Optimization of an advanced multi-junction solar-cell design for space environments (AM0) using nearly orthogonal Latin hypercubes

Abstract

This thesis focuses on the replacement of a genetic algorithm currently used to optimize multi-junction solar cells with Silvaco Atlas simulation software. It introduces the nearly orthogonal Latin hypercube (NOLH) design of experiments (DoE) as a means for exploring and optimizing solar cell designs in Silvaco Atlas. The general applicability of this approach has been proven, and compared to the genetic algorithm optimization technique, the following achievements have been made. The pre-generated simulation designs can now be processed in parallel, which drastically reduces the time required to conduct multiple simulations. Moreover, the data generated using the NOLH enabled a better understanding of the simulation input/output relationship and helped to focus the solar cell development by highlighting the design parameters that matter most. Using stepwise regression to build a metamodel helped in finding an optimal design and revealing the interactions among the input parameters. The initial simulation has already yielded promising results and has clearly shown the preeminence of the NOLH over the genetic algorithm by identifying a design with greater than 21% more power output than in previous designs. The NOLH DoE should become the new standard for optimizing solar cells in Silvaco Atlas.