Explainer series 4 of 5: What is quantum error correction and why is it the problem nobody talks about

[Bright Hermit]

Fourth in the series. This one covers the topic that experts say is the real gating factor for practical quantum computing, even more than qubit count. If you have followed the series so far you know that qubits are fragile, that superposition collapses under any environmental disturbance, and that entanglement breaks down through decoherence. Quantum error correction is the engineering response to all of that. It is also why the headline qubit numbers you see announced are largely meaningless without context.

A classical computer bit is reliable. It stays a zero or a one. If a cosmic ray flips a bit your computer has ways of detecting and correcting it. A qubit is nothing like that. It drifts. It accumulates errors continuously. The error rate on the best physical qubits today is somewhere around one error per hundred to one thousand operations. That sounds okay until you realise that Shor's algorithm for breaking real encryption might need billions of operations. The error rate compounds catastrophically. The solution is to spread the information from one logical qubit across many physical qubits, using the redundancy to detect and correct errors without measuring the actual quantum state, which would destroy the information you are trying to protect.

The numbers are brutal. Early estimates suggested you might need 1000 physical qubits per logical qubit. More recent work has pushed this toward 10,000 physical qubits per logical qubit for systems running at today's error rates. Breaking RSA-2048 is estimated to need around 20,000 logical qubits. Multiply those together and you get requirements for hundreds of millions of physical qubits, which is why the machine does not exist yet. Google's Willow chip demonstrated something genuinely important in December 2024: for the first time, adding more physical qubits to the error correction code actually reduced the error rate rather than increasing it. That is the threshold the field needed to cross and they crossed it.

There are an estimated 1800 to 2200 people in the world specialising in quantum error correction. That number comes from surveys of the research community and it is not a misprint. The entire global talent pool for the most critical technical area in quantum computing is smaller than a medium sized university department. McKinsey found there is one qualified candidate for every three open roles in the field. The error correction problem will be solved eventually, the physics says it can be, but the human capital constraint is almost as binding as the engineering one

[StuckOnDestiny]

The 1800 to 2200 people worldwide number is the one that genuinely stops me. I had no idea the specialist community was that small

[Rachel93]

It puts the timeline conversations in a different frame. It is not just about whether the engineering is possible. It is about how fast a very small community can work through a very hard problem

[MondayMoan51]

The Willow chip result from December 2024 being the threshold crossing that the field needed is something I did not understand before reading this. Why is below threshold error correction such a big deal specifically

[GoldbergFan]

Because before that result, adding more physical qubits to the error correction code was making things worse, not better. More qubits meant more errors introduced by the correction process itself. Willow showed that in the right regime, the error correction actually works as the theory says it should. The field was not sure it would

[SwiftQuarry]

Wait the error correction was introducing more errors than it was fixing. How is that possible

[EntangledOne]

The physical qubits in the error correction layer are themselves noisy. If your correction mechanism is noisier than the errors it is correcting, you end up worse off. You have to get the physical error rates below a threshold before error correction helps. That threshold is what Willow crossed

[SuperPosition78]

So current machines are in a regime where error correction does not reliably help yet

[EarlyBird]

Most current machines are near or approaching that threshold but not consistently across all their qubits and operations. Willow was significant because it demonstrated the scaling behaviour the theory predicts, which had not been shown cleanly before

[Violet Caitlin]

The logical versus physical qubit distinction is the thing I had been confused about in every quantum announcement. Most headlines just say qubits and do not specify

[SpinState52]

This is one of the most reliable signals of whether a quantum announcement is being reported accurately. Physical qubits is the raw count. Logical qubits is what actually matters for running algorithms. They are very different numbers and most coverage uses them interchangeably

[Brittle Ronan]

Is there a simpler way to achieve error correction than the current approaches. Are people working on fundamentally different methods

[Hollow85]

Yes. Microsoft is pursuing topological qubits which have error resistance built into their physical structure rather than requiring classical error correction overhead. PsiQuantum is building photonic systems where errors have different properties. Neither has demonstrated at scale yet but both could change the overhead numbers significantly if they work

[Daemon55]

The fact that error correction is the gating problem rather than raw qubit count explains why the IBM qubit count announcements feel less significant than they used to