Google Quantum AI Reveals Willow Quantum Computing Chip

[Music] introducing our latest Quantum Computing chip developed to learn and evolve like the natural world around us Willow from Google Quantum [Music] AI hi I'm Julian Kelly director of hardware at Google Quantum Ai and today on behalf of our amazing team I'm proud to announce Willow Willow is Google's newest and most powerful superconducting Quantum Computing chip and the next step in our path towards building large scale quantum computers and exploring their applications I have been fascinated with Quantum Computing since I first experimented with cubis in 2008 and since coming to Google in 2015 it has

been a dream to make our mission our reality building quantum computers for otherwise unsolvable problems we launched our first chip foxtail in 2017 followed by Bristol cone in 2018 and Sycamore in 2019 19 which powered our Milestone one the first quantum computer to surpass the best classical supercomputer on a computational task random circuit sampling over the years with Sycamore we have been able to squeeze a remarkable amount of performance from our Hardware including achieving a scalable logical cubit in our Milestone too but we've ultimately been limited by Quantum coherence times the length of time Cubit

maintain their intended state with Willow we've made a huge step forward we've increased Quantum coherence Times by a factor of five going from 20 microc in Sycamore to 100 microc in Willow and we've accomplished this all without sacrificing any of the features that made our systems so successful this advancement was enabled by our new dedicated super connecting Quantum chip fabrication facility in Santa Barbara one of only a few in the world and we're seeing exciting developments coming from Willow which has already surpassed sycamore's breakthrough demonstrations our logical cubits Now operate below the critical Quantum error

correction threshold a long sought-after goal for the quantum Computing field since the theory was discovered in the 9s and we've achieved it for the first time with Willow errors are exponentially suppressed neurological cubits as error rates are haved each time we add physical cubits in scale from distance 3 to 5 to 7 surface Cates additionally our logical Cubit lifetimes are now much longer than all of the lifetimes of the physical cubits that compose them this means that even as we make our Quantum chips larger and more complex by adding more cubits we can use quantum

error correction to actually improve their accuracy we've pitted Willow against one of the world's most powerful supercomputers with the random circuit sampling benchmark the results are pretty surprising by our best estimates a calculation that takes Willow under five minutes would take the fastest supercomputer 10 to the 25 years that's the one with 25 zeros following it or a time scale way longer than the age of the universe this result highlights the exponentially growing gap between classical and Quantum computation for certain applications let's talk about the hardware approach we've pioneered at Google Quantum AI that makes

these things possible our tunable cuits and couplers enable super fast Gates and operations to achieve low error rates reconfigurability to optimize Hardware in situ and run multiple applications and high connectivity to efficiently Express algorithms we leverage this tunability to enable reproducible high performance across the device let me explain a challenge in superc connecting cubits is that not all of them are created equal some are outliers with uncharacteristically high erors but here's where our Trin cubits really shine we're able to fix these outlier cubits by reconfiguring them to perform in line with the rest of the

device and we can go one step further by having our researchers use tunability to continuously develop new calibration strategies that push erors down across all key bits with software let's quantify this and nerd out for a minute on quantum computer teex specs we have number of cubits connectivity is the average number of interactions each Cubit can perform with its neighbors we quantify error probabilities for running simultaneous operations single Cubit Gates two cubic Gates and measurement coherence time measures how long each Cubit can retain its information measurement rate is how many computations we can run per

second an application performance is a full system Benchmark Willow hits a sweet spot across the full list it has a large number of cubits with high connectivity and can run diverse applications we measure low mean error rates across all operations with multiple native 2 cubic Gates we have greatly increased T1 times we have very high measurement rates and Willow is below the error correction threshold and performs random circuit sampling far beyond what is possible with classical computers looking to the future with Willow we continue our journey towards building large scale and useful aor corrected quantum

computers that will push the boundaries of Science and the exploration of nature with future commercially useful applications in areas like Pharmaceuticals batteries and fusion power we are excited to solve the otherwise unsolvable problems of tomorrow