BGA and QFN joints are inherently hidden by the package, and flat optics can't see the ball bottom. This plant combined height data with AI defect reading via DaoAI 3D AI-AOI, turning what was X-ray spot-checking into inline full inspection.
Communications and high-density modules use bottom-terminal packages like BGA and QFN heavily, with joints sitting under the component. Conventional 2D AOI sees only the part's outer edge and cannot judge voiding, bridging, or non-wetting in the balls themselves. The plant relied on X-ray sampling as a backstop, but sampling never covered all product and couldn't keep pace with the placement line; a batch-level solder anomaly often surfaced only at final functional test or even at the customer.
The plant brought in DaoAI 3D AI-AOI inspection hardware, building a three-dimensional height image of the joint area and letting an AI model read ball morphology. The 3D information gives a height-domain basis for defects—poor wetting, collapse, bridging—that are hard to separate reliably by grayscale alone, while APDT few-sample learning lets a new package type be modeled quickly when defect samples are scarce. The equipment runs inline at placement takt, moving inspection from sampling to full coverage.
Challenges and countermeasures
- Joints hidden by the package → 3D height imaging recovers ball-bottom morphology
- Voids/bridges/non-wetting look alike in grayscale → 3D features + AI reading separate them reliably
- Scarce defect samples on new packages → APDT few-sample learning models quickly
- X-ray sampling leaves gaps → inline 3D full inspection replaces sample-based backstop
Invisible balls are no longer a sampling gamble; every one is checked.
After go-live, combined detection of hidden-joint defects—voids, bridges, non-wetting—reached over 99%, batch-level solder anomalies were caught right after placement, solder-related failures found at functional test fell sharply, and X-ray shifted from primary backstop to confirming only difficult cases.