How Quantum Computers Will Fix Their Own Mistakes

A tennis ball is dropped above a net, it hits the net, and falls to one side or the other. Left or right, 0 or 1, the tennis ball always lands somewhere. A person calls out “left” or “right” depending on where the ball lands. But after every couple thousand of repetitions, the person mistakenly calls out the wrong side.

The ball acted appropriately. It landed on one of the sides like always, following the laws of gravity. The mistake happened in the reported side: there might’ve been a moment of distraction, a mishearing, a slip. That’s what a measurement error is in quantum computing. A measurement error happens when a qubit resolves to a definite state, but the hardware misreads which state it collapsed to. The machine ends up recording the wrong answer because the result gets blurred by noise on its way out.

This is just one of the many ways a quantum computer can fail.

You might be wondering: “Well, once every couple thousand tries doesn’t sound like a big deal, that’s so rare.”

Qubits (quantum bits) don’t run one operation and stop. A real quantum algorithm combines thousands of operations. An error once every couple thousand repetitions becomes enough to derail an entire calculation. This is what quantum error correction attempts to solve, and the only way to solve the error-rate is getting creative in ways that would be unnecessary in classical computing.

To put that “once every couple thousand” number in perspective: classical bits in your laptop’s memory are wrong around once every 10¹⁸ operations, that is once every one quintillion repetitions. Even the best quantum hardware today runs error rates around 1 in 10,000 operations. One of the most accurate results in a quantum test was Oxford Ionics’s 99.99% two-qubit gate fidelity, getting an error-rate of only 0.0084% (1 in 12,000).

Why Qubits Are Much Harder to Correct than Classical Bits

In classical computing, catching an error is much much simpler than quantum. Error correction in classical computing is straightforward: read the bit, compare it to what it should be, and correct it if it’s wrong. Computers have done this for decades through things like redundancy, parity checks, and error-correcting memory.

You cannot do these things with a qubit. As you probably know, a qubit can exist in superposition (a mixture of the states |0> and |1>). The exact moment you measure a qubit to check its state, you force it to collapse into a classical value, 0 or 1. So, whatever information the qubit was holding during its superposition, is gone the moment you try to measure and look at it. Thus, you can’t peek at a qubit to see if it made a mistake without discarding the information you are trying to protect.

So, what do we do? Well the obvious instinct is to think: if you can’t check the original, just make a backup qubit to read first. Unfortunately, you can’t do that either, as it’s a fundamental law of quantum mechanics that an unknown quantum state cannot be copied. Because of these issues and qubits’ errors being much harder to correct, engineers were met with a paradox. What did they do?

The Fix

Instead of trusting one fragile qubit to hold a piece of information, you can spread the same information across many physical qubits, linked through entanglement (read more about it in this article). Together, this group works as one logical qubit. So, no single qubit within this group holds the answer, the information is shared through the relationships between them.

That’s the theory behind quantum error correction, but in practice it looks a little different.

How It’s Going

Google’s Willow quantum chip was one of the most significant quantum accomplishments of the last two years due to its astonishing error-correction properties. Willow used an error-correction layout where physical qubits are arranged in a grid, and dedicated measurement qubits checked the relationships between neighboring qubits for signs of an error. Typically, as these grids become larger, the error-rate would in theory increase alongside it. However, as researchers increased the grid size from 3x3 to 5x5 to 7x7 physical qubits, the error-rate dropped. This is the first quantum chip that exhibits that increasing the number of physical qubits could actually reduce its error-rate. Willow ended up having a measured error rate of 0.143% per correction cycle for its largest, 105-qubit logical qubit.

Another example of quantum error-correction in practice was in 2024 when Microsoft and Quantinuum demonstrated their quantum systems. It was one of the most reliable quantum systems created by combining Quantinuum’s trapped-ion hardware with Microsoft’s error-correction software. The system used 30 physical qubits to create 4 logical qubits, and the team ran over 14,000 individual quantum circuits without detecting a single error.

While it may seem like we are far from reaching the scale needed for practicality and widespread quantum use, the demonstrations showed that quantum error correction isn’t just an idea. With recent developments, it can already suppress errors well enough to create highly reliable quantum systems. Day by day, quantum computing brings us closer to machines capable of solving problems beyond our comprehension. The quantum era is closing in on us, and with every article I write, we are one step closer to you reading this article on a quantum computer.