How to benchmark a 100-qubit quantum computer using fewer than 100 circuits

We propose a method for benchmarking quantum computers that is scalable and extremely efficient. It can be used to benchmark any number of qubits, and it requires very few resources. Our method consists of running a handful of randomized mirror circuits (RMCs) and then using machine learning methods to interpolate their performance. We show that RMCs form a well-motivated benchmark, because the average success rate of n-qubit RMCs decays exponential as a function of circuit depth, with a rate that is given by the average error rate of the tested quantum computer’s n-qubit circuit layers. We then show how to efficiently map out a quantum computer’s performance as a function of both circuit width (n) and depth using RMCs. We do so by running a few RMCs at a small set of circuit widths and depths, and using machine learning techniques to extrapolate to all widths and depths. We demonstrate our methods with 100+ qubit simulations and 10+ qubit experiments. This work was supported in part by the LDRD program at SNL. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.