Learning-Based Control of a Superconducting Processor for Quantum Error Correction

Quantum error correction (QEC) is essential for reaching the low error rates required to run quantum algorithms. For QEC to improve error rates, the fidelity of physical operations must exceed a code-specific threshold. This can be challenging because devices running QEC tend to be large and complex, with crosstalk and unpredictable noise. Reinforcement Learning (RL) has been used in similarly complex qubit calibrations previously, and in this work, we explore the use of RL to improve calibration for QEC with stabilizer codes. We calibrate a 9-qubit superconducting QEC device with the connectivity to run a [[6,3,2]] error detection code. The anticipated limiting factor for performance is two-qubit gates, largely due to two-level systems (TLSs) and ZZ crosstalk. In this talk, we explore RL-based calibration of this 9-qubit processor, with a focus on two-qubit gates. This is an initial step toward the broader goal of improving logical qubit lifetimes.