Bitcoin: Google reduces quantum errors by 40

Magic, but in small amounts. Google Quantum AI announced a breakthrough on Thursday, October 15 in the journal Science. His Willow chip operated a technique called “cultivation of magical states”. It reduces errors by around 40 times on one of the most expensive resources in quantum computing. On the quantum Bitcoin front, it is therefore one more brick in a wall that is being built quickly.

The threshold for breaking a secp256k1 key fell to 813 logical qubits at the end of August. But Willow only has 105 physical ones. And nine out of ten attempts end up in the trash.

Key Points

Magical states enable the operations that make a quantum computer a universal machine. They are consumed with use.

Google achieves 99.99% fidelity on 54 qubits, with a success rate of around 8% per attempt.

Breaking RSA-2048 would require states about a million times cleaner, according to the paper itself.

Magic States: the bottleneck that Google says it has unlocked

A quantum computer corrects its errors by distributing information over many physical qubits. It works for most operations. However, one essential family escapes: the so-called T doors. Without them, the machine does not exceed what a classic computer can simulate.

To execute them, you must first prepare magical states. These are special quantum states that are consumed only once. Some estimates thus reserved up to 95% of the capacity of a machine for their manufacturing alone. Sergei Bravyi and Alexei Kitaev proposed a distillation method in 2004. Since then, they have been produced in bulky qubit “factories”.

Cultivation, theorized by Craig Gidney and his colleagues in 2024, works like a filter. We inject a noisy state into a small area of ​​the chip, then we chain together coordinated measurements. Finally, any attempt that fails is discarded.

The team of Emma Rosenfeld and Craig Gidney obtained T-statements on Willow with a record fidelity of 99.99%. Or “a 40-fold reduction in errors compared to non-cultured states”. To achieve this, it went from a color code of distance 3 to a surface code of distance 5. The experiment used at most 54 of the 105 qubits on the chip. Only 8% of attempts passed all checks.

But Google itself warns that cultivation must “still be demonstrated on a larger scale.” Many technical challenges remain, in his own words.

Quantum Bitcoin: the threat advances, but not this week

The link with cryptocurrencies goes through resource estimates. In May 2025, Gidney reduced the cost of factoring an RSA-2048 key to less than a million noisy qubits. This is twenty times less than its 2019 estimate. Cultivation was already one of the tricks counted in this calculation.

In March, a Google white paper co-written with Justin Drake and Dan Boneh went further. He placed the cost of breaking Bitcoin's elliptic curve under 500,000 physical qubits. Thursday's demonstration therefore transforms a hypothesis of this model into a laboratory result.

There remains the gap between the laboratory and the attack. According to the Google post and associated paper, factoring RSA-2048 requires much cleaner states. The error must fall to one in ten billion, or about a million times better than the result obtained.

The good news for Google, and the bad news for the exposed ECDSA keys, lies in a scaling law. Cultivation errors decrease with the cube of hardware noise. Halving Willow's noise would therefore reduce the error to around two in 100,000, according to the same paper.