Recent advances in the development of semiconductor devices lead to more and more complex device structures. This concerns device geometry as well as the combination of different materials. Due to the rapid reduction of device geometries, the models describing the device physics increase in complexity. Traditional device simulation considered the behavior of isolated devices under artificial boundary conditions. To gain additional insight into the performance of devices under realistic dynamic boundary conditions imposed by a circuit, mixed-mode simulation has proven to be invaluable. However, the solution of this problem is very complex and only limited solutions have been available so far.
In this work different coupling strategies of device and circuit simulators are investigated. As the combined device-circuit simulator approach promised the best benefits the device simulator MINIMOS-NT has been extended with mixed-mode capabilities.
The classical Newton method provides quadratic convergence for an initial-guess sufficiently close to the final solution. This region of attraction can be enlarged by providing a damping algorithm for the solution variables. Completely different approaches are in use for circuit and device simulation. The usefulness of these algorithms is investigated for the problem of mixed-mode device simulation. A method is proposed which allows for solving medium sized circuits without a user-specified initial-guess with a small number of necessary iterations.
Due to the ongoing downscaling of semiconductor devices non-local effects become more and more pronounced. They can be modeled with a good accuracy using the hydrodynamic transport model. However, the convergence properties of the hydrodynamic transport model are inferior compared to the simpler drift-diffusion transport model which cannot cover non-local effects. To estimate the influence of these non-local effects comparisons between drift-diffusion and hydrodynamic simulations are necessary. Several conditions must hold for these comparisons to deliver meaningful results, an issue which is discussed in detail.
The new features of the simulator are tested with typical analog and digital circuits. The operating point could be found in many situations starting from the equilibrium without any initial-guess using the new embedding method of circuit elements. Drift-diffusion and hydrodynamic simulation results are compared pointing out the necessity of the hydrodynamic transport model for state-of-the-art devices.
As a final example thermal feedback of a complete operational amplifier is investigated. To model the thermal interaction between the transistors the lattice heat flow equation is solved in conjunction with a thermal network. This thermal network provides a connection of the input and output stage of the circuit thus approximating the temperature distribution along the chip. The complexity of this simulation can be considered well beyond the capabilities of commercially available simulators.