Demonstrating Scalable Benchmarking of Quantum Computers

Reliably testing multi-qubit quantum computers is notoriously challenging, but it is vital for measuring and aiding experimental progress, as well as for quantifying the capabilities of available processors. In this talk, I will present experimental results demonstrating simple, fast and scalable methods for benchmarking quantum computers. Our methods will work on essentially all current and near-term quantum computers, and they can be applied to hundreds of qubits with moderate error rates (around 0.1-1%) and thousands of high-quality qubits. Our core method is based on a class of randomized circuits, and it can be used to estimate the rate of errors in an average many-qubit circuit layer, using an analysis that will be familiar from randomized benchmarking. Our experiments reveal noise phenomena that only emerge at scale (e.g., crosstalk), quantify the divergence between the predictions of one- and two-qubit performance data and the actual behavior of many-qubit circuits, and provide high-level summaries of device performance versus circuit width and depth.