Self-heating is among the most significant performance and reliability limiters for power amplifier (PA) devices and MMICs. GaN-based HEMTs, today’s most promising technology for microwave PAs, are a case in point, since exploiting to the fullest extent the outstanding properties of the AlGaN/GaN system necessitates that the
thermal aspect be carefully addressed in the device design and manufacturing. Finite Element (FE) modelin can be of great help in this respect, given the complexity of the structures, which are wide, three-dimensional, made of different materials (showing significant temperature dependence of the thermal conductivity), and include top- and back-side features and thermal boundary conditions that, if overlooked, end up compromising the modeling results. However, FE models can be computationally heavy and do not lend themselves to integration in comprehensive design suites, unlike lumped-element models , . In this work we use a Finite Element (FE) simulation tool to solve the heat transport equation, specifically focusing on aspects that are often overlooked in the modeling of self-heating: the impact of die-attach and of finite backside heat-sinking, the thermal boundary resistance (TBR)  between GaN and the substrate, and the effect of metal lines and pads. The final goal of the work, however, is using FE simulation to develop lumpedelement models for insertion into self-consistent electro-thermal simulations , . We analyze a HEMT structure made of cells of 12 fingers each; each finger is 150 μm wide, and the separation between adjacent fingers is 30 μm. The overall gate periphery of each cell is therefore 1.8 mm, but thanks to symmetry planes (which can be replaced by adiabatic boundary conditions), only half of the fingers and half of the finger width need to be modeled. For the same reasons, a symmetric arrangement of cells allows to limit the study to a single cell. This abstract considers the case of a SiC substrate; the other relevant case of silicon substrate will be illustrated in the final paper. The simulated 12-finger cell is shown in fig. 1. It has a 2.5 μmthick GaN layer on top of the SiC substrate; we considered two values for the substrate thickness: tSiC = 250 μm and 125 μm. We include in our simulations the often neglected TBR between GaN and substrate, as a 50 nmthick interfacial layer with contact thermal conductivity kC TBR. The back of the SiC substrate is stuck to a 40 μm-thick Sn(96%)-Ag(4%) die-attach layer with thermal conductivity kDA. Top-side source and drain metals (another often overlooked feature that significantly impacts the thermal budget) are 4 μm-thick gold. Boundary conditions are adiabatic everywhere, except on the back of the die-attach and on the top metal pads, where contact thermal conductivities, kC BACK and kC TOP, respectively, are varied to simulate different combinations between isothermal (kC = ∞) and adiabatic (kC = 0) conditions. GaN and SiC thermal conductivities are temperature-dependent . The dissipated power is 5.4 W for the whole 1.8 mm cell (i.e., PD = 3 W/mm). Fig. 2 shows the temperature profile along a vertical line originating in the center of the hottest (innermost) finger, for the two SiC thicknesses and for two backside boundary conditions: the ideal isothermal case (kC BACK = 0), and a realistic case where kC BACK = 3.6·105 W·K-1·m-2, corresponding to a case-to-ambient thermal resistance of 10 K/W for the 12-finger cell. In this case there is no heat exchange from the front-side pads (kC TOP = 0). Note the sharp temperature drop on the TBR (kC TBR = 3.0·107 W·K-1·m-2 ) and the impact of the dieattach layer (kDA = 45 W·K-1·m-1). The effect of heat removal from the front-side metal pads is illustrated by fig. 3, for both substrate thicknesses, in the case of kC BACK = 3.6·105 W·K-1·m-2, kC TBR = 3.0·107 W·K-1·m-2, kDA = 45 W·K-1·m-1. Moving from the case of adiabatic top (kC TOP = 0) to better and better heat removal from the frontside pads, the peak channel temperature (TMAX) is drastically reduced. Fig. 4 shows the impact of the GaN/SiC TBR on TMAX. Here kC
BACK = 3.6·105 W·K-1·m-2, kC TOP = 106 W·K-1·m-2, tSiC = 125 μm, kDA = 45 W·K-1·m-1, and kC TBR varies between 107 and 108 W·K-1·m-2. Finally, fig. 5 shows the role of the die-attach thermal conductivity kDA (kC BACK = 3.6·105 W·K-1·m-2, kC TOP = 106 W·K-1·m-2, tSiC = 125 μm, kC TBR = 3.0·107 W·K-1·m-2). The full paper will include (1) the case of Si substrate, (2) dynamic (step response) simulations, (3) a study of via holes, and (4) the use the FE simulations to develop a lumped-element thermal model.