Scalable Parameter Design for Superconducting Quantum Circuits with Graph Neural Networks.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Materials Science

Background:

• Designing large-scale superconducting quantum circuits is complex and challenging for traditional computer-aided design methods.
• Simulating quantum systems, especially for large qubit counts, requires efficient and scalable computational approaches.

Purpose of the Study:

• To propose a novel parameter designing algorithm for large-scale superconducting quantum circuits using graph neural networks (GNNs).
• To demonstrate the algorithm's effectiveness in mitigating quantum crosstalk errors and improving design efficiency and scalability.

Main Methods:

• Developed a GNN-based parameter designing algorithm employing a 'three-stair scaling' mechanism with two neural network models: an evaluator and a designer.
• Trained the evaluator on small-scale circuits and the designer on medium-scale circuits for application to large-scale quantum chip design.
• Considered frequencies of single- and two-qubit gates simultaneously to mitigate quantum crosstalk errors.

Main Results:

• The GNN-based algorithm achieved a 51% reduction in errors compared to state-of-the-art methods for circuits with approximately 870 qubits.
• Design time was reduced from 90 minutes to 27 seconds, demonstrating significant improvements in efficiency and scalability.
• The algorithm effectively mitigated quantum crosstalk errors by optimizing qubit gate frequencies.

Conclusions:

• The proposed GNNs-based algorithm offers a more efficient, effective, and scalable solution for designing parameters of superconducting quantum chips.
• This work highlights the advantages of applying GNNs in the design and optimization of superconducting quantum computing hardware.