Small-Signal Analysis and Direct S-Parameter Extraction

We present a comparison of several device parameters obtained by fully two-dimensional physical device simulation of several device parameters and measurements. Several features for direct extraction of either extrinsic or intrinsic (de-embedded) parameters sets from AC-simulation were implemented in the three-dimensional device simulator Minimos-NT [1]. The scattering parameters (S-parameters) and other figures of merit of a complex heterostructure device have been obtained directly by using small-signal simulation. Thus, we use a combination of rigorous III-V group and IV group semiconductor materials modeling and the ability to simulate in the frequency domain. The results were verified both by analytical methods and by comparison with measurement data. Introduction Two-dimensional device simulation has proven to be valueable for understanding the underlying device physics and for improving the device reliability [2, 3, 4] of advanced heterostructure device, such as Heterojunction Bipolar Transistors (HBTs). The highly sophisticated physical models can also be used for high-frequency simulations, the results of which are required and commonly used for future device and circuit designs. For example, S-parameters are an extremely useful design aid provided by manufactures for high-frequency transistors. Normalized incident and reflected waves are used to characterize the operation of a two-port network. Thus, in contrary to the Y-parameters, no short circuit is required which often causes unstable devices and prevents measurements. Small-signal parameters are often taken to extract other figures of merit, such as the cut-off frequency for current gain ft or the maximum oscillation frequency fmax, which characterize a device and technology or emphasize their superiority, respectively. The simulation of these parameters can be based on several approaches, e.g. Fourier decomposition, applying quasi-static or equivalent-circuit parameter models [5, 6]. These approaches use a transient simulation mode and are both more CPU-time consuming and more inaccurate. Furthermore, they can perform only a limited number of time steps in reasonable CPU-time (basically, many time steps are necessary to obtain sufficient time-domain accuracy to generate sufficient frequency-domain accuracy). Usually, the transient effort must be reduced by extracting an equivalent circuit using the results of only one frequency. Our direct small-signal simulation mode is based on the SA (sinusoidal steady-state analysis) approach presented in [7], which is rigorously correct, regarded to be computationally inexpensive and very accurate. The high accuracy is due to the formal linearization of the device and because of doing the calculations directly in the frequency domain. Fig.1 sketches comparison of the different approaches. Since the simulator offers both simulation modes, we were able to directly compare the computational effort. The paper includes a description of the implemented small-signal simulation mode and a short overview about the physical modeling used in the simulator. We present the current state of development by showing six simulation examples followed by an analysis of the computational effort.