| The objectives of this thesis were to develop accurate and useful models for the MBE process and to use these models to synthesize novel process recipes. These goals were accomplished by: (1) the use of statistical experimental design to assist in experimental planning and analysis of measurement results; (2) the identification of the main effects and factor interactions; (3) the physical interpretation of significant effects; and (4) the innovative use of neural networks and genetic algorithms to model and optimize the process.; To successfully utilize the statistical experimental design approach to build models for the MBE process, two design methods were pursued. The fractional factorial experimental design was employed for AlGaAs/InGaAs SQW growth. For GaN growth, a D-optimal design was used. These methods facilitated the gathering of knowledge about the important input factors, the resultant responses, and the range over which these factors should be varied, allowing one to derive a better physical understanding and draw conclusions objectively by studying the various physical phenomena. Among a few new observations is the remarkable difference of the properties of reconstructed films and nonreconstruced films during the GaN growth study.; For AlGaAs/InGaAs SQW process modeling, a novel modeling method has been developed using time-series neural networks to build a model for RHEED intensity variation. The importance of this accomplishment was that models were derived using real-time RHEED intensity measurement with substrate rotation, providing a better practical solution for manufacturing environment than non-rotated substrate.; Detailed experimental procedures and their results were discussed based on the neural network models. Physical interpretations based on the neural network model were also presented. The neural network models not only agreed with known information published in the literature, but also revealed a few new relationship between the interplay of growth conditions and the final material qualities.; Finally, the optimization of the MBE process characteristics was performed by means of genetic algorithms. This genetic optimization approach revealed GaN growth conditions that can result in a great improvement in the final material quality. |
