Democratizing an electroluminescence imaging apparatus and analytics project forwidespread data acquisition in photovoltaic materialsJustin S. Fada, Nicholas R. Wheeler, Davis Zabiyaka, Nikhil Goel, Timothy J. Peshek, and Roger H. French Citation: Review of Scientific Instruments 87, 085109 (2016); doi: 10.1063/1.4960180 View online: http://dx.doi.org/10.1063/1.4960180 View Table of Contents: http://scitation.aip.org/content/aip/journal/rsi/87/8?ver=pdfcov Published by the AIP Publishing Articles you may be interested in Quantitative analysis of electroluminescence images from polymer solar cells J. Appl. Phys. 111, 024505 (2012); 10.1063/1.3677981 Electroluminescence imaging of organic photovoltaic modules Appl. Phys. Lett. 97, 233303 (2010); 10.1063/1.3521259 Analytic findings in the electroluminescence characterization of crystalline silicon solar cells J. Appl. Phys. 101, 023711 (2007); 10.1063/1.2431075 Analysis of images from videocameras in the Frascati Tokamak Upgrade tokamak Rev. Sci. Instrum. 75, 4082 (2004); 10.1063/1.1789262 Mission to Planet Earth: Who provides, controls, and owns the data? AIP Conf. Proc. 387, 175 (1997); 10.1063/1.51988

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 129.22.124.78 On: Fri, 12 Aug

2016 20:38:14

REVIEW OF SCIENTIFIC INSTRUMENTS 87, 085109 (2016)

Democratizing an electroluminescence imaging apparatus and analyticsproject for widespread data acquisition in photovoltaic materials

Justin S. Fada,1,2 Nicholas R. Wheeler,2,3 Davis Zabiyaka,2,3 Nikhil Goel,2,4

Timothy J. Peshek,2,3,a) and Roger H. French2,31Department of Mechanical and Aerospace Engineering, Case Western Reserve University,Cleveland, Ohio 44106, USA2Solar Durability and Lifetime Extension Research Center, Case Western Reserve University,Cleveland, Ohio 44106, USA3Department of Materials Science and Engineering, Case Western Reserve University,Cleveland, Ohio 44106, USA4Department of Chemical and Biochemical Engineering, Case Western Reserve University,Cleveland, Ohio 44106, USA

(Received 16 March 2016; accepted 19 July 2016; published online 10 August 2016)

We present a description of an electroluminescence (EL) apparatus, easily sourced from commerciallyavailable components, with a quantitative image processing platform that demonstrates feasibility forthe widespread utility of EL imaging as a characterization tool. We validated our system using aGage R&R analysis to find a variance contribution by the measurement system of 80.56%, whichis typically unacceptable, but through quantitative image processing and development of correctionfactors a variance contribution by the measurement system of 2.41% was obtained. We furthervalidated the system by quantifying the signal-to-noise ratio (SNR) and found values consistent withother systems published in the literature, at SNR values of 10-100, albeit at exposure times of greaterthan 1 s compared to 10 ms for other systems. This SNR value range is acceptable for image featurerecognition, providing the opportunity for widespread data acquisition and large scale data analyticsof photovoltaics. Published by AIP Publishing. [http://dx.doi.org/10.1063/1.4960180]

I. INTRODUCTION

As of 2015 the United States has installed 20 GW ofphotovoltaics (PV), and worldwide there is approximately55 GW installed annually, currently totaling 177 GW of globalinstalled PV.1 This vast number of modules has enabled a bigdata approach to the study of these systems’ performance togain insights into the mechanisms of performance. Advancesin data informatics allow for inclusion of new datasetsincluding imaging and hyperspectral mapping.2,3 Drone-basedinfrared imaging, for example, is becoming commonplaceand is serving as a general tool for warranty complianceverification.4 Here we address several common barriers to aluminescence based imaging apparatus that can be deployedon a wide scale for a large number of researchers or as afield-relevant data acquisition tool.

Electroluminescence is typically the band-edge radiativerecombination of carriers injected into a pn or pin junction.Luminescence imaging was developed in the 1960’s tospatially resolve the radiative recombination in silicon devicesusing an infrared (IR) detector.5 Fuyuki et al.6 employedelectroluminescence (EL) imaging, whereby injected carriersare provided by applying a voltage in forward-bias to thedevice junction, to image radiative recombination in siliconsolar cells using a “commercial off-the-shelf” (COTS) siliconcharge-coupled device (CCD) camera, which remains sensi-tive to the band-edge emission of silicon devices and therefore

a)tjp3@case.edu

bypasses the need for special IR detectors. Microcracks andother electrical breakages in the cells are easily observed inthe EL imaging as dark areas where radiative recombinationis low, and as such this tool has become highly usefulamong researchers in reliability/durability, collecting ELimages at points in time.7,8 Shunts and other leakage pathsare also visible as darker areas, and combined with darklock-in thermography (DLIT), it has been shown that localrecombination current and local series resistance can bederived yielding a comprehensive picture of recombinationpathways in solar cells.9–11

All of these techniques take advantage of measuringthe solar cell at different working points (voltage and/orillumination), by measuring only spectral fractions of theluminescence using optical filters or by analyzing the lateraldistributions.

Quantitative analytics are a critical element to theutilization of EL imaging as a research, quality assurance,or durability analysis tool. Quantitative approaches to ELhave been developed including measuring effective diffusionlengths,12,13 and the local junction voltage variation.14,15

These methods have led to theoretical models showing theimage distribution of series resistances,11,16–18 and saturationcurrent densities,15,19 which were derived from physics-basedsingle diode models of solar cells that may not be able tocapture the complexity of nonuniform contact resistivity orheterogeneous cracking.20

Hinken et al.21 presented a rapid and highly sensitiveluminescence imaging process and tool using electrical oroptical stimulation that is useful for photovoltaic cells. Their

0034-6748/2016/87(8)/085109/6/$30.00 87, 085109-1 Published by AIP Publishing. Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 129.22.124.78 On: Fri, 12 Aug

2016 20:38:14

085109-2 Fada et al. Rev. Sci. Instrum. 87, 085109 (2016)

instrument demonstrates a high signal-to-noise ratio (SNR)given a fast exposure time of 30 ms. This interesting researchtool includes entry barriers for other researchers based on thecomplexity of the tool, and the difficulty in processing largequantities of information.22,23

In support of laboratory-based PV materials research wehave developed a platform for electroluminescent (EL) imag-ing based on ease of use and easily obtainable components.This platform is coupled with advanced image processingsoftware using the “R” open source language to account forsystem heterogeneity and subsequent image correction, as wellas preliminary data analysis of PV cells. The image processingsoftware is an alternative to engineering controls and advancedhardware solutions to achieve a high degree of repeatability,but also enables an open and reproducible science componentto quantitative EL imaging with COTS components that areeasily sourced, in a similar way to Uchida.5,24,25 The advent ofopen-source codes and techniques presents an opportunity tounderstand quantitatively and statistically model the behaviorof a large number of PV cells using EL imaging as an input.Therefore, the platform we describe has a reduced barrierto entry for the global cohort of PV researchers to acquire,quantitatively analyze, and share results through, e.g., anopen-source data repository and analytics project,26,27 anddemonstrates the potential for COTS tools and advanced imageprocessing to enable a large-scale, non-centralized acquisitionof EL images.

II. DEVICE DESCRIPTION

A. EL camera modification

To obtain an imaging system sensitive to silicon band edgeluminescence, a Samsung NX100 compact body mirror-lesswhite light camera with a large 23.4 × 15.6 mm coupled metal-oxide semiconductor (CMOS) sensor, having an effective pixelcount of 14.6 megapixels (MP), was acquired. This particularmodel meets many ideal criteria, including large number ofpixels, which directly corresponds to the number of individualdata points used for quantitative image analytic processes, easeof remote data acquisition, and easily removable IR filters.28

To allow for the capture of photons emitted from the1.11 eV (high energy infrared) band gap silicon solar cell, twoinfrared (IR) filters were removed.

With the back panel removed, the overlaid printablecircuit boards (PCBs) were separated with band connectorsremoved. The PCB containing the sensor can be seen heldin place with 3 screws. These screws are used to registerthe position and alignment of the sensor. It is critical thatthe sensor is placed back to the original orientation uponreassembly after the preceding steps. This was ensured bycounting the number of turns of the screw during removal.With the internal PCBs removed as seen in Figure 1(b), thelayered sensor system can be seen with the ultra-sonic cleaningsystem (U-SCS) as the “top” layer, appearing with a red hue.The four outermost screws on the PCB containing the sensorcan be removed to decouple the sensor from the U-SCS. Themounts were removed and the top layer completely detachedby disconnecting the ribbon cabling. The blue colored IR

FIG. 1. (a) IR filter on left and U-SCS with IR coating on right. (b) Camerainternals including PCB’s and U-SCS with IR coating.

filter and U-SCS, seen in Figure 1(a), can then be completelyremoved. The camera was reassembled ensuring realignmentof the sensor. A ring shaped artifact was noticed in the outputimages post-modification. It appears at a fixed radius fromthe image center as seen in the upper right and upper leftcorners of Figure 5(e) and upper right corner of Figure 5(f).This non-ideality does not impede current image processingmethods or significantly impact the device’s repeatability orreproducibility, as shown in more detail below. The artifactwill be addressed via an engineered solution in the future, asthis current instrument is in use for multiple studies.

B. EL setup and considerations

A custom aluminum sample holder, seen in Figure 2(Element B), was designed to rigidly hold PV mini-modulesand bare cells. Keeping samples at fixed registration formeasurement repeatability, at a location to maximize sample

FIG. 2. Concept rendering of sample indexing apparatus and surroundingenclosure frame. (a) Camera, (b) sample platform/fixture, (c) 4-cell mini-module, (d) single-cell, (e) imaging enclosure frame.

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 129.22.124.78 On: Fri, 12 Aug

2016 20:38:14

085109-3 Fada et al. Rev. Sci. Instrum. 87, 085109 (2016)

area in the image for highest pixel density (Figures 2(a),2(c), and 2(d)) and adjustable to future sample types, werekey considerations. Two THOR optics 95× 95 mm rails weremounted vertically to compatible THOR base plates. The baseplates were secured to a Newport RPR Reliance optical tableatop Newport Stabilizer pneumatic laminar flow vibrationdamping legs. THOR dovetail clamps were affixed to thevertical rails and 2 × 1 10S series 80/20 aluminum extrusionswere extended parallel to the table surface. 1 × 1 10S series80/20 extrusion cross linked the 2 × 1 members, serving asmounting surfaces for the THOR kinematic bases supportingthe sample holders. The sample holders were designed to fitthe specific sample type using 80/20 aluminum and customdesigned 3-D printed poly-carbonate parts to interface with thesample precisely. Implementing rapid prototyping 3-D printedparts allows for specific sample interfaces for any sample type.The camera was affixed to a THOR kinematic base and viaTHOR 1/2” posts to the Newport table top.

Constructed from the generally ubiquitous materials of1/2” PVC tubing and corresponding joints, multi-purpose3.5 mil black polyethylene plastic sheeting, 3M high-strengthVelcro, steel tubing clips, and heavy-duty 17 mil opaqueadhesive tape, the enclosure was designed for ease of materialacquisition and assembly (Figure 2(e)).

To create a light impermeable space for mini-modules, sin-gle cells or any laboratory samples for imaging we constructeda box frame of PVC to 30 in. (height) × 38 in. (width) × 42 in.(depth). Using 3.5 mil polyethylene plastic sheeting, the framewas wrapped with excess material skirting encompassing theedge adjacent to the table top. This black 3.5 mil polyethyleneplastic was tested for spectral transmission using a Cary6000iSpectrophotometer and the resulting spectra indicated that theaverage transmission value for light of 300-1700 nm was lessthan 1%. Heavy-duty adhesive tape was used to seal the over-lapping material to itself and to the table top. Continuous 3MScotch Velcro was used to seal the end flaps for accessing thecamera and ease of exchanging samples.

As an additional precaution, all room lights are extin-guished during imaging sessions.

III. DATA COLLECTION

To acquire an image, a PV mini-module is securelyindexed onto a sample holder and the camera is mountedon a kinematic base attached to a vibration damping opticaltable. Electrical leads from a Magna-Power Electronics TS50-300/208 programmable power supply are connected to themodule in forward-bias.

The camera settings for typical image acquisition aretuned to ISO 800, aperture F5.6, and shutter speed of 30 s. TheSamsung 1:3.5-5.6, 20-50 mm ED lens is set to the 20 mm (nooptical zoom) position.

To generate high quality data, “fine JPEG” and/or “RAW”file saving format is selected. The power supply is set between6 and 10 A, and typically is set to 8 A for 4-cell multi-crystalline silicon mini-modules. The shutter is triggered viaa remote control to prevent vibration in the system induced bydepressing the shutter button on the body of the camera.

IV. SYSTEM VALIDATION

We validated the EL imaging system as an analyticalresearch tool by employing Gage R&R analysis and charac-terized the response function of the system by measuring theaverage signal-to-noise ratio for images in comparison to otherEL systems.29

A. EL Gage R&R

Gage R&R is a standard analytical technique used toassess the repeatability and reproducibility of a measurementsystem (equipment, operators, procedures).29 In a Gage R&Rstudy, a set of samples is measured multiple times with mul-tiple operators. The resulting dataset contains variation fromboth sample-to-sample differences as well as measurementsystem variation from within and between operators’ measure-ments. An analysis of variance (ANOVA) random effectsmodel is used to identify contributions to the total variabilityobserved coming from the measurement system in the formof repeatability (within operator variation) and reproducibility(between operator variation), as opposed to sample-to-samplevariation coming from the samples themselves (Figure 3). Inthis way, the precision of a system can be characterized andcompared to the magnitude of the signal being measured, toassess the capability of a system for measuring what it isintended to.

Successive Gage R&R studies were performed, with 10samples, 3 operators, and 3 measurements apiece, to improvethe precision of electroluminescence measurement systemdescribed in this work. Modifications to the setup were madeto minimize incident stray light into the system, as wellas to standardize the sample loading and camera operation

FIG. 3. EL Gage R&R results (10 samples, 3 operators, 3 measurements)showing study data and variance contributions from repeatability, repro-ducibility, and sample-to-sample variation.

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 129.22.124.78 On: Fri, 12 Aug

2016 20:38:14

085109-4 Fada et al. Rev. Sci. Instrum. 87, 085109 (2016)

procedure. The final best results that were obtained, usingthe total pixel sum of unmodified measurement images as thequantified data value, showed a variance contribution by themeasurement system of 80.56%. After implementing imagecorrection and using the ratio of the pixel sum of near busbarcell areas (Figure 5, white boxed regions) to between busbarareas (Figure 5, black boxed regions) in the isolated cellimages as the quantified data value, a variance contributionby the measurement system of 2.41% was obtained. This isconsidered to be well within acceptable limits of variationintroduced by the measurement system, which shows thatrepeatable and reproducible measurements can be obtainedwith this system by implementing image correction and featureextraction.

B. Signal-to-noise

Average signal-to-noise ratios (SNR) were calculated asa means of device validation and benchmarking. Our methodfor SNR acquisition followed closely the method outlined inHinken et al.21 A 4-cell multi-crystalline silicon mini-modulewas fixed inside the imaging chamber. Powering the modulewith 2.89 V (720 mV per cell) and 8.6 A with the camera setto ISO800 and aperture F5.6, M = 10 successive images werecaptured at each of the following exposure times (in seconds):1, 5, 10, 15, 20, 25, and 30. Using image processing techniquesoutlined below, each RGB image channel was weighted by 1/3and summed into a single gray channel to create a gray-scaledimage. From these M image matrices, local average pixelvalues are calculated,

Ns =1M

i

Ns, i. (1)

And local standard deviation,

σs =

1M

i

(Ns − Ns, i). (2)

Resulting in local SNR

SNR =Ns

√M

σs. (3)

This procedure across all pixels yields a matrix of SNRvalues. By averaging the SNR values in a large luminescentarea of the module, a representative SNR was obtained foreach exposure time.

Our camera’s SNR values are compared to other camerasused for electroluminescent imaging in Figure 4. These cameradata are presented in Hinken et al.21 These data were gatheredfor a single silicon solar cell operating at 550 mV. The camerascompared were the PCO Sensicam QE (silicon CCD, 1.4 MP),Hamamatsu C9100-13 (back-illuminated silicon CCD withelectron multiplier (EM) gain, 0.26 MP) in normal and EM-gain modes, and Xenics Cheetah (Indium Gallium Arsenide(InGaAs) CMOS, 0.33 MP) in high gain (HG) mode. Ourcamera exhibits SNR values in the same range as thesecameras. The exposure time of our device is longer witha maximum of 30 s contained in these data. However, this

FIG. 4. Signal-to-noise ratio as a function of exposure time for this workcompared with the PCO Sensicam QE, Hamamatsu C9100-13 (normal andEM-gain modes) and Xenics Cheetah presented in Hinken et al.

duration is reasonable for lab-based solar studies and will bea non-inhibiting factor to device employment.

V. DISCUSSION

The images provide a wealth of data to be studied. Ina single image captured using the camera outlined herein,approximately 45 × 106 observations are generated across theRGB color channels.

Using the fine-JPEG file, one can utilize standardpackages to perform analysis.30,31 Basic techniques involveexample methods such as histogram style binning, row/columnsumming, and multi-directional slicing. These simple func-tions normally used upon the entire image can provide usefullocalized information at a specified grid resolution.

A key advantage to using quantitative image analyticswith collected images is that one can readily perform unbiasedanalysis. For instance, if we were categorizing the size of somefeature(s) present in a series of images, a standard approachmay be to have the experimenter record the size visuallyas say “small,” “medium,” or “large.” Categorization in thismanner can result in variance from subjective evaluation andcan be difficult conveying a convincing result. However, usingquantitative image analytics it is straightforward to evaluatethe size of the feature and set hard cutoff values to the desiredcategories, thus increasing repeatability of characterizationand reducing any perceptive bias. Using code sharing toolscombined with this laboratory-based measurement apparatuswith COTS hardware, the barrier to quantitative EL research isreduced thus enabling a large population of data to be acquiredby non-centralized researchers.

The EL image data shown in this work showed undesiredvariability introduced in the form of sample orientation as wellas overall image brightness. Orientation variability occurredfor the overall modules as well as for the individual photovol-

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 129.22.124.78 On: Fri, 12 Aug

2016 20:38:14

085109-5 Fada et al. Rev. Sci. Instrum. 87, 085109 (2016)

FIG. 5. Image processing, starting with original images ((a) and (e)), rotating and cropping to single brightness adjusted cell images ((b) and (f)), and examiningquantified cell features through the summing of pixel rows ((c) and (g)) and columns ((d) and (h)). The black and white boxed regions imposed on image (f),which are also represented by open circles and open triangles on image (h), represent the between busbar regions and the near busbar regions, respectively.

taic cells themselves within the assemblies. Overall brightnessvariability, independent of important sample features, was alsointroduced by an uncontrolled embedded camera algorithm.Image processing algorithms were implemented to reduce thisundesired variability.

Probability density distributions of pixel brightness areused to evaluate the overall image brightness and determinethe correction factors. Summing pixel totals in the x andy directions (Figures 5(c) and 5(g) and 5(d) and 5(h),respectively) is used to identify image features for assessingimage orientation. Summing row and column brightnessvalues, cell edges are located, and the distance between themminimized during a series of rotations to achieve a bestorientation. This process was performed on the entire moduleassembly featuring multiple cells, then again on isolated cellareas to arrive at well aligned and cropped individual cellimages.

As an example method of quantifying mini-module cellimages we summed the pixels in regions between the busbars.The locations of the busbars are found easily by locatingminima in the y direction pixel sum of a cell image. Withthe coordinates of the busbars identified, a cell image can bedivided into regions representing the areas between busbars,and the pixels in these regions summed to create quantifiedvalues reflecting the electrically active area of the cell.

VI. CONCLUSION

We demonstrated that a COTS camera with 14.6 MPresolution can be modified to create a useful laboratory-basedresearch tool. Acceptability as a research tool was shownthrough a Gage R&R analysis and by comparison of theSNR values to currently employed cameras in the literature.Image processing techniques allowed for post-processing ofimages to normalize tool variability and reduce ANOVA-

based variability measures to an acceptable 2.41%. This workachieved an SNR value range of 10-100 varying with exposuretime, which falls directly in line with those of other ELcameras in the literature. Our device does require an exposuretime at the upper end of those cameras compared, however,for laboratory work the maximum exposure time tested forour device (30 s) does not prove to be an inhibiting factorto use as a research tool. With a cost effective and widelydistributed camera, paired with quantitative image analytics,electroluminescent imaging can become more commonplaceand routine in laboratory-based materials research in supportof the future of PV.

ACKNOWLEDGMENTS

The authors are grateful for funding of aspects of thisproject from the BAPVC (Bay Area Photovoltaic ConsortiumPrime Award No. DE-EE0004946) and the Department ofEnergy (Award No. DE-EE-0007140). The authors would alsolike to thank Andreas Meisel, Mason Terry, Thomas Dang,and Christopher Alcantara of Dupont Photovoltaic Solutions(Silicon Valley Technology Center) for providing the mini-modules used to conduct the Gage R&R study.

1IEA Key World Energy Statistics, 2015.2R. H. French, R. Podgornik, T. J. Peshek, L. S. Bruckman, Y. Xu, N. R.Wheeler, A. Gok, Y. Hu, M. A. Hossain, D. A. Gordon, P. Zhao, J. Sun,and G.-Q. Zhang, “Degradation science: Mesoscopic evolution and temporalanalytics of photovoltaic energy materials,” Curr. Opin. Solid State Mater.Sci. 19(4), 212–226 (2015).

3J. C. Schlothauer, K. Grabmayer, G. M. Wallner, and B. Röder, “Correla-tion of spatially resolved photoluminescence and viscoelastic mechanicalproperties of encapsulating EVA in differently aged PV modules,” Prog.Photovoltaics: Res. Appl. 24, 855 (2016).

4M. Aghaei, “Unmanned aerial vehicles in photovoltaic systems monitoringapplications,” in 29th European Photovoltaic Solar Energy Conference andExhibition (EU PVSEC 2014), Amsterdam, The Netherlands, 2014.

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 129.22.124.78 On: Fri, 12 Aug

2016 20:38:14

085109-6 Fada et al. Rev. Sci. Instrum. 87, 085109 (2016)

5I. Uchida, “Observation of the recombination radiation from silicon p-njunction,” Jpn. J. Appl. Phys. 2(9), 561–564 (1963).

6T. Fuyuki, H. Kondo, Y. Kaji, T. Yamazaki, Y. Takahashi, and Y.Uraoka, “Electroluminescence,” in Proceedings of the 31st PhotovoltaicSpecialists Conference, Lake Buena Vista, FL (IEEE, New York, 2005),pp. 1343–1345.

7M. Kontges, S. Kurtz, C. Packard, U. Jahn, K. Berger, K. Kato, T. Friesen,H. Liu, and M. Van Isehegam, “IEA-PVPS {task 13}: Review of failures ofPV modules,” Technical Report IEA-PVPS T13-01:2014, 2014.

8J. Kasewieter, F. Haase, and M. Kontges, “Model of cracked solar cell met-allization leading to permanent module power loss,” IEEE J. Photovoltaics6(1), 28–33 (2016).

9O. Breitenstein, J. Bauer, T. Trupke, and R. A. Bardos, “On the detectionof shunts in silicon solar cells by photo-and electroluminescence imaging,”Prog. Photovoltaics: Res. Appl. 16(4), 325–330 (2008).

10M. Kasemann, D. Grote, B. Walter, W. Kwapil, T. Trupke, Y. Augarten, R.A. Bardos, E. Pink, M. D. Abbott, and W. Warta, “Luminescence imagingfor the detection of shunts on silicon solar cells,” Prog. Photovoltaics: Res.Appl. 16(4), 297–305 (2008).

11K. Ramspeck, K. Bothe, D. Hinken, B. Fischer, J. Schmidt, and R. Brendel,“Recombination current and series resistance imaging of solar cells by com-bined luminescence and lock-in thermography,” Appl. Phys. Lett. 90(15),153502 (2007).

12P. Würfel, T. Trupke, T. Puzzer, E. Schäffer, W. Warta, and S. W. Glunz,“Diffusion lengths of silicon solar cells from luminescence images,” J. Appl.Phys. 101(12), 123110 (2007).

13J. A. Giesecke, M. Kasemann, and W. Warta, “Determination of localminority carrier diffusion lengths in crystalline silicon from luminescenceimages,” J. Appl. Phys. 106(1), 014907 (2009).

14K. Bothe, P. Pohl, J. Schmidt, T. Weber, P. Altermatt, B. Fischer, and R.Brendel, “Electroluminescence imaging as an in-line characterization toolfor solar cell production,” in 21st European Photovoltaik Solae EnergyConference, Dresden, 2006.

15M. Glatthaar, J. Haunschild, M. Kasemann, J. Giesecke, W. Warta, and S.Rein, “Spatially resolved determination of dark saturation current and seriesresistance of silicon solar cells,” Phys. Status Solidi RRL 4(1-2), 13–15(2010).

16T. Trupke, E. Pink, R. A. Bardos, and M. D. Abbott, “Spatially resolvedseries resistance of silicon solar cells obtained from luminescence imaging,”Appl. Phys. Lett. 90(9), 093506 (2007).

17D. Hinken, K. Ramspeck, K. Bothe, B. Fischer, and R. Brendel, “Seriesresistance imaging of solar cells by voltage dependent electrolumines-cence,” Appl. Phys. Lett. 91(18), 182104 (2007).

18H. Kampwerth, T. Trupke, J. W. Weber, and Y. Augarten, “Advanced lumi-nescence based effective series resistance imaging of silicon solar cells,”Appl. Phys. Lett. 93(20), 202102 (2008).

19D. Hinken, K. Bothe, K. Ramspeck, S. Herlufsen, and R. Brendel, “Deter-mination of the effective diffusion length of silicon solar cells from photo-luminescence,” J. Appl. Phys. 105(10), 104516 (2009).

20T. Peshek, J. S. Fada, Y. Hu, Y. Xu, M. Elsaeiti, E. Schnabel, M. Koehl,and R. French, “Insights into metastability of photovoltaic materials at themesoscale through massive I-V analytics,” J. Vac. Sci. Technol., B (to bepublished).

21D. Hinken, C. Schinke, S. Herlufsen, A. Schmidt, K. Bothe, and R. Brendel,“Experimental setup for camera-based measurements of electrically andoptically stimulated luminescence of silicon solar cells and wafers,” Rev.Sci. Instrum. 82(3), 033706 (2011).

22PIXIS CCD Cameras for Imaging & Spectroscopy, http://www.princetoninstruments.com/products/PIXIS-CCD, 2016, accessed: 2016-05-15.

23LumiSolarProfessional Inline, http://www.greateyes.de/en/systems.html?mCtrl=Model&mOp=View&m_Products%5BmodelId%5D=56, 2016, ac-cessed: 2016-05-15.

24OSTP Initiatives, https://www.whitehouse.gov/node/232081, accessed:2016-03-14.

25“Announcement: Reducing our irreproducibility,” Nature 496, 398 (2013).26D. M. Dryden, J. C. Hopkins, L. K. Denoyer, L. Poudel, N. F. Steinmetz,

W.-Y. Ching, R. Podgornik, A. Parsegian, and R. H. French, “van der Waalsinteractions on the mesoscale: Open-science implementation, anisotropy,retardation, and solvent effects,” Langmuir 31(37), 10145–10153 (2015).

27R. H. French, Gecko Hamaker: An Open-Science Software Tool for theCalculation of van der Waals-London Dispersion (vdW-Ld) Interactions,2015, see http://geckoproj.sourceforge.net.

28S. Paul, Removing IR/AA filter from Samsung NX200, 2012.29R. K. Burdick, C. M. Borror, and D. C. Montgomery, Design and Anal-

ysis of Gauge R&R Studies: Making Decisions with Confidence Inter-vals in Random and Mixed ANOVA Models, ASA-SIAM Series on Statis-tics and Applied Probability (Society for Industrial Applied Mathemat-ics, Philadelphia, PA; American Statistical Association, Alexandria, VA,2005).

30R Core Team, R: A Language and Environment for StatisticalComputing (R Foundation for Statistical Computing, Vienna, Austria,2015).

31G. Pau, F. Fuchs, O. Sklyar, M. Boutros, and W. Huber, “EBImage—AnR package for image processing with applications to cellular phenotypes,”Bioinformatics 26(7), 979–981 (2010).

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 129.22.124.78 On: Fri, 12 Aug

2016 20:38:14