Detection of sub-500-μm cracks in multicrystalline silicon wafer using edge-illuminated dark-field imaging to enable thin solar cell manufacturing

Highlights

• Sub-mm cracks are not reliably detected by most of the commercially-available crack detection tools

• Our prototype captures the images of edge cracks in multicrystalline silicon wafers based on the dark-field imaging technique with vicinal illumination

• Edge-coupling images show a higher contrast than the rear side illumination technique

• Physical mechanisms of NIR crack detection via 1D optical simulation indicate that the robustness of the vicinal illumination is due to the diffuse reflectance

Abstract

High capital expenditures associated with manufacturing thin silicon wafers make it difficult for the industry to scale up and prevent novel technologies from entering the market. Thin wafers fail largely due to breakage during solar cell processing and handling. One of the root causes for breakage is sub-mm edge cracks in the silicon wafer, and these cracks cannot be reliably detected by most commercially-available crack detection tools. In this work, we first investigate the correlation between wafer thickness and critical crack length, and explain the importance of detecting sub-500-μm edge cracks as the wafer thickness is reduced. Secondly, we extend our previous work of micro-crack detection to demonstrate an edge illumination technique using a near-infrared laser to image edge cracks less than 500 μm in length in multicrystalline silicon. Thirdly, we investigate two fundamental edge illumination mechanisms based on dark-field imaging; namely, direct and vicinal illumination. We will then compare these methods to a state-of-the-art rear illumination method. The advantages and disadvantages of both illumination methods are presented and provide an in-depth analysis of light-crack interaction. In particular, we find that the robustness of vicinal illumination is due to diffuse reflectance. The diffuse reflectance has less dependence on crack configurations, while direct illumination has more dependence on the crack configurations because it utilizes specular reflectance. Our results show that our proposed prototype can detect sub-mm edge cracks in multicrystalline silicon wafers, which is an important step in enabling thin silicon wafer manufacturing.

Introduction

Enabling thin silicon wafer manufacturing at high yield in today's manufacturing lines is a promising way to reduce solar cell and module fabrication costs, which then directly translates to a capital expenditure (capex) reduction [[1], [2], [3]]. Therefore, one path to reducing capex is the reduction of silicon usage per wafer by either reducing the wafer thickness or growing silicon wafer via kerfless technologies. However, there are several barriers which hinder the adoption of thin wafers in industry manufacturing lines. Although the benefits of thin silicon wafer are clear, technological innovation is needed to realize thin silicon wafers in practice. Thin silicon adoption has a chicken-and-egg problem: the yields of drop-in manufacturing for thin silicon wafers are low, thus industry has not widely adopted equipment that reliably produces, handles, and monitors thin silicon wafers. Compounding this are historically low polysilicon feedstock prices.

Handling and processing thin Si wafers (e.g. 80–120 μm) is difficult [4] and manufacturing yield in today's manufacturing lines is unacceptably low. Fundamentally, one major root cause of breakage of silicon wafers is due to through-cracks present in the wafer [5]. These cracks are in the sub-mm range and can propagate from the wafer edge during handling, and processing steps [6] leading to wafer breakage somewhere downstream in the solar cell production process [7]. Manufacturing steps that could induce cracks include sawing and cutting silicon ingots, contact metallization, soldering of contacts, and wafer handling [8]. Due to the brittle nature of silicon, these micro-cracks are increasingly critical in thinner wafers. These cracks reduce the strength of the wafer, causing the wafer to break with substantially less applied force when compared to thicker wafers [5]. In addition, cracks located at the edge of the wafer require a lower force to propagate than cracks that are located in the interior of a silicon wafer [9].

To date, there are various approaches for detecting cracks; ideally, these methods are used before cracked cells are integrated into the cell string. The most common methods include photoluminescence (PL) [10,11], electroluminescence (EL) [12,13], optical transmission [14,15], edge illumination [16], transflection [[17], [18], [19], [20]], infrared lock-in thermography (LIT) [21], resonance ultrasonic vibration (RUV) [22], scanning acoustic microscopy (SAM) [23,24], radiant heat thermography (RHT) [25], dye inspection, eddy current inspection [26] or acoustic inspection [27]. All of the listed methods have advantages and disadvantages which will be described in the following paragraphs.

The major disadvantages of using PL and EL are large interferences with other defects such as scratches or grains. In addition, EL can only be used for finished solar cells despite the high screening throughput that can be achieved. Another disadvantage of luminescent methods is that recombination defects (dislocation) reduce the lifetime of the wafer, leading to false interpretations. Dye inspection and/or acoustic methods are destructive and might induce cracks during the measurement procedure. Since the wafer has to be covered with a liquid, long acquisition time is required, making the method applicable only as a standalone system. LIT and RHT can detect cracks which are less than <1 mm in length; however, they require a long acquisition time. Optical transmission has the advantage of a high wafer screening throughput, but cannot be applied to finished solar cells due to metallization hindering the usage of rear illumination.

As summarized, significant effort has been invested into developing new and improved crack detection techniques for increasing wafer screening throughput while maintaining a high precision without false positive results. In addition adding to accuracy, machine learning algorithms have been developed which are able to differentiate between cracks and recombination active defects like grain boundaries and dislocations [28,29]. However, the only tools which would meet all these criteria (and account for the high-throughput requirement of 1 s/wafer) for a fully automated in-line production line are PL, EL, optical transmission and edge illumination. Whereas the first two mentioned methods are based on luminescence imaging, the latter two rely on near-infrared (NIR) imaging. Luminescence imaging utilizes the process of non-radiative recombination by measuring the minority carrier lifetime of the wafer, where a decrease in the image contrast then indicates a crack. However, the detection of edge cracks is challenging because the edge of the wafer itself reduces the carrier lifetime, especially at early stages in the solar cell fabrication process. The development of detection algorithms has aided the technique in reaching its potential; however, it is still very challenging to detect cracks at a high detection rate. In a previous study, it was shown that only 65% of the cracks could be detected, which was attributed to the crack width (less than 1 μm) being much smaller than a single pixel resolution [29]. In contrast, NIR imaging is based on light-crack interaction where the IR light scattering behavior is detected, either in bright- or dark-field. Since this method can also be applied to image edge cracks, we will focus on NIR imaging in this paper.

We have previously shown that a dark-field imaging approach, in combination with a broad-response InGaAs camera and a microscope objective lens, can be used to image edge cracks approx. 1 mm in length in multicrystalline silicon (mc-Si) wafers [30]. Here, an incandescent light was employed as a light source. However, due to the low intensity of the light source, only a relatively low signal-to-noise ratio (SNR) could be achieved which prevents the imaging of edge cracks shorter than 1 mm in length. Ortner et al. [16] have shown that a high SNR can be achieved if a focused laser is used in an edge-coupling geometry with direct illumination. Cracks with a minimum length of 3 mm could be measured with high accuracy using their crack detection tool ‘edge-light’ with the high throughput requirement of 1 s/wafer. We further extended the work of Ortner et al. [16] and propose the edge-coupling geometry with vicinal illumination as an alternative to reliably detect edge cracks, even for those parallel to the laser propagation direction [31]. Our proposed prototype combines the advantages of industrially-available NIR dark-field imaging methods from Refs. [16,20]. To achieve the high resolution required coupled to a high SNR, we combine the focused laser edge-coupling unit with a broad-response InGaAs camera and a 5× microscope objective lens [30]. Our approach showed the potential for scalability and the high throughput requirement of 1 s/wafer for an industry in-line crack detection tool [31].

In this study, we further demonstrate the capability of detecting small edge cracks less than 500 μm in length in mc-Si wafers, and provide an understanding of the fundamental principles of NIR imaging. First, to show the importance of detecting sub-500-μm cracks, we investigate the correlation between the critical crack length and the silicon wafer thickness via material point method (MPM) simulations. Secondly, we show how an improved edge-coupling illumination can be used to advance state-of-the-art NIR crack detection tools. We compare and evaluate bright- and dark-field edge illumination approaches by showing the capabilities and weaknesses of both methods. Thirdly, to further address the questions of 1) are there differences in operation modes in the dark-field edge-light geometry, and 2) how small of a crack can edge-light detect, we perform crack scattering imaging in two configurations, i.e. direct and vicinal illumination. Our results reveal that the robustness vicinal illumination compared to direct illumination is based on the usage of diffuse light. In addition, we perform optical transfer-matrix simulations to study the light-crack interaction in more depth to gain a physical understanding.