syndrome cycle and a worst-case latency within one syndrome cycle. We further reduce hardware cost using input-graph filtering and post-training quantization. Based on these optimized models, we propose an FPGA-based architecture for low-latency inference and real-time decoding. Evaluated on surface codes up to distance 7 under circuit-level noise at physical error rate p=10−3p=10^{-3}p=10−3, our decoder outperforms MWPM in decoding accuracy for both average-latency and max-latency settings. It reduces logical error rate by 40% at 1μs1μs1μs average latency, and by 13% under a strict 1μs1μs1μs deadline.
Fault-tolerant quantum computing requires classical decoders that keep pace with the underlying hardware, translating syndrome measurements into corrections fast enough to avoid an exponential backlog. To meet this real-time constraint, pre-decoders have emerged as part of a hierarchical decoding approach to resolve simple, local errors before passing a sparser residual syndrome to a strong decoder. While pre-decoding should, in theory, speed up the strong decoder, in practice, the speedup is only marginal, since existing strong decoders are designed to decode dense syndromes and cannot exploit the sparsity provided by pre-decoders. To address this, we present Zero-G, a strong decoder designed for use alongside pre-decoders. As a stochastic approximate minimum-weight perfect matching (MWPM) decoder, Zero-G exploits sparse residual syndromes, dynamically trading latency for accuracy rather than relying on an all-or-nothing runtime-accuracy trade-off. By decoupling hardware control from the decoding core itself, we enable heterogeneous deployment across both FPGAs and CPUs without maintaining separate implementations. Zero-G achieves a 10×10\times10× latency improvement over existing strong decoders at matching accuracy, with worst-case sub-350ns decoding at code distances up to d=15, while scaling to 640 logical qubits on a single 128-core CPU and 32 logical qubits on a single AMD Versal V80 FPGA.
Quantum low-density parity-check (QLDPC) codes represent a promising approach for error correction in quantum computing. The recently proposed Tesseract decoder uses the A* search algorithm that guarantees finding the most likely error pattern. However, practical implementations of Tesseract often involve an extremely large graph, and the inherently sequential nature of the search results in high computational overhead and long runtime. To improve decoding efficiency, we propose a two-stage decoding framework. First, a belief propagation (BP) decoder efficiently processes the syndrome. This step generates hard decisions (a binary error vector) and soft information (per-qubit confidence levels). In non-convergent BP cases, a gating mechanism examines the BP decoder's output to identify and filter out qubits with oscillating confidence values. In these cases, the refined output serves as input to Tesseract, which then attempts to identify and correct any residual errors. This hybrid approach leverages the high speed of BP decoding and delegates the more challenging decoding instances to Tesseract. Numerical results demonstrate a substantial reduction in overall decoding complexity while maintaining the logical error rate (LER) of the stand-alone Tesseract. Across all tested physical error rates, the proposed method achieves at least 5x reduction in the number of expanded nodes in the Tesseract, with a peak reduction of nearly 15x at a physical error rate of p = 0.05 for the [[126, 12, d < 11]] T1 code, and approximately 8.8x at p = 0.05 for the [[72, 12, 6]] bicycle bivariate (BB) code.
Belief propagation (BP) combined with ordered statistics decoding (OSD) provides a good balance between accuracy and complexity for many quantum error correction (QEC) codes, making it nearly universal. However, the complexity of OSD can limit real-time decoding, particularly for superconducting qubits, and the limits of classical hardware decoders have not been fully explored. Therefore, it is important to assess the architecture of OSD for different code families, such as surface codes and bicycle bivariate codes, under realistic assumptions like the detector error model. This paper introduces a BP + OSD parallel architecture implemented on FPGA and ASIC for surface codes (distances 3–21) and bicycle bivariate codes (distances 6–24). Results show that for surface codes up to distance 9, the OSD post-processor fits into a single VCU129 FPGA, achieving a frequency of 200 MHz with a worst-case latency of 134 μ s. For bicycle bivariate codes, the limit is distance 12, with a frequency of 244 MHz and a worst-case latency of 84 μ s. In ASIC, with 45 nm technology, latency improves by 31%, but area resources grow significantly, making parallel implementation beyond distance 12 impractical on a single chip. The designs were verified using a hardware emulator, ensuring that the decoder’s behavior matches software simulations and revealing interesting results like potential error floors at low logical error rates.
Quantum error correction (QEC) is essential for enabling quantum advantages, with decoding as a central algorithmic primitive. Owing to its importance and intrinsic difficulty, substantial effort has been made to QEC decoder design, among which neural decoders have recently emerged as a promising data-driven paradigm. Despite this progress, practical deployment remains hindered by a fundamental accuracy-latency tradeoff, often on the microsecond timescale. To address this challenge, here we revisit neural decoders for surface-code decoding under explicit accuracy-latency constraints, considering code distances up to d=9 (161 physical qubits). We unify and redesign representative neural decoders into five architectural paradigms and develop an end-to-end compression pipeline to evaluate their deployability and performance on FPGA hardware. Through systematic experiments, we reveal several previously underexplored insights: (i) near-term decoding performance is driven more by data scale than architectural complexity; (ii) appropriate inductive bias is essential for achieving high decoding accuracy; and (iii) INT4 quantization is a prerequisite for meeting microsecond-scale latency requirements on FPGAs. Together, these findings provide concrete guidance toward scalable and real-time neural QEC decoding.