Towards Automated Superconducting Circuit Calibration using Deep Reinforcement Learning

This paper presents an innovative way of quantum circuit optimization; we propose an automated superconducting circuit calibration using deep reinforcement learning (ASCC-DRL). A simplified model of a transmon qubit coupled to a cavity resonator is used to demonstrate a quantum circuit. We analyzed the circuit through its equivalent two-mode lumped-element representation of the distributed circuit and derived the equivalent Hamiltonian translation of the circuit with the non-linearity of the Josephson Junctions. We adopted the parametrized qubit junctions and used the component physical value to tune the circuits to maximize the fidelity of the desired entangle state and the resulting quantum circuit state. However, quantum systems are susceptible to environmental parasitics and noise; therefore, a reliable optimization technique is essential to maintain high fidelity across the system. For optimization, we integrate deep reinforcement learning using deep deterministic policy gradient techniques as our mechanism for our AI module to optimize the fidelity of the superconducting quantum circuit. We demonstrate the feasibility of this approach by training the agent in three distinct environments to achieve the highest possible best rewards. We achieved an average fidelity (best rewards) of 0.92, which is 94% of the statically tuned circuit.