Materials in small device structures underly certain parameter fluctuations as much as bulk materials, but due to the small dimensions of the material regions, the impact of the fluctuations is much higher as compared to bulk material. Hence, fluctuations have to be considered from the beginning of the design. Even if an optimal device has been designed, the characteristic after its fabrication might be completely different. In order to minimize these discrepancies, certain technology-specific constraints have to be introduced which have to be considered within the optimization frameworks to improve the characteristics. To perform the optimization tasks the state-of-the-art simulation and optimization framework Simulation Environment for Semiconductor Technology Analysis (SIESTA) [44] is used and refined in this thesis which provides an open interface that allows to easily add new software tools. SIESTA can be used with several optimization strategies for specific optimization tasks. The optimizer varies and proposes values for the unknown or uncertain parameters. The framework sends the parameters from the optimizer in an appropriate format to the simulators. The simulator may be arranged in a simulation tool flow where the output of one simulator is submitted as the input to another simulation tool. At the end of the simulation flow, the quality of the final simulation result is determined by an objective function which returns a score value which is a quantified representation of the quality of the simulation result. The following presents typical applications in which optimization is used.

Parameter extraction can be used to identify model parameters which are not accurately known [47]. The required input data for this task includes the simulation software with the appropriate models, as well as measurements or reference data to which the simulation result can be compared, and a score function (or objective function) that determines the quality of the simulation result. This extraction mechanism uses the inverse modeling technique [48,49,50], which is often performed to characterize novel device structures and new materials as well as material compositions in order to develop compact models at a specified scope.

Calibration is a special case of the parameter extraction [49]. The range of the uncertain parameters can be further constrained which enables in general faster convergence to complete the calibration task. The main difference between parameter extraction and calibration is that calibration needs a much higher accuracy because the initial guess is normally very close to the optimum, but should be further improved, if for instance a sample has to be calibrated to a certain set of measurements to minimize the model error. Due to the higher quality demands, the determination of the quality of the simulation result is a very critical issue for calibration. These quality criteria (objective or score functions) have to be specified by the user for each particular problem class and tuned for each individual problem. This function can include comparisons of absolute and relative values to calculate a significant metric to determine the quality of the simulation result with respect to reference data.

General optimization is the most general approach and can be used for arbitrary purposes. The optimization is performed until a certain quality criterion has been reached. There exists a wide range of applications for the optimization related to TCAD or electronic devices [51,52]. More general electronic design purposes have been discussed in [53,54], and specific optimizers and application for other regimes for instance in economics have been discussed in [55,56].

With a rigorous implementation of the major aspects occurring in a particular setup problem the optimization framework is able to minimize or maximize certain figures of merit within user-defined specifications. Hence, many trade-offs can be optimized together to obtain a reasonable solutions for the specified problem.

Stefan Holzer 2007-11-19