Infrared Thermal Imaging Detection and Image Segmentation of Micro-Crack Defects in Semiconductor Silicon Wafer Scanned by Laser

Single-crystal silicon wafers play a key role in photovoltaic technology and microelectronics manufacturing due to their good semiconductor characteristics. In order to meet the demand of high-tech industries, the production technology of silicon wafer is supposed to meet the high-precision standard, and if the micro-cracks produced during grinding are not detected on time, the yield of a useful product will be reduced. In order to achieve more efficient detection of micro-cracks in silicon wafers, a scanning laser thermal nondestructive testing system was developed. Using the pseudo static matrix reconstruction algorithm, the experimental data has been converted into static images to provide easier defect detection and evaluation. The influence of geometric characteristics (length, width and depth) of micro-cracks and laser excitation power on surface temperature signals in the laser scanning tests has been studied. The image enhancement techniques, such as linear gray scale transformation, basic function transformation and histogram equalization have been compared. The effectiveness of using super-pixel segmentation, dual threshold segmentation, iterative threshold segmentation and UNet3+ network for improving micro-crack detection efficiency has been explored. Common segmentation techniques have not proven to be useful in the image enhancement because of the presence of noise. Better results in image segmentation have been achieved by using a UNet3+ network, which ensured identification accuracy of about 90% in the segmentation of micro-crack defects.