What Is Quantum Error Correction?

A physical qubit is the actual hardware realization of a qubit—usually a component or element on a chip: a superconducting circuit, trapped ion, neutral atom, or other quantum system. Physical qubits are prone to errors in quantum gates, phase-flip events, measurement errors, and decoherence—meaning their quantum states decay unpredictably over time, making computations of any practical duration impossible with them alone.

A logical qubit is an abstract qubit, encoded in a group of physical qubits using a quantum error-correction code. There is no single physical location in a quantum processor that one can point to for a logical qubit. Rather, it is both a collection of physical qubits along with the error-correction protocols running to coordinate them. While a single physical qubit may have a physical error rate of 0.1%–1%, a well-implemented logical qubit can achieve a logical error rate (errors occurring when error correction fails) many orders of magnitude lower—making long, complex quantum computations feasible with logical qubits. Reaching utility-scale quantum computing likely requires building systems with at least hundreds or thousands of high-quality logical qubits.

An important concept in quantum error correction (and error correction more generally) is that of an error syndrome. An error syndrome is the information read out from a system that one hopes to use to infer the presence of errors. The key achievement of quantum error correction schemes is that these detectable signatures, the error syndromes, can be measured in a way that, although it destroys the quantum information in some of the physical qubits, does not destroy the overall quantum information encoded in the collective logical qubit. 

There are many different QEC codes, ranging from the foundational Shor code developed by Peter Shor to more modern variants. Each different code has a varying prescription for how to orchestrate physical qubits into logical qubits and how to collect syndromes. For any given QEC code, there are usually also different decoding algorithms that can be employed to infer error locations from a given QEC code’s syndrome data. Two QEC codes are currently very popular in the development of fault-tolerant quantum computing:  

• Surface Codes: The most widely studied QEC code. Physical qubits are arranged in a 2D lattice—well suited for on-chip arrangements of physical qubits; error syndromes are local and easy to measure. Surface codes require a relatively large number of physical qubits per logical qubit but are well-suited to many hardware architectures, often using a process called lattice surgery to perform logical gate operations with the code.
• qLDPC Codes (Quantum Low-Density Parity-Check): A wider class of codes inspired by classical LDPC codes. New qLDPC codes can have significantly lower qubit overhead than surface codes—requiring far fewer physical qubits to encode each logical qubit. qLDPC codes can present other challenges (for example, they may require physical qubits located further away from each other to interact, which can be challenging for some hardware). NVIDIA CUDA-Q™ supports qLDPC decoding, enabling research and deployment of these next-generation codes.
• Color Codes: These are another type of topological code that are well understood. Although they require more physical qubits for memory than surface codes to achieve the same target logical failure rate, they can perform logical gates more efficiently than surface codes due to their ability to do all Clifford gates transversally and the symmetry of their representation resulting in simpler lattice surgery operations. With fast and accurate decoders, it is likely that parameter regimes exist where logic with color codes could be more efficient than logic with surface codes. Get more details in this paper.

Once syndromes are collected, a classical decoding algorithm interprets them and determines the most likely location errors that occurred. This allows error locations to be tracked and, at some point, eventually corrected. The decoder must be fast enough to keep pace with the operation of the quantum processor—a requirement that places enormous demands on classical computing. Decoding too slowly means errors accumulate faster than they can be corrected.

The leading decoding algorithm is BP-OSD (Belief Propagation-Ordered Statistics Decoding), which is particularly effective for qLDPC codes, whereas minimum-weight perfect matching is often used for surface codes. NVIDIA CUDA-Q QEC accelerates BP-OSD decoding with a 29x–35x speedup over industry-standard decoders for a single decoding shot, and up to 42x additional speedup in high-throughput scenarios. Researchers at the University of Edinburgh’s Quantum Software Lab (AutoDEC project) achieved a 2x improvement in both speed and accuracy by running BP-OSD through CUDA-Q QEC.

AI-based decoders are emerging as a complementary approach. QuEra Computing developed a transformer-based decoder using NVIDIA PhysicsNeMo™ and cuDNN™ that delivers a 50x decoding speed boost with improved accuracy—demonstrating how modern AI infrastructure can radically accelerate the classical half of quantum error correction.

QEC only improves logical qubit performance when the physical error rate is below the “fault-tolerance threshold” of the chosen code. Above this threshold, more physical qubits make things worse; below it, more physical qubits make the logical qubit more reliable. Modern superconducting and trapped-ion hardware is approaching or crossing this threshold for leading codes, making real QEC cycles increasingly practical.

Practical QEC requires the classical decoder to receive syndrome data from the QPU, process it, and return correction instructions—all within the qubit’s coherence time. This demands an ultra-low-latency, high-bandwidth link between the quantum processor and the classical decoding system.

NVIDIA NVQLink is an open architecture that provides exactly this: a < 4.0 microsecond round-trip latency between FPGA, GPU, and QPU, with up to 400 Gb/s GPU–QPU throughput. NVQLink supports three key workloads—QPU calibration, QEC decoding, and logical orchestration—and is backed by more than 25 ecosystem partners across QPU hardware, quantum control electronics, and supercomputing infrastructure.