Using SOM neural network for X-ray inspection of missing-bump defects in three-dimensional integration

• A nondestructive method for missing-bump inspection in 3D integration is presented.
• Vectors with 4 features based on gray values representing microbumps are extracted.
• The SOM network with X-ray images realizes accurate classification and recognition.

Three-dimensional integration has been a key technology in scientific research and industrial production of integrated circuits, where microbumps bridge multiple layers of chips. Microbump defect inspection, especially for missing-bump, is of major significance. We introduce self-organizing map network combined with X-ray imaging, and demonstrate a non-destructive method for rapid and effective inspection of missing microbump defects. 2D X-ray images of samples with microbumps are segmented, and vectors consisting of four features as representatives of microbumps are extracted. A self-organizing map network is constructed, and vectors of microbumps selected randomly from four samples are inputted into the network. Clusters of the defective bumps and the normal bumps are distinguished obviously. Then the other microbumps from the same samples are used for testing. The trained network can recognize the defective and normal microbumps through the clustering areas with no error. Microbumps from a different sample are inputted to the network for further verification, and high recognition accuracy is achieved. These prove the feasibility of using self-organizing map network for X-ray inspection of missing-bump defects.