Home / Quantum Computing / Gothenburg and Chalmers Researchers Develop Modular Graph Neural Decoder for…
Gothenburg and Chalmers Researchers Develop Modular Graph Neural Decoder for Surface Code Memory and Lattice Surgery
quantumcomputingreport.com Oct 29, 2026

Gothenburg and Chalmers Researchers Develop Modular Graph Neural Decoder for Surface Code Memory and Lattice Surgery

Curated by · TechCapsules ·

AI-summarised brief · reviewed before publication

Researchers from the University of Gothenburg and Chalmers University of Technology have introduced a modular graph‑neural‑network (GNN) decoder for surface‑code quantum error correction that can be transferred across different superconducting hardware, noise conditions and multi‑patch logical operations. Validated on IBM’s Nighthawk r1 (ibm_berlin) and Google’s sub‑threshold Sycamore processors, the decoder outperforms minimum‑weight perfect matching in memory tests and successfully decodes a two‑patch ZZ‑merge lattice‑surgery operation. The architecture uses a renormalization‑inspired hierarchy: local GNNs process 3 × 3 × 2 syndrome clusters, followed by lightweight convolutional aggregation and a gated recurrent unit for temporal integration. Parameter count stays near 2‑3 million for code distances d = 3, 5, 7, 9, enabling training on a single commercial GPU while remaining scalable. The team released the preprint on arXiv and open‑source code on GitHub, highlighting a shift toward reusable, hardware‑agnostic neural decoders.

💡 Why It Matters

  • · By breaking the monolithic model barrier, the decoder paves a practical path toward real‑time, fault‑tolerant quantum computing on existing superconducting platforms.
Capsules Quantum Computing
Read full article on quantumcomputingreport.com
More in Quantum Computing →
Zumo and Quantum Chain Partner on Quantum-Resilient Finance
thequantuminsider.com · Oct 29

Zumo and Quantum Chain Partner on Quantum-Resilient Finance

Zumo and Quantum Chain signed a memorandum of understanding to develop quantum‑resilient infrastructure for the UK’s digital asset and blockchain finance sector. The partnership will test quantum‑safe solutions across private networks, tokenised deposits, payments, and institutional digital assets, including a potential quantum‑resistant GBP stablecoin. Both firms aim to provide cryptographic agility and post‑quantum security as the National Cyber Security Centre urges a full migration to post‑quantum cryptography by 2035, amid rising fears of cryptographically relevant quantum computers.

💡 By proactively building quantum‑safe infrastructure, the collaboration addresses the imminent threat of quantum computers that could break current public‑key systems, protecting the integrity of the UK’s growing digital finance ecosystem.

University of Sydney Demonstrates Multi-Qubit Logical Entanglement and 170-Round Stability on IBM Square-Lattice Nighthawk r2 Processor
quantumcomputingreport.com · Oct 29

University of Sydney Demonstrates Multi-Qubit Logical Entanglement and 170-Round Stability on IBM…

Researchers at the University of Sydney demonstrated fault‑tolerant logical operations on IBM Quantum’s 120‑qubit ibm_phoenix processor, the first multi‑qubit logical benchmark on the platform 54 days after its fleet integration. The team validated distance‑3 and distance‑5 rotated surface code memories, prepared logical Bell states via lattice surgery, and performed state and gate teleportation across three logical qubits. They introduced a dynamic “walking” syndrome circuit that alternates data and ancilla roles every round, reducing leakage errors and achieving 170‑round stability.

💡 The work shows that a commercial superconducting system can sustain deep logical operations, a critical step toward scalable quantum computing. It also provides a practical leakage‑mitigation strategy that could be adopted across future quantum processors.

IonQ Receives $7.5 Million from AFRL for Quantum Networking
thequantuminsider.com · Oct 29

IonQ Receives $7.5 Million from AFRL for Quantum Networking

IonQ secured $7.5 million in Phase 2 funding from the Air Force Research Laboratory to upgrade a quantum network first deployed in Rome, New York. The investment adds free‑space optical links, quantum frequency converters, enhanced hardware, and automated control software, enabling connections between diverse quantum systems over standard telecommunications infrastructure and open air. The enhancements aim to create an interoperable research platform for metropolitan and defense environments, advancing AFRL’s goal of practical, heterogeneous quantum networking.

💡 The upgrade demonstrates the feasibility of deploying quantum communication beyond laboratory settings, potentially reshaping secure data exchange for defense and commercial networks.

Yonsei University Researchers Develop Mamba State-Space Neural Decoder to Enable Real-Time Quantum Error Correction
quantumcomputingreport.com · Oct 29

Yonsei University Researchers Develop Mamba State-Space Neural Decoder to Enable Real-Time Quantum…

Yonsei University researchers introduced a Mamba‑based state‑space neural decoder that cuts deep‑learning decoding complexity from O(d⁴) to O(d²) for quantum error correction. Published in npj Quantum Information, the model matches Transformer accuracy on Google Sycamore data while eliminating quadratic self‑attention bottlenecks that hinder real‑time feedback on superconducting hardware. The decoder processes syndrome sequences efficiently, maintaining error‑suppression advantages under simulated latency constraints. Implementation will require porting the architecture to low‑latency FPGAs or ASICs to meet sub‑microsecond cycle budgets.

💡 The new decoder enables practical, real‑time quantum error correction, a critical step toward scalable fault‑tolerant quantum computers.

Guest Post: Why Dedicated Compute Matters for AI and Quantum Computing
thequantuminsider.com · Oct 29

Guest Post: Why Dedicated Compute Matters for AI and Quantum Computing

Christopher Miglino, CEO of Axe Compute, argues that as AI infrastructure expands, it will also underpin future quantum computing. He stresses the need for dedicated GPU environments to meet enterprises’ demands for performance, security, and compliance, especially for sensitive workloads like finance and pharma. Miglino explains that shared cloud models will remain viable, but dedicated compute offers control over cryptographic implementations and network segmentation—critical as quantum threats could compromise current public‑key systems. Axe Compute’s services, Axe Access and Axe Build, provide immediate, scalable dedicated GPU capacity without building on‑premises facilities.

💡 Dedicated compute gives firms the autonomy to safeguard data now and prepare for quantum‑resistant encryption, ensuring continuity of high‑stakes operations that cannot tolerate shared‑resource volatility.

BlueQubit Benchmarks Quantum Computational Advantage in Random-Circuit Sampling on IBM Nighthawk r2 Hardware
quantumcomputingreport.com · Oct 29

BlueQubit Benchmarks Quantum Computational Advantage in Random-Circuit Sampling on IBM Nighthawk r2…

Researchers from BlueQubit, EPFL and the XPRIZE Foundation demonstrated quantum computational advantage using random‑circuit sampling on IBM’s 120‑qubit ibm_phoenix processor, which employs the square‑lattice Nighthawk r2 architecture. The experiment ran forward RCS on 61 active qubits with native CZ gates, submitting circuits via standard Qiskit Runtime without custom calibration. At a depth of 36 cycles (918 CZ gates), the team achieved an XEB fidelity of approximately 2.3 × 10⁻³, generating 10⁶ samples in 19 seconds of QPU time. Reproducing the same output by tensor‑network rejection sampling on Oak Ridge’s Frontier supercomputer would require about 1.2 × 10²⁷ FLOPs, estimated as 110 years of exascale compute. The study also showed a per‑cycle fidelity of 0.872 on Nighthawk r2, a marked improvement over the earlier r1 generation, and confirmed resistance to classical “severing attacks,” underscoring the benchmark’s placement beyond the noise‑induced phase transition.

💡 The result proves that commercially available superconducting hardware can now outperform the world’s fastest classical supercomputers on a specific sampling task, tightening the gap between research‑grade and production‑grade quantum advantage.