Surface Codes Explained: The Path to Fault-Tolerant Quantum Computing

A surface code is a quantum error-correcting code that lays qubits out on a flat grid and checks each qubit only against its neighbours. It powered Google’s 2024 error-correction milestone, and it descends from the Shor code, the first quantum error-correcting code (Peter Shor, 1995).

• Note: Shor’s code is not Shor’s algorithm. They come from the same researcher but are different work. The 1994 algorithm factors integers, and the 1995 code protects qubits.
• On the grid, data qubits hold the protected information. Measurement qubits sit between them. These are also called ancilla qubits, meaning helper qubits used only for checking.
• Each measurement qubit repeatedly checks the combined parity of its neighbouring data qubits. If the answer changes, that signals an error nearby, and the check never reveals the stored value.

As the grid grows, the code distance grows with it. Code distance is how much damage the code can take and still recover.

The Path to Fault-Tolerant Quantum Computing

Fault-tolerant quantum computing means getting reliable answers from noisy qubits and gates. The path there starts with the Shor code, the first quantum error-correcting code (Peter Shor, 1995), and runs to surface codes working below threshold on Google hardware in 2024.

• Shor’s code is not Shor’s algorithm. The 1994 algorithm factors integers; the 1995 code protects qubits.
• The path has five steps. Each one closes a gap the previous step left open, and the table below maps the route in order:

Milestone	What it added
1. Three-qubit bit-flip code	Catches bit flips
2. Three-qubit phase-flip code	Catches phase flips
3. Shor’s nine-qubit code (1995)	Catches any single-qubit error
4. Surface code	Local checks, adjustable protection
5. Google’s below-threshold result (2024)	Bigger codes, fewer errors

Why Quantum Computers Need Error Correction

A qubit can fail in more ways than a bit can. It can suffer a bit flip, a phase flip (a sign change that bits don’t have), both at once, or any small rotation in between.

The classical fix of copying and voting doesn’t work, because an unknown qubit can’t be copied and looking at it destroys it. Every code on this path needs redundancy without copying and checks without looking.

What Is Shor’s Nine-Qubit Quantum Error-Correcting Code?

Shor’s code is the first quantum error-correcting code, published by Peter Shor in 1995. It protects one qubit of information by spreading it across nine physical qubits. It can correct any single-qubit error: a bit flip, a phase flip, or both at once.

Its parameters are [[9,1,3]]: nine physical qubits, one logical qubit, and code distance 3. Source: Shor’s original 1995 paper, P. W. Shor, “Scheme for reducing decoherence in quantum computer memory,” Phys. Rev. A 52, R2493 (1995).

Shor’s code protects qubits. Shor’s algorithm factors numbers.

How Shor’s Code Combines Bit-Flip and Phase-Flip Error Correction

Shor’s code is easiest to understand as two small codes stacked together. Each one fixes a single kind of error, and neither is enough on its own.

How the Three-Qubit Bit-Flip Code Detects and Corrects Errors

This code stores |0⟩ as |000⟩ and |1⟩ as |111⟩. A superposition becomes the matching mix of both, which is entanglement rather than copying.

You never measure the qubits themselves. Ancilla qubits (helpers used only for checking) measure the parity of pairs: do qubits 1 and 2 agree, and do qubits 2 and 3 agree? The two answers form the syndrome:

1 and 2 agree?	2 and 3 agree?	Diagnosis
Yes	Yes	No bit flip
No	Yes	Flip qubit 1 back
No	No	Flip qubit 2 back
Yes	No	Flip qubit 3 back

This is the key idea. The syndrome tells you which qubit disagrees, but never what any qubit holds. You learn about the error without learning the value, so the superposition survives. Every later step builds on this.

This code’s blind spot is phase flips.

How the Three-Qubit Phase-Flip Code Protects Quantum Information

A phase flip changes the sign between 0 and 1 instead of swapping them. The fix is to change which question you ask, so that a phase flip looks like a bit flip.

Hadamard gates rotate qubits into the |+⟩ / |−⟩ basis, and a phase flip swaps those two states. If you store logical 0 as |+++⟩ and logical 1 as |−−−⟩, the same pair checks from Step 1 now catch phase flips.

This code’s blind spot is bit flips.

How Concatenating Bit-Flip and Phase-Flip Codes Creates Shor’s Nine-Qubit Code

Shor’s key step was concatenation, which means nesting one code inside another. Each code then covers the other’s blind spot.

Start with the three-qubit phase-flip code. Replace each of its three qubits with a three-qubit bit-flip block. Three blocks of three qubits gives 3 × 3 = 9.

The checks inside each block catch bit flips. The checks across blocks compare the blocks’ signs and catch phase flips. Any single-qubit error, even a small rotation, can be broken into these two parts, and the code catches both.

Why Shor’s Code Is Difficult to Scale for Quantum Hardware

Shor’s code solved the problem on paper, but two limits kept hardware from being built around it:

• Overhead with no dial. Nine qubits only buy distance 3. The only way to go further is to concatenate again, which multiplies the qubit count.
• Connectivity. Each of its sign checks spans six qubits across two blocks. Chips where qubits can only talk to their neighbours struggle to wire that.

Shor’s code proved that quantum error correction was possible at all, which was its job. Everything after it is engineering built on that proof.

What Is a Surface Code in Quantum Computing?

The surface code, introduced by Alexei Kitaev in the late 1990s, is a different construction: a topological code on a flat grid. It keeps Shor’s core trick of checking parity without reading data, and it fixes both limits above.

How Surface Code Lattices Use Data Qubits and Ancilla Qubits

On the grid, data qubits hold the protected information and measurement (ancilla) qubits sit between them. Each measurement qubit checks the parity of its neighbouring data qubits, so every check stays local.

As the grid grows, the code distance grows with it, so protection becomes a dial rather than a fixed setting.

Why Surface Codes Are Suited to Superconducting Quantum Processors

Its neighbour-only checks match what superconducting chips can wire, and it tolerates comparatively high noise. The standard reference is Fowler, Mariantoni, Martinis and Cleland, “Surface codes: Towards practical large-scale quantum computation,” Phys. Rev. A 86, 032324 (2012).

The cost is qubits. In Google’s layout, a distance-d code uses 2d² − 1 physical qubits per logical qubit. That works out to 17 qubits at distance 3, and a projected 1,457 at distance 27 for a one-in-a-million error rate per cycle.

The table below compares the two codes on the points that decided which one hardware adopted.

Feature	Shor’s code	Surface code
Qubits per logical qubit	9, fixed	2d² − 1, grows with distance
Qubits per check	Two, or six across blocks	Up to four neighbours
Hardware use today	Small demonstrations	Leading superconducting design

Google’s 2024 Willow Experiment: Surface Code Error Correction Below Threshold

Every qubit you add to a code is also another place for errors, so a bigger code can perform worse. The threshold is the physical error rate below which this reverses and bigger codes start to win.

On 9 December 2024, Google Quantum AI reported crossing that threshold in Nature, in “Quantum error correction below the surface code threshold” (vol. 638, issue 8052). On its Willow processor, Google reports:

• An error suppression factor Λ = 2.14 ± 0.02 each time the code distance rose by 2.
• A 101-qubit distance-7 logical qubit with a logical error rate of 0.143% ± 0.003% per correction cycle.
• A logical qubit that outlived its best physical qubit by 2.4 ± 0.3×.

“Below threshold” means one specific thing: on this hardware, adding qubits to the code now helps rather than hurts. It does not mean quantum computers are error-free.

Google’s paper notes that these error rates are still orders of magnitude away from what practical algorithms need. It also notes that the experiment protected stored information (a memory), not a full computation.

 How to Learn Quantum Error-Correcting Codes and Surface Code Decoding

For most readers, not as a specialisation. The field runs largely on postgraduate physics and mathematics, and dedicated roles are few. Three things still make it worth an afternoon:

• It is the most honest measure of how far quantum hardware is from production.
• Its maths (vectors, matrices and tensor products) is the same linear algebra that machine learning uses.
• Decoder work is one of the few ways into the field that starts from software rather than physics. Decoders are the classical software that turns syndromes into corrections.

The maths behind all of this is linear algebra and basic probability. 

If you’re ready to Take Your Quantum Computing Knowledge Further?

Understanding quantum error correction is one step toward exploring how quantum computers can solve real-world problems. If you’re ready to move beyond the theory, the Certification in Applied Quantum Computing & AI by CEP, IIT Delhi offers 156 hours of learning, Qiskit-based hands-on labs, and capstone projects covering quantum computing, AI/ML, optimisation, and cybersecurity.

Build practical quantum computing skills and work toward applying them to real-world challenges.

Explore the Applied Quantum Computing & AI Programme by IIT Delhi

Frequently Asked Questions