Google Demos Magic State Cultivation, Errors Cut 40-Fold

A quantum computer cannot make progress unless it keeps producing components that are consumed during the calculation. Those components are special quantum states called “magic states,” which are indispensable for performing calculations while correcting errors. On October 15, Google Quantum AI announced that it had demonstrated experimentally a method for producing these components on its “Willow” quantum processor. Compared with running without the method, errors fell by about one-fortieth.

A magic state cannot be reused once it has been consumed, so such states must be produced one after another during a calculation. Producing high-quality ones also requires enormous resources. Some estimates suggest that up to 95% of a quantum computer’s capacity would have to be devoted to generating them, and researchers have spent about 20 years looking for ways to lower this cost. Conventional “magic state distillation” refines high-quality states from many noisy ones, consuming a large number of qubits and a great deal of time.

The new approach takes a different tack. Noisy states are injected into a small region of the chip, and measurements are repeated to check for errors. Attempts in which errors are detected are discarded, and the remaining states are grown. This filter-like mechanism was published theoretically in 2024 by Google Quantum AI researchers. In the new experiment, the fidelity of the remaining states reached at least 99.99%, which the company says is a record. However, only about 8% of all attempts remained after passing the error checks.

During the demonstration, the team obtained measurement data that appeared to contradict physical laws and had to investigate the cause. Gradual drift in the characteristics of electronic components over time, along with noise caused by temperature changes, had undermined the assumptions of conventional evaluation methods. The team added controls to suppress the effects of noise and developed a new evaluation method that uses the cultivation process itself as a measurement. It also succeeded in “grafting” states created with a “color code,” an error-correction scheme well suited to cultivation, onto a larger region based on the more established “surface code.” The company describes this as a first step toward showing that cultivation is compatible with its quantum computer designs.

So what would change as a result? Google says the results bring it a step closer to practical, error-corrected quantum computers, but it concedes that larger-scale demonstrations will be needed before such computers become practical. If the cost of producing magic states falls, the scale of the hardware required for practical calculations could be greatly reduced. However, the announcement does not include any figures indicating when that might become reality.