.png){kind=icon}
.png){kind=icon}
Return
Adaptive transformer-augmented network for multi-type defect detection in ball grid array assemblies
.png){kind=icon}
PRE
AI
DOI：10.1007/s10845-026-02856-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Ball Grid Array (BGA) packages have become the ideal choice for many high-performance electronic products due to their high density, good thermal management, and stable mechanical performance advantages. Their corresponding solder quality detection technology has also been rapidly developed. However, the invisibility of internal solder balls, the diversity of BGA chip models, and the complexity of soldering defects have significantly increased the technical challenges of detecting defects in on-board BGA chips. Traditional manual inspection methods can no longer meet the efficiency and accuracy demands of modern industrial production. In response, this study thoroughly explores BGA defect detection technologies, focusing on the challenges arising from the diversity of BGA chip models and the complexity of soldering defect types. First, we construct a comprehensive X-ray dataset containing various BGA defect types, which comprises 29 distinct chip models. Secondly, an improved DeepLabv3+ model is proposed. By integrating a Transformer encoder into the original architecture, this approach boosts the segmentation precision for BGA solder balls and void defects by 4.49% compared to the baseline. Finally, to overcome the limitations of previous methods that focus on voids detection while ignoring other critical defect types, an integrated scheme was developed to detect three major BGA solder joint defects: voids, offset, and ball size variations. This expands the range of defects that can be detected and better meets the needs of practical detection.
Keywords:
Ball Grid Array
Defect detection
Transformer
Semantic segmentation
X-Ray
Journal
IF:
7.4
Papers:
3.5K
Citations:
1.1W