Error Correction
Quantum LDPC Code
A quantum error-correcting code whose checks each involve only a few qubits, promising much lower qubit overhead than the surface code.
Definition
Quantum LDPC codes are error-correcting codes in which every check touches only a handful of qubits and every qubit is touched by only a handful of checks. This sparse structure keeps the circuits simple, and the codes can protect many logical qubits at once with far fewer physical qubits than the surface code needs. IBM's roadmap to a fault-tolerant machine is built on this idea, though it requires longer-range connections between qubits than a simple grid provides.
In depth
The surface code is easy to build because it needs only nearest-neighbor connections on a grid, but it is wasteful: a logical qubit with distance d needs about 2 d squared physical qubits, and the rate of logical qubits per physical qubit goes to zero as d grows. Classical LDPC codes, introduced by Gallager in 1962, achieve constant rate and good distance with sparse checks, and the quantum community searched for analogs for decades. Tillich and Zémor's hypergraph product codes in 2009 gave constant rate with distance growing as the square root of n, and Panteleev and Kalachev's 2021 result produced the first asymptotically good qLDPC codes, with both rate and distance growing linearly.
IBM's 2024 Nature paper on high-threshold and low-overhead fault-tolerant quantum memory proposed bivariate bicycle codes that fit practical sizes. The gross code stores 12 logical qubits with distance 12 in 144 data qubits and 144 check qubits, roughly a tenfold saving over the surface code for similar protection. The price is connectivity: the checks need long-range couplers, and the qubit layout is not planar, so chips need extra wiring layers. IBM's Loon processor, delivered in 2025, was built to test those components.
Decoding qLDPC codes is a challenge too, because the structure is less regular than a grid. Decoders based on belief propagation combined with ordered statistics, known as BP-OSD, work well in simulation, and IBM has developed faster variants for real-time use. Performing logical gates on these codes also needs new techniques, such as generalized lattice surgery and code switching, which are active research topics.
Technical Definition
A quantum low-density parity-check (qLDPC) code is a stabilizer code, often of CSS type, whose stabilizer generators have constant weight and each qubit participates in a constant number of generators. Good qLDPC codes, shown by Panteleev and Kalachev in 2021, have constant rate and linear distance. IBM's bivariate bicycle codes include the [[144, 12, 12]] gross code, which uses 288 physical qubits including check qubits to encode 12 logical qubits.
Visual Explanation: An Analogy
The surface code is like protecting each valuable item in its own safe, one safe per item. A qLDPC code is like a well-designed shared vault where many items are protected together by a network of simple locks, which needs far less steel but requires a more complicated floor plan.
Key facts
| Item | Detail |
|---|---|
| Property | Constant-weight checks, with each qubit in a constant number of checks |
| Gross code | [[144, 12, 12]], using 288 physical qubits including check qubits |
| Saving vs surface code | About 10 times fewer qubits for comparable protection |
| Good codes | Panteleev and Kalachev, 2021: constant rate and linear distance |
| Hardware need | Long-range couplers or non-planar connectivity |
Where the idea comes from
Classical LDPC codes date to Gallager's 1962 thesis and became widely used in communications in the 1990s. Quantum versions were proposed in the 2000s, the existence of good codes was proven in 2021 and 2022, and IBM adopted bivariate bicycle codes as the core of its 2025 roadmap to the Starling fault-tolerant machine.
Real-World Use Cases
- Reducing the physical qubit count of fault-tolerant machines.
- Designing chip architectures with long-range couplers.
- Developing decoders and logical gate schemes for non-planar codes.
- Comparing hardware roadmaps that rely on different codes.
How it differs from related ideas
- Surface codes need only local connections and many qubits, while qLDPC codes need nonlocal connections and far fewer qubits.
- Neutral atoms and ions can implement nonlocal connections more naturally than fixed superconducting chips.
- Concatenated codes also achieve good rates but have lower thresholds than qLDPC codes in practice.
Common Misconceptions
- Sparse does not mean simple to build. The connectivity pattern may be hard to wire.
- A 10-fold saving applies to quantum memory. Logical gates add overhead.
- qLDPC codes are not a different physics. They are stabilizer codes with special structure.
Frequently asked questions
What is a qLDPC code?
A quantum error-correcting code with low-weight checks that protects many logical qubits with relatively few physical qubits.
What is the gross code?
IBM's bivariate bicycle code that encodes 12 logical qubits with distance 12 using 288 physical qubits, about ten times fewer than the surface code for similar protection.
What is the catch?
The codes need long-range connections between qubits, which are difficult on fixed planar chips.
Related Terms
Keep exploring
- Algorithm: Quantum Error-Correcting Codes (Shor Code, Surface Code)
- Algorithm: Quantum Belief Propagation
- Hardware: IBM Heron (r1, r2, and r3)