Deep Adaptive Design for Characterisation of Superconducting Qubits

Modern superconducting quantum computers are severely limited not by the number of qubits but by the high error and noise characteristics. The detailed system characterization required to understand the underlying error sources is an arduous process and impractical with increasing chip size. Typical textbook characterisation routines do not scale efficiently to large multi-qubit chips, requiring the development of advanced techniques based on statistical and information theoretic foundations. We present a Bayesian Experiment Design process that adaptively recommends the most optimal experiments at every step to maximise the expected information gain about the system, thus enabling optimal identification of arbitrary system model parameters. The cost of calculating expensive Bayesian posteriors is amortised by the use of Reinforcement Learning assisted Deep Adaptive Design techniques. A high-fidelity differentiable digital twin that models the open quantum dynamics, complete electronic control stack and noise & transfer functions for various superconducting quantum devices lies at the heart of this closed loop adaptive calibration and characterisation process. The practical usability of this Bayesian Experiment Design method is demonstrated on a multi-qubit superconducting transmon chip.