Google's Willow chip cultivates magic states with about 40x fewer errors

The 40x and the 95% measure different things. The 40x is a ratio of error rates between cultivated and un-cultivated states on Willow ^[5]. The 95% is an older estimate of the share of a quantum computer's capacity that might go to making magic states ^[12].

Getting from one to the other takes three inputs: the error rate each finished state must reach, how many attempts the filter discards to reach it ^[4], and how many qubits sit occupied while it runs. The 99.99% logical fidelity ^[6] works out to an infidelity of 0.01%, or 1 in 10,000 ^[11]. The post does not report the discard rate.

The path to the 40x included months of results that did not make sense. When the team first tested the protocol, research scientist Emma Rosenfeld saw data that seemed to defy the laws of physics ^[9]. After months of confusion, simulations and consistency checks, she traced it to an incorrect assumption in the team's go-to measurement technique: noise from electrical equipment drift and temperature variation was interfering with the benchmark's validity ^[9]. In our view, that check is what makes the later number believable. The team then added extra control to the system to reduce the noise and invented a new benchmarking technique that uses cultivation to measure itself ^[10].

Every figure here comes from Google's own post about its Science paper ^[2]. The post is direct about the limit. Google wrote that "we still need to demonstrate magic state cultivation at a larger scale to make it realistic" ^[8].