edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 26 NO. 3 4 EDFAAO (2024) 3:4-11 1537-0755/$19.00 ©ASM International® NONDESTRUCTIVE DEFECT DETECTION IN 3D X-RAY MICROSCOPY DATA OF BALL GRID ARRAY SOLDER FOR VOID DETECTION IN SOLDER JOINTS USING DEEP LEARNING Kishansinh Rathod1, Sankeerth Desapogu1, Andreas Jansche1, Timo Bernthaler1, Daniel Braun2, Stephan Diez2, and Gerhard Schneider1 1Materials Research Institute – Aalen University, Aalen, Germany 2BMW AG, Munich, Germany kishansinh.rathod@hs-aalen.de VOID DETECTION IN SOLDER JOINTS In the world of microelectronics assembly, soldering is a fundamental and crucial process for connecting semiconductor devices with PCBs.[1,2] The solder balls present in ball grid array (BGA) packages play a crucial role in creating this connection by forming solder joints with the corresponding pads on the PCB. The quality of a solder joint directly impacts the functionality and reliability of electronic devices.[3] One of the significant challenges faced in electronic assembly is the presence of voids within solder balls.[1,4] These voids can be formed because of many factors such as poor flux coverage, improper substrate outgassing, and reflow conditions.[5,6] The presence of these voids can have a negative impact on the performance of microelectronic devices. For example, voids can lead to increased electrical resistance, thermal impedance, and a negative impact on mechanical stability and stress concentra- tions.[4–7] Eventually, these effects can lead to breakdown or malfunctioning of the device. However, it is important to note that all voids cannot be classified as defects. For example, the current IPC A-610 standard specifies that voiding that exceeds 30% (25% as a previously applied IPC A-610 standard) of the x-ray image area is considered a defect.[8–10] Moreover, the negative impact of voids is based on their location within solder joints.[9] Therefore, it becomes very necessary to detect the voids accurately in order to quantify them correctly. X-ray inspection has emerged as a widely used tech- nique for void detection in electronic assemblies. Traditional 2D x-ray methods are commonly used due to their speed and cost efficiency.[11] A variety of traditional and machine learning-based methods and approaches have been proposed and used for the BGA solder ball’s void detection in 2D x-ray images. Peng et al.[12] proposed a novel approach using a blob filter for the automatic detection of voids in BGA x-ray images. The proposed blob filter uses the local image gradient magnitude and different-sized average box filters for multi-scale analysis. It compares the brightness differences between a blob region and its neighbors for blob enhancement. The method also uses a contrast enhancement technique and a morphological method for region growing and connecting open regions. The voids are detected based on the roundness of the region. Said et al.[13] proposed an algorithm that uses histogram and morphological-based segmentation methods for solder balls in 2D x-ray images. A voting procedure is used to segment the occluded balls. The algorithm also used an independent edge detection method to identify voids in solder balls. Ahuja et al.[14] proposed void detection based on a multidirectional scanning algorithm where the Laplacian of Gaussian is used to detect the edges. Connected component labeling is applied to separate each void region. Krammer et al.[15] investigated different traditional image processing-based methods such as Canny edge detection, global thresholding, adaptive thresholding, and blob detection for void detection. Xia et al.[16] used a 2D x-ray-based void detection algorithm that uses a dynamic enhancement algorithm to pre-process the background interference image and perform the threshold-based segmentation.[16] Schiele et al.[17] proposed an approach to segment voids in microelectronics using a convolutional neural network (CNN). Various CNN-based architectures such as U-Net and Mask-RCNN are implemented and compared.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==