A study on the prediction of MoS₂ transistor electrode contact characteristics based on transfer learning methods

Data scarcity is one of the key bottlenecks in the application of machine learning in the field of materials discovery. In this challenge, transfer learning can leverage existing consistent large-scale data to assist in property prediction on small datasets, thereby opening up more possibilities for materials development. With the discovery of many-dimensional materials and the challenges posed by the miniaturization of transistors, the range of electrode materials available for transistors is extremely broad, but it is difficult to explore them through traditional experimental methods. Therefore, in the face of scarce data on electrode contact characteristics, this study proposes a cross-scale hybrid transfer learning framework that integrates first-principles calculations with a 2D materials database. By utilizing large-scale potential height data obtained through PBE functional calculations, the framework achieves high-precision predictions of DFT-1/2 method and HSE06 functional calculation results, with an MSE controlled within 0.04 eV. The research results indicate that this learning framework accelerates the process of screening electrode materials for MoS₂, providing important theoretical guidance and technical support for the design and optimization of new electronic devices.