 Astronomy
 Biomedical Optics & Medical Imaging
 Defense & Security
 Electronic Imaging & Signal Processing
 Illumination & Displays
 Lasers & Sources
 Micro/Nano Lithography
 Nanotechnology
 Optical Design & Engineering
 Optoelectronics & Communications
 Remote Sensing
 Sensing & Measurement
 Solar & Alternative Energy
 Sign up for Newsroom EAlerts
 Information for:
Advertisers




Solar & Alternative Energy
Photovoltaic informatics accelerates processdesign speed
Changwon Suh, Stephen Glynn, John Scharf, Miguel A. Contreras, Rommel Noufi, Kristin Munch, Wesley B. Jones and David Biagioni
Dataintensive processes optimization is effected by simultaneously uncovering multiple interdependencies between properties and growth conditions of an essential solardevice component.
7 April 2011, SPIE Newsroom. DOI: 10.1117/2.1201103.003577
Enhancing photovoltaic (PV) functionality at low cost and by highrate processes is critically needed. Current challenges in PV development primarily arise from limited understanding of processproperty relationships, since PV data is increasingly heterogeneous, complex, and multiscaled. Therefore, an accelerated complex decisionmaking process is crucial for finding quantitative processingpropertyperformance relationships (QPPPRs). Our goal is to optimally combine processing routes to meet multiple PV performance requirements, regardless of laboratorytofabrication scaling. The immediate goal of copper indium gallium selenide (CIGS, CuIn_{x}Ga_{1−x}Se_{2}) devices is to achieve benchscale results, i.e., approximately 20% efficiency. To this end, our current research strategy is to identify QPPPRs based on PV informatics within each layer, from the molybdenum backcontact layer to the aluminumdoped zinc oxide (AZO) layer of the transparent front contact, and fuse them together for optimal CIGSdevice functionality. Our research focuses on experimentally and computationally finding ad hoc process recipes—for synthesizing stable AZO films—that meet specific multiple requirements for an AZO layer based on a CIGS device of 20% efficiency.^{1} PV informatics uses data mining to extract and visualize hidden information in massive arrays of data sets.^{2} Data mining is typically performed by socalled data origami, which focuses on folding, cutting, and pasting data for QPPPR extraction. The two primary advantages of PV informatics with respect to conventional data analysis for AZO data are simultaneous identification of complex QPPPRs and the ability to transform highdimensional AZO data into lowdimensional information visualization. Figure 1 illustrates the parallel coordinate,^{3} a highdimensional visualization approach—to visualize numerous correlations of crosslinked variables in AZO datasets—for overcoming dimensionality. The abscissa in parallel coordinates represents each variable and the ordinate represents ten different levels of variables. Interesting associations between factors (i.e., process conditions) and responses of AZO thin films are more easily visualized. For instance, the figures of merit at 500 and 550nm (#33 and #34, respectively) of the AZO films^{4, 5} result from noticeably low values of #2 (forward power of sputtering target) and #3 (process pressure of sputtering chamber). They also result from an increase in #13 (mobility) and #8 (film thickness), with decreasing #11 and #12 (film resistivities). Effects of #2 and #3 on thinfilm properties have been reported.^{6,7} We will soon publish parallelcoordinate results. Figure 1. Parallel coordinates simultaneously elucidating quantitative processingpropertyperformance relationships (QPPPRs) of aluminumdoped zinc oxide (AZO) films. Blue polygons represent the reference AZO sample, based on a copper indium gallium selenide (CIGS, CuIn _{x}Ga _{1x}Se _{2}) device of 20% efficiency. ^{1} Discontinuous polygons imply missing values of corresponding processing conditions in need of optimization. To more effectively elucidate QPPPRs, it is often useful to organize variables based on similarity. In Figure 2(a), we demonstrate the use of hierarchical agglomerative clustering (HAC) algorithms, representing clustering of reorganized AZO variables based on their pairwise similarities.^{8} Figure 2(b) is a heat map that visualizes the results of Figure 2(a). Our adapted HAC algorithm is a much more robust and systematic dataanalysis technique than the conventional HAC approach. It emphasizes local distance between points in the highdimensional feature (variable) space and captures nonlinearities in the data by diffusionmap embedding.^{9} Figure 2. (a) Hierarchical cluster tree generated from diffusionmap embedding. (b) Heatmap representation for QPPPRs of AZO data. The colors in the heat map represent function values, arranged according to the dissimilarity measure ρ(x_{i}, x_{j}) = 2  2corr(x_{i}, x_{j}). ρ: Dissimilarity function. x_{i}, x_{j}: Random variables of AZO data. Corr(x_{i}, x_{j}): Correlation between x_{i} and x_{j}. In summary, PV informatics efforts based on linear data mining have recently become pervasive, ranging from combinatorial experiments to defining search/design spaces of transparent conducting oxides.^{10,11} Linear projectionbased clustering of the features space is often inadequate, because PV data may be nonlinearly interrelated. Consequently, we also characterize the control process by nonlinear regression techniques.^{12} In future work, an accurate model of this form will expand our virtual metrology tools, explicitly optimizing desired features and exploring functional relationships between process controls and PV properties.
Changwon Suh, Stephen Glynn, John Scharf, Miguel A. Contreras, Rommel Noufi, Kristin Munch, Wesley B. Jones National Renewable Energy Laboratory Golden, CO David Biagioni Department of Applied Mathematics, University of Colorado Boulder, CO
References: 1. I. Repins, S. Glynn, J. Duenow, T. J. Coutts, W. K. Metzger, M. A. Contreras, Required material properties for highefficiency CIGS modules, Proc. SPIE 7409, pp. 74090M, 2009. doi: 10.1117/12.8283652. J. R. Rodgers, C. David, Materials informatics, MRS Bull. 31, pp. 9751021, 2006. doi: 10.1557/mrs2006.2233. A. Inselberg, Parallel Coordinates: Visual Multidimensional Geometry and Its Applications, Springer, 2009. 4. G. J. Exarhos, X.D. Zhou, Discoverybased design of transparent conducting oxide films, Thin Solid Films 515, pp. 70257052, 2007. doi: 10.1016/j.tsf.2007.03.0145. G. Haacke, New figure of merit for transparent conductors, J. Appl. Phys. 47, pp. 40864089, 1976. doi: 10.1063/1.3232406. R. Das, K. Adhikary, S. Ray, Comparison of electrical, optical, and structural properties of RF sputtered ZnO thin films deposited under different gas ambients, Jpn J. Appl. Phys. 47, pp. 15011506, 2008. doi: 10.1143/JJAP.47.15017. O. Kluth, G. Schöpe, J. Hüpkes, C. Agashe, J. Müller, B. Rech, Modified Thornton model for magnetron sputtered zinc oxide: film structure and etching behaviour, Thin Solid Films 442, pp. 8085, 2003. doi: 10.1016/S00406090(03)0094908. T. Hastie, R. Tibshirani, J. H. Friedman, The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Springer, 2009. 9. M. Maggioni, R. R. Coifman, Multiscale analysis of data sets using diffusion wavelets, Proc. Data Mining Biomed. Inf., 2007. Conf. presentation. 10. C. Suh, C. W. Gorrie, J. D. Perkins, P. A. Graf, W. B. Jones, Strategy for the maximum extraction of information generated from combinatorial experimentation of Codoped ZnO thin films, Acta Mater. 59, no. 2, pp. 630639, 2011. doi: 10.1016/j.actamat.2010.09.06811. C. Suh, K. Kim, J. J. Berry, J. Lee, W. B. Jones, Data miningaided crystal engineering for the design of transparent conducting oxides, 2010. NREL rep. CP2C0050079 12. G. Beylkin, J. Garcke, M. J. Mohlenkamp, Multivariate regression and machine learning with sums of separable functions, Soc. Indust. Appl. Math. J. Sci. Comput. 31, no. 3, pp. 18401857, 2009. doi: 10.1137/070710524


