The research objective of this award is to establish statistics-transformed nanostructure growth process models and efficient experimental strategies for improving process repeatability in the fabrication of nanostructures for the application in photovoltaic cells. To achieve repeatable fabrication of photovoltaic cells with respect to yield (productivity) and uniformity (quality), it is essential to identify and optimize the growth conditions rooted on predictive process models. These models will capture the mechanisms of nanostructure growth under process uncertainties. Since most of the current growth kinetics models are deterministic, the research will first devise statistics-transformed nanostructure growth process models that account for uncertainties. Based on the process model, optimal experimental strategies will be established for model estimation and validation with a high degree of precision under cost and time constraints. The methodology will be validated through controlled growth of nanowires and fabrication of photovoltaic cells.
Successful completion of this research will lead to new tools and methods for improving process repeatability and yield in nanomanufacturing, particularly in the large scale fabrication of photovaic cells. Reducing cost in photovoltaics gives prospect of achieving highly efficient and low-cost solar energy conversion, increasing the utilization of clean and renewable energy, and creating green job opportunities. This truly interdisciplinary project will promote training of a new breed of workforce excelling at nanomanufacturing process modeling and optimization and contributing to the sustainable growth of US economy. The educational goal will be achieved through (1) creating interdisciplinary nanomanufacturing curricular materials, (2) enhancing the existing research and education collaborations between University of Southern California and Harvard University, and (3) involving women/minority students through REU (Research Experience for Undergraduates) program. The generated knowledge will be broadly disseminated through leading journals, conferences, websites, and collaborators.
Nanotechnology is a buzzword in the field of science and technology, and nanowires are building blocks of nano-devices. To scale-up the synthesis process of nanowires from the laboratory to an industrial setting, it is of utmost importance to understand the growth conditions better, relate these conditions to potential sources of noise in a scaled-up environment, and predict how scaled-up systems should be designed for higher yield, quality and throughput. Such an understanding can be facilitated by conducting experiments in the laboratory under a vast multitude of conditions, building predictive models and predicting the yield under different conditions as well as the uncertainty associated with those predictions. There are, however, serious resource constraints associated with these experiments in terms of time and cost. The focus of this project was to develop novel statistical principles for designing efficient, yet effective experiments, which would lead to good prediction models with limited resources without wasting many experimental runs. The research done in the past three years has addressed the aforementioned problem. The key results of the research are expected to provide nano-scientists with a principled approach to conduct economic experiments to study growth-curve of nanowires. Specifically, it provides guidelines on (a) how to utilize available physical knowledge to conduct efficient experiments that lead to good prediction models (optimal designs) (b) how to integrate physics and statistics to build models from experimental data, and (c) how to sequentially "learn" from the experiments and conduct subsequent experiments that are more "promising" in terms of potential high yield or better quality and (d) how to analyze data from nanostructure synthesizing experiments better. Among the above four points, (c) can be illustrated by the attached two-dimensional figure, where the experimenter is interested in conducting experiments by varying two inputs x1 and x2 in the blue box. However, the "effective" region where some meaningful yield takes place is a small banana-shaped sub-region of this box, which the experimenter is totally unaware of. In the absence of the proposed methodology, experimenters would have to conduct experiments at arbitrary points within the box, and waste plenty of runs before getting a sense of which regions are promising. Our methodology is able to "carve out" such regions quickly, and generate most points within the region of interest, increasing the efficiency of experimentation many-fold. As a byproduct, the research has also led to some methodological developments in the field of statistics and computing. Simulating models that capture uncertainty of physical systems is a rich area, receiving plenty of attention these days. Some of the statistical methodology developed by our research team can be used in performing such simulation experiments more efficiently, but generating outcomes only within specific regions of interest. They are expected to find applications in some challenging statistical computing problems as well. As a consequence of the collaboration with the University of Southern California, the research has indirectly led to the development of new tools and methods for improving process repeatability and yield in nanomanufacturing, particularly in the large scale fabrication of photovoltaic cells. Reducing cost in photovoltaics gives prospect of achieving highly efficient and low-cost solar energy conversion, increasing the utilization of clean and renewable energy, and creating green job opportunities. There have been extensive and productive collaborations among PIs from Harvard and USC. As an example, investigators from the University of Southern California (USC) and Harvard University organized a one-day research workshop at USC in January 2012 in which 6 PhD students (2 from Harvard) participated. The synergy generated by the workshop was manifested not only in the form of joint publications, but also in the form of the rich inter-disciplinary research based on the integration of statistical and engineering thinking, to which these doctoral students were exposed. The graduate students involved in this project have gained substantial experience and expertise in conducting inter-disciplinary applied research at the interface of statistics and physics/engineering. The doctoral thesis of one graduate student at Harvard was motivated by this project. The student acquired a substantial amount of research experience by working on this project and by presenting his research findings at conferences.
