xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 002007325
007 cr mnu|||uuuuu
008 090618s2008 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002779
Resonance ultrasonic vibrations and photoluminescence mapping for crack detection in crystalline silicon wafers and solar cells
h [electronic resource] /
by Andrii Monastyrskyi.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 65 pages.
Thesis (M.S.E.S.)--University of South Florida, 2008.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
ABSTRACT: The solar energy, or photovoltaic (PV) industry, driven by economic competition with traditional fossil energy sources, strives to produce solar panels of the highest conversion efficiency and best reliability at the lowest production cost. Solar cells based on crystalline silicon are currently the dominant commercial PV technology by a large margin, and they are likely to remain dominant for at least one decade. The problem of improvement mechanical stability of silicon wafers and finished solar cell is one of the most critical for entire PV industry. Mechanical defects in wafer and cells in the form of periphery or internal cracks can be initiated at various steps of the manufacturing process and becomes the trigger for the fracture. There are a limited number of characterization methods for crack detection but only a few of those are able to satisfy PV industry needs in sensitivity of the crack detection incorporated with the analysis time.The most promising are a Resonance Ultrasonic Vibrations (RUV) technique and Photoluminescence (PL) imaging. The RUV method was further developed in this thesis project for fast non-destructive crack detection in full-size silicon wafers and solar cells. The RUV methodology relies on deviations of the resonance frequency response curve measured on a wafer with peripheral or bulk millimeter-length crack when it is compared with identical non-cracked wafers. It was observed that statistical variations of the RUV parameters on a similarly processed silicon wafers/cells with the same geometry lead to false positive events reducing accuracy of the RUV method. A new statistical approach using three independent RUV crack detection criteria was developed and applied to resolve this issue. This approach was validated experimentally. Crack detection using RUV technique was applied to a set of production-grade Cz-Si wafers and finished solar cells from the Isofoton's S.A. (Spain) production line.Cracked solar cells rejected by the RUV method using the statistical approach were imaged with room temperature PL mapping and independently controlled with Scanning Acoustic Microscopy (SAM). A comparison of three independent techniques for crack detection, RUV, PL and SAM, was performed on selected samples. A high accuracy and selectivity of the RUV method to identify mm-size cracks in wafers and cells was confirmed.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Co-advisor: John Wolan, Ph.D.
Co-advisor: Sergei Ostapenko, Ph.D.
x Chemical and Biomedical Engineering
t USF Electronic Theses and Dissertations.
Resonance Ultrasonic Vibrations and Photoluminescence Mapping for Crack Detection in Crystalline Silicon Wafers and Solar Cells by Andrii Monastyrskyi A thesis submitted in partial fulfillment of the requirements for the degree of Master of Scie nce in Chemical Engineering Department of Chemical and Biomedical Engineering College of Engineering University of South Florida Co Major Professor: John Wolan Ph.D. Co Major Professor: Sergei Ostapenko Ph.D. Scott Campbell, Ph.D. Date of Appro val: October 1, 2008 Keywords: renewable energy, photovo ltaic, crack s detection acoustic standard deviation Copyright 2008 Andrii Monastyrskyi
Acknowledgements First of all I would like to thank my major professor Dr. Sergei Ostapenko for his help, technical guidance and expertise, patience during my research work that led to this Master, and for all o ther support that I have received. I would also like to thank Dr. John Wolan who was of great help and shared his expertise with me while motivating me to work and study harder and obtain better resu lts. And last but certainly not least, I wo uld like to thank Dr. Igor Tarasov, who had had th e patience to help me in many aspects of resonance ultrasonic vibrations area as well as photoluminescence segment of this work.
i Table of Contents List of Tables ii List of Figures i ii Abstract v i Chapter 1 : Introduction and Background 1 1.1 Defining the need for renewable e nergies 1 1.2 Silicon for photovoltaic c ells 2 1.2.1 Mono crystalline s ilicon 6 1.2.2 Polycrystalline s ilicon 7 1.2.3 Amorphous s ilicon 8 1.3 Fracture and cleavage in s ilicon 9 1.4 Crack d etection m ethods 10 1.4. 1 Optical t ransmission 12 1.4. 2 Photoluminescence and electroluminescence i maging 14 1.4. 3 Infrared lock in ultrasound t hermography 16 1.4 4 Scan ning acoustic m icroscopy 1 7 1.4. 5 Impact t esting 19 1.5 Resonance ultrasonic v ibration 20 1.5.1 Development of the RUV s ystem 21 1.5.2 Current system d escription 24 Chapter 2 : Experiments 29 2.1 RUV system s etup 29 2.2 PL s ystem s etup 29 2.3 SAM system s etup 31 Chapter 3 : Results 33 3.1 RUV s tatistics 38 3.2 Experimental and production grade crack detection 39 3. 3 RUV, PL, SAM comparison 46 Chapter 4 : Conclusions and Recommendations 55 References 58
ii List of Tables Table 1.1 Comp arisons of crack detection methods 12 Table 3.1 Definition of thresholds 36 Table 3.2 Summary of RUV/SAM comparison on 125 mm x 125 mm wafers and cells 37
iii List of Figures Fi gure 1.1 Czochralski method of silicon growth 4 Figure 1.2 Float Zone method of silicon cry stalline growth 5 Figure 1. 3 crack detection a pparatus 13 Figure 1. 4 crack widths and detection m echanisms 13 Figure 1.5 EL images of an industrial screen printed solar cell 15 Figure 1. 6 SAM image of solar cel l sample showing a crack along the periphery. Image sizes 160mm x 160mm, crack length is about 100 mm total 19 Figure 1.7 Impact testing apparatus including impact hammer, wafer, and m icrophone 20 Figure 1. 8 3 D image of the piezoelectric transducer with cross hatching for vacuum c ontact, positioned beneath a sample w afer 22 Figure 1.9 SR850 Lock In functional b lock diagram 23 Figure 1.10 A physical schematic of the experimental RUV s ystem. 24 Figure 1 11 An electrical schematic of the RUV system 25 Figure 1.12 RUV program user interface screen shot 26 Figure 1.13 RUV parameter shift due to decreasing number of points 28 Figure 2.1 Photoluminescence setup for room temp measurements of Si wafers and cells 30 Figure 2.2 Principal setup for SAM operation 32 Figure 3.1 Si wafer/cell with crack (open marks) can be separated from a
iv regular wa fer/cell (closed marks) using one of three rejection criteria: (1) reduced amplitude, (2) increased bandwidth (BW), and (3) resonance downward frequency shift 33 Figure 3.2 Statistics of the bandwidth distribution on a set of as cut cast waf ers. Solid curve is an approximation of the histogram with a normal distribution: mean value = 90.4 Hz, standard deviation = 33 Hz. Wafers with potential 3 4 Figure 3.3 RUV statistics of the three parameters of the set of 65 cells. 36 Figure 3.4 Diagram illustrating full statistical evaluation of the Cz Si wafers and cells 38 Figure 3.5 Deviation of the RUV peak bandwidth from the mean value versus crack length measured by SAM on a set of 125mm Cz Si cells 39 Figure 3.6 RUV parameters shift within in case of cracked wafer and SAM conformation of it. The image size is 158x158 mm, resolution is 100 micron 40 Figure 3.7 RUV of 125mm production grade Cz Si cells 42 Figure 3.8 SAM images of rejected cells 43 Figure 3.9 t hreshold and 3 threshold of amplitude, bandwidth a nd peak frequency as a function of process step 44 Figure 3.10 Comparison of the frequency distribution of as cu t and finished 125125 mm 2 cells 45 Figure 3.11 a normalized by intensity PL spectrum measured in the range of 1050 1580 nm at room temperature in a single crystal silicon wafer without crack (opened marks) and same wafer in the cr ack area (closed marks) ; b not normalized spectrum in the range of 1050 1580 nm at room temperature without crack (opened marks) and the same sample in the crack area (closed marks) 47 Figure 3.12 a PL map of crack a rea 29 x 29 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 260 mV red; b SAM image of the same area 29 x 29 mm, resolution 100 microns 48
v Figure 3.13 a PL map of area before crack 40 x 40 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 100 mV red; b PL map with crack 17.7 x 24.5 mm of the same area; c SAM image 40 x 40 mm, resolution 10 microns; d RUV parameters shift on the same wafer be fore the crack closed marks) and after the crack (opened marks) 49 Figure 3.14 a PL map of crack area 20 x 20 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 40 mV red; b SAM image of the same are a 20 x 20 mm, resolution 10 microns 50 Figure 3.15 a,c PL maps of crack area 20 x 20 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 140 and 47 mV red, the actual total crack lengths are ~19 mm and 21 mm; b,d SAM image of the same area respectively, resolution 100 microns 51 Figure 3.16 a SAM image 40 x 40 mm wafer area with 25 mm crack; b PL map of the crack area 40 x 40 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 200 mV red; c RUV parameters shift on the same wafer before the crack (closed marks) and after the 25 mm crack (opened marks) 53 Figure 3.17 a,b SAM images 20 x 20 mm from the back and f ront respectively, resolution 10 microns; b PL map of the scratch area 20 x 20 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 33 mV red; c RUV frequency scan on the same wafer before the scratch (closed marks) was introduced and with ~13 mm scrat ch (opened marks) 54
vi Resonance Ultrasonic Vibrations and Photoluminescence Mapping for Crack Detection in Crystalline Silicon Wafers and Solar Cells Andrii Monastyrskyi ABSTRACT The solar en ergy, or p hotovoltaic (PV) industry, driven by economic competition with traditional fossil energy sources, strives to produce solar panels of the highest conversion efficiency and best reliability at the lowest production cost. Solar cells based on crysta lline silicon are currently the dominant commercial PV technology by a large margin, and they are likely to remain dominant for at least one decade. The problem of improvement mechanical stability of silicon wafers and finished solar cell is one of the mos t critical for entire P V industry. Mechanical defects in wafer and cells in the form of periphery or internal cracks can be initiated at various steps of the manufacturing process and becomes the trigger for the fracture. There are a limited number of char acterization methods for crack detection but only a few of those are able to satisfy PV industry needs in sensitivity of the crack detection incorporated with the analysis time. The most promising are a Resonance Ultrasonic Vibrations (RUV) technique and P hotoluminescence (PL) imaging. The RUV method was further developed in this thesis project for fast non destructive crack detection in full size silicon wafers and solar cells. The RUV methodology relies on deviations of the resonance frequency response curve measured on a wafer with peripheral or bulk millimeter length crack when it is compared with identical non cracked wafers. It was observed that statistical variations of the RUV
vii parameters on a similarly processed silicon wafers/cells with the same geometry lead to false positive events reducing accuracy of the RUV method. A new statistical approach using three independent RUV crack detection criteria was developed and applied to resolve this issue. This approach was validated experimentally. Crack detection using RUV technique was applied to a set of production grade Cz Si wafers and finished solar A. (Spain) production line. Cracked solar cells rejected by the RUV method using the statistical approach were imaged with room temperature PL mapping and independently controlled with Scanning Acoustic Microscopy (SAM). A comparison of three independent t echniques for crack detection, RUV, PL and SAM, was performed on selected samples. A high accuracy and selectivity of the RUV method to identify mm size cracks in wafers and cells was confirmed.
1 Chapter 1 : Introduction and Background 1.1 Defining the n ecessity for r enewable s ources of e nergy The fossil energy is non renewable source of energy as it takes million s of years to form. The defining feature of global energy markets remains high i n 2007 and unsteady prices reflecting a tight balance of supply and demand . World energy consumption is projected to expand by fifty percent from 2005 to 2030 in the International Energy Outlook 2008 reference case projection [ 2 ] Although high prices for oil and natural gas, which are expected to continue throughout the period, are likely to slow th e growth of energy demand in the long term, world energy consumption is projected to continue increasing violently as a result of strong economic growth and e nlarging The complete revision of t he world fossil sources of energy usage has to be initiated This already has put issues such as energy security, energy trade and alternative energies at the forefront o f the political agenda worldwide. Environment issues linked closely with the move towards a more sustainable development path are leading the drive to develop and deploy cleaner technologies. Renewable energy technologies are well placed to contribute to improving environmental concerns. Over the past five years renewable energy prices have fallen to a
2 point where most are now cost competitive with fossil technologies when all values (environment, jobs, security, etc.) are considered. The present cost of r enewable sources of energy is affordable for certain markets but it is still too high to actually compete with conventional energy sources T his leave s space for further research and development The energy of sun, wind and water has to be used as the leve l of the world energy consumption is constantly high. Electricity is the most versatile form of energy we have and as long as light is shining on the solar cell it generates electrical power which help s decrease our dependence on conventional energy source s 1.2 Silicon for photovoltaic c ells Photovoltaic (PV) cells are devices that convert directly sunlight into electricity bypassing thermodynamic cycles and mechanical generators. Cells are packaged in photovoltaic modules which are electrically connect ed making solar photovoltaic arrays. PV production has been doubling every two years, increasing by an average of 48 percent growing energy technology . At the end of 2007, global solar cell productio n was 7,911 megawatts. At present the vast majority of photovoltaic cells are made from silicon. Crystalline s ilicon in a form of single crystal, polycrystalline or ribbon wafer s, was responsible for almost 90% of worldwide PV production in 2001. The conversion efficiency for single silicon commercial modules ranges between 1 8 20%. Currently t he average price for single crystal modules is $ 2.5 3. 8 per peak watt. The energy conversion efficiency for a commercial module made of polycrystalline silicon ranges
3 between 1 4 to 1 6 % with average price of $3.91 per peak watt. Single silicon modules are also very reliable for outdoor power applications. The wafering in which ingots are cut into less than 2 00 m wafers is most expensive and time consuming stages. It is clear that resulting wafers are very fragile. Despite of dominant position of single silicon on PV market, it was recognized that other semiconductors could make good solar cells. Most of them exist in a form thin semiconductor layer of PV material s that is deposited on a low cost supporting layer such as glass, metal or plastic foil. Thin film solar cells can be expected to provide cost reduction and energy savings in cell manufacture. Of the large variety of solar cell types investigated the most popular are gallium arsenide [GaAs], cadmium telluride [CdTe], and copper indium diselenide [CuIn Ga Se 2 ]. Copper Indium Diselenide thin film materials ha ve the highest energy conversion efficiency of up to 18% (in individual cells) but its complexity make s it manufactur ing more complicated 1.2.1 Mono crystalline s ilicon Manufacture of the mono crystalline silicon solar cells is an outgrowth of the methods used for microprocessors fabrication. A major difference is a high purity of row polysilicon whi ch is critical for electronic device manufacturing and is less important for PV cell production. Although various techniques have been utilized to grow single crystals of silicon, two techniques have dominated in the production of silicone single crystals: Czochralski method and Float Zone method. The Czochralski (C Z ) method is the most common method of growing single
4 crystal ingots. The seed crystal is lowered in molten silicon doped with a p type impurity such as boron and drawn upward under tightly contr olled conditions of temperature, and linear and rotational speed [3 ] Figure 1.1 Czochralski method of silicon growth [4 ] Manufacture of the mono crystalline silicon solar cells is an outgrowth of the methods used for microprocessors fabrication. A m ajor difference is the purity of silicon which is critical for electronic device manufacturing and is not needed for PV cells production. Therefore the major source for silicon solar cell production is the waste material from the microelectronics device fa brication. CZ method produces cylindrical ingots of typically 100 200 mm diameter, although ingots of the largest silicon ingots produced today are 400 mm in diameter and 1 to 2 meters in length PV s ilicon wafers with approximately 0.2 mm thickness are t hen cut from the initial ingots. Wh ile silicon is grown by the C Z method the melt is contained in a quartz crucible and the contamination by oxygen in the silicon ingot from the walls of the silica should not exceed 10 18 atoms per cm 3 [ 5 ] Sensitive to ox ygen
5 impurity accessories must be developed using other crystal growth methods [6 ] Figure 1.2 Float Zone method of silicon crystalline growth . The Float Zone (FZ) method is originated from zone melting which was used to refine binary alloys and was invented by Theuerer [8 ]. The first FZ crystal was grown by Keck and Golay in 1953 [7 ]. In the FZ method a polysilicon rod is converted into a single crystal ingot by passing a molten zone from one end of the rod to the other The ingot is moved linear ly and rotationally under controlled conditions and a high purity, low oxygen containing single crystal could be formed with the help of the seed crystal The
6 dimensions of FZ wafers are generally less than 150mm due to the surface tension limitations duri ng growth The molten silicon does not contact any substances other then ambient gas in the growth chamber and FZ crystal contains less then 10 16 atoms of oxygen per cm 3 Inherently then FZ crystal is dist inguished by its higher purity and resistivity [ 8 ] Nowadays FZ silicon is employed for premium high efficiency cell applications and CZ Si is used for higher volume, lower cost applications. 1.2.2 Polycrystalline s ilicon Polycrystalline silicon (poly Si) wafers are capable of producing cells of abou t 80% of the performance of cells buil t on a single crystalline wafer. Today 50% of the world's supply of polysilicon is being used for production of solar power panel The poly Si wafer has higher packing density due to rectangular geometry which makes pol y Si PV modules competitive on the market. The polycrystalline silicon wafers are also less fragile and can be cut into one third the thickness of CZ material. The starting material for high purity pol y Si is silica. The melting and reduction of silica gives metallurgical grade silicon (MG Si) with purity about 98 99%. However, to be useful as a semiconductor material in solar cells, silicon must be refined to a purity of 99.99%. Thus the next step is to purify MG Si to the level of semiconductor grade purity. The basic concept is that powdered MG Si is reacted with anhydrous HCl to form various chlorosilane compounds in a fluidized bed reactor. Then the silanes are purified by distillation and chemical vapor deposition (CVD) to form s emiconductor grade silicon
7 [ 8 ]. There are several methods of producing poly Si wafers but most accepted are sheet or ribbon silicon growth and direct ingot casting. In the casting process molten silicon is directly cast into a mold and allowed to solidify into an ingot. I n order to overcome surface stress within parallel aligned carbon molds it uses a centrifugal casting approach. The cooling rate is one factor that determines the final size of crystals in the ingot and the distribution of impurities. Despite initially pr omising results this appro ach is no longer used widely [40 ]. The second commercially developed is ribbon growth approach which is modification of the Edge defined Film fed Growth (EFG) technique. This method involves pulling of sili con ribbon to heights o f up to 6 m from the graphite dies with the source of molten silicon. A frame holds the thin sheet of material when pulled from the melt. In the EFG method the ribbon is pulled in the form of a polygon (8 or 12 facets) in which the sides could be cut by a laser to indivi dual wafers (12.5 12.5 cm2) [40, 41 ]. In 1997 Evergreen Solar Inc developed and commercialized string ribbon technology which implies use of two high temperature resistant strings drawn in the melted silicon. The pulling of silicon to a bout 7 8 mm is allowing the crystallization which becomes the ribbon. In this case temperature control at liquid solid phase is less demanding in comparison to the EFG method [40 ] Despite the fact of lower manufacturing cost of poly Si it is offset b y the lower cell efficiency.
8 1.2.3 Amorphous s ilicon Si) is another allotropic form of the silicon with Si and they contain large numbers of structural a nd bonding defects. In 1974 amorphous silicon beg a n to be use d in photovoltaic devices by properly controlling the conditions under which it was deposited and by carefully modifying its composition, for example by plasma enhanced chemical vapor deposition (PECVD). Si solar cells individual layers are deposited in a high frequency glow discharge reactor. Hydrogen and silicon are split from the mixture of silane (SiH 4 Si can then be transferred onto met al or glass. The a ddition of diborane (B 2 H 6 ,) or phosphine (PH 3 ) are used for appropriate p and n doping . Thin film can absorb 90% of the usable solar energy because for a given layer thickness a morphous silicon absorbs solar radiation 40 times mo re efficiently than single crystal silicon does. This is one of the most important factors affecting its potential for low cost. Other principal economic advantages are that amorphous silicon can be produced at a lower temperature and can be deposited on l ow cost substrates. These characteristics make amorphous silicon the leading thin film PV material. Despite achieved efficiency of 13% (low defect density deposition technique and use of anti Si solar cells is t heir stability. Weak silicon hydrogen bonds tend to be broken in the amorphous material due to the recombination of light generated charge particles. This degradation was first researched
9 by Staebler and Wronski and the ef fect was named after them . To day, amorphous silicon is commonly used for solar powered consumer devices that have low power requirements calculators, watches. 1.3 Fracture and cleavage in s ilicon One of the major current technological problems for the PV industry is to identify and eliminate potential sources of mechanical defects such as microcracks leading to the loss of wafer integrity and ultimate breakage of processed silicon wafers and finished solar cells . Generally speaking fracture occurs when the energy availab le for crack elasticity states that the amount by which a material body is deformed (the strain) is linearly related to the force causing the deformation (the stress). The re are two different approaches to fracture analysis: the energy criterion and the stress intensity approach. Griffith and Irwin found an expression for the constant G in terms of the surface energy of the crack : equation 1.3.1 where f is the fracture stress, E moment of fracture critical energy release rate G is a measure of fracture toughness. Silicon fracture has been studied extensively for the last few decades and i s
10 reported to have two principal cleavage planes: (111) and (110) [15 17]. The lowest fracture energy at which a crack propagated in (111) plane is 2.2 J/m2 which is twice the surface energy density. The energy needed for fracture of already defected sil icon material is even lower . Fracture can be characterized by crack direction , propagation speed , and also the atomistic mechanisms of fracture . Wafer breakage during PV processing is a very high cost issue. Propagation of cracked cel ls into the end modules would cause failure of the end unit. Downtime of several minutes caused by wafer failing and cleaning of scattered parts is not acceptable at a high throughput production level typically a few seconds per wafer Additional stress at the process step of already cracked wafer s cause s crack elongation followed by wafer breakage. Electrical failures during cell and module tests is also common. It was shown that cracks in readily processed solar cells lead to a weak recombination current. Cracks are also causing serious ohmic shunts especially if wafer has it before processing or screen pri nting the contact metallization . The shunts described above may also emerge if there are any holes presented in a cell, e.g., resulting from laser cutting . 1.4 Cra ck detection m ethods Currently mono as well as poly crystalline Si wafers are sliced thinner with thicknesses down to 15 0 microns in order to reduce production costs and compensate for the feedstock shortage. Wafers having a high l evel of residual elastic stresses behave extremely unpredictably during processing and handling. Usually cracks are introduced at the sawing stage of block s /ingots. Those cracked wafers processed as normal and even
11 more cracks are added during the later s tages such as solar cell processing. About 0.5% undetected defective cells end up in modules which makes approximately 74,000 defective cells per year with yearly los s close to $1M. To improve the economics of cell manufacturing, the PV industry requires a development of specialized inspection and quality control tools for integration into the production process. Dallas in 2006 [ 39 ] identify that this in line tool should allow (1) rejection of mechanically unstable Si wafers after ingot slicing before waf ers are introduced into further cell processing, (2) identification of wafers with mechanical defects (such as cracks) during production to avoid their in line breakage, (3) detection of cracked cells before they will be laminated into modules to avoid pan el efficiency reduction and product return from the field. The testing tool must possess the following features at a minimum: (i) high speed data acquisition and analysis, matching the approximately 2 seconds per wafer throughput rate of typical cell lines ; (ii) high stability (reliability and duty cycle) of the hardware performance including wafer loading/unloading and parts movement; (iii) easy integration into a belt conveyor configuration or cell testing station, an (iv) user friendly algorithm for waf er/cell rejection with a minimum number of false positives. There are numbers of experimental results published for crack detection in Si wafers and a majority of them ha ve been based on imaging techniques. The most interesting are optical and ultrasonic m ethods such as, optical transmission , photoluminescence  and electro luminescence imaging , infrared lock in ultrasound thermography , impact testing , and scanning acoustic microscopy [26, 27]. In Table 1 .1 we compared different metho ds for crack inspection currently under
12 investigation and prototyping. A n alternative new approach for stress control and crack detection in solar grade Si wafers and cells using the Resonance Ultrasonic Vibrations (RUV) system was developed by a research group at the University of South Florida [33 ]. Table 1 .1 Comparison of crack detection methods 1.4. 1 Optical t ransmission crack detection method was developed at the beginning of the century at the University of Konstanz, Germany and is said to be viable as an inspection Method Strength Weakness Scanning Acoustic Microscopy High spatial resolution, 10 microns Wafer must be immersed in water, Long data acquisition time above 10 minutes per whole wafer ; IR thermography High spatial resolution (below 1 mm) imaging technique Long acquisition time to use for in line control ( > 1 min ute per wafer ) Luminescence High throughput, snap shot imaging Interference with other defec ts (scratches, dislocations) Closed cracks are hidden due to diffraction limit. Optical transmission High throughput, high sensitivity Not applicable in case of closed cracks and final cells with back contacts Impact testing High throughput Minimum cr ack length is above 25 mm Resonance Ultrasonic Vibrations (RUV) High throughput (<2.0 seconds per wafer); Applicable for in line control; No interference with scratches and other defects Sensitivity to crack length is limited by wafer statistics; Do not identify crack location, (basic model)
13 tool for inline characterization of wafers and finished cells with throughput rates of up to 2.500cells/hr [19,20]. In this method the silicon wafer is placed above a broad spectrum flash light with an intensity of approximately 1000 suns. A high resolution CCD camera is used to detect the optical transmission through the wafer (Figure 1.3). Fi gure 1. 3 Optical transmiss ion crack detection apparatus [19 ] Individual wafer thickness and crystal structure affect the infrared portion of the light passes through the sample. CCD camera is detecting those modulation and can detect 1 5 M cracks while for cracks smaller than 1 M infrared detection system is sufficient (Figure 1.4). Figure 1.4 crack widths and detection m echanism [19 ]
14 However, small peripheral wafer cracks would not be visible using optical transmission. Also due to the optical diffraction limi t closed cracks within the wafer's surface with 1 m width would not be detectable. Complete solar cells and wafers having backside aluminum contact are also not suitable for crack detection using this technique due to metal interference. 1.4. 2 Photolum inescence and electroluminescence i maging Luminescence is a process in which light is emitted from a material at different wavelengths after external energy was delivered and absorbed. Luminescence is somewhat opposite to basic PV operation where photon energy exceeding a certain band gap energy, is applied to a valence electron, causing the bonds to brake and the electron to move to the conduction band. Various t ypes of luminescence are identified which differ by a source of external energy, such as phot oluminescence (external light), electroluminescence (electric field), cathod o luminescence (electron beam), etc. Photoluminescence (PL) is a process in which a chemical compound absorbs a photon with a wavelength in the range of visible or UV electromagne tic radiation, thus transitioning to a higher electronic energy state, and then radiates a photon back out, returning to a lower energy state. The period between absorption and emission is typically extremely short, on the order of 10 nanoseconds . Tru pke et al [21, 22] used photoluminescence imaging to view process defects in silicon wafers. In order to get PL image a 15 W /815 nm diode laser was used to illuminate the sample. Lifetime distribution can then be analyzed within an area of
15 interest. PL s ignal was detected using a cooled CCD one mega pixel camera with a 1000 nm longpass filter located between the sample and CCD camera to prevent incident laser light from contributing to the measured PL intensity. Thus PL is contactless and can be applied t o monitoring silicon wafers carrier lifetime at different stages of production. Authors also reported data acquisition time of 1.5 s and spatial resolution of 130 m The minority carrier diffusion length distribution could also be analyzed using electr oluminescence imaging . Electroluminescence (EL) is a form of luminescence which is the result of radioactive electron hole recombination in semiconductor material. The excited electrons then release their energy as light (photon). Fuyuki et al  ha d found that the intensity distribution of light emission agreed with the mapping of minority carrier diffusion length in silicon active layers. Emitted from the silicon wafer infrared light was collected by the cooled CCD camera in the 300 1100 nm region. Emission intensity increased linearly with the diffusion length. The spatial resolution was 250 m (Figure 1.5). Figure 1.5 EL images of an industrial screen printed solar cell .
16 Both PL as EL techniques are fast and contactless but are lacking th e capability of identification and differentiation between actual crack and surface scratch which will be shown later. 1.4. 3 Infrared lock in ultrasound t hermography Rakotoniaina et al [25 ] had employed the Ultrasound lock in t hermography ( ULT ) method f or detection of cracks in silicon wafers a nd solar cells not long ago The principle of crack detection by ULT is based on the heat created by the cracks flanks because of their friction caused by the ultrasound driven into the silicon wafer. The special r esolution of the method depends on the quality of IR camera incorporated into the ULT setup. Their lock in thermography system allowed imaging of periodic surface temperature modulations at frequencies up to 54 Hz having an effective value as low as 10 mK using 1/2 hour measure time. This long acquisition time is required for signal averaging due to low infrared intensity and makes ULT unsuitable for in line crack detection where no more than a couple of seconds are accepted for quality control. Also optic ally invisibl e cracks could be detected, etched cracks do not lead to local heat sources and might require c overing the wafer surface with black paint which considerably enhances the IR signal
17 1.4.4 Scanning acoustic microscopy The use of ultraso und methods had been established in industry in the early 1900s. The acoustic waves with frequency over 20 kHz are referred as ultrasonic waves widely in military and m edicine. For example medical sonography nondestructive diagnostic medical imaging technique used to visualize muscles, tendons, and internal organs, their size, structure and any pathological lesions is purely based on acoustic phenomena. Lemons and Qua te developed the first scanning acoustic microscope (SAM) in 1974 at Stanford University . But before that the C scan was used as a nondestructive testing (NDT) method since the 1950s. Modern SAMs are the hybrid tools with characteristics of both the f scan. SAM was used by the IC industry extensively for a long time, although the PV manufacturers began to employ acoustic microscopy technique for crack detection in silicon wafers and solar modules for a relatively short time . SAM is a pulse echo microscope that operate s a focused transducer which generate s and receive s the ultraso nic pulses reflected beneath the front surface of the sample  The piezoelectric crystal acts as active element in the transducer which gen erates an ultrasonic impulse. Concave lens are attached to the tip of the transducer. Image visualization is produced by passing the transducer across the sample. Scan time varies from seconds to minutes depending on the desired resolution and the area of scan. The coalescence of longitudinal and flexural waves interference is reflected from the
18 surface an d from the back side of the surface. The reflection contrast is patterned when separate layers within a structure isolated using time separation Ultrason ic waves are very sensitive to the density variations (such as voids or air gaps) so a coupling medium is required. Immersion tank is normally employed to run most of the inspections. SAM is a non de structive package analysis when imaging the internal feat ures of the sample The area of distortion is identified as acoustic impendence discontinuity at damaged silicon wafer or cell. A coustic impedance is a ratio of the acoustic pressure to the particle velocity per unit area  and can be expressed as : Equation 184.108.40.206 where Z i is an acoustic impedance of the material in the i th layer, i is a density of the m aterial in the i th layer, and v i = velocity of sound in the i th layer. The magnitude and time of travel of the reflected signal provide crucial information about the material. SAM has proven to be an accurate method for identify ing cracks and microcracks in wafers and partially processed solar cells. It can provide information about where cracks and microcracks are occurring and be a useful tool in quantitative analysis of those (Figure 1.6) However, each SAM wafer measurement to ok 20 minutes for sample set up and data collection. This is clearly not an in line production process. The data from this effort was very useful as the sample set evaluated by SAM can now be used to calibrate and verify the accuracy of any new methods dev eloped for crack detection.
19 Figure 1.6 SAM image of solar cell sample showing a crack along the periphery. Image sizes 160mm x 160mm, crack length is about 100 mm total. 1.4. 5 Impact t esting Recently at USF reported new tec hnique for crack detection in silicon wafers [ 30 ]. In this method the audible response in the frequency range up to 2,000 Hz after a mechanical impact on a Si wafer is analyzed. In the experiment setup a piezoelectric hammer with a vinyl tip appl ies the m echanical impact to t he si licon wafer which is placed on a piece of convoluted foam. The microphone is mounted 2 cm above the wafer picking up the impact response and is allow ed to calculate the frequency response with the impact force defined as the input and the sound pressure
20 level as the output. A picture showing the impact testing apparatus is shown in F igure 1. 7 Figure 1. 7 Impact testing apparatus including impact hammer, wafer, and microphone [30, 31 ] Cracked wafer s showed a decrease in natur al frequency, a decrease in peak magnitude, and a larger bandwidth which corresponded to crack response from the RUV technique  Peak amplitude has been found to be most sensitive to the cracked wafers. Impact testing allows identif ication of cracks w ith total length of 10 mm only. 1.5 Resonance ultrasonic v ibrations The Resonance U ltrasonic V ibrations (RUV) is a technique for non destructive stress control and crack detection in full size silicon wafers and solar cells [33, 35] In a case of crack detection t he RUV methodo logy relies on deviation of the resonance frequency response curve of a wafer with peripheral or bulk millimeter length crack
21 versus regular non cracked wafers. Three RUV frequency curve crack rejection criteria were proposed : (1) shift of the peak position downward ; (2) increase of the bandwidth, and (3) reduction of the amplitude. 1.5.1 Development of the RUV s ystem The RUV as a method of detecting deficient wafers was developed at USF in 2002 [32, 33]. This technique involves exciting entire silicon wafer with tunable frequency and recording the response from it at defined regions. The deviation of the frequency response curve of a wafer with a periphery crack versus undamaged wafers is analyzed The wafer is positioned on the transducer which is connected to the vacuum pump through the cross section in the center of the transducer (Figure 1.8). Once the wafer is placed on the proper size base the transducer goes up on the moving Z stage triggering the vacuum pump to turn on wh ich create s small negative pressure from the back side. This pressure (about 50 kPa) keeps the wafer in a specific location and allows the transducer to induce resonance vibrations in the form of standing waves into the sample A transducer frequency can b e swept in the ultrasonic range from 20 kHz to 100 kHz using a function generator (WaveTec 10 MHz DDS 29) [31 34] and amplified by a power amplifier (Samson Servo 260) Ultrasonic vibrations are propagated into the wafer and detected using a high sensitive broadband ultrasonic probe at the edge of the wafer
22 Figure 1.8 3 D image of the piezoelectric transducer with cross hatching for vacuum contact, po sitioned beneath a sample wafer. Thin layer of chemical mechanical planarization ( CMP ) pad material is attached to the probe active side in order to control applied force. The vibrations are then fed into the lock in amplifier ( Stanford Research Systems SR 850 )  The Lock in amplifier is used to detect a very small AC signals from the probe and separa te it from the noise (Figure 1.9). It detects signal based on modulation at some known frequency A reference frequency is essential for the measurement as lock in encounter s the response from the frequency swe e p and separate s component s of the signal at a specific reference frequency and phase using p hase sensitive detection technique Noise signals, at frequencies other than the reference frequency, are cancelled out and do not affect the measurement The accurate positions of the wafer, transducer and the probe are regulated by
23 s tepper motors. The whole RUV system is PC controlled and operated by Windows based original software Figure 1.9 SR850 Lock In functional b lock diagram [ 42 ] The RUV measures single resonance curve and then analyzes each of the rejections parameters of defected wafer in comparison to undamaged one Each silicon wafer however differs in physical properties such as size, weight, and thickness. This alteration induce s changes in peak position, peak bandwidth, and peak amplitude and small variations of these parameters are acceptable. However it also might provoke a generation of false positive events when undamaged wafer is recognized as a potentially cracked one. RUV technique is also not capable of providing information of ex act location of the crack.
2 4 1.5.2 Current system d escription The c urrent model of RUV system (RUV 2) provides a controlled delivery of the ultrasonic vibrations in the form of a frequency sweep into a solar grade silicon (Si) wafer or solar cell using external ultrasonic transducer. High sensitive ultrasonic probe measures the resonance ultrasonic vibrations of the wafer at a specific frequency range, which depends on the wafer size and shape. Resonance vibrations generated by the wafer provide a sensi tive feed back on mechanical quality of the wafer or solar cell. Frequency scans identify the wafer mechanical problems such as stress or cracks. Figure 1.10 A basic schematic of the experimental RUV system. The RUV 2 (basic version) is composed of t wo parts (Figure 1.10) : (a) measuring unit, which include the X and Z moving stages with ultrasonic transducer and ultrasonic probe; and (b) computer controlled electronics, including PC with Windows base d software (Figure 1 .12 ). The m easuring unit is a pl ace where the measurements occur. The
25 unit consists of a broad band ultrasonic transducer located on a moving Z stage, the ultrasonic probe (high frequency microphone) attached to the moving X stage, and a wafer holder. The stages are controlled by rack mo unted electronics comprised of (a) lock in amplifier; (b) stepping motor controller; (c) power amplifier, (d) DC power supply, (e) oscilloscope, and (f) vacuum pump [31, 35] The computer with GPIB interface card and original software operates the RUV 2 system. Buil t in function generator in Stanford SR 850 lock in amplifier is used to generate the frequency sweep. The signal from the probe is directly fed to the lock in without amplifying which significantly reduces acquisition time. Figure 1.11 An electrical schematic of the RUV system The RUV 2 system is also a PC controlled unit with Windows based software providing system operation, data acquisition and data processing.
26 T he RUV user interface id shown in Figure 1.12 Figure 1.12 RUV program user interface screen shot stage moves up and pick s up the wafer, vacuum switch opens and appl ies negative pressure between the wafer and transducer, the p robe located on the X stage moves toward the scan runs and display s in the window (Figure 1.12 ). The row f ures and displays the value of negative pressure in mm Hg. A principal RUV peak is approximated by a Lorentzian curve. Parameters of the approximation (amplitude, peak position and band width) are stored in the selected data file. Consecutive repeated mea surements add new approximation parameters to the data
27 file. At the end of the multiple runs the data can be exported to Microsoft Excel in the File menu for statistical evaluation. The RUV 2 system also allows performing a calibration for follow up crac k detection using calibrated set of wafers In the calibration protocol RUV 2 system finds statistic al paramet distribution (average value and standard deviation) on a calibrated set of similar wafers/cells which are measured sequentially. These statistical parameters for resonance peak amplitude, bandwidth and peak position are saved. The p arameters are used to quickly identify wafers/cells with f scan characteristics diff erent from average values by 1 (selected in sub menu). The duration of the measuring cycle can be varied by changing the number of data points per f scan o r lock in amplifier integration time. In Figure 1.13, the frequency sweeps performed with different data points per f scan (from 100 to 10) and cycling time from 12.4 seconds down to 2.0 seconds per cycle. T hroughput rate of 2 seconds is an achievable rate for the RUV 2 system. Reduction of the data points per scan provides only a small variation of the f scan parameters.
28 Figure 1.13 RUV parameter shift du e to decreasing number of points This allows RUV 2 prototype unit be integr ated into an automatic belt moving solar cell production line or used as a stand alo ne testing system for quality control.
29 Chapter 2: Experiments 2.1 RUV system s etup In Figure 1.9 a schematic of the RUV system with details published elsewhere [31, 35] is show n RUV 2 system was used for all experiments as described above. The entire RUV measurement cycle consisted of the consecutive steps: wafer loading from a home position on the transducer by vacuum coupling the wafer and transducer, lifting the transducer wi th the wafer to a measuring position using computer controlled Z stage, controlled X stage, data acquisition by measuring f scan, Lorentzian fit of the experimental data, and wafer unloading to the home position. Data stored in .txt files then was analyzed using scientific graphing and data analysis software Origin 6.0. 2.2 PL system s etup AlGaAs laser diode (pulsed) with emission wavelength at 804 nm and maximum output power 150 mW in pulse (model Spectra Diode Labs, SDL 800) was used as the excitation sources in the PL experiments. The photoluminescence signal was dispersed
30 with a 0.5 m SPEX 500M grating spectrometer possessing a reciprocal dispersion of 3.2 nm/mm (2nd order) with a 600 line s/mm diffraction grating. The dispersed signal was registered with a liquid nitrogen cooled Ge detector (North Coast Scientific Corp.) in the range of 1050 1550 nm. AC signal from the detectors was fed to Lock in amplifier EG&G Model 5209 and collected b y a computer. The PL mapping experiment was done with the use of an X step precision. All PL experiments were performed at room temperature. A schematic of the PL setup for PL measurement is shown on Figure 2.1. Figure 2.1 Photoluminescence setup for room temp measurements of Si wafers and cells
31 2.3 SAM system s etup Sonix, Inc. was used for all SAM experiments. T he emitting transducer of SONIX also serves as the re ceiving transducer upon reflection of the ultrasound pulses (DPR002S digital broad band pulser/receiver with emitting frequency up to 150 MHz). The general setup of HS1000 is shown on Figure 2.2. The instrument is also equipped with 1 M encoder for enhan ced resolution. Silicon wafer or cell is emerged into the coupling media which is DI water. Step motors (SMD1000C) then are allowing to map pulse magnitude in transverse plane by moving the stage with a transducer . The speed and scan area can be cont rolled by Win original software version 3.5.1. The cracks as small as 10 M can be identified and located using SAM technique. The reflection from back surface only analyzed. The resolution of 100 M was sufficient to visualize such a damaged wafer. The average time for scanning 158x158 mm wafer/cell is 24 minutes.
32 Figure 2.2 Principal setup for SAM operation [ 29 ]
33 Chapter 3: Results 3. 1 RUV s tatistics A crack introduced into Si wafer alters the RUV peak parameters: amplitude, bandw idth and peak posit ion. This is illustrated in Figure 3 .1 for two identical 125 mm size Cz Si 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 non cracked wafers. Figure 3.1 Si wafer /cell with crack (open marks) can be separated from a regular wafer/cell (closed marks) using one of three rejection criteria: (1) reduced amplitude, (2) increased bandwidth (BW), and (3) resonance downward frequency shift 
34 One of the technological ch allenges for using the RUV method as in line production tool occurred due to the fact that wafers (cells) of even the same size and shape are not identical. They show a statistical variation of the RUV peak characteristics caused by variations of the wafer size, thickness, internal stress, etc. The example of thi s variation is presented in Figure 3.2 on a set of production grade as cut 125 mm cast wafers. Figure 3.2 Statistics of the bandwidth distribution on a set of as cut cast wafers. Solid curve is a n approximation of the histogram with a normal distribution: mean value = 90.4 Hz, standard deviation = 33 Hz. Wafers with potential cracks are located above
35 The histogram represents a statistical distribution of the wafer bandwidth (BW ) and a fit by a normal distribution with characteristic mean value and a standard deviation l non cracked wafers, which statistic ally possess a large BW and may c this issue we suggested using p arallel statistical approach applied to all three independent RUV parameters simultaneously. It was assumed that such parallel statistics will dramatically y of the RUV method. These experiments were performed on a set of 125 mm and 156 mm Cz Si solar cells, which show a strong statistical scattering of the RUV parameters. As an example the BW standard deviation was 78 Hz compared to 19 Hz in similar size as cut Cz Si wafers. In a case of screening multiple wafers or cells with identical geometry, a statistical algorithm has been developed and implemented into the RUV system. In this algorithm, the RUV software generates a mean value (M) and standard deviation RUV parameters, i.e. amplitude, BW and peak using initial (reference) set of RUV parameter the system calculates three thresholds for accept or r eject command to represents a minimum or maximum allowable value of the RUV parameter. In the case of osition, and
36 Table 3.1 Definition of thresholds which further improves an accuracy of the threshold calculations. In the experiment we teste 3. 3 we show a measur ed RUV parameter on Cz three parameters. Figure 3.3 RUV statistics of the three parameters of the set of 65 cells. Cells with potential cracks are rejected using Amplitude BW Peak 3 case M 3 /2 M + 3 /2 M 3 /2 2 case M M + M
37 and 62. All suspects were removed from the batch of cells and measu red using Scanning Acoustic Microscopy (SAM). SAM mapping provided a clear confirmation of true positive events revealed by the RUV method. In Table 3.2 we summarize the statistical analyses on 125 mm wafers and cells. Note that percentage of errors which Table 3.2 Summary of RUV/SAM comparison on 125 mm x125 mm wafers/cells Concurrently, the number of confirmed by SAM is increased. It has been found that the RUV method provided identification of cells with cracks length down to 3 mm. Based on this study we con cluded that the RUV method offers a high probability of crack detection with 91% success rate and 9% of errors as a total of false positive and false negative events. We illustrate in Figure 3. 4 the results of the statistical analysis on Process Num ber of Waf ers Number of RUV Rejects Number of True Posi tives Number of False Positives Number of False Negatives As cut 112 1 1 1 1 0 0 0 Texturing 98 5 5 5 5 0 0 0 Diffusion 100 10 10 10 10 0 0 0 AR coating 99 3 3 3 3 0 0 0 Solar cells 110 8 12 7 11 1 5 1 Total 519 27 31 2 6 30 1 5 1
38 both sets of 125 m m and 156 mm wafers and cells. Note that this high success rate of the RUV method will lead to a substantial 10 fold reduction in wafers and cells that contain cracks and interfere with production, reducing line throughput and increasing module cost. We pr opose that different crack rejection coefficients must be incorporated into RUV system software, allowing a production manager to optimize crack inspection depending on the particular technological step of the cell manufacturing. Figure 3.4 Diagram il lustrating full statistical evaluation of the Cz Si wafers and cells. It was further proposed that the larger deviation of the rejected parameter (e.g. BW) from the mean value or corresponding threshold the larger a damage to the wafer cased by cracks. In Figure 3. 5 we present the result of quantitative analyses of the crack
39 length based on the rejected 125 mm cells. We observed a strong correlation of the BW deviation versus crack length. This experiment demonstrates that the RUV method can be also used f or estimation of the wafer damage and therefore serve as a crack characterization technique for in line application. Figure 3.5 Deviation of the RUV peak bandwidth from the mean value versus crack length measured by SAM on a set of 125mm Cz Si cells. 3.2 Experimental and production grade crack d etection The RUV method relies on a measurement of the resonance ultrasonic vibrations generated in a full size Si wafer/cell. It includes quantitative analyses of three RUV parameters that characterize the R UV resonance curve amplitude, bandwidth (BW) and peak position in a frequency scale (peak). Crack is expressed in the RUV method as a reduction of peak amplitude, shift of peak to lower frequency and increase of its BW (Figure 3.6 ) The hardware details are described earlier.
40 Figure 3.6 RUV parameters shift within in case of cracked wafer and SAM conformation of it. The image size is 158x158 mm, resolution is 100 micron.
41 In a case of screening multiple wafers or cells with identical geometry, sta tistical algorithm is developed and implemented in the RUV 2 system. In this algorithm, the RUV software generates a mean value (M) and standard deviation ( ) for each of the RUV parameters, i.e. amplitude, BW and peak as described above To reduce the nu mber of false positiv e events during crack detection, we suggest using a combination of at least two thresholds to identify and reject the wafers with cracks. In this project we tested both 3 To prove our assumption two size s of Cz Si wafers 125 mm x 125 mm, and 156 mm x 156 mm were used. The wafers were randomly selected at the solar cell producer company Isofoton (Spain) after different process steps from as cut wafers up to finished solar cells. The RUV system and statisti cal approach was applied separately to wafers of the same size and the same process step. After the blind test at Isofoton using the RUV system, all rejected wafers and cells were measured using Scanning Acoustic Microscope (SAM) with 100 microns spatial resolution to correlate with RUV data. In some cases to get better confidence of the crack using SAM the local area with crack was remeasured with 10 microns resolution (Figure 3.8) The results of RUV rejection and SAM images are compared.
42 Figure 3.7 RUV of 125mm production grade Cz Si cells
43 Figure 3.8 SAM images of rejected cells Image sizes 1 27 x 127 mm or 158 x 158 mm. Resolution is 100 microns. The last one is 20 x 20 mm, resolution 1 0 microns Since the characteristic ultrasonic vibration change after each process step, a mean value has to be defined for each step. Therefore, a set of about 100 wafers has been employed for each step to measure the amplitude, bandwidth and frequency of the characteristic peak of the ultrasonic vibration, ob t ain t thresholds.
44 In Figure 3.9 the change in mean value, 1 thresholds from one process step to the next has been depicted for all parameters. In all cases we can note that the mean values change. The amplitude var ies for both sample sizes without any clear trend. The same holds for the bandwidth for the larger sample size. For the 125125 mm 2 devices, the bandwidth slightly decreases until the final cell step and then increases abruptly. The same trend, but with op posite signs, is observed for the peak frequency for both sample sizes. In all cases the gap between the mean and threshold widens for solar cells indicating a broadening of the distribution. Figure 3.9 t hreshold and 3 threshold of amplitude, bandwidth a nd peak frequency as a function of process step
45 This is further highlighted in Figure 3.10 where the distribution of the peak frequency is plotted for as cut and finished cells of the 125125 mm 2 group. Note also that the mean value does not coincide with the peak of the distribution if a tail exists at one side of it. This shifts the threshold value towards the tail side if the tail lies on the same side as the threshold. The threshold will shift away from it, if the tail lies on the opposite side as the threshold. A better threshold may be determined by fitting the main peak of the distribution with a norm al distribution and using the 3 value associated with it. Figure 3.10 Comparison of the frequency distribution of as cut and finished 125125 mm 2 cells. The rejection value of the RUV system is based on a statistical deviation from a mean value. Theory and experiments indicate that with increasing crack length, the amount of deviation from the mean value increases. A crack has thus a higher probability
46 to be detected, if the distribution is narrow. If the distribution of the values is dispersed over a broad range, the 3 threshold m ay overlook cracked cells. The ultrasonic vibration also strongly depends on crystal orientation, which is manifested by a strong change in peak frequency and amplitude. In this case at least 8 of 100 wafers had a different crystallographic orientation tam pering the mean v alue and 3 threshold. On the one side this leads to a great number of apparently rejected wafers, on the other side this may push the threshold too far away from the real mean value, effectively excluding a certain number of broken wafers from rejection. The use of wafers of different crystal orientation is, nevertheless, very rare and therefore this extreme case should not emerge in a production environment. 3.3 RUV, PL, SAM comparison In this section we performed a correlation study of cracks on Si wafers and ce lls using three independent methods: RUV, SAM, and photoluminescence (PL). A room temperature PL spectrum in the range of 1050 1580 nm measured on Cz Si wafers is presented in Figure 3.11a b. The PL spectrum consists of a single peak with a maximum at ~11 25 nm corresponding to the band to band transitions in silicon . The spectral position and the shape of this peak are identical in different regions of the wafer, including the area around the crack (Figure 3.11a). Contrary to this, the amplitude of th e peak is various across the wafer and also is related to the crack (Figure 3.11b). This will be further shown by the PL mapping technique.
47 Figure 3.11 a normalized by intensity PL spectrum measured in the range of 1050 1580 nm at room temperature i n a single crystal silicon wafer without crack (opened marks) and same wafer i n the crack area (closed marks) ; b not normalized spectrum in the range of 1050 1580 nm at room temperature without crack (opened marks) and the same sample in the crack area ( closed marks) An illustrative result of the peripheral crack in Cz Si as cut wafe r measured by the PL mapping and SAM is presented in Figure 3.12a b T wo images were measured on the same part of a silicon wafer containing ~20 mm crack using PL and SAM. Th e crack was produced by applying the pressure to the sample edge using the stainless needle. The cracks propagated perpendicular to the sample edge along the <110> cleavage direction. The crack appearance is confirmed by SAM measured with 20 microns resol ution on the same sample Both methods give consistent results in this case. We noticed however, that the PL map in Figure 3.12a contains various artifacts such as reddish area around the crack area which can be attributed to the wafer damage or stress var iation caused by a crack This obviously may complicate crack identification using PL mapping or imaging technique s
48 Figure 3.12 a PL map of crack area 29 x 29 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 260 m V red; b SAM image of the same area 29 x 29 mm, resolution 100 microns In Figure 3.13 a d we are showing PL maps before crack was introduce (a) and actual map of the crack area (b crack length ~ 17.7 x 24.5 mm) at room temperature and wavelength of 1126 nm. The crack was produced applying the punctual pressure at 45 degrees to the wafer edge. The SAM image measured with 10 microns spatial resolution confirms presence and shape of the crack.
49 Figure 3.13 a PL map of area before crack 40 x 40 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 100 mV red; b PL map with crack 17.7 x 24.5 mm of the same area; c SAM image 40 x 40 mm, resolution 10 microns; d RUV parameters shift on the same wafer before the c rack (closed marks) and after the crack (opened marks). I n one of the solar cells using the RUV method we were able to diagnose a partially hidden crack, which was barely seen in the optical technique due to blocking by the busbar (Figure 3.14a b) Thus R UV method offered a clear advantage in th e case of cracks with similar location, which are quite common due to stress applied by contact printing technique.
50 Figure 3.14 a PL map of crack area 20 x 20 mm at room temperature and 1126 nm wavelength, inte nsity shown from 0 mV blue to 40 mV red; b SAM image of the same area 20 x 20 mm, resolution 10 microns. As a part of Isofoton study we also involve PL mapping as one of the three techniques for identifying the cracks in finished cells. Figure 3.1 5a d shows PL maps confirmed also by SAM of cracked area. The crack was introduced by hitting the cell with a diamond pin falling on the cell surface from about 30 mm height. The internal cross cracks with total length of 15 mm were obtained. PL map clearl y shows the presence of the crack with reduced PL intensity along the crack lines. We noticed that the crack observed in the PL image is in reality almost double in length compared to the crack in SAM (c and d). This can be explained that in the PL image t he real crack length is misinterpreted caused by artificial extension without penetration through the wafer thickness. This will be further demonstrated in scratch experiment below. Interesting observation in the PL mapping technique is that close to the c rack area we observed an evident increasing of the PL intensity (shown as red areas along the crack). This effect
51 could be explained by a local change of the geometry of the cracked surface or alternatively by increased of the local stress around the crack which contributes to the PL intensity. More experiments are necessary to identify the mechanism. Figure 3.15 a,c PL maps of crack area 20 x 20 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 140 and 47 mV red, th e actual total crack lengths are ~19 mm and 21 mm; b,d SAM image of the same area respectively, resolution 100 microns In the next experiment we show how the surface scratch could be mis interpreted by the PL method as a crack. Although scratches are o n e of silicon cell surface defects, t hey ha ve only a minor impact on the wafer/cell breakage in PV manufacturing. The
52 scratch was d esigned by a needle in the direction perpendicular to the wafer edge and tilted at 45 degrees with respect to cleavage directi on. This allowed to reliably distinguishing the scratch from the crack. In Figure 3.16a c we show example s of such a crack. The crack with length approximately 25 mm was introduced by applying the punctual pressure at 45 degrees to the wafer edge. The dire ction and shape of the crack was confirmed with SAM and PL (Figure 3.16a, b) as well as the wafer was rejected by RUV technique (Figure 3.16c). Thus decreasing of the peak position of no crack versus cracked wafer measurement is 1807 Hz, increasing in BW i s 238 Hz and decreasing of amplitude is 2.92 mV. Figure 3. 17 a c shows how a scratch defect is revealed by each of three methods on the same silicon wafer A scratch is clearly visualized in the PL map and has also been found by SAM image measured at the fr ont surface reflection (Figure 3 17a,b) In SAM the wafer was scanned from both sides and analyzed pulses reflected from the front and rear surface of the wafer. This SAM study allowed confirmation that ickness and is not classified as a crack (Figure 3.17c). In contrast to the PL there are no changes in the RUV peak parameters caused by the scratch (Figure 3.17d). This experiment justifies high selectivity of the RUV technique to provide diagnostics and characterization of cracks propagated through the wafer thickness.
53 Figure 3.16 a SAM image 40 x 40 mm wafer area with 25 mm crack; b PL map of the crack area 40 x 40 mm at room temperature and 1 126 nm wavelength, intensity shown from 0 mV blue to 200 mV red; c RUV parameters shift on the same wafer before the crack (closed marks) and after the 25 mm crack (opened marks).
54 Figure 3.17 a,b SAM images 20 x 20 mm from the back and front r espectively, resolution 10 microns; b PL map of the scratch area 20 x 20 mm at room temperature and 1126 nm wavelength, intensity shown from 0 mV blue to 33 mV red; c RUV frequency scan on the same wafer before the scratch (closed marks) was introd uced and with ~13 mm scratch (opened marks).
55 Chapter 4: Conclusions and Recommendation The RUV technique is adapted to fast and non destructive crack detection in full size Si wafers for solar cells. The RUV methodology relies on deviation of t he resonance frequency vibration curve of the wafer with periphery crack versus non cracked wafers It includes quantitative analyses of three RUV parameters that characterize the RUV resonance curve amplitude, bandwidth and peak position in a frequency scale. Crack is expressed in the RUV method as a reduction of peak amplitude, shift of peak to lower frequency and increase of its BW. A new statistical approach is developed and tested to reduce a number of false positive events when screening silicon wa fers with randomly distributed RUV characteristics. The statistical approach is based on parallel statistics to identify cracks in high volume production test. In this algorithm, the RUV software generates a mean value (M) and standard deviation ( ) for each of three RUV parameters By this means, 6 statistical parameters of M and are calculated. For each RUV pa rameter the system or reject command to pass the wafer as a good wafer or to rejec t as a crack suspect The threshold represents a minimum allowable deviation of the RUV parameter from its mean value In the case of screening multiple wafers or cells the statistical algorithm was applied successfully at single crystal silicon wafers an d finished cells from the Isofoton
56 production line Both sizes of wafers used by Isofoton 125 mm and 156 mm were consistently measured. Resonance peaks for RUV are clearly separated from other features in the f scans. RUV was applied to all wafers star ting from as cut and including finished cells. T h e effect of statistical distribution of wafers and cells can be minimized using combined statistics of three RUV parameters. The RUV crack detections w ere confirmed by SAM mapping with success rate of 95% u sing 3 threshold. Thus from more than 1000 wafers and cells (125 mm and 156 mm) scanned 27 were rejected by RUV and confirmed as a cracked with the SAM; 4 wafers/cells were rejected by RUV but there were no cracks in SAM; and 3 wafers/cells were not rejected by RUV but showed the presence of a crack in SAM image. It has been also experimentally shown that 2 rejection is sufficient interval to identify cracks above 3 mm length. Moreover, 1 mm crack length can be accepted as a limit for 2 rejection. This ca n be improved by using 1 reject ion with potential increase of false positive events. The RUV method can distinctly be used for as cut and processed wafers including finished solar cells. The PL and SAM technique s were employed in the current theses to co nfirm the RUV methodology. A comparison of three independent techniques for crack detection, RUV, SAM and PL, was performed on selected samples. A high accuracy and selectivity of the RUV method to identify mm size cracks in wafers and cells was confirmed. In contrast to optical inspection techniques it was experimentally shown that RUV method is not sensitive to surface scratches and therefore provides a firm identification of opened cracks which have the highest probability to initiate the wafer and cell breakage.
57 Future work to improve the RUV system would consist of: Exploring applicability of the RUV method for cells sequentially connected in cell strings. Performing theoretical analyses of the resonance vibration modes in mechanically connected silic on plates Performing experimental justification of the string concept using RUV approach.
58 References 1. www.bp.com 2. Jonathan G. Dorn, Solar Cell Production Jumps 50 percent in 2007 in Ear th Policy Institute, 2007 3. Kasap, S.O., Principles of Electronic Materials and Devices. 2nd ed. 2002: McGraw Hill. 68. 4. www1.eere.energy.gov/.../ images/illust_czoch.gif 5. Goetzberger, A., J. Knobloch, and B. Voss, Crystalline Silicon Solar Cells. 1998: Joh n Wiley & Sons, Inc. 136 137. 6. Schmidt, J., A.G. Aberle, and R. Hezel. Investigation of Carrier Lifetime Instabilities in CZ grown Silicon. in 26th IEEE Photovoltaics Specialists Conference. 1997. Anaheim, CA. 7. Keck, P.H. and M.J.E. Golay, Crystallization of silicon from a floating liquid zone. Phys. Rev., 1953. 89: p. 1297. 8. Shimura, F., Semiconductor SIlicon Crystal Technology. 1989: Academic Press, Inc. 128. 9. www.tf.uni kiel.de/.../ advanced/t6_1_3.html 10. Kalejs, J.P. An overview of new developments in cr ystalline silicon ribbon material technology for solar cells. in 3rd World Conference on Photovoltaic Energy Conversion. 2003. Osaka, Japan. 11. Wronski, C.R., et al. Progress in amorphous silicon based solar cell technology. in RIO 02 World Climate & Energ y Event. 2002. Rio de Janeiro, Brazil. 12. Winer, K., Defect formation in a Si:H. Physical Review B, 1990. 41(17): p. 12150. 13. Staebler, D. and C.R. Wronski, Reversible conductivity changes in discharge produced amorphous Si. Applied Physics Letters, 1977. 31( 4): p. 292. 14. Anderson, T.L., Fracture Mechanics Fundamentals and Applications. 2nd ed. 1995: CRC Press, Inc. 39,282. 15. Perez, R. and P. Gumbsch, Directional anisotropy in the cleavage fracture of silicon. Physical Review Letters, 2000. 84(23): p. 5347 5350.
59 16. Hauch, J.A., et al., Dynamic fracture in single crystal silicon. Physical Review Letters, 1999. 82(19): p. 3823 3826. 17. Bailey, N.P. and J.P. Sethna, Macroscopic measure of the cohesive length scale: fracture of notected single crystal silicon. Physical R eview B, 2003. 68: p. 205204. 18. O. Breitenstein, J. Rakotaniana, M. Al Rifai, M. Werner, Prog. Photovolt: Res. Appl., vol. 12, pp. 529 538, 2004. 19. Rueland, E., et al. Optical u crack Detection in Combination with Stability Testing for in line Inspection of Wafers and Cells. in Proceedings of 20th EU PVSEC 2005. Barcelona. 20. Rueland, E., et al. u crack detection and other optical characterisation techniques for in line inspection of wafers and cells. in 19th European Solar Energy Conference & Exhibition. 2004. Paris, France. 21. Trupke, T., et al., Photoluminescence imaging of silicon wafers. Applied Physics Letters, 2006. 89: p. 044107. 22. Trupke, T., et al. Fast Photoluminescence Imaging of Silicon Wafers. in Proceedings of WCPEC 4 2006. Hawaii. 23. Fuyuki, T., et al ., Photographic Surveying of Minority Carrier Diffusion Length in Polycrystalline Silicon Cells by Electroluminescence. Applied Physics Letters, 2005. 84: p. 262108. 24. Lu, J., et al. Investigation of Electrical Activity of Dislocation and Grain Boundary in Polycrystalline Float Zone Silicon. in NCPV and Solar Program Review Meeting. 2003. Denver, Colorado. 25. Rakotoniaina, J.P., et al. Detection of Cracks in Silicon Wafers and Solar Cells by Lock in Ultrasound Thermography. in Proceedings of PV Solar Conferenc e 2004. Paris. 26. Connor, Z.M., et al., Using scanning acoustic microscopy to study subsurface defects and crack propagation in materials. JOM, 1998. 50(11). 27. Quate, C.F., A. Atalar, and H.K. Wickramasinghe, Acoustic microscopy with mechanical scanning a review. Proceedings of the IEEE, 1979. 67(8): p. 1092 1114. 28. Wickramasinghe, H.K., Scanning Acoustic Microscopy: A Review. Journal of Microscopy, 1983. 129(1): p. 63 73.
60 29. SONIX, i., The A,B,C's of Ultrasonics. 30. Hilmersson, C., et al., Crack Detection in si ngle crystalline silicon wafers using impact testing. Ap plied Acoustics, 2006 31. Dallas, W., Resonance Ultrasonic Vibrations for Crack Detection in Silicon Wafers for Solar Cells Master's Thesis in Electrical Engineering. 2006 University of South Florida: Tampa, FL. 32. Belyaev, A., et al., Crack Detection and Analyses using Resonance Ultrasonic Vibrations in Full Size Crystalline Silicon Wafers. Applied Physics Letters, 2006. 88: p. 111907. 33. B elyayev, A., Resonance Acousti c Diagnostics in Silicon Wafers Mas ter's Thesis in Electrical Engineering. 2002, University of South Florida: Tampa, FL. 34. Scofield, J.H., A frequency domain description of a lock in amplifier. American Journal of Physics, 1994. 62(2): p. 129 133. 35. Byelyayev, A., Stress diagnostics and crack detection in full size silicon wafers using resonance ultraonic vibrations, Doctoral Dissertation in Electrical Engineering. 2005, University of South Florida: Tampa, FL. 36. Wasmer, K., et al., Cleavage Fracture of Brittle Semiconductors from the Nanometer to the Centimeter Scale. Advanced Engineering Materials, 2005. 7(5): p. 309 317. 37. Belyaev, A., et al., Resonance ultraso nic vibration diagnostics of ela stic stress in full size silicon wafers. Semiconductor Science and Tec hnology, 2006. 21: p. 254 260. 38. Tarasov, I., et al., Defect monitoring using scanning photoluminescence spectroscopy. Materials Science and Engineering B, 2000. 871: 51 55. 39. det (2007), 18, pp. 852 858. 40. W. Koch, D. Franke, C. Hassler, J. Kalejs, H. J. Moeller, Bulk Crystal Growth and Wavering for PV: Handbok of photovoltaic Science and Engineering, Ed. By A. Lique et al., Wiley and Sons, 2003. 41. J.P. Kalejs, Silicon Processing for Photovoltaics II, ed. by C.P. Khattak, K.V. Ravi (Elsevier, Amsterdam, 1987), p. 185
61 42. MODEL SR850 DSP Lock In Amplifier User Manual. Stanford Research Systems. Revision 1.8 (02/2007)