Abstract

In modern semiconductor technology ion implantation has turned out to be the most important technique to introduce dopant atoms into semiconducting materials. The major advantage of the ion implantation technique is the high controllability and reproducibility of the process parameters influencing the doping distributions. Furthermore, very shallow doping profiles can be formed, which are a prerequisite for ULSI (ultra large scale integration) technology.

Since it is mainly ion implantation which determines the distribution of the dopants and thereby the electrical properties of the semiconductor devices highly accurate simulation methods for ion implantation processes are required to be able to predict and optimize the behavior of integrated circuits. In recent years successively shrinking device dimensions and new design concepts have shown the necessity of a full three-dimensional treatment of simulation problems, e.g. the simulation of MOS transistors with narrow gates, or vertical transistors.

Three-dimensional simulations obviously require large computation times and a lot of memory. Therefore, it is a waste of computational resources if a three-dimensional simulation would be applied to all applications. Several problems, like the buried layer or the well formation of an MOS transistor can be analyzed as accurate by simpler two-dimensional or even one-dimensional simulations. Since it should be easy to switch the dimension of the simulation without recalibrating a simulator, it is not desirable to use different simulators, which eventually use different models, for the simulation of one-dimensional, two-dimensional and three-dimensional problems.

The goal of this work was to further improve a Monte-Carlo ion implantation simulator developed over the last fifteen years within the scope of several PhD theses. As part of this work several new models and methods have been developed and implemented to improve the accuracy and the efficiency of the simulator, in order to be applicable to modern ULSI technology. Besides an enhancement of the models, the simulator has been redesigned in a way to cover one-dimensional, two-dimensional and three-dimensional problems. Thereby the simulator has been made adaptable to the specification and requirements of a certain problem, which allows to avoid a waste of computational resources.

The functionally of the simulator and the theoretical background is presented in detail, especially focused on methods and algorithms developed as a part of this work. Worth mentioning among them is first the Follow-Each-Recoil method which allows a very accurate simulation of the implantation induced damage. By using the Follow-Each-Recoil method it is possible to provide very accurate input data for the simulation of rapid thermal annealing processes. As well the precise distribution of point defects and the formation of amorphous areas can be simulated. Furthermore the full and the simplified molecular methods have been developed and implemented. These methods enable to treat the implantation of molecular ions and atom clusters and thus the implantation of BF$_2$, which is a widely used for the doping with boron atoms. By providing two methods for the simulation of molecular ions the functionality of the simulator can be adapted to the problem requirements. While the simplified molecular method needs less computation time, the full molecular method provides more precise results.

Another part of this work was the design and the implementation of a point response interface method. It allows to interface Monte-Carlo simulation results to an analytical ion implantation simulator. Thereby the flexibility and the accuracy of an analytical simulator can be significantly improved, because the calibration of an analytical function for certain process conditions is not necessary.

For the reduction of the simulation time two new methods have been developed. On the one hand side the Trajectory-Reuse method, which enables a significant speed up of the simulation. Especially in case of three-dimensional simulation problems and if the simulation domain consists of large amorphous areas the reduction in computation time is remarkable. On the other hand side the simulator has been parallelized in a way to be able to perform the simulation on a cluster of workstations which is the typical TCAD (technology computer aided design) environment. If the load of the workstations participating in the simulation is approximately constant an almost linear performance gain could be achieved by the parallelization method, even if a fairly slow network connects the workstations.

Finally, the developed Monte-Carlo ion implantation simulator is applied to a set of examples making use of some of the special features of the simulator. Additionally a small operating manual for the simulator is included in the appendix.