Machine Learning the Decoherence Property of Superconducting and Semiconductor Quantum Devices from Graph...

Area of Science:

• Quantum Computing
• Machine Learning
• Materials Science

Background:

• Quantum computing faces decoherence and noise challenges, limiting practical algorithm implementation.
• Understanding collective qubit behavior and its relation to connectivity architecture is crucial but computationally complex.
• Existing methods struggle to systematically characterize the link between quantum architecture and noise susceptibility.

Purpose of the Study:

• To develop a machine learning framework for predicting quantum device decoherence lifetime directly from connectivity patterns.
• To bridge graph theory features with quantum device characterization for noise-optimized quantum processor design.
• To identify platform-specific relationships between topological features and decoherence mechanisms.

Main Methods:

• Representing quantum architectures as connected graphs with 14 topological features.
• Utilizing supervised learning models to predict decoherence lifetime from graph features.
• Analyzing sensitivity to global connectivity (superconducting) and system scale (semiconductor).

Main Results:

• Achieved accurate lifetime predictions (R2>0.96) for superconducting and semiconductor platforms using graph features.
• Identified distinct decoherence sensitivities: superconducting qubits sensitive to global connectivity, semiconductor qubits to system scale.
• Demonstrated complete failure of cross-platform model transfer, highlighting platform-specific design needs.

Conclusions:

• Machine learning can accurately predict decoherence lifetime from quantum architecture connectivity.
• Connectivity design for noise optimization is platform-specific, requiring tailored approaches for superconducting and semiconductor qubits.
• This framework offers rapid assessment of quantum architectures, guiding practical quantum processor development.