Scholarlink

<p>The transport properties of Josephson junctions (JJs) directly influence quantum tunneling effects and play a critical role in the performance of superconducting qubits. However, accurately predicting the electrical properties of JJs based on three-dimensional atomic models remains a considerable challenge, and further revealing the relationship between structure and electrical properties from a complete junction model proves even more difficult. To address this challenge, this study adopts the Al/AlO<sub>X</sub>/Al JJ as a model system, constructs a dataset comprising defect-free and defective atomic models along with their corresponding electron transport properties, and proposes a voting regression ensemble strategy, termed Lrmr. By comparing the conductance prediction performance of five baseline machine learning algorithms, the Lrmr framework demonstrates significantly improved prediction accuracy and stability and identifies key features governing the electrical properties of the junctions. This data-driven framework provides an effective tool for deepening the understanding of the electronic behavior of JJs and for optimizing the performance of superconducting devices.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The prediction of transport properties in Josephson junctions based on first principle and machine learning

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Junling Qiu,
  • YuQing Xing,
  • Yuan Ma,
  • Min Ma,
  • Xiaodong Ding

摘要

The transport properties of Josephson junctions (JJs) directly influence quantum tunneling effects and play a critical role in the performance of superconducting qubits. However, accurately predicting the electrical properties of JJs based on three-dimensional atomic models remains a considerable challenge, and further revealing the relationship between structure and electrical properties from a complete junction model proves even more difficult. To address this challenge, this study adopts the Al/AlOX/Al JJ as a model system, constructs a dataset comprising defect-free and defective atomic models along with their corresponding electron transport properties, and proposes a voting regression ensemble strategy, termed Lrmr. By comparing the conductance prediction performance of five baseline machine learning algorithms, the Lrmr framework demonstrates significantly improved prediction accuracy and stability and identifies key features governing the electrical properties of the junctions. This data-driven framework provides an effective tool for deepening the understanding of the electronic behavior of JJs and for optimizing the performance of superconducting devices.