Google's "Willow" quantum processor: New RCS record and first error correction below the surface code threshold

Using the principles of quantum state superposition and entanglement, quantum computing has been proven to be able to tackle problems that are hard for state-of-the-art supercomputers. Thirty years ago, when Shor’s algorithm and Grover’s algorithm were proposed and proven to have great acceleration on solving some problems, including factoring and searching, quantum computing was only a beautiful scientific dream. Today, quantum computing is advancing at an incredible pace. Over 10 years ago, Devoret and Schoelkopf said something similar in their review:

...we have witnessed so many advances that successful quantum computations, and other applications of quantum information processing (QIP) such as quantum simulation and long-distance quantum communication, appear reachable within our lifetime, even if many discoveries and technological innovations are still to be made.

Recently, we witnessed a new breakthrough brought by the Google Quantum AI team. On December 10, 2024, Kelly from the amazing team announced their new generation quantum processor, named “Willow.” While the scale of the processor has been extended to 105 qubits, the average coherence time has improved by a factor of 5 when compared to their previous generation, “Sycamore.” Together with advances including readout efficiency, optimizer, decoder, etc., Willow enables two great outcomes.

Random circuit sampling benchmarking

Five years ago, the Google Quantum AI team announced their first milestone of achieving “quantum supremacy” in their first-generation quantum processor, Sycamore.2 Using 53 qubits that nearest-neighbor connected by tunable couplers, they ran a random circuit with a depth of 20, including 1,113 single-qubit gates and 430 two-qubit gates, and a measurement on each qubit. The benchmark, which is called random circuit sampling (RCS), takes 200 s to run the circuit a million times, while the equivalent task would take about 10,000 years for “Summit,” the state-of-the-art classical supercomputer at that time. Although they assert that this is the first time a quantum computer extends the Church-Turing thesis, doubts soon arose from other teams, including IBM quantum. They assessed the task and estimated that, with optimizations including storage and graphics processing unit (GPU) acceleration, classical computers could complete the task in 25 days. Later in 2021, Feng Pan and Pan Zhang from the Institute of Theoretical Physics, Chinese Academy of Sciences, even proved that, by using the tensor network method, one can finish the task in 5 days with only a small GPU cluster and obtain a much higher fidelity than that run on Sycamore. In 2024, researchers from USTC reported an energy-efficient classical simulation algorithm that was 7 times faster than the Sycamore 53-qubit experiment. However, John Martinis, who was then the chief scientist of the Google Quantum AI team, said that it is easy to maintain the supremacy by increasing only several qubits, since the complexity of the problem increases exponentially.