Speaker
Description
Topological superconductors hosting Majorana zero modes offer a promising pathway to fault-tolerant quantum computing, yet material realization remains a critical bottleneck. This study investigates the Bi₂Se₃/Nb heterostructure while employing machine learning to accelerate discovery of materials with analogous properties. Recent experimental work on Bi₂Se₃/NbSe₂ Josephson junctions reveals anomalous peak-dip structures and nonreciprocal supercurrent flow near a flux quantum. These signatures were attributed to nonuniform supercurrent distribution rather than Majorana bound states, challenging prior expectations. Concurrently, theoretical modeling of lateral Bi₂Se₃/Sb₂Te₃ heterostructures demonstrates that interface-bound states can be described via Bogoliubov-de Gennes formalism, with chemical potential engineering enabling control over topological phase transitions. These findings underscore both the promise and complexity of TI-based superconducting platforms. To address the discovery bottleneck, we leverage machine learning for topological material identification. Hybrid frameworks integrating chemical heuristics, physical descriptors, and language model embeddings have improved classification accuracy for topological insulators. Generative models with reinforcement fine-tuning have identified novel candidates, including Ge₂Bi₂O₆ with a 0.26 eV band gap. Graph neural networks predict spin-orbit coupled Hamiltonians, enabling high-throughput screening of thousands of materials. Our approach combines physics-informed machine learning models to predict topological character from compositional descriptors, with experimental validation on Bi₂Se₃/Nb devices. This framework aims to identify alternative systems exhibiting robust topological superconductivity while circumventing complexities observed in Bi₂Se₃-based junctions. Preliminary results suggest machine learning-guided discovery can substantially reduce computational cost, accelerating the pathway toward scalable, fault-tolerant quantum computing platforms.
| Topic | Quantum-enhanced robustness in machine learning models |
|---|---|
| Additional Topic | Quantum state reconstruction from data |