Jin Ming Koh, Dax Enshan Koh, Jayne Thompson (Jun 13 2024).
Abstract: Current-day quantum computing platforms are subject to readout errors, in which faulty measurement outcomes are reported by the device. On circuits with mid-circuit measurements and feedforward, readout noise can cause incorrect conditional quantum operations to be applied on a per-shot basis. Standard readout error mitigation methods for terminal measurements which act in post-processing do not suffice in this context. Here we present a general method for readout error mitigation for expectation values on circuits containing an arbitrary number of layers of mid-circuit measurements and feedforward, at zero circuit depth and two-qubit gate count cost. The protocol uses a form of gate twirling for symmetrization of the error channels and probabilistic bit-flips in feedforward data to average over an ensemble of quantum trajectories. The mitigated estimator is unbiased and has a sampling overhead of ${\sim} 1 / (1 - 2 r)^m$ for $m$ total measurements and characteristic readout error rate $r$ per measurement. We demonstrate the effectiveness of our method, obtaining up to a ${\sim} 60%$ reduction in error on superconducting quantum processors for several examples of feedforward circuits of practical interest, including dynamic qubit resets, shallow-depth GHZ state preparation, and multi-stage quantum state teleportation.