Google Reduces Quantum Computing Error Rate by 40-Fold with New Method
Google Quantum AI has demonstrated a technique that significantly improves efficiency when generating "magic states," which are essential for general-purpose quantum computers. While this achievement is an important step in addressing cost issues, challenges remain before practical implementation.

Logical fidelity achieved in generating T states on the Willow processor.3
Percentage of attempts that passed checks via synchronous measurement and were transferred to a larger error-correcting code.3
Total capacity of a quantum computer that could be consumed by generating high-quality magic states essential for universal quantum computing.3
Story
Improving the Efficiency of Magic State Generation in Quantum Error CorrectionGoogle Quantum AI has demonstrated a new technology to reduce errors in quantum computing using its Willow processor.3 This technique, dubbed "magic state cultivation," succeeded in reducing the error rate to about one-fortieth compared to cases without cultivation.3 This 40-fold reduction rate has also been reported by media outlets such as ITmedia.6 Details of this significant research achievement were published as a paper in the scientific journal Science.3
This cultivation process begins by preparing multiple noisy initial states in a restricted area on a quantum chip.3 Next, leveraging synchronized measurement techniques, it inspects these states one by one to identify defective attempts.3 States deemed imperfect as a result of the inspection are completely discarded from the process.3 Only the states that pass this rigorous screening procedure are retained for subsequent processing.3
This screening process was extremely strict, with only about 8% of the initial attempts passing inspection and proceeding to the next stage.3 These high-quality states that passed inspection are transferred into a larger, more robust error-correcting code.3 Through this experiment, an extremely high logical fidelity of 99.99% was achieved in a specific quantum state known as a T state.3 This figure indicates that the cultivation method can generate quantum states of very high quality.3
Magic states are fundamental building blocks that must be continuously supplied to operate future general-purpose quantum computers.3 A major limitation in their use is that once used in a calculation, the state is consumed and cannot be reused.3 Furthermore, the process of generating error-corrected, high-quality magic states consumes computational resources significantly.3 In some cases, as much as 95% of the total capacity of a quantum computer may be spent solely on generating these magic states.3
Google's research team first proposed this method of magic state cultivation as a theoretical approach in 2024.3 At the time, this concept was positioned as a promising solution to overcome the problem of enormous resource costs associated with magic state generation.3 Sundar Pichai stated that this experiment marks a memorable milestone where the theory was demonstrated on actual physical hardware for the first time.3
In addition to the success of magic state cultivation, this research report also includes another important technical demonstration.3 The research team also succeeded in safely moving quantum states between two different types of error-correcting codes.3 As a specific operation, a quantum state protected by a distance-3 color code was transferred to a distance-5 surface code.3 This state transfer between codes was achieved through a special process called "grafting."3
Google's Sundar Pichai particularly emphasized that this marks the first instance of this technology transitioning from a theoretical concept to a demonstration on physical hardware.3 However, the company itself has taken a cautious stance, indicating that whether this method will function similarly in larger-scale quantum computing systems still requires future verification.3 Google has also officially acknowledged that technical challenges remain to be resolved toward practical application.3
It is necessary to clearly distinguish that this announcement regarding quantum computing is unrelated to other large-scale Google projects reported during the same period.3 For example, one report conveyed that Google commenced construction of a massive 1-gigawatt data center in Visakhapatnam, India.4 This data center plan is an entirely separate venture from this demonstration of quantum technology.34
Similarly, another report that Google is funding a 100-megawatt virtual power plant in the PJM Interconnection region is unrelated to this matter.5 This energy-related investment is an independent project separate from quantum computing research and development.35 These reports concern distinct operations and do not serve as direct evidence for evaluating the achievements of the quantum research.3
Structure
Who is connected to whom- 1Alphabet
- invests in →Fact
- uses →Fact
- partners with →Fact
- agreement with →Fact
- produces →Fact
History
How it came to this- 2024Theoretical proposal of magic state cultivationA Google team proposed the method as a theoretical approach to the challenge of resource costs in magic state generation.
- NowAbout 1/40 Error reduction ratio from magic state cultivation (compared to without)
Impact
Spreading outward, level by level- Level 1Demonstration of state generation and inter-code transfer on hardware
Magic state cultivation was demonstrated on physical hardware for the first time, and the team also succeeded in moving quantum states from a distance-3 color code to a distance-5 surface code via grafting.3
Fact - Level 2Easing resource bottlenecks
By curbing the generation cost of magic states, which cannot be reused and are continually consumed, this may contribute to solving the resource-consumption challenges of quantum computers.
Analysis - Level 3Evaluating challenges toward practical scale
This result is a laboratory demonstration showing improvements in state preparation, and testing in larger-scale environments as well as overcoming remaining technical challenges will be required.
Analysis
Sources
What each source supportsWritten by AI from the sources listed below: every fact was checked word for word against its source, and inference is marked apart. How we write