Google Claims 40x Reduction in Error Rate for Quantum Magic State Preparation

Author: Jose Antonio Lanz, Decrypt

Compiled by Deep潮 TechFlow

DeepChain Summary: A recent paper published by Google's Quantum AI team in *Science* shows that its Willow chip successfully reduced the error rate of the scarce and rapidly consumed "magic states" by 40 times. This breakthrough, known as "magic state distillation," significantly lowers the physical cost of building practical quantum computers and brings the theoretical threat to RSA-2048 and Bitcoin’s elliptic curve cryptography closer to reality.

Google Quantum AI announced on Thursday that a technique called "magic state cultivation" reduced error rates by approximately 40 times on its Willow chip, producing magic states with a fidelity of up to 99.99%.

Magic states are essential, single-use resources required to run high-value quantum programs. Google states that, according to some academic estimates, the cost of producing magic states previously accounted for up to 95% of a quantum computer’s total computational power.

“Technique cultivation” is also one of the key factors behind Google’s previous prediction that the number of noisy physical qubits required to break RSA-2048 encryption could be reduced to under one million; however, at the level of cryptanalysis, the error rate of magic states still needs to approach one in ten billion.

Google says it has turned one of the most promising theoretical concepts in quantum computing into a technology that runs on actual chips.

In a paper published Thursday in the premier academic journal Science, Google’s Quantum AI team demonstrated that magic state distillation reduced error rates by approximately 40 times on its latest Willow processor.

Making quantum computers cheaper to build will have far-reaching implications beyond the realm of basic physics laboratories: it will not only significantly accelerate the practical application of quantum computing in fields such as drug design and new material development, but also hasten the day when it becomes capable of breaking the encryption algorithms that secure bank login credentials and Bitcoin wallets.

Quantum computers operate on qubits. Unlike classical bits, which can only be 0 or 1, qubits can exist in a superposition of both 0 and 1 simultaneously, enabling them to explore a vast number of possible outcomes in parallel. However, qubits are extremely fragile: even minor heat or faint electromagnetic noise can cause a qubit to flip, leading to complete computational failure. This is why the entire quantum industry is intensely focused on “quantum error correction”—the process of distributing a single piece of information across many physical qubits to detect and correct errors in real time.

Quantum error correction can effectively handle basic logical operations, but a commercially valuable complex program requires a higher-order class of operations that cannot be protected in the same way. To perform these operations, the machine must first generate "magic states." These function like single-use fuel cells for these complex computations: once used, they are immediately consumed.

Producing high-purity, low-noise magic states has historically been extremely costly. Google cites estimates indicating that the cost of generating magic states can sometimes reach up to 95% of the total computational capacity of an entire quantum computer, leaving very little capacity available for actual business computations.

As early as 2004, renowned physicists Sergei Bravyi and Alexei Kitaev proposed the standard solution known as "Magic State Distillation." This method collects a large number of noisy, low-quality magic states and, through multiple stages of filtering, purifies them into a small number of high-purity magic states. Although effective, distillation requires building large, specialized physical qubit "factories" dedicated solely to purification.

In 2024, Google researchers proposed a more cost-effective alternative named "Magic State Cultivation." Unlike massive factories, this technology functions like a precision filter: it injects a crude initial magic state into a tiny region of a chip, then applies a series of cascading checks, immediately discarding every flawed attempt and only preserving and continuously cultivating the flawless surviving states. The research team even boldly speculated in their original paper that, in practical engineering, traditional second-round distillation may never again be necessary.

On Google Quantum AI’s Willow processor, our protocol achieved an extraordinary 40-fold reduction in error rate compared to the unoptimized initial state. We obtained T states with a record-breaking logical fidelity of 99.99%,” Google stated.

In the experiment, only about 8% of attempts successfully passed all tests, meaning the chip must actively discard over 90% of initial attempts for each successfully identified high-purity magic state.

The magic state distillation technique is one of the key foundations for researcher Craig Gidney’s significant prediction. In 2025, he predicted that the number of noisy qubits required to break mainstream RSA-2048 encryption could be reduced to fewer than one million—down 20-fold from his 2019 estimate of 20 million. Prior to this experiment, this distillation mechanism had only been tested in pure theory and supercomputer simulations.

In March of this year, a white paper co-authored by Google Quantum AI, Ethereum Foundation researcher Justin Drake, and renowned Stanford cryptographer Dan Boneh further tightened this prediction: the number of physical qubits required to break Bitcoin’s elliptic curve cryptography—the mathematical foundation that prevents anyone from deriving a private key from a public key—could be fewer than 500,000.

However, no one’s cryptocurrency wallet has been compromised as a result of this experiment. The experiment used at most 54 of Willow’s 105 qubits, and the paper explicitly states that factoring an RSA-2048 key would require a magic state error rate as low as approximately one in ten billion—nearly a million times purer than Google’s recently achieved 99.99% fidelity.

Google stated that the next step is to run magic state distillation on a larger hardware array. The team expects that reducing the underlying hardware noise of the Willow processor by half would further lower the error rate of magic states to approximately two in 100,000, since the error rate in this process scales cubically with the chip’s noise level.