Accurately predicting the temperature of semiconductor devices is very important in the initial design of power electronics converter. RC thermal models derived from well-known methods have some ability to predict the temperature. However, the accuracy is boundary condition specific, hence, these methods cannot be used in the reliability analysis. To make the thermal model more accurate and robust the factors contributing to discrepancies need to be analyzed carefully. These are powermodule-materials' non-linear properties, thermal grease layer and the cooling system (i.e., liquid-cooled cold plate). In this work, estimation of accurate RC parameters from FEA thermal model is demonstrated in COMSOL. The electrical model having temperature dependent power loss model is coupled to refined thermal model and solved in a circuit simulator, PLECS. The proposed method is applied in two applications: assessing thermal interaction between IGBTs and anti-parallel diodes in a halfbridge power module, and assessing thermal interaction among the discrete switches in an interleaved bidirectional DC-DC converter. Results show that the impact of material non-linearity, thermal grease layer and cooling boundary conditions are significant for accurate prediction of IGBT and diode temperatures. The proposed model is consistent to FEA results and differs by 2-6.5% comparing to the experimental results.