Abstract: In terms of the electro-thermal design, condition monitoring, and failure analysis, the junction-case thermal resistance is a key specification of the double-sided cooling (DSC) power module. Nevertheless, due to the inherited single-channel thermal path of the single-sided cooling power module, the traditional thermal model and measurement standard are challenged by the emerging DSC power module with multiple-channel thermal paths. As a result, the intrinsic thermal impedance of the DSC power module is missed. On the basis of the machine learning method, the intrinsic thermal impedance model of the DSC power module is proposed by using the neural network (NN). The concept of the intrinsic thermal impedance is clarified for the DSC power module. Besides, the feasibility and uniqueness of the reconfigured intrinsic thermal impedance are confirmed. Moreover, the difference between the traditional and proposed thermal impedances is assessed. Concerning the commercial DSC power module, the thermal impedance database of the DSC power module is created by using the multiphysics tool. Furthermore, the NN-based intrinsic thermal impedance of the DSC power module is trained and cross-checked with aid of the created massive training and testing datasets. Additionally, the measured and characterized thermal impedances of several commercial DSC power modules are demonstrated. In this way, the generalization and compatibility features of the proposed NN model to reconfigure the intrinsic thermal impedance are ensured. The findings might support the modeling and characterization of the multiple-channel thermal impedance with insightful research routine and technology solution.