Skip to content
Quantum.world

The Signal

Anyon says it ran real-time decoding over NVIDIA NVQLink

Anyon says it ran a real-time error-decoding loop for superconducting quantum processors using NVIDIA NVQLink. The proof-of-concept architecture sends mid-circuit measurement results to a host for decoding and returns corrections to the controller for the next round.

Anyon says it ran real-time decoding over NVIDIA NVQLink

Anyon Computing says it has run a real-time error-decoding loop for superconducting quantum processors using NVIDIA NVQLink. The company describes the work as a proof of concept: measurement results travel from the controller to a host server for decoding, and the resulting correction is applied in the next round.

Measurements to corrections

In the demonstrated cycle, the control system takes mid-circuit measurements and sends their results over a low-latency RDMA network to a real-time host. NVIDIA CUDA-Q Realtime passes the results to a decoder, after which the controller applies the decoder’s correction in the following round.

The setup uses CUDA-Q Realtime to connect GPU-accelerated computing with the processor’s control system. Controllers write measurement results directly into host memory in the format the software expects, so it can read them in place without copying. They connect to the host using RDMA over Converged Ethernet, or RoCE; each controller implements the networking technology in programmable logic and works with standard Linux RDMA software and NVIDIA ConnectX network adapters.

Decoding runs as software on the host rather than as fixed control hardware. Anyon Computing says decoders can be downloaded, installed, updated or replaced—for example, to use a different error-correction code—without changing the control hardware. The same interface is designed to run decoders on either GPUs or CPUs in the host servers.

Built for multiple processors

The broader architecture is designed for systems made from multiple quantum processors connected by quantum interconnects that distribute entanglement. In each error-correction cycle, measurements from the processors can be decoded together on GPUs, with corrections returned to their control systems in real time.

The proof of concept includes controller-side feedback for conditional operations that do not wait for a host computer, and multi-controller operation in which controllers share a reference clock and exchange measurement results over the network. Its listed capabilities also include surface-code logical qubits and lattice surgery between logical qubits on different controllers, as well as union-find, minimum-weight matching and neural-network decoders.

Other listed functions include detecting leaked qubits and accounting for them during decoding. The system is designed to work with Slurm: a real-time host runs as a Slurm node, and jobs can request a quantum processor as a resource alongside GPUs. Programs can be written with NVIDIA CUDA-Q, OpenQASM or QIR.

Architecture planned for sharing

The demonstration builds on a real-time quantum control system Anyon introduced in September 2026.

Anyon cofounder and CEO Jie (Roger) Luo said the demonstration was a step toward interconnected processors coupled to GPU supercomputers, and said the company would share the architecture for other builders to use. NVIDIA Director of Quantum Product Sam Stanwyck said the approach shows how new decoding algorithms can be deployed as software in a real-time control loop.

“Running a real-time error decoding loop with NVIDIA NVQLink and CUDA-Q Realtime is a step toward those systems.”

Jie (Roger) Luo, cofounder and CEO of Anyon Computing

Source

  1. Anyon Computing

Related stories

Explore

More articles