Thermal Modeling of Wire-bonded Power Modules Considering Non-Uniform Temperature and Electric Current Interactions M. Akbari a , A.S. Bahman b , P.D. Reigosa b , F. Iannuzzo b , M.T. Bina a a Power Electronics Laboratory, K. N. Toosi University of Technology, Tehran, Iran b Center of Reliable Power Electronics (CORPE), Aalborg University, Aalborg, Denmark Abstract To assess power devices’ reliability, it is crucial to have a relatively accurate thermal approach which provides valid temperature estimates. In this paper, for two commonly used Si IGBT and SiC MOSFET power modules, the electric current-induced effects on bond wires and the correlation between the non-uniform temperature distribution and electrical conductivity of the sensitive constituent materials are studied. In addition, a more realistic active area of the modules is defined by excluding inactive regions, i.e., the gate area, gate runners, and termination ring. Also, the electric current distribution among parallel bond wires attached to the dies’ metalization pads is investigated. Comparisons made between an approach which includes all the above aspects with a conventional one where uniform heat is injected into semiconductor dies, although showing acceptable error in Si IGBTs, result in a very significant difference in SiC MOSFETs. 1. Introduction In the operation of power electronics converters, semiconductor devices are often the most vulnerable component. For example, in [1] it is presented that a significant proportion, almost 25%, of power converters’ failures may arise in the semiconductor devices. Accordingly, the lifespan reliability of semiconductor devices becomes of great importance in early stages of development of power converters. Among failure causes, the operating life of semiconductor devices is strongly influenced by the thermal cycling coming from a mission profile. For example, in [2] it is demonstrated that the temperature causes over 50% of failures where bond wire lift-off and solder joint fatigue are two primary thermal failure mechanisms due to the temperature swing (ΔT) and the difference in coefficients of thermal expansions (CTEs) of the constituent materials [3]. Consequently, an accurate knowledge of the semiconductor devices’ temperature is indispensable to improve the converter reliability. The advantage of that can be better understood if one considers that a 10°C change in a power device’s temperature can result in approximately 50% change in the estimated lifespan [4]. Various approaches have been introduced in the literature to find power devices’ surface temperature so far. One way is to integrate an NTC resistor or an on-die thermal diode into a device [5, 6], but it needs fundamental design modifications and auxiliary external pins, which increase the manufacturing cost and introduce new reliability concerns. Another way which has been introduced in some research works is the use of temperature-sensitive electrical parameters (TSEPs), e.g., the on-state voltage at low (sense) current levels [7]. However, the accuracy of TSEP approaches is an intrinsic limit together with the measurement circuit complexity. Furthermore, TSEPs provide an average temperature of the die rather than the surface temperature distribution [8]. One may use optical methods such as optical fibers [9], and infrared cameras [10] in order to map the devices’ temperature. Nevertheless, they require an intrusive modification of the device (e.g., removal of the package or dielectric insulating gel) such that they are not applicable for on-line measurement during the converter operation. Many research works have focused on the use of numerical methods such as finite element method (FEM) [11], finite difference method (FDM) [12], and finite volume method (FVM) [13], in order to mimic as accurate as a possible experimental condition. The difference between these methods is the mathematical models or governing equations used for the computation. FEM is used by most simulators as it allows a relatively fair approximation with lower calculations when compared with two other ones [14]. Although numerical methods give an accurate temperature distribution of devices, limitations have been found, which are rarely addressed in the literature so far. One of the limitations is that the power loss/heat is typically assumed to be independent of electrical variables, hence uniformly