High-speed Micro-crack Detection of Solar Wafers with Variable Thickness

Transcription

1 High-speed Micro-crack Detection of Solar Wafers with Variable Thickness T. W. Teo, Z. Mahdavipour, M. Z. Abdullah School of Electrical and Electronic Engineering Engineering Campus Universiti Sains Malaysia Penang, Malaysia ttw13 Abstract A new approach for high-speed micro-crack detection of solar wafers with variable thickness is proposed. Using a pair of laser displacement sensors, wafer thickness is measured and the lighting intensity is automatically adjusted to compensate for loss in NIR transmission due to varying thickness. In this way, the image contrast is maintained relatively uniform for the entire size of a wafer. An improved version of Niblack segmentation algorithm is developed for this application. Experimental results show the effectiveness of the system when tested with solar wafers with thickness ranging from 125 to 170 µm. Since the inspection is performed on the fly, therefore, a high throughput rate of more than 3600 wafers per hour can easily be obtained. Hence, the proposed system enables rapid in-line monitoring and real-time measurement. I. INTRODUCTION In recent years, the photovoltaic (PV) industry has seen a boost in production of solar cells to meet the demand for clean renewable energy due to growing environmental concerns. Among the various types of solar cells in the market, crystalline silicon based solar cells appear to be dominant. The base of a crystalline silicon solar cell can come in either monocrystalline or polycrystalline form. The former is grown from a single silicon crystal while the latter contains several types of silicon crystals during the formation of the wafer ingot. A major issue currently faced in the production of solar cells is the quality assurance of the base material, especially defects such as micro-cracks. Micro-cracks which are not detected early in the production phase would not only weaken the structural integrity of the finished solar cells but could also cause potential disruptions to a production line due to breakages and neglect. About 5 to 10 percent of solar wafers break during production and this could potentially cause machine jams and damage the production machine leading to lengthy delays and expensive repairs. Micro-cracks are regularly defined as cracks with a size smaller than 30 µm in width and can be classified as visible or invisible to naked eyes [1] and they can occur in both monocrystalline and polycrystalline wafers as shown in Fig. 1. Surface inspection techniques utilising illumination in the visible light spectrum are capable of detecting visible microcracks depending on the size of the crack [2]. However similar optical techniques will not be able to detect invisible ones Fig. 1. Examples of defective solar wafer images. monocrystalline, polycrystalline. which are hidden from view beneath the surface of a solar wafer. Several inspection techniques which are suitable for in-line production inspection has been developed to detect invisible micro-cracks in solar wafers; popular among them are the optical transmission [1], the resonance ultrasonic vibration [3], and the photoluminesence imaging [4] method. Taking into consideration the cost, speed and accuracy of these inspection techniques in a high-speed in-line production environment, the optical transmission method appear to be the most promising. The optical transmission technique relies on the principle of transmission of near infrared (NIR) light through silicon, of which are what solar wafers are made from. NIR light penetrating through the solar wafer will scatter when encountering the micro-cracks, producing low gray level regions with high enough contrast to differentiate between defective and intact regions. These low gray level regions can be segmented fairly accurately provided that there s enough contrast with the background. Hence, it is critically important to ensure there is sufficient NIR light transmission through the solar wafer. The transmission of NIR light largely depends on the thickness of the wafer, the thicker the wafer, the lower the transmission. In production, solar wafers tend to vary in thickness from one wafer to another. Even within a same sample, the thickness of a wafer varies. This results in images with varying contrast which will make micro-crack detection difficult. An example /14/$ IEEE

2 Fig. 2. Example of a polycrystalline solar wafer image with varying thickness. The thickness increases from left to right resulting in poorly contrast image. Fig. 3. The schematic showing the geometry of the laser displacement sensors for measuring the wafer thickness. of such a sample is shown in Fig. 2. Reference [5] proposed a method which uses a photo diode to measure the transmission of the NIR light and adjusting the exposure of the camera to compensate for the wafer thickness. They have shown that by varying the exposure of the camera, they re able to successfully detect micro-cracks from previously low contrast images of wafers with variable thickness when the camera is set with a constant exposure. However, their method may not be suitable for a high-speed inline inspection as it requires a stepping mode type of conveyor system with varying stoppage time due to the inconsistent camera exposure, which may result in lengthy wafer transfer time. The proposed method by [5] is also not capable of handling solar wafers which contain varying thickness in different parts of the same wafer. To resolve these limitations, we propose a different solution by deploying a pair of laser displacement sensors to measure the wafer thickness, and dynamically adjusting the NIR light intensity based on thickness measurement feedback. Since the measurement including image capturing are performed without stoppage, the methods and procedures would enable highspeed monitoring and rapid in-line inspection. II. METHODOLOGY A. Wafer thickness measurement A pair of off-the-shelf high-accuracy CCD laser displacement sensors is used to measure the thickness of solar wafers. These sensors operate by measuring the displacement of the reflected laser, and by triangulation, an accurate measurement of proximity can be determined [6]. An example of such a set-up is shown in Fig. 3. Referring to this figure, the wafer thickness n can be calculated as follows: n = d t (d 1 + d 2 ) (1) where d 1 is the measured distance from the top surface of the wafer to the top sensor, d 2 the measured distance from Fig. 4. The schematic of the overall imaging system showing important elements. the bottom surface of wafer to the bottom sensor and d t the predetermined and calibrated distance between the top and bottom sensors. This set-up utilises a pair of sensors at approximately the same measurement position at the top and bottom of the wafer which minimises the effects of vertical vibrations interfering with measurements while in motion on a conveyor. The conveyor is a dual belt type and the pair of sensors is placed in the gap in between the two belts with the schematics as shown in Fig. 4. This enables the thickness measurements to be performed along the profile of the wafer without stoppage. B. Proposed imaging system To facilitate image capturing while the solar wafer is in motion, a NIR sensitive line-scan camera is used. A line-scan

3 camera captures a single row of pixels instead of a matrix and subsequently joins them together to form an image. In this system, the camera s line-scan rate is synchronised with the conveyor speed with a rotary encoder and produces images of pixels in size with 8-bit gray levels. Field of view is designated to mm which will cater to the 2 standard solar wafer sizes of mm and mm. A LED light bar 200 mm in length is used as the back light source. The LED light bar consists of high powered NIR LEDs with peak wavelengths at 940 nm. This specific wavelength was chosen due to its excellent ratio of light penetration and signal acquisition [5]. The light output intensity of the NIR backlight is adjustable by varying the output current of the lighting controller and this is determined by the wafer thickness as measured by the pair of laser displacement sensors. Optimal lighting output current A for the back light in ampere is determined in accordance with, A = 18.23e n (2) where n is the measured wafer thickness in µm. Equation (2) is determined experimentally by sampling 15 pieces of monocrystalline wafers average thickness ranging from 110 µm to 270 µm. For each solar wafer sample, the lighting intensity is manually adjusted to produce the same amount of contrast in the captured image and the resulting trend is as shown in Fig. 5. where k is the weight for adjusting and controlling the effect of standard deviation. In the original Niblack algorithm, k is rigidly fixed to 0.18 [7]. D. Proposed method for micro-crack detection The algorithm works by firstly calculating the weight k in (3) which is required in Niblack s filtering. Instead of rigidly fixing the k parameter as in the original Niblack filter, here, this parameter is adaptively calculated for a given image. Mathematically: k = [µ (I) σ (I) ]/[µ (I) σ (I) ] (4) where µ (I) and σ (I) are the mean value and standard deviation respectively of the input image, I. Niblack s original method does not work well when segmenting poorly contrast images, producing many false edges and other noise artefacts. To solve this problem, secondly, a parameter z is introduced. This parameter is based on the mean and standard deviation of gray levels of the image. Parameter z is calculated as follows: z = [µ (I) σ (I) ]/[ µ (I) σ (I) ] (5) The third step is the parameter x which is calculated as follows: x = µ (I) /σ (I) (6) Finally, the thresholding value is adaptively calculated as follows: ] T = [µ (I) + kσ (I) xz (7) Two parameters k and z are added to Niblack s original method (3). These parameters are used to suppress the effects of standard deviation influencing the calculated threshold value. Due to these two parameters, a more effective threshold can be calculated to enable separation of micro cracks from the background. Once T value is obtained, a pixel within a micro-crack region of a image will contain low gray level while noncrack pixels will contain high gray level. The threshold for this classification is presented as: Fig. 5. Variation of lighting output current (A) requirements against wafer thickness (µm). C. Image Processing We have developed an improved version of the Niblack s filter with adaptive segmentation. The proposed method is based on Niblack s method [7] where the main idea is to create a threshold surface, based on the local mean µ (i,j) and the local standard deviation σ (i,j), of gray values calculated over a small neighbourhood around each pixel in the form of: T = µ (i,j) + kσ (i,j) (3) M (i, j) = { 0 if I(i,j) T 255 Otherwise Wiener filter is then used after thresholding to remove unwanted noises. III. EXPERIMENTAL RESULTS A. Wafers with consistent thickness An inspection process using the proposed system was firstly conducted using a monocrystalline solar wafer containing a micro-crack and an average thickness of 155 µm. The wafer was travelling along the conveyor at a speed of 300 mm/s, which is equivalent to the throughput rate of approximately 3600 wafers per hour. (8)

4 Monocrystalline solar wafers were chosen in this experiment as they are homogeneously textured and therefore will provide a fairly constant gray level across the surface of the solar wafer image. This will minimise the effects of polycrystalline wafers interfering with intensity readings due to its heterogeneous texture. It was discovered in the experiment that the micro-crack has an average grey level of 90, while the rest of the solar wafer has an average gray level of 140. Inspections was performed using wafers with thickness ranging from 125 to 170 µm. This range represents the extreme ends of acceptable tolerances for a single batch of wafers produced commercially. All the solar wafers inspected contain micro-cracks. TABLE I AVERAGE 8- BIT GRAY LEVEL RESULTS OF DETECTED MICRO - CRACKS AND ITS BACKGROUND FROM SOLAR WAFERS OF DIFFERENT THICKNESS (µm ) CAPTURED WITH VARYING LIGHTING INTENSITY Wafer thickness Micro-crack Background As shown in Table I, the gray levels of the background for a given lighting intensity remain relatively constant even though the wafer thickness varies significantly. The average intensity of micro-cracks tend to vary as light scattering rate changes depending on the type of micro-crack present. However, as there s enough contrast between the micro-cracks and the background, they were all successfully extracted. To verify the effectiveness of the system, both 170 µm and 125 µm average thickness wafers was inspected again without the varying lighting intensity system with the 155 µm average thickness wafer being used as the benchmark. The lighting intensity was manually adjusted so that the wafer with 155 µm average thickness will produce images which contain micro-cracks with average grey level of 90 and average grey level of 140 for the rest of the wafer. This lighting intensity setting remains constant throughout the inspection of all solar wafers. The gray level results of images captured with constant lighting intensity is tabulated in Table II. TABLE II AVERAGE 8- BIT GRAY LEVEL RESULTS OF DETECTED MICRO - CRACKS AND ITS BACKGROUND FROM SOLAR WAFERS OF DIFFERENT THICKNESS (µm ) CAPTURED WITH CONSTANT LIGHTING INTENSITY Wafer thickness Micro-crack Background Without the use of the varying lighting intensity system, the contrast became so low that the micro-crack is barely visible and can no longer be extracted as shown in Fig. 6. This confirms the effectiveness of the varying lighting intensity system in improving the contrast of solar wafers of variable thickness along with its capability to perform a high-speed in-line inspection. Fig. 6. Examples of defective monocrystalline solar wafers with average thickness of 155 µm (left), 170 µm (middle) and 125 µm (right). image captured using constant lighting intensity, image captured using varying lighting intensity. Fig. 7. Example of a polycrystalline solar wafer with varying thickness. image captured with constant lighting intensity, image captured with continuously varying lighting intensity. B. Wafers with variable thickness The intensity of the lighting system does not need to remain constant while the line-scan camera is capturing. By utilising the rotary encoder information, the specific thickness along a part of the solar wafer can be determined along with its specific location. The variable lighting system can then be continuously adjusted accordingly while the wafer is being captured by the line-scan camera. As shown in Fig. 7, by continuously adjusting the lighting intensity in accordance to wafer thickness information, the contrast consistency is much improved. The horizontal line profile graphs is given in Fig. 8, showing an improvement in average gray level consistency. However, it should be noted that this method will only work if the thickness of the solar wafer varies perpendicularly to the direction of travel of the conveyor. The wafers therefore must be placed on the conveyor based on the slicing direction of the silicon ingot as that s where the variations to the wafer thickness tend to occur.

5 Fig. 8. Horizontal line profile graph of a solar wafer. The thickness increases from the left to the right regions. image captured with constant lighting intensity, image captured with continuously varying lighting intensity. C. Image thresholding Fig. 9 compares the results of the proposed method with standard Niblack s algorithm. In this case, both monocrystalline and polycrystalline solar wafers are inspected. It can be seen from Fig. 9 that the original Niblack s algorithm produced very noisy results. Close examination of these images revealed the difficulty in visually distinguishing between micro-crack and intact pixels. In contrast the proposed algorithm resulted in improved performance as as evident from Fig. 9 (c). Clearly the micro-crack pixels have accurately been segmented, leading to a more accurate representation. These results suggest that the proposed algorithm is more accurate in partitioning solar wafer images into micro-crack and intact regions. IV. CONCLUSION A new approach for high-speed micro-crack detection of solar wafers with varying wafer thickness has been presented. In the proposed system, the transmission intensity of the NIR source is adaptively matched to wafer thickness, resulting in an optimal image contrast. The image processing is performed using an improved version of Niblack s filter. Results from images captured on monocrystalline and polycrystalline solar wafers suggest that the proposed method outperformed the original Niblack operator, leading to a more accurate representation and defect detection. (c) Fig. 9. Comparison of Niblack s method against the proposed method. original image Niblack s method (c) proposed method. ACKNOWLEDGMENT This work has been supported by the Collaborative Research in Engineering, Science and Technology (CREST), grant 304/PELECT/ /C121. REFERENCES [1] Y. C. Chiou and J. Z. Liu, Micro crack detection of multi-crystalline silicon solar wafer using machine vision techniques, Sensor Review, vol. 31, no. 2, pp , [2] D. M. Tsai, C. C. Chang, and S. M. Chao, Micro-crack inspection in heterogeneously textured solar wafers using anisotropic diffusion, Image and Vision Computing, vol. 28, no. 3, pp , [3] A. Belyaev, O. Polupan, W. Dallas, S. Ostapenko, D. Hess, and J. Wohlgemuth, Crack detection and analyses using resonance ultrasonic vibrations in full-size crystalline silicon wafers, Applied Physics Letters, vol. 88, no. 11, pp , [4] T. Trupke, R. A. Bardos, M. C. Schubert, and W. Warta, Photoluminescence imaging of silicon wafers, Applied Physics Letters, vol. 89, no. 4, pp , [5] S. S. Ko, C. S. Liu, and Y. C. Lin, Optical inspection system with tunable exposure unit for micro-crack detection in solar wafers, Optik - International Journal for Light and Electron Optics, vol. 124, no. 19, pp , [6] J. Sun, J. Zhang, Z. Liu, and G. Zhang, A vision measurement model of laser displacement sensor and its calibration method, Optics and Lasers in Engineering, vol. 51, no. 12, pp , [7] W. Niblack, An Introduction to Digital Image Processing. Birkeroed, Denmark, Denmark: Strandberg Publishing Company, 1985.