Resource-efficient fully randomized benchmarking

Randomized benchmarking is a commonly used technique for estimating the average fidelity of quantum logic operations, measuring the decay of probability of returning to a final fiducial state as a function of the length of a sequence of randomly selected gates. Typically, repeated trials of specific pre-computed random gate sequences are used to reduce the memory requirements on control hardware, limiting the randomness of sequences. By generating pseudorandom numbers on the control hardware itself we can construct gate sequences on the fly; only the instantaneous ideal stabilizers are tracked, reducing memory requirements substantially and enabling gate sequences of arbitrary length. The just-in-time sequence generation allows each sequence to be independently random, which enforces the assumption of depolarizing error channels. Sequence lengths are chosen to extract maximum information in an allotted wall time, making the most efficient use of experiment resources.