HBT Model Scaling for Different Epi Materials and Geometries Yingying Yang, Peter J. Zampardi* and Kai Kwok Skyworks Solutions, Inc., 2427 Hillcrest Drive, Newbury Park, CA 91320 yingying.yang@skyworksinc.com , (805) 480-4289 *Formerly of Skyworks Abstract — In this paper, we extend our previous physics- based scalable models to include the scaling of transit time with base thickness and base doping based on a design of experiment (DOE) and to include the variation of ledge length as a device design parameter. Index Terms — HBT, Modeling. I. INTRODUCTION The rapid growth of communication applications and increasingly stringent specifications pushes HBT fabs to support many HBT geometries (shapes and sizes) with a variety of materials (doping layers and profiles) [1]. Given shorter product design cycle-times, there is increasing pressure on model turn-time too. Previously, we demonstrated a unified modeling approach (sharing model parameters among devices that share the same Epi layers, e.g. HBT and BE diode, BC diode and base resistors) generating DC [2] and base resistance [3] scalable model equations for our III-V model library (containing all on- chip devices for number of epi material types). This modeling approach provides designers with the flexibility Fig. 1. τ intercept and peak f T versus calculated τ B . The lines are the fitting results. The red line is for W C of 0.9 ȝm and the blue line for W C of 0.75 ȝm. The solid lines are for τ intercept and dashed lines are the peak f T . The data is shown as symbols. to select optimal devices (sizes and shapes) for their applications (frequency band, operation mode, power level, etc.), and to easily simulate physical process variation with minimum simulation runs to ensure high yield even before production ramps [4]. One of the “missing pieces” in the previous work was a set of expressions for the transit time that reflect intentional changes or process variation in the base layer parameters (doping, N B , or thickness, W B ). This is particularly important to help with process development/comparison where base parameters are very different between two processes/foundries. It also, obviously, is important for process variation simulation. A second extension is the incorporation of ledge length variation in the models – which can significantly affect DC current gain, β. This allows the ledge length to be used as a design parameter to allow application dependent trade-offs. In this paper, we extend the previous statistical models, to include emitter, base transit time scaling for different epi materials (with a wide range of β), and emitter ledge impact on DC parameters scaling. These enhance the previous modeling approach (one basic physical model for different epi materials and device geometries) and eliminate fitting for each individual epi material and individual device. II. BASE TRANSIT TIME SCALING FOR DIFFERENT EPI MATERIALS A. Obtain Scaling Equation In most epi-material designs, the active emitter layer is not modified (although emitter contact/setback layers can be different). In contrast, the N B and W B , collector doping, doping profile, and thickness, W C , are optimized for different design specifications. Rather than fitting the base transit time, τ B , for each epi material and device, we use material DOE to intentionally generate materials that span a wide range of N B and W B , and W C (shown in Table 1). This is especially useful for the Agilent HBT model [5] since τ B is a stand-alone parameter. The bases and collectors in the DOE are all uniformly doped. A single- device-size f T is measured for all splits. Using Cooke’s method [6], the low-current-density fitted y-intercepts 978-1-4799-8494-7/15/$31.00 ©2015 IEEE