Machine Learning Assisted Superconducting Qubit Readout – Part II: Deep Reinforcement Learning*

Qubit-state readout of contemporary superconducting quantum processors is a significant error source. For a qubit dispersively coupled to a resonator, quick resonator ring-up and ring-down ensure fast readout and limited qubit dephasing in future operations. In an efficient, frequency-multiplexed readout of multiple qubits, effects such as drive crosstalk increase the complexity of optimal readout pulse shapes, requiring computationally intensive methods to discover high-fidelity pulse shapes. We present a pulse shaping optimization module using deep reinforcement learning (DRL). We compare DRL to conventional methods in readout pulse shaping experiments of multi-qubit devices and evaluate future generalized use of DRL methods in quantum computing.