OmniQEC
build reliability.

OmniQEC is an open research platform for discovering, evaluating, and deploying quantum error-correction stacks — from code search to syndrome circuits, decoders, and hardware execution.

A framework for autonomous quantum error correction.

OmniQEC’s AI agents discover codes that suit real processors, build decoders accurate and fast enough to run in real time, and deploy the resulting stack on hardware. Every step is judged end to end, under circuit-level noise and hardware constraints, rather than layer by layer.

Code discovery

Discover

An AI scientist that searches for quantum error-correcting codes suited to real processors.

Preprint · arXiv, July 2026

OmniQEC

Discovering practical quantum error-correcting codes by an AI scientist

Ge Yan, Shanchuan Li, Pengyue Ma, Qixin Zhang, Pingchuan Ma, Jianping Wang, Min-Hsiu Hsieh, Yuxuan Du

Background

A code is practical only if it works on real hardware, and that depends jointly on its structure, the processor, syndrome extraction and decoding, demands that often conflict.

OmniQEC is an AI scientist for this search. An LLM-based orchestrator coordinates code generation, code-level screening, syndrome-extraction synthesis and decoder-based circuit evaluation in a two-speed loop: a fast loop screens many candidates with inexpensive code-level proxies, and a slow loop evaluates the promising ones at the circuit level and feeds the results back into the search.

Read the paper ↗PDF · arXiv:2607.25865
Against BB codes
[[72,12,6]] · [[144,12,12]]
Both outperformed at complete-implementation budgets of 98 and 240 physical qubits.
Scaling
Better with budget
Logical-error suppression improves as the physical-qubit budget grows.
Search
14
Physical-qubit budgets per LLM backend, across four qLDPC families and three backends.
Neural decoding

Decoders

Neural decoders accurate enough for fault tolerance and fast enough to run in real time.

Preprint · arXiv, June 2026

NTU decoder

Efficient foundation decoders for fault-tolerant quantum computing

Ge Yan, Shanchuan Li, Shiyi Xiao, Pengyue Ma, Hanyan Cao, Feng Pan, Yuxuan Du

Background

Foundation decoders, high-capacity neural decoders, are leading candidates for fault-tolerant quantum computing, but building them at large code distances runs into a steep scaling barrier: syndrome generation and neural optimisation costs rise quickly.

Neural transfer unification (NTU) aligns decoding tasks across code distances using the algebraic structure shared by scalable code families, so what a decoder learns on smaller codes speeds up training on larger ones. It is implemented as NTU-Transformer for planar surface codes and bivariate bicycle codes.

Read the paper ↗Project pageCode
Surface code
[[361,1,19]]
Outperforms correlation-aware matching under circuit-level noise.
Transfer
[[625,1,25]]
Reached through transfer adaptation, exceeding standard matching.
BB code
[[72,12,6]]
Surpasses Relay-BP in the low-physical-error regime.
ICML 2026

Rethink

Rethink the Role of Neural Decoders in Quantum Error Correction

Ge Yan, Shanchuan Li, Yuxuan Du

INT4

Background

Neural decoders can decode surface codes accurately, but to be useful they must also meet the fundamental accuracy–latency trade-off of real hardware, often on a microsecond timescale.

This work unifies and redesigns representative neural decoders into five architectural paradigms, and builds an end-to-end compression pipeline to test their deployability and performance on FPGA hardware, at code distances up to d = 9 (161 physical qubits).

Read the paper ↗ICML 2026 poster
Data
Scale beats complexity
Near-term decoding performance depends more on data scale than on architectural complexity.
Design
Inductive bias matters
An appropriate inductive bias is necessary for high decoding accuracy.
Latency
INT4 on FPGA
INT4 quantization is a prerequisite for microsecond-scale latency on FPGAs.
Hardware deployment

Deploy

Carrying discovered codes and their decoders onto quantum hardware.

Coming soon