Watch Google Quantum AI Director of Hardware, Julian Kelly, demonstrate Google’s new Quantum Computing Chip’s capabilities. The chip performs a standard benchmark calculation in less than five minutes, which would take a modern supercomputer 10 septillion years.
Introducing the latest quantum computing chip, designed to learn and adapt like the natural world. Julian Kelly is the director of hardware for Google Quantum A I. I am Willow. Today, on behalf of my amazing team, I am proud to announce Willow Willow as Google’s latest and most powerful superconducting chip for quantum computing. The next step on our journey to building large-scale quantum computers and exploring applications is the release of Willow Willow. Since 2008, when I first tried Cubis, I was fascinated by quantum computing. Since joining Google in 2015, the dream of building quantum computers to solve otherwise unsolvable issues has become a reality. We launched our first chip, foxtail, in 2017, followed by Bristol Cohen, in 2018, and Sycamore, in 2019. This chip powered our milestone one, which was the first quantum supercomputer to surpass the best classic supercomputer in a computational task called random circuit sampling. Over the years, with sycamore we have been able squeeze a remarkable amount performance from our hardware, including achieving a scaleable logical cubit. We were ultimately limited by the quantum coherence multiplied by the time it takes cubist to maintain their intended state. With Willow, we have made a big step forward. We have increased quantum coherence time by a factor five, going from 20 microseconds to 100 microseconds. We’ve done this without sacrificing the features that have made our systems so successful. This breakthrough was made possible by our dedicated superconducting chip fabrication facility, one of the few in the entire world, in Santa Barbara. We’re also seeing exciting developments from Willow, whose breakthrough demonstrations have already surpassed those of Sycamore. Our logical qubits are now operating below the critical quantum-error correction threshold. Since the discovery of quantum computing in the 1990s, this goal has been a long-sought after one for the field. We’ve achieved this for the first with willow errors as error rates have been halved. We add physical qubits at a scale of 3 to 5 to 7, each time. Our logical cubits now have a much longer lifetime than the lifetimes of all the physical qubits they are composed of. Quantum error correction can be used to improve accuracy even as we increase the complexity of our quantum shifts. We have pitted Willow’s random circuit sampling benchmark against one of the most powerful supercomputers in the world. Our best estimates show that the results are quite surprising. A calculation that Willow can complete in under five minutes, would take a supercomputer between 10 and 25 years. This is a one followed by 25 zeros or a timescale that is longer than the age the universe. This result highlights the exponentially increasing gap between classical and quanta computation for certain applications. Let’s talk about hardware. We’ve developed a quantum A I at Google that makes this possible. Our returnable cubits, couplers and hardware enable fast gates and operations for low error rates. They can be reconfigured to optimize hardware on-site and run multiple applications. We leverage this tun capability to enable reproducible, high performance across all devices. I’ll explain that superconducting cubits are not all created equal. Some are outliers, with unusually high ears. Here’s where the trainable cubits shine. We can fix these cubits that are outliers by reconfiguring the device to make them perform like the rest. We can even go one step beyond by having our researchers use the tune ability to continually develop new calibration strategies which push errors down across all of the cubits using software. Let’s nerd out and quantify this for a moment. Number of cubits connectivity measures the average number interactions that each Cuba can have with its neighbors. We quantify the error probabilities of running simultaneous operations. A full system is an application’s performance. Benchmark. Willow is a great choice for the entire list. It can run a variety of applications and has a high number of cubits. We measure low mean errors across all operations using multiple native two cubic gate. We have increased the measurement rate t one time and willow is under the error correction threshold. It also performs random circuit sample. Willow is a quantum computer that goes beyond what was possible with classical computers. We continue to work towards building large-scale and useful error-corrected quantum computers, which will push the limits of science and exploration of nature. Future commercially useful applications include pharmaceuticals, batteries, and fusion energy. We are excited to solve otherwise unsolvable future problems.
