Neural Network Modeling of Size-Dependent Thermal Conductivity in Semiconductor Nanostructures

Accurate prediction of thermal conductivity in semiconductor nanostructures is essential for the design of nanoscale electronic and thermoelectric devices, where heat transport is strongly influenced by size effects. In this study, a physics-guided neural network framework is developed to model the size-dependent thermal conductivity of semiconductor nanostructures. A physically motivated model is employed to generate reference data, capturing the gradual suppression of phonon–boundary scattering with increasing characteristic size. The neural network is trained using this dataset to learn the underlying nonlinear relationship between thermal conductivity and characteristic size. The results demonstrate that the proposed model successfully reproduces the global trend predicted by the physical model, while maintaining high predictive accuracy across a broad size range. Parity analysis confirms strong agreement between predicted and reference values, with minor deviations at higher conductivity levels. The proposed approach provides an efficient and flexible alternative to purely analytical models and can be extended to data-driven or hybrid frameworks for nanoscale thermal transport analysis.

کلیدواژه ها:

Thermal conductivity ، Semiconductor nanostructures ، Size effect ، Neural networks ، Phonon boundary scattering ، Physics-guided modeling