Training neural network local decoders for circuit-level quantum error correction

Abstract

Techniques for training local decoders for use in a local and global decoding scheme for quantum error correction of circuit-level noise within quantum surface codes such that the decoding schemes have fast decoding throughout and low latency times for quantum algorithms are disclosed. The local decoders may have a neural network architecture and may be trained using training data sets comprising simulated rounds of syndrome measurements for respective simulated quantum surface codes in addition to information such as syndrome differences, qubit placements, and temporal boundaries within the simulated rounds of syndrome measurements in order to train the local decoders for arbitrarily sized quantum surface codes and arbitrary numbers of rounds of syndrome measurements. Following a local decoding stage in which a large number of data errors have been corrected by a local decoder, error correction for remaining errors may continue with a more efficient global decoding stage.