Quantum computers are highly sensitive to noise, and quantum error correction is widely viewed as a necessary foundation for scaling quantum systems from experimental demonstrations toward reliable, useful computation. QXL’s work is focused on one of the central challenges in this transition: developing AI-assisted decoders that can interpret quantum syndrome data efficiently and accurately, and that can continue improving as quantum hardware advances.

The latest results were generated using Google’s public surface-code dataset from a real quantum-hardware experiment. QXL evaluated its updated decoder on a public surface-code configuration using the same cross-validation approach used for Google’s published decoder comparisons.

In this test, QXL’s updated decoder demonstrated improved performance against matching-family benchmarks, including Google’s published correlated-matching and PyMatching benchmark results for the same configuration. Importantly, QXL’s model was trained exclusively on synthetic samples and was not trained on real hardware shots from the Google dataset.

The result supports a key principle behind QXL’s technical roadmap: quantum error-correction decoders should not only perform well in controlled simulations, but should also be able to generalize toward real experimental syndrome data. This synthetic-to-real transition is a critical step toward practical QEC workflows that can support future low-latency and eventually real-time decoding.

"These results are important because they bring us closer to the point where AI-driven quantum error correction can be evaluated against real hardware behavior, not only simulation," said Prof. Nir Sharon, Chief Quantum Technology Scientist at Quantum X Labs. "Our updated decoder improved performance against matching-family benchmarks in this experiment while training only on synthetic data. That is a meaningful validation point for our roadmap toward trusted quantum error correction. At the same time, we remain disciplined: this is one benchmark configuration, and our next objective is to replicate and extend the result across additional device centers and code configurations."

QXL’s updated decoder combines quantum-code structure, syndrome information and AI-based error weighting to improve decoder performance while preserving a practical path toward efficient implementation. The latest result supports the relevance of this approach for real-hardware syndrome data and scalable QEC workflows.

The AI component is designed for GPU acceleration and integration into broader QEC workflows, supporting QXL’s roadmap toward low-latency and eventually real-time decoding. This roadmap includes real-hardware data evaluation, workflows with NVIDIA accelerated computing and NVIDIA CUDA-Q and planned IQCC syndrome experiments to advance more reliable and scalable quantum-computing systems.