System optimization of superconducting qubit readout for quantum error correction

An accurate measurement of a qubit register is a crucial component of quantum algorithms and error correction. For mid-circuit measurements there is a tradeoff between accuracy and duration, since other parts of the processors might decohere during the measurement.

The parameter space for superconducting qubit readout is too large to map out fully, and effects such as qubit-qubit crosstalk impose non-locality. In other words, we can’t optimize each qubit separately and expect good results for simultaneous qubit readout. Instead, we need to consider the performance of the system as a whole and employ a global optimization algorithm to find the minimum of the sum of all readout errors.

In this talk, we discuss optimizing a 49 qubit readout suitable for quantum error correction by performing an offline and model based optimization. The search space is ~10^249 and we achieve a mean readout error of 1.95% for a total readout time of 500 ns, with a minimal amount of calibrations performed on the quantum processor itself.

Additionally, there are subtle and detrimental effects from the measurement process that don’t show up in the readout error numbers but can increase logical error rates. By accurately modeling these effects and including them as cost functions in the optimization we are able to significantly reduce their impact, which was a key part in our recent demonstration of a distance-5 surface code.