Mapping and Imaging of Thin Films on Large Surfaces

Thin films covering large surfaces are used in a very wide range of applications from displays through corrosion resistance, decoration, water proofing, smart windows, adhesion performance to solar panels, and many more. Scaling up existing thin‐film measurement techniques requires a high speed and the redesign of the configurations. The aim of this review is to give an overview of recent and past activities in the area, as well as an outlook of future opportunities.


Introduction
Methods of in-depth thin-film characterization [1] based on either depth profiling by sputtering, complex modeling based on nondestructively measured whole-depth data, or preparing cross sections destructively and time-consumingly are all mature to an extent that only requires incremental adaptations and improvements for new tasks such as large-area mapping. According to Jones et al., [2] further methods for the determination of coating thickness are reviewed with in-line and offline capabilities. There are intrinsic features of those methods that make only a few of them capable of collecting the amount of data needed for large-area measurements within an acceptable time frame. The intention of this review article is to give an overview of methods used and potentially usable for large-area mapping of thin films, with special attention to ellipsometry, the core competence of the authors. The key features include not only lateral and vertical resolution, speed, and sensitivity, but also availability, price, and nondestructive capabilities.
In this work, we categorize area (structure) sizes in the context of optical characterization, that is, compared to the wavelength. 1) In case of structure sizes smaller than the wavelength, usually the effective medium approximation (EMA) holds, considering the materials optically homogeneous, but as a mixture of known reference dielectric functions. 2) If the characteristic feature sizes are comparable to the wavelength, scattering has to be modeled, which can be used to determine sizes of periodic structures if the period size is in the range of the wavelength. 3) If the feature sizes are larger than the wavelength, imaging can be applied having a typical resolution determined by the wavelength and the numerical aperture. 4) The last case is the area size which is much larger than the wavelength (macroscopic, frequently "industrial", meter-sized areas)-in this case, both imaging and mapping are applied. Mapping means in our nomenclature that a spot of individual measurements is moved relative to the surface. In this study, we mainly focus on case (4), but partly also consider case (3) because many of them have the potential of being scaled up depending on the application.
Methods of thin-film and interface characterization can be classified by numerous different aspects such as the way depth profiling is achieved, [1] according to illumination and detection, [3] or based on in-line capabilities. [2] In Section 2, we briefly summarize the many opportunities and in the rest of the paper focus on those potentially applicable to large-area mapping. At the end of this article, we discuss some applications and conclude with a short summary of the major characteristics of the methods.

Methods of Thin-Film Characterization
Abou-Ras and coauthors published a study [1] comparing methods of in-depth thin-film characterizations. Three main groups of the approaches have been identified: 1) depth profiling by sputtering (e.g., secondary ion mass spectrometry [ Sputtering-based depth-profiling and cross-sectional methods require a long measurement time and tedious sample preparation, therefore they are not capable of practical large-area mapping. Nondestructive methods such as SE, RBS, ERDA, and AXES require a few minutes or even a few times 10 min for taking a spectrum at one spot. However, the measurement time of some optical configurations can be reduced to even a few seconds per spot, or a few minutes for a whole mapped or imaged area if using simultaneous measurements at all illuminated lateral positions. We therefore focus primarily on optical methods in this review.

Ellipsometry
Ellipsometry determines angle-of-incidence-dependent relative amplitude ratios and phase difference shifts of orthogonal electric field components typically defined as parallel and perpendicular to the plane of incidence upon specular reflection of light from a planar surface. [4] Thus, collimated light beams are conventionally used with a well-defined angle of incidence at the reflecting surface. An established method for large-area mapping has been to move both the source and detector units of an ellipsometer in a specular reflection configuration over the surface. This is a widely used standard method for measuring large panels (e.g., solar or display with sizes even over meters) in industrial projects. [5][6][7][8] When measuring in a roll-to-roll (RtR) configuration, the movement of the substrate can also be used as a means of moving the spot on the surface. [9] A typical application of point-by-point mapping is photovoltaic (PV) because the quality of panels is largely dependent on the lateral uniformity. Razooqi Alaani et al. [10] showed an example of PV optimization by mapping using the CdS/CdSe/CdTe multilayer structure (Figure 1a-d). The measurement of one point does not take more than a few seconds even using spectroscopy; however, it adds up to hours when measuring thousands of points.

Reflection, Transmission Spectroscopy, and Scatterometry
A major application of spectroscopic mapping is the characterization of ultrathin 2D materials due to emerging application in optoelectronics and many other areas. An overview of spectroscopic mapping techniques that provide abundant spatial and spectral information can be found in ref. [11]. Different forms of graphene, molybdenum disulfide (MoS 2 ), and black phosphorus can be measured by this method. Spatial adsorption and emission, excitonic behavior, light sensitivity, and plasmonic effects can be investigated and mapped. Hyperspectral linescanning spectroscopy has been applied for MoS 2 by Dong et al. [12] for thickness mapping and spectral characterization. Hyperspectral imaging microscopy (HSIM) provides abundant spectral information on a surface area rather than a single spot. HSIM has first been applied for microscale imaging of cells and tissues in life sciences, [13] food industry, [14] and perovskite energy materials. [15] The 3D datacube obtained by HSIM is similar to that measured by divergent light source ellipsometry, but in the latter case, the data cube is created in one-directional Reproduced with permission. [10] Copyright 2021, Elsevier.
www.advancedsciencenews.com www.pss-a.com scan of a line on the surface, [16,17] whereas in the usual working modes, HSIM includes both line scanning and area scanning options.
A spatially resolved transmission technique was used by Carr et al. [18] for demonstrating a method for large-area quality checking in graphene production. A focused beam passes through the sample and is compared with a reference beam obtained by a beam splitter, while the sample is moved by a mapping stage with a spatial resolution of %10 m. Using this technique, areas over 1 cm 2 can effectively be mapped.
A spatially resolved reflection technique for larger thicknesses (hundreds of microns) is confocal microscopy [19,20] demonstrated by Choi et al. [21] for raster mapping areas as large as 370 mm by 470 mm. Transparent double-layer films have been measured taking into account the numerical aperture of the focusing optics. The nondestructive capability has been emphasized suggesting possible applications for the online inspection of multilayer transparent coatings on flexible substrate during manufacturing. Some optical tomography techniques are also capable of measuring layer thicknesses in tissues in the hundreds of microns thickness range. [22] Scatterometry is applied for mapping surfaces from very small to extremely large [23] scales. Modeling of the diffraction patterns allows to reconstruct complex grating [24] and overlay [25] structures. Point-by-point mapping of an area is also utilized in microscopy by scanning the focused spot while measuring the scattered response-for example, to measure, find, and characterize individual plasmonic nanoparticles by spectroscopy. [26] Imaging scatterometry has been applied for a range of materials from butterfly wings [27,28] to plasmons. [29] The technique is capable of point-by-point mapping of surfaces, especially because of the imaging technique, allows the measurement of the scattered light quickly. However, it has been in most cases only used for one spot. [27,29,30] It has been shown that the advanced, phase-detecting version of the method can even be extended to utilize the ellipsometry approach. [31] Single-particle spectra can be measured using an upright optical microscope combined with an aberration-corrected imaging spectrometer. Particle size is well below the applied wavelengths if the average distance between the particles is bigger than the focus size. The sample is mounted on an XYZ piezo stage for precise sample positioning. Illumination and scattered light collection can be performed in epi-illumination using a 100X dark-field objective ( Figure 2).
The optical measurements can be correlated to SEM. Accurate sample positioning was achieved by an XYZ piezo stage mounted on the microscope stage, and the actual size and shape of the measured particles was determined by SEM in nanometer scale. Even complex-shaped (only partially covered gold/silica "mushroom") nanoparticles featuring structural or surface chemical patchiness were studied. [32] These straightforward methods readily available for gold and silica surface modification using range of different (bio)molecules make these well-defined nanoscale objects excellent candidates to study fundamental processes of programmed self-assembly or application as theranostic agents. [33] Another example is gold nanorods on indium-tin oxide substrates, which are modified by the combination of nanosphere lithography and ion-bombardment to create a nanopattern with sharp boundaries between the irradiated and masked regions.
The single-particle scattering spectra of gold nanorods distributed over the nanopattern were adequately interpreted in terms of a classical damped harmonic oscillator model, taking the chemical interface damping into account. The damping variations with the related intensity changes can indicate the substrate inhomogeneity at nanometer length scales. [34]

Other Methods
A nondestructive tool for the mapping of surface properties is the capillary bridge method. It is related to the measurement of contact angle formed on the solid-liquid-vapor triple line, which has been one of the most important surface characterization techniques for a long time. The measurement of the advancing and receding contact angles under static conditions on a solid surface enables the quantification of its wettability and the determination of surface free energy, and it provides information about its rough or heterogeneous character. Undoubtedly, the most popular measuring technique is the sessile drop method due to its relative simplicity. Its precision was typically AE2 in the past, which was significantly improved (< AE1°) by the modern axisymmetric drop shape analysis. [35] For low contact angles (typically < 10°), the precision of the sessile drop method is not sufficient.
An elegant solution to this problem can be the captive bubble method, [36] which inherently fulfills the requirement of saturated vapor. The drawback of this method is that the needle usually must be left in the bubble to hold it in the right place; therefore, the analysis of the bubble's shape is less accurate. The Wilhelmy balance method is another good solution to determine low contact angles. It has high sensitivity and accuracy due to the measurement of wetting force, and it enables to study adsorption processes. [37] There are important requirements to reach this accuracy: the length of the contact line should be Figure 2. Setup of single-particle mapping spectrometer. Reproduced with permission. [30] Copyright 2018, The American Chemical Society.
www.advancedsciencenews.com www.pss-a.com known with high precision, that is, the sample geometry is restricted; furthermore, all surfaces along the contact line must be identical.
To avoid the restrictions of the captive bubble and the Wilhelmy plate techniques, the best method is the contact angle determination on hydrophilic or superhydrophilic surfaces by using r-θ-type capillary bridges under equilibrium conditions. It enables to determine even ultralow contact angles with high precision without prewetting the investigated surface. In this case, the capillary bridge of the test liquid is formed from a pendant drop and used as a probe. The contact angle is determined from the measured capillary force and liquid bridge geometry by using analytical solution. Contact angles less than 1 can be measured with the uncertainty down to 0.1 . [38] This method can be used as a rapid mapping method, too.
Mercury probe Schottky capacitance-voltage (MCV) method is another surface mapping technique for large-area profiling of near-surface carrier density and resistivity in Si epitaxial layers. It is a crucial method in high-performance semiconductor manufacturing. [39] MCV systems allow to eliminate the need for costly metal and poly deposition processes by using a pneumatically controlled, non-damaging probe design and a top-side mercury contact. MCV is superior technique both for bulk/epitaxial-and dielectric layer characterization, although the surface preparation technique is crucial for a good measurement. [40] There are a wide range of treatments from wet to dry methods. MCV tool often include electrical characterizations to determine the thickness, k-value, and more for dielectrics. Semiconductor-dielectric interface parameters can also be determined from the C-V analysis for compound semiconductors and other special applications.

Ellipsometry
Although microimaging ellipsometry is a well-established technique. [41][42][43] much fewer efforts have so far been made for large-area imaging of thin films. Our group has proposed a divergent light source configuration for large-area imaging. Using this arrangement, the surface is illuminated from a divergent point source, and the reflection from the measured surface is imaged by a spherical or parabolic mirror and correction optics to a charge coupled device (CCD). All the pixels on the CCD correspond to one point of the sample surface. The modulation of the polarization creates a periodic variation of the intensity in each pixel, from which the map of (Ψ, Δ) pairs can be determined at the given wavelength of illumination. Spectroscopic macro-imaging ellipsometry can only be realized in this configuration by measuring maps at different wavelength one by one. [16,44] The "pin hole-version" of this kind of instrument uses an illumination by an almost diffuse, "divergent beam" of light that provides a collection of rays with diverse angles of incidence at every point of the sample. Precise "angle-selection" is performed on the detector side by a pinhole camera (Figure 3). The pinhole works as an "angle-of-incidence filter," selecting only one single light beam from every direction (or sample position). The angular resolution of this type of camera is dependent on the diameter of the pinhole. In the case of a single wavelength measurement, this instrument is appropriate for measuring a large-area sample with possible nonuniform properties such that the pinhole camera selects a single angle of incidence corresponding to each sample point, but that angle varies across the sample. If the sample is moved parallel to its surface and parallel to the plane of incidence along a line during the measurement; however, each sample point will experience multiple angles of incidence successively, which provides additional information to assist in data analysis. This additional information can be useful in the case of anisotropic samples or layers that are inhomogeneous in depth. By repeating the measurement with light of various wavelengths, multi-angle, multi-wavelength (but not continuously spectroscopic) ellipsometry data can be acquired. [16,44] By confining the number of measured sample points to a narrow range along a line only, we can generate continuous spectroscopic data at each point along the line image using white light illumination with a spectral dispersion (grating) after the pinhole. This instrument produces spatial information in 1D of the CCD array simultaneously with spectroscopic information in the perpendicular direction of the array ( Figure 4A,B). Both approaches (multi-angle/multi-wavelength and continuous spectroscopic) are useful, for example, in the analysis of a product moving along a coating line [16,17,[44][45][46] such as in case of an RtR configuration. [47]

Reflectometry
In many cases, complex modeling or nanometer accuracy is not required for the surface mapping. For example, when deviations from an optimal layer structure must be followed, usually a fixed setting for the polarization state and the deviation from this measured by the detected intensity can be followed. It has been shown in several studies that spectroscopic imaging reflectometry [43] is capable of the quantitative measurement of surface layer properties, such as the area fraction of metallic gold layer in ref. [48] or the layer thickness and roughness of ZnSe films in ref. [49].

Scatterometry
Textured surfaces are usually characterized by time-consuming scanning probe or electron microscopy techniques. Imaging scatterometry is a powerful technique to measure extended www.advancedsciencenews.com www.pss-a.com (larger than a focused spot) areas of structured surfaces with nanometer accuracy. [50] An in-line metrology tool has been demonstrated by Skovlund Madsen et al. [51] (Figure 5) for the dimensional characterization of RtR-imprinted nanostructures. It combines real-space imaging with spectral scatterometry. 1D line gratings have been measured using this method. The depth of the structures was accurately characterized with uncertainties of a few nanometers. Light scattering methods have also been applied in medical optics, which includes a wide range of approaches, for example, fluorescence and inelastic light scattering, coherent effect time-resolved and spatially modulated spectroscopy and tomography, polarization-sensitive techniques, optical coherence tomography, and more. [52,53] In case of lowdensity plasmonic nanoparticles, subwavelength objects can be imaged, [30] preferably in dark-field configuration. [54]

Interferometry
Speckle is an interference pattern created by reflection from a non-specular (rough) surface. When a surface is illuminated by laser light, in the reflected image captured by the camera, a granular pattern is obtained. If the particles, certain parts or object on the surface are moving, and a time-dependent speckle pattern is generated at each pixel of the camera. The variation of the intensity recorded by each pixel of the camera contains information on the scattered particles. This method has been demonstrated on the 2D mapping of blood flow in the vessels of Sprague Dawley rats. [55] The method has later been reviewed by Boas and Dunn. [52] There are many ways of improving speckle imaging, including the multi-exposure technique. [56]

Applications
The capabilities of expanded-beam SE have been demonstrated by numerous prototypes using uncollimated (nonparallel, diffuse) illumination with a detection system consisting of an angle-of-incidence-sensitive pinhole camera for high-speed, large-area imaging/mapping applications. The speed of the measurement system makes the instrument suitable for use on production lines. The precision enables the detection of sub-nanometer thicknesses, and refractive index and extinction coefficient changes of 0.01. Alternative commercial instruments for SE mapping must translate the sample in 2D. Even a sample with a size of 150 mm by 150 mm with a resolution of 10 mm requires >200 measurements and at least 15 min. By imaging along 1D in parallel, the expanded-beam system can measure with similar resolution in 1-2 min. The movement in the second dimension can be performed in robotic arm ( Figure 6A-C) or RtR [47] (Figure 7B-D) configurations. It has been demonstrated that 300 mm diameter Si wafers covered by amorphous NiSi (a-NiSi) and ion-implanted using plasma immersion can be measured on a robotic arm in a clean room environment ( Figure 6). The agreement of the determined layer thickness compared to the point-by-point measurement of a commercial Semilab-SOPRA ellipsometry is good (within one nanometer), while the duration of the measurement is at least an order of magnitude shorter on a divergent light source mapping ellipsometer. NiSi layers on a 300 mm diameter Si wafer are interesting in microelectronic fabrication where the silicide layers with controlled thickness can be used as ohmic-or Schottky contacts. The band is illuminated along a line, whereas the second dimension of the illuminated CCD detector contains the spectroscopic information. By moving the roll, each lateral position of the band is measured spectroscopically. Reproduced with permission. [47] Copyright 2014, IEEE. Figure 5. Setup of in-line scatterometer for the dimensional analysis of gratings on an RtR embossing line. The inset shows that the spectroscopic data is continuously collected in the y-direction as the sample is moving to the x-direction. Reproduced with permission. [51] Copyright 2021, OSA.
www.advancedsciencenews.com www.pss-a.com The homogeneity is important if we consider that the fabricated chip parameters should be within the fabrication tolerance limits of the chip.
Expanded-beam SE was applied for the imaging/mapping analysis of large-area spatial uniformity of multilayer-coated substrates in RtR thin-film PV. [47] The instrument allows widthwise  . The measurement has been made in an RtR configuration. B) Measured (symbols) and fitted (lines) ellipsometry spectra used for the calculation of the sample properties, the thickness of which is shown in (A). The inset shows the optical model in which the dispersions of the a-Si and ZnO layers were described using the Cody-Lorentz [57] and the Sellmeier þ critical point models, [58] respectively. Reproduced with permission. [47] Copyright 2014, IEEE.
www.advancedsciencenews.com www.pss-a.com imaging and lengthwise mapping for uniformity evaluation at the high linear substrate speeds required for real-time, in situ, and online analysis in RtR thin-film PV: Ag, ZnO, and n-type hydrogenated amorphous Si (a-Si:H) layers of an a-Si:H n-i-p structure deposited on a flexible polyimide substrate in the RtR configuration. SE data across a line image were collected as the substrate was translated by an RtR mechanism. A coated area of 450 mm by 120 mm was analyzed for layer thickness and optical properties by applying the appropriate analytical models for the complex dielectric functions of the Ag, ZnO, and n-type Si:H layers. In the case of the a-Si/ZnO/Ag/plastic-foil structure, we measured first the ZnO layer only on the {Ag layer}/{plastic foil} RtR sample and moved the cassette RtR model machine into the a-Si deposition chamber. After the a-Si deposition, the cassette RtR model machine was moved back to the measurement chamber (without breaking the vacuum) and measured again by the expanded-beam mapping SE (see Figure 7). Optical model of similar complexity was demonstrated for a ZnO/Mo/glass structure. This structure is also used in the photovoltaics where the ZnO layer can be used as a transparent conductive layer, which also serves as an optimized back reflector. The investigated sample was a glass plate covered by Mo and a ZnO layer ( Figure 8A). The investigated area was 300 mm by 300 mm (Figure 8B), demonstrating the compatibility with the current 300 mm wafer technology of the semiconductor industry. A reference measurement was performed on the same sample using a Woollam M-2000DI ellipsometer ( Figure 8C). Using this instrument, the maximum area for mapping is 150 mm by 150 mm ( Figure 8D). The thickness determined by the expanded-beam instrument agrees within a few nanometers; however, the duration of the measurement for the same resolution is an order of magnitude smaller.
The compatibility of divergent-source mapping ellipsometry with the 450 mm wafer technology was demonstrated by measuring an area covered by individual thermally oxidized 4 inch wafers. The thicknesses of the SiO 2 layers were different in different rows showing that even highly inhomogeneous large-area samples can be mapped easily. Figure 9A shows an overview of the fit quality (mean square error [MSE]) on the whole surface, demonstrating that the optical model of Figure 9 is valid for the regions covered by the 4 inch wafers. Figure 9 zooms into specific areas by plotting the thickness of the SiO 2 layer determined from the model. Note that a good agreement was obtained between the measured and the fitted curves by only using two fit parameters ( Figure 9E), the thickness and the nonuniformity.
The size of the illuminated area of the expanded-beam ellipsometer [17] can be adjusted for the task. Figure 10 shows an extreme case in which the area of mapping was 1000 mm by 500 mm on a glass panel used for solar applications. The glass is covered by a-Si. Each map represents a set of ellipsometric spectra that are analyzed and fitted by an optical model. In Figure 10, maps of the parameters of thickness ( Figure 10A) and bandgap ( Figure 10B) are plotted. The accuracy of the measurement has been verified using a Woollam M2000 ellipsometer, and the agreement was within a few percent, similar to other cases comparing the expanded-beam ellipsometer with reference measurements on commercial tools. The thin-film PV technology is more cost-effective than using bulk material (the same W peak costs less because less energy is needed to prepare the thin films than the bulk crystalline silicon, therefore, the energy payback time is also better), but it requires a good lateral homogeneity. Commercial instruments using 1D detector arrays for SE must translate either the sample or the ellipsometer in 2D to map large-area samples. A sample size of 150 mm by 150 mm requires more than 200 measurements and at least 15 min of measurement time to achieve a spatial resolution of 10 mm. By using a 2D detector array and imaging along 1D in parallel, the sample needs only be translated in 1D to map large-area samples. As a result, the expandedbeam system that provides a line image of SE (wavelength from 350 to 1000 nm) can measure with similar spatial resolution in less than 2 min. Incorporating the results of real-time SE measurements, which provide proper initial parameters in least-squares regression for the analysis of the imaging/mapping data, expanded-beam ellipsometry can be an effective in-line monitor of uniformity in PV production lines using RtR or rigid plate substrates. The goal of near future efforts is to increase the speed by 4. Then, 1800 points could be mapped in a 1 min traverse of a PV panel over an area as large as 1000 mm by 500 mm using a continuous measurement with a simultaneous evaluation.

Conclusions
Methods of large-area thin-film characterization have been summarized. According to other studies, [2] only optical methods are capable of a high-speed, high-sensitivity, and nondestructive characterization of thin films over large surfaces (although some potential future candidates have also been mentioned, such as  www.advancedsciencenews.com www.pss-a.com the thermoelectric method with a magnetic readout approach, THz-TDS, CCM, infrared thermography, as well as XRR or XRF). In case of optical methods, we discussed techniques including spectroscopic reflectometry, transmission, scattering, speckle imaging, and SE in all cases when the areas under study were larger than the wavelength of light. Only two methods have been identified that are capable of mapping or imaging meter-sized areas. One of them moves an ellipsometry head over the surface utilizing point-by-point mapping. [10] Divergent light source macro-imaging SE [17,47] is, however, capable of increasing the speed of large-area measurement by at least an order of magnitude utilizing imaging, measuring multiple points over the surface simultaneously by SE, and maintaining the capabilities of complex modeling and nanometer-scale sensitivity ( Table 1). In the applications section, a broad range of examples have been demonstrated about the capabilities of this method.