Transactions of the Institute of Measurement and Control http://tim.sagepub.com/
Automatic detection of micro-crack in solar wafers and cells: a review Mohd Israil, Said Amirul Anwar and Mohd Zaid Abdullah Transactions of the Institute of Measurement and Control published online 10 October 2012 DOI: 10.1177/0142331212457583 The online version of this article can be found at: http://tim.sagepub.com/content/early/2012/10/10/0142331212457583
Published by: http://www.sagepublications.com
On behalf of:
The Institute of Measurement and Control
Additional services and information for Transactions of the Institute of Measurement and Control can be found at: Email Alerts: http://tim.sagepub.com/cgi/alerts Subscriptions: http://tim.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav
>> OnlineFirst Version of Record - Oct 10, 2012 What is This?
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
Article
Automatic detection of micro-crack in solar wafers and cells: a review
Transactions of the Institute of Measurement and Control 0(0) 1–13 Ó The Author(s) 2012 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0142331212457583 tim.sagepub.com
Mohd Israil, Said Amirul Anwar and Mohd Zaid Abdullah
Abstract This paper presents a review of the machine detection systems for micro-crack inspection of solar wafers and cells. To-date, there are various methods and procedures that have been developed at various laboratories around the world to inspect solar wafers and solar cells for manufacturing defects. This paper on micro-crack detection offers a comprehensive and an up-to-date review of different detection strategies, describing the main features of each technique together with its strengths as well as weaknesses. This paper will benefit researchers and practitioners, especially those who want to develop automatic inspection systems for inspecting solar wafers or solar cells for micro-cracks.
Keywords Machine vision, micro-crack, NIR imaging, solar cell, solar wafer.
Introduction It is important to recognize that the silicon wafer is a large contributor, up to 75%, to the overall cost of the solar cell (Zimmermann, 2006) and the silicon raw material price has increased exponentially due to a worldwide shortage of polycrystalline silicon. To compensate for the feedstock shortage of silicon, solar wafer manufacturers are slicing silicon thinner with thicknesses down to order of 100 mm or less (Zhang and Ciftja, 2008). Figure 1 shows typical flow in the production of wafers from silicon. Wire saw technology is being used by Zhang and Ciftja (2008); it is the technology for slicing thin wafers from a large diameter crystalline ingot of silicon. A wire saw must be balance precisely to achieve higher productivity while minimizing the breakage problems in the wafer. In addition to the reduction of the thickness, wafer manufacturers are also increasing the size of the wafer in order to reduce the overall production cost. Solar wafers of size up to 2103210 mm square are now available on the current market. These technological trends in production make wafer handling more challenging, as the processes can potentially reduce the yield due to increased wafer and cell breakage. Typically, handling or mishandling may lead to some physical defects in the wafers, like cracks or scratches. These cracks may vary from macro-level to micro-level; generally, cracks of width of the order \ 100 mm are considered micro-cracks. Both polycrystalline and monocrystalline solar wafers/cells occasionally contain micro-cracks. Figure 2 illustrates an example of a polycrystalline solar cell with micro-crack. This figure shows the micro-crack indicated by an arrow. Low grey level and high gradient magnitudes are two main features of the micro-cracks in solar wafers. Due to their size, naturally this type of defect cannot be seen by the naked eye.
Consequently, this may result in the production of inferior quality solar panels if this defect in solar wafers or cells goes undetected. In the worst case, the cell might even fail and this leads to the potentially malfunctioned photovoltaic (PV) modules (Kontges et al., 2011). These authors concluded that cracks, regardless of their size, would affect the performance of the cell if they cause some areas in the cell to become inactive or alter its electrical integrity. Also, it can be seen from the Figure 2 that the polycrystalline solar cell shows multiple grains of different shapes and sizes, and therefore it is very hard to differentiate between a micro-crack and a grain boundary by simple machine vision learning. So it is important to develop an inspection system for the detection and evaluation of such a defect. Moreover, there is great potential for automation of solar cell inspection, since millions of solar cells are being manufactured daily worldwide. According to recent statistics, the growth rate of solar PV reached a record high in 2011, generating more than US$93 billion in revenue with multi-crystalline cells constituting more than 50% of the world production (Solarbuzz, 2012). Preferably, such a system should be non-contact in order to ensure that the surface and subsurface integrity of silicon wafers is preserved before and after assessment, and from the start of the production process until completion (Dunlop and Halton, 2004). The main objective of this paper is to review some of the well known and emerging technologies for micro-crack detection School of Electrical and Electronic Engineering, Engineering Campus, Universiti Sains Malaysia, Penang, Malaysia Corresponding author: Mohd Zaid Abdullah, School of Electrical and Electronic Engineering, Engineering Campus, Universiti Sains Malaysia, 14300 Penang, Malaysia. Email:
[email protected]
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
2
Transactions of the Institute of Measurement and Control 0(0)
Figure 1 Typical process flow in the production of crystalline silicon wafers. Reproduced with kind permission from Elsevier (Zhang and Ciftja, 2008).
solar wafers and solar cells. The most common methods that have been investigated include laser beam induced current (LBIC) (Carstensen et al., 2003; Sites and Nagle, 2005; Vorster and Dyk, 2007; Zook, 1980), electron beam induced current (EBIC) (Brietenstein et al., 2008; Kittler et al., 1999; Meng et al., 2010), optical testing such as photoluminescence (PL) (Giesecke et al., 2011; Trupke et al., 2006a, 2006b), electroluminescence imaging (Fuyuki and Kitiyanan, 2009), radiometric pulse or thermal imaging (Dunlop and Halton, 2004), and Hyper spectral imaging (Li et al., 2010). Meanwhile other promising techniques such as ultrasonic vibration (Belyaev et al., 2006; Dallas et al., 2007; Monastyrskyi et al., 2008), electrical characterization (Konovalov et al., 1997), and infrared lock in ultrasound thermography (Rakotoniaina et al., 2004) will also be discussed. In this paper, all the aforementioned methods will be reviewed, highlighting some of their salient characteristics including merits and demerits. For completion and thoroughness, some image processing techniques for the shape and size detection of micro-cracks will also be discussed. Figure 2 Example of polycrystalline solar cell with micro-crack.
Laser beam induced current (LBIC) of solar wafers. Some of the salient features of these methods are identified and critically discussed, aiming to provide useful guidance to new and existing researchers wishing to venture into this interesting research area.
Micro-crack inspection in solar wafers/cells To date, various researchers have experimented with various methods and techniques for the detection of micro-cracks in
LBIC is a non-destructive optical testing for the characterization of semiconductors (Bajaj and Tennant, 1990; Bajaj et al., 1993). The basic LBIC system set-up is shown in Figure 3. As shown in this figure, the light source is selected from laser diodes of different wavelengths between 638 and 850 nm, and an electrical current to the laser diode is electronically modulated to produce an AC laser beam; the modulation also provides the reference signal for a lock-in amplifier. When a light beam is scanned over the surface of a photosensitive device, it
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
Israil et al.
3
creates electron–hole pairs in the semiconductor causing the dc current to flow, which in turn is measured using suitable devices (Carstensen et al., 2003; Sites and Nagle, 2005; Vorster and Dyk, 2007; Zook, 1980). Such measurements are repeated for different position of the laser beam to obtain LBIC image of the sample. The variations in the current are recorded and converted into variation in contrast forming the LBIC image. More variation in the current indicates that the cell is more defected. In a typical set-up, the LBIC technique consists of a calibrated measurement of current and reflection coefficient. This information allows the internal quantum efficiency (IQE) of the solar cell to be assessed (Thantsha et al., 2009). The IQE is defined as the fraction of incident photons transmitted into the solar cell that contribute to the generation of electron– hole pairs. Mathematically it is given by Mazer (1996): 1 h c Isc IQE = 1 R e l IL
ð1Þ
where R is the reflection coefficient, h is Planck’s constant, c is the velocity of light, e is the electron charge, l is the wavelength of the illuminating light, ISC is the measured short circuit current and IL is the intensity of the illuminated light. The quantum efficiency is the photon-to-electron conversion efficiency of the solar cell. Hence, lower efficiency of the cell indicates that the cell is more defective. Figure 4 shows the current distribution map of the cell obtained through LBIC imaging. In this figure, the dark irregular lines correspond to the active performance degrading the grain boundaries. Figure 5(a) shows the LBIC reflection map corresponding to the darker areas of Figure 4. It is evident that the current distribution, as expected for a multicrystalline material, is not uniform, as illustrated in regions marked A–C. The uniformity is compromised by the reflection and absorption of different grains at the surface of the polycrystalline silicon solar cell. Light is reflected more in region C than the neighbouring regions A and B. In Figure 5(b), a reflective line scan is depicted, which further indicates the high current response in region C. This region is expected to decrease the efficiency of the solar cell when it is in operation. The feature indicated by X corresponds to the
Beam-steering mirrors
Laser Diode Single-Mode Fiber Attenuator
Intensity Monitor Beam G-T Polarizer Sampler
Reflected beam Reflection Monitor
Objective mounted on Z stage Sample on X-Y translation stage
Figure 3 Laser beam induced current (LBIC) measurement set-up. Reproduced with kind permission from Cambridge University Press (Hiltner and Sites, 2001).
Figure 4 Laser beam induced current (LBIC) map of polycrystalline silicon solar cell. Reproduced with kind permission from Elsevier (Thantsha et al., 2009).
grain boundary, which clearly reflects more incident light, as do the contact fingers. LBIC methods have been investigated both for fast line scan techniques and for detailed surface mapping (Agostenelli et al., 2001). The major drawback of this method lies in the necessity for electrical contacts, making this technique nearly impossible to apply for wafer inspection and technically difficult for non-tabbed solar cells. Furthermore, the scanning needs to be performed on the entire wafer area and this process is prohibitively time consuming even though the accuracy of the LBIC is acceptable.
Electron beam induced current (EBIC) EBIC analysis, as the name implies, is a semiconductor analysis technique that employs an electron beam to induce a current within a sample, which may be used as a signal for generating images that depict characteristics of the sample, among others showing the locations of p–n junctions in the sample, highlighting the presence of local defects and mapping doping non-homogeneities (Leamy, 1982). Since a scanning electron microscope (SEM) is a convenient source of electron beam for this purpose, most EBIC techniques are performed using an SEM. A typical EBIC imaging system consisting of an SEM, low noise current amplifier and display unit is shown in Figure 6. When an electron beam from the SEM strikes the surface of the solar cell, it generates electron– hole pairs within the volume of beam interaction over the cell. With proper electrical contact with the sample, the movement of the holes and electrons generated by the SEM’s electron beam can be collected, amplified and analysed, such that variations in the generation, drift or recombination of these carriers can be displayed as variations of contrast as in the LBIC image discussed previously. EBIC imaging is very sensitive to electron–hole recombination. This is the reason why EBIC analysis is very useful for finding defects that act as recombination centres in semiconductor materials.
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
4
Transactions of the Institute of Measurement and Control 0(0)
Figure 5 (a) Reflection map and (b) reflection current map corresponding to image in Figure 4. Reproduced with kind permission from Elsevier (Thantsha et al., 2009).
Figure 6 Electron beam induced current (EBIC) imaging systems.
The EBIC current (IEBIC) collected is many times larger than the primary beam current absorbed by the sample (Iab), and is given by the equation Eb IEBIC = Iab 3 3 n Eh
ð2Þ
where Eb is the primary beam energy or the SEM’s accelerating voltage, Eh is the energy needed to create an electron–hole pair (about 3.6 eV for Silicon) and n is the collection efficiency. The accelerating voltage belongs to the extremely high tension (EHT) category, ranging from tens to hundreds of keV. Thus, assuming a collection efficiency of 100%, and an EHT of 20 keV, the collected EBIC current would be about 5556 times larger than Iab. EBIC currents are usually in the
nanoampere to microampere range, while Iab is in the picoampere range. In areas around the p–n junction where physical defects exist, electron–hole recombination is enhanced, thus reducing the collected current in those defected areas. Hence, if the current through the junction is used to produce the EBIC image, the areas with physical defects will appear to be darker in the EBIC image than areas with no physical defects. EBIC imaging is therefore a convenient tool for finding sub-surface and other difficult-to-see damage sites. Referring to Figure 6, the wire that carries the current away from the top contact can be seen in the lower left. The solar cell is slowly scanned and the EBIC current given by Equation (2) is then measured. This current is displayed in colour. The measured EBIC current was small when the beam fell on the metal contact but was larger when it fell on the active region of the solar cell. Figure 7 shows a secondary electron image of a polycrystalline silicon solar cell. Within the active region of the solar cell, there are large variations in the current. This is due to a variation in the density of defects, which causes the electron–hole pairs to recombine before they are separated by the built-in electric field. Figure 8 illustrates a typical EBIC image when the electron beam energy is 20 keV (Meng et al., 2010). The crack can be clearly seen in the image. Therefore, this technique is useful for detecting the presence or absence of micro-cracks in a solar cell or solar wafer. EBIC and LBIC are powerful tools for mapping distribution of recombination active defects and impurities in solar cells. The operation of both EBIC and LBIC is based on local injection of minority carriers and their subsequent collection by a p–n junction or a Schottky diode fabricated on the sample surface; the measurement closely mimics the actual operation of a solar cell. LBIC, which has somewhat lower resolution than EBIC, is usually used to map the whole cell, whereas EBIC is better suited for high-resolution imaging of
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
Israil et al.
5
Figure 7 Electron beam induced current (EBIC) current map of a polycrystalline silicon solar cell.
Figure 8 Example of electron beam induced current (EBIC) image captured at 20 keV excitation. Reproduced with kind permission from the IEEE (Meng et al., 2010).
small areas of the wafer. The analysis of temperature dependence of EBIC contrast enables one to distinguish shallow and deep recombination centres, but no further parameters of the traps can be determined. Additionally, the depth of the analysed layer is shallow, typically several microns from the surface in EBIC, and several tens or hundreds of microns in LBIC, depending on the wavelength of the illuminator. Therefore, only a small fraction of the sample volume in which electron–hole pairs are generated can be analysed
Electroluminescence (EL) imaging technique Luminescence imaging is a very attractive idea for the microcrack detection for the solar cells and wafers. Luminescence in the semiconductor is the result of the electron–hole recombination by electron excitation. Electroluminescence (EL) is
the form of luminescence in which electrons are excited into the conduction band through the use of electrical current by connecting the cell in forward bias mode. This technique could be applied not only to the finished cell but also to the module and solar panels. The typical set-up for an electroluminescence-based inspection system is shown in Figure 9. It shows the solar cell sample connected to a power supply; a silicon- charge couple device (CCD) camera used to capture the picture, which is then processed by the work station. The EL method requires the solar cells to be in the forward bias condition in order for it to emit infrared radiation. The luminescence ranges from 950 to 1250 nm with the peak occurring at approximately 1150 nm. Emission intensity is dependent on the density of defects in the silicon, with fewer defects resulting in more emitted photons. The EL system should be placed in a dark room, as the image of the cells is being taken by cooled CCD camera. Figure 10(a) shows the sample of optical image of the defected monocrystalline silicon solar cell, whereas Figure 10(b) shows the EL image of the same cell. The presence of horizontal line can easily be seen in the bottom part of the Figure 10(b). This horizontal line is a crack present in the cell, which cannot be seen in the Figure 10(a). Meanwhile Figure 10(d) shows an EL image of the polycrystalline silicon cell in which the grain boundaries became visible; those are not visible in the optical image as shown in Figure 10(c). The beauty of this system is that it can be applied for the wafer cell as well as PV module. Figure 11 shows an EL image of the monocrystalline PV module reported by Sander et al. (2010). The CCD image of the monocrystalline PV module acquired at delivery is shown in Figure 11(a), while Figure 11(b) shows the corresponding EL image. The presence of manufacturing defects like cracks in the module is not clearly visible in Figure 11(a). From the results given above, it is clear that the EL imaging is a good technique for inspecting the defects in the solar cell, but this method also requires electrical contacts between the cell and the leads supplying currents from an external power supply. Therefore, this method works well for cells and modules, but not for wafers. However, with wafers the radiation can also be induced by illuminating it with a source of smaller wavelength: so-called photoluminescence (PL). The details are explained in the following section.
Photoluminescence (PL) imaging technique As explained in previous section, the EL is very efficient technique for locating defects in the solar cell but it can be applied for the finished cell or module only. This method cannot be applied in the case of a solar wafer. PL is a versatile nondestructive tool for inspecting silicon wafers and solar cells. More importantly, this method eradicates the need for an electrical contact with the device under test. Moreover, it can be applied not only at the end of the cell’s production; it can be slotted in during the processes of producing solar cells (Abbott et al., 2006). PL is the result of the electron–hole recombination, in which the electron is excited to the conduction band after absorption of photon. The imaging set-up is very similar to
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
6
Transactions of the Institute of Measurement and Control 0(0)
Figure 9 A typical set up for electroluminescence. Reproduced with kind permission from Springer (Fuyuki and Kitiyanan, 2009).
that for EL. The only difference is that the electrons are excited by means of a laser source, as shown in Figure 12. The PL image is detected using a cooled CCD camera with a 1000-nm long-pass filter to remove the reflected and scattered laser light. The physics behind the PL imaging is that most of the photon-generated electrons give up their energy as heat, but a small fraction of the electrons recombine with a hole, emitting a photon (radiative recombination). The PL intensity depends on the rate of recombination of electron–hole pairs, which depends on the excess carrier density and the doping concentration in the semiconductor. If we consider the case of p-type solar wafer with doping concentration NA and Dn is the excess minority carrier density, then the intensity of the PL current is given as follows (Herlufsen et al., 2008): IPL a R ’ B Dn ðDn + NA Þ
ð3Þ
where R and B are the radiative recombination rate and radiative recombination coefficient, respectively. PL intensity is proportional to the carrier concentration. Therefore, bright areas in general indicate higher minority-carrier lifetime regions, whereas dark areas indicate higher defect concentration. More defects in the silicon will result in more energy lost as heat, and fewer emitted photons. In contrast, fewer defects in the silicon will result in more radiative recombination and more emitted photons. An example of a PL image of the polycrystalline silicon solar cell is given in Figure 13. PL imaging is an efficient technique, as it does not require any electrical contact and the
image taken by this technique is free from series resistance. It can be applied to a wafer, cell or module.
Mechanical stress testing The mechanical stress tests work on the principal that when the silicon wafer has defects, then the mechanical strength of the wafer is reduced. In the stress test, the wafer is tested by applying stress to it. Micro-cracks in the solar wafer/cell may lead to fracture under a critical stress level. A number of tests have been investigated by various researchers (Gustafsson et al., 2008); few of them are twist bending, double ring bending, point bending and line bending. The twist tester used by Gustafsson et al. (2008) is given in Figure 14. In this test, a load of 1.5 N is applied to the multicrystalline wafer of size 1563156 mm and thickness 200 mm. A maximum bending is recorded for the wafer that passed the test and breakage force was recorded for the wafer that broke. The maximum stress applied to the wafer is dependent on the crack length. Cracks longer than 20 mm have a breakage force usually smaller than the shorter cracks of 10 mm in length. For the cracks shorter than 10 mm, the breakage force is normally 1 N or higher (Gustafsson et al., 2008). According to these researchers, more than 90% of wafers with a crack length shorter than 10 mm passed the test, but cracks over 20 mm would eventually break or damage the samples. The stability testing unit will be used as an inline tool for incoming inspection of wafer stability and cell stability testing
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
Israil et al.
7
Figure 10 (a) Optical image of a defected monocrystalline silicon solar cell; (b) the corresponding electroluminescence (EL) image of (a); (c) optical image of defected polycrystalline silicon solar cell; (d) the corresponding EL image of (c). Reproduced with kind permission from Springer (Fuyuki and Kitiyanan, 2009).
before sorting. The micro-crack vision system can be used for crack detection and characterization of cracks as incoming inspection (as-cut) or after texturization and thermal processing. The results for finished cells were much less significant due to the aluminium layer. Also, when applying the stress to the wafer, one must be very accurate and able to maintain the stress level, otherwise a wafer may be damaged.
Vibration testing Several researchers used vibration testing for the detection of micro-cracks in the wafer (Belyaev et al., 2006; Dallas et al., 2007). They applied ultrasonics to the solar wafer and then examined the frequency response resulting from it. Ultrasonic vibrations are generated with the help of a resonance
piezoelectric transducer. The transducer contains a central hole, which provides vacuum coupling to the wafer and a transducer, as shown in Figure 15. From this measurement, they deduced whether or not the wafer is defected with microcracks. The same principle was adopted by Hilmersson et al. (2008), who had used a piezoelectric hammer in the examination. The method given by Belyaev et al. (2006) and Dallas et al. (2007) is called resonance ultrasonic vibration (RUV), which is fast and accurate enough. The RUV system provides accurate measurement of the resonance frequency and the bandwidth of the resonance curve using a computer control frequency sweep that varies from 10 to 100 kHz. Initially, the standard wafer frequency spectrum in terms of bandwidth and peak is recorded. After that, the wafer is inspected for defects using shape and size analysis. The deviation from the peak and the broadening or
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
8
Transactions of the Institute of Measurement and Control 0(0)
Figure 11 Electroluminescence (EL) images of a photovoltaic (PV) module: (a) at delivery status; (b) after exposed to temperature change.
Figure 12 Typical photoluminescence imaging set-up.
narrowing of bandwidth of the spectrum indicates defects in the wafer. This is illustrated in Figure 16, as reported by Monastyrskyi et al. (2008) for two identical 125-mm sized Czochralski silicon wafers. Specifically, the crack in the wafer shows the following features: 1) a frequency shift of the peak position; 2) an increase of the bandwidth; and 3) a reduction of the amplitude. Therefore, the RUV approach is essentially based on fast measurement and analyses of a specific resonance peak and rejection of the wafer if peak characteristics deviate from the normal and good wafers.
Another RUV technique is very efficient technology for micro-crack detection, but it requires a contact type mechanism, which can potentially cause damage to solar wafers.
Computer vision The extraction of the micro-crack in the case of a single crystalline silicon wafer is easy because of its homogenous background, but in the case of polycrystalline silicon, the problem is very challenging because of heterogeneous background of
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
Israil et al.
9
Figure 15 Experimental set-up of resonance ultrasonic vibrations. Reproduced with kind permission from IOP Publishing (Dallas et al., 2007).
Figure 13 Example of photoluminescence image of a polycrystalline silicon solar cell . Reproduced with kind permission from the IEEE (Trupke et al., 2006b).
Figure 16 Silicon wafer/cell with crack can be separated from a regular wafer. Reproduced with kind permission from the IEEE (Monastyrskyi et al., 2008).
Figure 14 Mechanical twist testers. Reproduced with kind permission from the WIP (Gustafsson et al., 2008).
the image and grain boundaries. Therefore, there is a need for advanced image processing software for accurate detection and analysis. Computer vision is an important tool that utilizes image processing techniques to solve a wide range of vision-based problems (Abdullah et al., 2005; Nixon, 1989; Ponsa et al., 2011). Recently, Tsai et al. (2010) proposed a simple machine vision-based system for capturing the image of a solar wafer by using a front light emitting diode (LED) and an off-the-shelf CCD camera. In so doing, this author has developed an image processing algorithm known as the anisotropic diffusion technique. Following Tsai et al. (2010)’s work, Chiou et al. (2011) proposed a set-up that utilizes a near-infra-red (NIR) camera illuminated with infra-red (IR) LED backlighting to improve detection. A simple image
processing algorithm is also implemented. In addition to micro-cracks, Li and Tsai (2011) proposed the schemes that detect saw-marks on the surface of the polycrystalline solar wafers. A number of researchers submitted their ideas for the extraction of micro-cracks in solar wafers. Image segmentation is very popular technique, since it allows the desired part of the image to be separated from the background. The most common methods in the segmentation include thresholding, clustering, region-based segmentation and edge-based segmentation methods. The resulting segmented image is in the form of binary image where the desired (cracks) pixels are represented by one value and the background pixels by another. The morphological operations are usually performed on the binary image to improve its representation (Chiou et al., 2011, Udrea and Vizireanu, 2007; Vizireanu, 2007). Thresholding is very common and mostly used method in the segmentation, and can be applied locally, i.e. within a neighbourhood of
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
10
Transactions of the Institute of Measurement and Control 0(0)
each pixel, or globally. However, it is clear from the results of few researchers (Hill et al., 1983; Otsu, 1979; Ridler and Calvard, 1978; West et al., 1988) that the local thresholding is better than global segmentation for the micro-crack detection. As in global segmentation methods, a fixed threshold to segment an entire image is used, but in local segmentation methods different thresholds are used in different regions. One of the popular local thresholding methods is based on Niblack’s algorithm. The details are described in the following section.
The Niblack’s algorithm Niblack suggested a technique for calculating a threshold surface by shifting a window across the image, and uses local mean and standard deviation for each pixel in the window (Farid and Ahmed, 2009). The threshold value for a pixel within a fixed neighbourhood is a linear function of the mean and standard deviation of the neighbourhood pixels, with a constant gradient, which is highly tuneable to separate objects. The size of the neighbourhood is highly tuneable, as it is chosen so that it will be small enough to preserve local details, and large enough to suppress noise. Examples of image processing results produced by this method are illustrated in Figure 17. It can be seen from this figure that the presence of cracks has been located in the wafers. The performance and accuracy of the Niblack’s algorithm is illustrated in Figure 17, which shows two different types of wafers with micro-cracks. Figure 17(a) is the original image, while Figure 17(b) is result after Niblack’s segmentation before filtering. It can be seen from this figure that the
presence of small objects in the form of noise and unwanted objects are enhanced after image processing. Close examination of this figure indicates that there are few small objects in the binary image, which are not necessarily the defects. These artefacts can be eliminated by using size filtering or by blob analysis. The results are shown in Figure 17(c).
Region growing method For high-speed inline inspection systems, it is required that the image processing algorithm should be fast enough to extract the micro-crack. It has been reported that the methods based on region growing are twice as fast as Niblack’s method (Chiou et al., 2011). Originally developed by Adams and Bischof (1994) and Mehnert and Jackway (1997), the region growing technique has been further developed, resulting in much more sophisticated techniques like the region splitting (Dutra et al., 2009), region merging (Moscheni et al., 1998), split and merge (Liu and Sclaroff, 2004; Manousakas et al., 1998), and the watershed techniques (Meyer and Beucher, 1990; Vincent and Soille, 1991; Shafarenko et al., 1997). The region growing started with the gradient operation to sharpen the image. Then the component labelling and size filtering are used to remove the noise. The objective behind the use of gradient operation is to enhance the edge of the image. The gradient magnitude p and the gradient direction f of the two dimensional image f (x, y) are given by Equation (4) sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ! 2 2ffi ∂f= ∂f ∂f ∂y 1 p= + , f = tan ∂f= ∂x ∂y ∂x
ð4Þ
Figure 17 (a) Sample image of polycrystalline silicon wafer; (b) Niblack’s segmentation results before labelling and size filtering; (c) after labelling and size filtering (right). Reproduced with kind permission from Emerald Publishing (Chiou et al., 2011).
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
Israil et al.
11
Figure 18 Region growing images: (a) original; (b) gradient; (c) region growing; and (d) results after component labelling and size filtering. Reproduced with kind permission from Emerald Publishing (Chiou et al., 2011).
In region growing, a seed pixel is considered first, then the region is grown by adding neighbouring pixels that are similar to seed pixels. When the growth of the one region stops, another seed is then selected and the same process is repeated for other regions. After that, a merging operation is used to merge the neighbouring regions into a large region. Chiou et al. (2011) applied this technique to polycrystalline silicon wafers and the results are presented in Figure 18. Referring to this figure, three original polycrystalline silicon wafers with micro-cracks are given in Figure 18(a). The respective gradient image of the samples is given in Figure 18(b). In Figure 18(c), the images after region growing are re-produced. There are some small objects in form of noise, which can also be seen in Figure 18(c). These are obviously artefacts and their presence can be eliminated by using size filtering or by blob analysis. The results after filtering are provided in Figure 18(d).
Future work and conclusions In this review article, we discussed a number of imaging set-ups to capture the image; future work in this direction can be done by choosing a good technique for imaging with good image processing. The NIR imaging set-up is the most successfully imaging device for the solar wafer/cell because in a forward bias condition the solar device also emits NIR light. Information related to micro-crack defects can be extracted by means of digital filtering and image processing. Popular techniques include the region growing and segmentation algorithms, though other emerging techniques such as the wavelet analysis are slowly being investigated for inspecting solar wafers and cells.
Funding This research has been supported by the Malaysia Fundamental Research Grant Scheme, grant number 6071216. Mohd Israil acknowledges financial support from the Centre of Excellence for Electrical & Electronics, Penang, Malaysia.
References Abbott MD, Cotter JE, Trupke T, et al. (2006) Application of photoluminescence to high-efficiency silicon solar cell fabrication. Proceedings of the 4th World Conference on Photovoltaic Energy Conversion, Waikoloa. Volume 1, pp. 1211–1214. Abdullah MZ, Fathinul-Syahir AS and Mohd-Azemi BMN (2005) Automated inspection system for colour and shape grading of starfruit (Averrhoa carambola L.) using machine vision sensor. Transactions of the Institute of Measurement and Control 27: 65–87. Adams R and Bischof L (1994) Seeded region growing. IEEE Transaction on Pattern Analysis and Machine Intelligence 16: 641–647. Agostinelli G, Friesen G, Merli F, et al. (2001) Large area fast LBIC as a tool for inline PV module string characterisation. Proceedings of the 17th European Photovoltaic Solar Energy Conference, Munich, pp. 410–413. Bajaj J, Bubulac LO, Newman PR, et al. (1993) Spatial mapping of electrically active defects in HgCdTe using laser beam induced current by scanning laser microscopy. Semiconductor Science and Technology 5: 3186–3189. Bajaj J and Tennant WE (1990) Remote contact LBIC imaging of defects in semiconductors. Journal of Crystal Growth 103: 170–178. Belyaev A, Polupan O, Dallas W, et al. (2006) Crack detection and analyses using resonance ultrasonic vibrations in full-size crystalline silicon wafers. Applied Physics Letters 88: 0111907–0111909.
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
12
Transactions of the Institute of Measurement and Control 0(0)
Breitenstein O, Bauer J, Kittler M, et al. (2008) EBIC and luminescence studies of defects in solar cells. Scanning 30: 331–338. Carstensen J, Popkirov G, Bahr J et al. (2003) CELLO: an advanced LBIC measurement technique for solar cell local characterization. Solar Energy Materials and Solar Cells 76: 599–611. Chiou YC, Liu JZ and Liang YT (2011) Micro crack detection of multi-crystalline silicon solar wafer using machine vision techniques. Sensor Review 31: 154–165. Dallas W, Polupan O and Ostapenko, S (2007) Resonance ultrasonic vibrations for crack detection in photovoltaic silicon wafers. Measurement Science & Technology 18: 852–858. Dunlop ED and Halton D (2004) Radiometric pulse and thermal imaging methods for the detection of physical defects in solar cells and Si wafers in a production environment. Solar Energy Materials and Solar Cells 82: 467–480. Dutra TA, Pires AP and Bedrikovetsky PG (2009) A new splitting scheme and existence of elliptic region for gasflood modelling. SPE Journal 14: 101–111. Farid S and Ahmed F (2009) Application of Niblack’s method on images. Proceedings of International Conference on Emerging Technologies, Islamabad, pp. 280–286. Fuyuki T and Kitiyanan A (2009) Photographic diagnosis of crystalline silicon solar cells utilizing electroluminescence. Applied Physics A: Materials Science & Processing 96: 189–196. Giesecke JA, Michl B, Schindler F, et al. (2011) Minority carrier lifetime of silicon solar cells from quasi-steady-state photoluminescence. Solar Energy and Materials 95: 1979–1982. Gustafsson J, Larsson H, Solheim HJ, et al. (2008) Mechanical stress tests on mc-Si wafers with micro-cracks. Proceedings of the 23rd European Photovoltaic Solar Energy Conference, Valencia, pp. 1–4. Herlufsen S, Schmidt J, Hinken D, et al. (2008) Photo conductancecalibrated photoluminescence lifetime imaging of crystalline silicon. Physica Status Solidi (RRL) 2: 245–247. Hill WJ, Norton-Wayne L and Finkelstein L (1983) Signal processing for automatic optical surface inspection of steel strip. Transactions of the Institute of Measurement and Control 5: 137–154. Hilmersson C, Hess DP, Dallas W, et al. (2008) Crack detection in single-crystalline silicon wafers using impact testing. Applied Acoustics 69: 755–760. Hiltner JF and Sites JR (2001) Micron-resolution photocurrent of CdTe solar cells using multiple wavelengths. MRS Proceedings 668: H9.8. Kittler M, Kveder VV and Schroter W (1999) Temperature dependence of the recombination activity at contaminated dislocations in Si: a model describing the different EBIC contrasts behaviour. Diffusion and Defect Data Pt. B.: Solid State Phenomena 69: 417–422. Konovalov IE, Breitenstein O and Iwig K (1997) Local current-voltage curves measured thermally (LIVT): a new technique of characterizing PV cells. Solar Energy Materials and Solar Cells 48: 53–60. Kontges M, Kunze I, Kajari-Schroder S, et al. (2011) The risk of power loss in crystalline silicon based photovoltaic modules due to micro-cracks. Solar Energy Materials & Solar Cells 95: 1131–1137. Leamy HJ (1982) Charge collection scanning electron microscopy. Journal of Applied Physics 53: R51–R80. Li Qingli, Wang Weisheng, Ma Chao, et al. (2010) Detection of physical defects in solar cell by hyperspectral imaging technology. Optics and Laser Technology 42: 1010–1013. Li WC and Tsai DM (2011) Automatic saw-mark detection in multicrystalline solar wafer images. Solar Energy Materials and Solar Cells 95: 2206–2210. Liu L and Sclaroff S (2004) Deformable model-guided region split and merge of image regions. Image and Vision Computing 22: 343–354.
Manousakas IN, Undrill PE, Cameron GG, et al. (1998) Split-andmerge segmentation of magnetic resonance medical images: performance evaluation and extension to three dimensions. Computers and Biomedical Research 31: 393–412. Mazer JA (1996) Solar Cell: An Introduction to Crystalline Photovoltaic Technology. Boston, MA: Kluwer Academic. Mehnert A and Jackway P (1997) An improved seeded region growing algorithm. Pattern Recognition Letters 18: 1065–1071. MEMC (June 2011) http://www.memc.com/index.php?view=SolarManufacturing. Meng L, Nagalingam D, Bhatia CS, et al. (2010) SEAM and EBIC studies of morphological and electrical defects in polycrystalline silicon solar cell. Proceedings of IEEE International Reliability Physics Symposium, Singapore, pp. 503–507. Meyer F and Beucher S (1990) Morphological segmentation. Journal of Visual Communication and Image Representation 1: 21–46. Monastyrskyi A, Ostapenko S, Polupan O, et al. (2008) Resonance ultrasonic vibrations for in-line crack detection in silicon wafers and solar cells. Proceedings of 33th PVSC Photovoltaic Specialists Conference, Tampa, FL, pp. 1–6. Moscheni F, Bhattacharjee S and Kunt M (1998) Spatiotemporal segmentation based on region merging. IEEE Transactions on Pattern Analysis and Machine Intelligence 20: 897–915. Nixon M (1989) Automatic shape analysis of crystal glasses. Transactions of the Institute of Measurement and Control 11: 63–69. Otsu N (1979) A threshold selection method from graylevel histograms. IEEE Transactions on Systems, Man and Cybernetics 9: 62–66. Ponsa D, Serrat J and Lo´pez AM (2011) On-board image-based vehicle detection and tracking. Transactions of the Institute of Measurement and Control 33: 783–805. Rakotoniaina JP, Breitenstein O, Al-Rifai MH, et al. (2004) Detection of cracks in silicon wafers and solar cells by lock-in-ultrasound thermography. Proceedings of Photovoltaic Solar Energy Conference and Exhibition, Paris, pp. 640–643. Ridler TW and Calvard S (1978) Picture thresholding using an iterative selection method. IEEE Transactions on Systems, Man and Cybernetics 8: 630–632. Sander M, Henke B, Schweizer S, et al. (2010) PV module defect detection by combination of mechanical and electrical analysis methods. Proceedings of 35th PVSC Photovoltaic Specialists Conference Photovoltaic Specialists Conference (PVSC), Hawaii, pp. 1765–1769. Shafarenko L, Petrou M and Kittler J (1997) Automatic watershed segmentation of randomly textured color images. IEEE Transactions on Image Processing 6: 1530–1544. Sites JR and Nagle TJ (2005) LBIC analysis of thin-film polycrystalline solar cells. Proceedings of 31st PVSC Photovoltaic Specialists Conference, Fort Collins, CO, pp. 199–204. Solarbuzz (2012) World Solar Photovoltaic Market Grew to 27.4 Gigawatts in 2011, Up 40% Y/Y. http://www.solarbuzz.com/ news/recent-findings/world-solar-photovoltaic-market-grew-274gigawatts-2011-40-yy. Thantsha NM, Macabebe EQB, Vorster FJ, et al. (2009) Opto-electronic analysis of silicon cell by LBIC investigations and current– voltage characterization. Physica B 404: 4445–4448. Trupke T, Bardos RA, Schubert MC, et al. (2006a) Photoluminescence imaging of silicon wafers. Applied Physics Letters 89: 0441071–0441073. Trupke T, Bardos RA, Abbott MD, et al. (2006b) Fast photoluminescence imaging of silicon wafers. Proceedings of 4th IEEE world conference on Photovoltaic Energy Conversion, Sydney, Volume 1, pp. 928–931.
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
Israil et al.
13
Tsai DM, Chang CC and Chao SM (2010) Micro-crack inspection in heterogeneously textured solar wafers using anisotropic diffusion. Image and Vision Computing 28: 491–501. Udrea RM and Vizireanu N (2007) Iterative generalization of morphological skeleton, Journal of Electronic Imaging 16: 010501. Vincent L and Soille P (1991) Watersheds in digital spaces: an efficient algorithm based on immersion simulation. IEEE Transaction on Pattern Analysis and Machine Intelligence 13: 583–598. Vizireanu DN (2007) Generalizations of binary morphological shape decomposition. Journal of Electronic Imaging 16: 013002. Vorster FJ and Dyk EEV (2007) High saturation solar light beam induced current scanning of solar cells. Review of Scientific Instrument 78: 013904–013907.
West GAW, Ellis TJ and Finkelstein L (1988) Image processing and understanding for automatic inspection. Transactions of the Institute of Measurement and Control 10: 265–272. Zhang L and Ciftja A (2008) Recycling of solar cell silicon scraps through filtration. Part I: experimental investigation. Solar Energy Materials and Solar Cells 92: 1450–1461. Zimmermann CG (2006) The impact of mechanical defects on the reliability of solar cells in aerospace applications. IEEE Transactions on Device and Materials Reliability 6: 486–494. Zook JD (1980) Effects of grain boundaries in polycrystalline solar cells. Applied Physics Letter 37: 223–226.
Downloaded from tim.sagepub.com at Saudi Digital Library on November 26, 2012
