During the last decades numerical device simulation has proven to be invaluable for characterizing semiconductor devices. The economic impact is enormous because results of expensive experiments can be predicted by employing one or more simulation tools. Furthermore, the results can be used to optimize the investigated device structures and to single out unpromising variations in advance. The application of simulation tools has been significantly intensified because of the computational power and the available memory in today's average computers.
To meet the requirements for the simulation of advanced devices, ongoing effort is put into extension of these tools by implementations of state-of-the-art descriptions of all relevant physical effects. Besides of that modeling the simulators have also to be extended by new simulation modes. In course of this work, the requirements for small-signal simulations have been identified and new features have been added to the general-purpose device and circuit simulator MINIMOS-NT.
After a short introduction including a motivation and overview of the current market situation, the analytical problem of the respective simulations is derived. As advanced device structures cannot be correctly simulated by the drift-diffusion transport model any more, higher-order transport models such as four- and six moments models are also taken into consideration. In addition, the small-signal systems are derived which are based on the sinusoidal steady-state approach. Afterwards, the various new capabilities of the simulator are presented and applied for typical simulation tasks.
All features have been used to characterize advanced devices, such as InGaP/GaAs and SiGe HBTs, a wide-bandgap SiC MESFET, and double gate MOSFETs. The latter require higher-order transport models in order to accurately extract the steady-state and small-signal device quantities. Furthermore, the mixed-mode small-signal features have been used to simulate a Colpitts oscillator.
Since the small-signal simulation mode is directly based in the frequency domain, the solution of one complex-valued equation system per frequency step is required. For that reason, the numerical core modules have been extended to handle both real-valued and complex-valued quantities. The application programming interface of the new modules allows an efficient and user-friendly assembling and solving of linear equation systems.
In order to profit from new developments of mathematical code, the solver module has been extended by an interface to external solvers. Several promising linear solvers which have particular strengths for different kinds of simulations have been coupled to the solver module. In the course of a subsequent performance evaluation, these strengths have been identified and efficient solver hierarchies have been constructed.