Intelligent Software Package for Predicting Thermal Resistance of Semiconductors
Abstract
Intelligent Software Package for Predicting Thermal Resistance of Semiconductors
Incoming article date: 11.08.2025The article presents the architecture and implementation of an intelligent software package (ISP) for predicting the thermal resistance of semiconductor devices, in particular MOSFET transistors, at the design stage. The developed system combines physical and mathematical modeling of multilayer heat-conducting structures with machine learning methods, which allows for an accurate prediction of thermal parameters based on engineering characteristics and case design. The ISP implements a mechanism for automatically supplementing incomplete data using a knowledge base of typical parameters of domestic and foreign devices. Models were trained on a synthetically expanded sample formed taking into account the thermal conductivity of structural materials and layer geometry. Among the algorithms used are ensembles of random forests and gradient boosting, as well as neural network models. An analysis of the importance of features was carried out, key parameters that determine them were identified, and the possibility of using the ISP for early assessment of thermal conditions in CAD and CAE environments was demonstrated.
Keywords: thermal resistance, MOSFET, machine learning, intelligent software system, multilayer structure, predictive model, CAD, thermal conductivity