Y. Zhao
Please Note
1 records found
1
焊点可靠性通常采用电气失效准则进行评估。然而，在疲劳损伤初期，电阻通常保持稳定，仅在裂纹显著扩展后才会增加。因此，传统的监测方法能够提供关于失效前内部退化的信息有限。虽然三维X射线显微镜（3DXRM）能够对内部裂纹和空隙进行无损观察，但其庞大的图像体积使得一致的人工分析变得困难。
本论文开发了一种基于人工智能的BGA、FCBGA和QFN封装在热循环作用下的可靠性表征工作流程。采用YOLO11n-Seg模型检测重建的3DXRM图像中的裂纹、空隙和焊点结构。分割结果被转化为定量退化指标，包括裂纹体积比和主裂纹面积比。此外，还通过重复性、再现性、灵敏度和不确定性分析评估了整个测量工作流程的稳定性。
对于BGA样品，所提出的裂纹指标的变化趋势与电阻变化以及基于Arivis的现有结果一致。对于QFN样品，采用威布尔分析比较了-40℃至125℃和-40℃至150℃热循环下的退化情况。实验得到的退化尺度比为1.58，95%置信区间约为1.12-2.24，与已建立的热机械加速模型相符。
该工作流程进一步扩展到扫描电镜图像筛选、空洞分析和焊料体积测量。总体而言，所提出的框架减少了人工图像分析的工作量，并能在焊点完全失效之前，提供关于焊点劣化更早、可重复且具有实际意义的信息。
Solder-joint reliability is commonly evaluated using electrical failure criteria. However, electrical resistance often remains stable during early fatigue damage and increases only after substantial crack propagation. Therefore, conventional monitoring provides limited information about internal degradation before failure. Although three-dimensional X-ray microscopy (3DXRM) enables non-destructive observation of internal cracks and voids, its large image volume makes consistent manual analysis difficult.
This thesis develops an AI-assisted reliability-characterization workflow for BGA, FCBGA, and QFN packages subjected to thermal cycling. A YOLO11n-Seg model was used to detect cracks, voids, and solder-joint structures in reconstructed 3DXRM images. The segmentation results were converted into quantitative degradation indicators, including crack volume ratio and main-crack area ratio. The stability of the complete measurement workflow was also evaluated through repeatability, reproducibility, sensitivity, and uncertainty analyses.
For BGA samples, the proposed crack indicators showed trends consistent with electrical resistance changes and existing Arivis-based results. For QFN samples, Weibull analysis was used to compare degradation under −40 ℃ to 125 ℃ and −40 ℃ to 150 ℃ thermal cycling. The experimental degradation-scale ratio was 1.58, with an approximate 95% confidence interval of 1.12–2.24, showing consistency with established thermo-mechanical acceleration models.
The workflow was further extended to SEM image screening, void analysis, and solder-volume measurement. Overall, the proposed framework reduces manual image-analysis effort and provides earlier, repeatable, and physically meaningful information about solder-joint degradation before complete electrical failure.
...
焊点可靠性通常采用电气失效准则进行评估。然而，在疲劳损伤初期，电阻通常保持稳定，仅在裂纹显著扩展后才会增加。因此，传统的监测方法能够提供关于失效前内部退化的信息有限。虽然三维X射线显微镜（3DXRM）能够对内部裂纹和空隙进行无损观察，但其庞大的图像体积使得一致的人工分析变得困难。
本论文开发了一种基于人工智能的BGA、FCBGA和QFN封装在热循环作用下的可靠性表征工作流程。采用YOLO11n-Seg模型检测重建的3DXRM图像中的裂纹、空隙和焊点结构。分割结果被转化为定量退化指标，包括裂纹体积比和主裂纹面积比。此外，还通过重复性、再现性、灵敏度和不确定性分析评估了整个测量工作流程的稳定性。
对于BGA样品，所提出的裂纹指标的变化趋势与电阻变化以及基于Arivis的现有结果一致。对于QFN样品，采用威布尔分析比较了-40℃至125℃和-40℃至150℃热循环下的退化情况。实验得到的退化尺度比为1.58，95%置信区间约为1.12-2.24，与已建立的热机械加速模型相符。
该工作流程进一步扩展到扫描电镜图像筛选、空洞分析和焊料体积测量。总体而言，所提出的框架减少了人工图像分析的工作量，并能在焊点完全失效之前，提供关于焊点劣化更早、可重复且具有实际意义的信息。
Solder-joint reliability is commonly evaluated using electrical failure criteria. However, electrical resistance often remains stable during early fatigue damage and increases only after substantial crack propagation. Therefore, conventional monitoring provides limited information about internal degradation before failure. Although three-dimensional X-ray microscopy (3DXRM) enables non-destructive observation of internal cracks and voids, its large image volume makes consistent manual analysis difficult.
This thesis develops an AI-assisted reliability-characterization workflow for BGA, FCBGA, and QFN packages subjected to thermal cycling. A YOLO11n-Seg model was used to detect cracks, voids, and solder-joint structures in reconstructed 3DXRM images. The segmentation results were converted into quantitative degradation indicators, including crack volume ratio and main-crack area ratio. The stability of the complete measurement workflow was also evaluated through repeatability, reproducibility, sensitivity, and uncertainty analyses.
For BGA samples, the proposed crack indicators showed trends consistent with electrical resistance changes and existing Arivis-based results. For QFN samples, Weibull analysis was used to compare degradation under −40 ℃ to 125 ℃ and −40 ℃ to 150 ℃ thermal cycling. The experimental degradation-scale ratio was 1.58, with an approximate 95% confidence interval of 1.12–2.24, showing consistency with established thermo-mechanical acceleration models.
The workflow was further extended to SEM image screening, void analysis, and solder-volume measurement. Overall, the proposed framework reduces manual image-analysis effort and provides earlier, repeatable, and physically meaningful information about solder-joint degradation before complete electrical failure.