Gradient-Based Optimization of Superconducting Quantum Circuit Designs -- Part 2

Superconducting circuits are among the most versatile emerging platforms for quantum information processing. However, designing and discovering these circuits can be challenging due to the vast parameter space and the numerous criteria that need to be satisfied for these circuits to operate optimally. Therefore, we believe that using optimization methods can automate and facilitate the design and discovery process.In this presentation, we highlight the new auto-differentiation features of our software, SQcircuit. This open-source package is designed for simulating superconducting quantum circuits. SQcircuit is not only capable of calculating circuit properties, but it can also compute the gradient of these properties with respect to circuit parameters. This is achieved using the back-propagation mechanism, which is an essential building block for gradient-based optimization.Automating the calculation of the gradient enables the optimization of arbitrary metrics derived from the eigenspectrum and circuit parameters. These metrics include (but are not limited to) operation frequency, anharmonicity, flux and charge sensitivity, and decoherence times. We demonstrate the continuous improvement of these metrics, framed as a scalar loss function, across a variety of circuit topologies targeting an experimentally realistic parameter range. Furthermore, the circuit discovery problem can also be formulated as an optimization problem. We illustrate how to design and discover qubits using SQcircuit's auto-differentiation capabilities.

Part 2: In this part of a multi-part talk, we showcase trends and results from performing optimization over a selection of circuit topologies.