Study on Enhancing Flip-Chip-Chip Scale Package (FCCSP) Reliability Testing using Deep Learning (DL)Assisted Scanning Acoustic Microscopy

This study integrates Scanning Acoustic Microscopy (SAM) A-Scan data with a Deep Learning (DL) network in an effort to improve the reliability assessment of Flip-Chi-Chip Scale Packages (FCCSPs). Due to its small size, high integration density, and exceptional electrical performance, FCCSP technology has attracted a lot of attention. However, it is extremely difficult for manufacturers to guarantee the dependability of FCCSPs under diverse operating circumstances. By combining DL algorithms with SAM, defects can be identified efficiently and quickly, reducing the disadvantages of manual inspection procedures. The study used both molded package and P-Lapped sample as test devices to train various sequential networks for classification. The study also emphasizes training DL networks to distinguish across various layers of the sample in addition to fault detection. Due to the complex signal data produced while scanning from the molded side, the networks performed better on the p-lapped sample to verify defects. This research collected echo and through scan mode data (SAM) from molded sample and p-lapped sample. All these data were trained with ID Convolutional Neural Network (CNN), Long Short Term Memory (LSTM) and Transformer Networks. For molded sample, LSTM networks predicted defects with an accuracy of more than 98% on validation dataset. In P-lapped sample, ID CNN network predicted defects with an accuracy of 99%.