AI improves 3D X-ray imaging for package failure analysis

Technology News |
By Nick Flaherty

Zeiss has developed machine learning AI algorithms to improve its 3D X-ray imaging for failure analysis of semiconductor packages. 3D X-ray analysis is increasingly vital for 3D packaging of multiple chips in a single package with vias and interconnnect.

The Advanced Reconstruction Toolbox for the Zeiss Xradia Versa series of non-destructive 3D X-ray microscopes (XRM) and its Xradia Context 3D X-ray micro-computed tomography (microCT) systems is based around two modules: OptiRecon for iterative reconstruction, and DeepRecon, the first commercially available deep learning reconstruction technology for microscopy applications.

3D XRM is used for imaging defects to aid root cause investigation of package failures as it enables visualization of features that are not visible in 2D X-ray projection images. In package failure analysis, both fast results and high success rates are important. Consequently, decreasing imaging time while maintaining image quality is of very high value. Typically, Feldkamp-Davis-Kress (FDK) filtered back-projection algorithms are used to reconstruct the 3D dataset from many 2D projections acquired at different sample rotation angles. When image exposure times or numbers of projections are reduced in an effort to improve throughput, the FDK techniques often lead to degraded image quality.

The OptiRecon and DeepRecon engines enable higher scanning speeds while maintaining or even increasing image quality with improved contrast-to-noise ratios for semiconductor advanced packaging failure and structural analysis. In addition to electronics and semiconductor packaging, the Advanced Reconstruction Toolbox can be used for materials research, life sciences and advanced battery development.

OptiRecon is aimed at analysing a broad range of semiconductor packages and is suitable for both research and development applications, as well as FA. It uses iterative reconstruction, where differences are calculated between real and modeled projections through multiple iterations until convergence. This means fewer projections are needed with and less acquisition time compared to FDK. Scanning speeds as twice as fast with similar or better image quality can be achieved for semiconductor packages.

Next: Failure analysis benefits 

The higher speed enables multiple benefits: regions of interest can be enlarged, analysis jobs can be completed in a single work shift, and the sample’s radiation dose can be reduced. At throughput levels similar to FDK, OptiRecon can provide improved image quality enabling better contrast-to-noise ratios, easier defect visualization and reduced eye fatigue for the analyst. It includes an easy-to-use interface to optimize reconstruction parameters and an advanced high-performance offline workstation for fast and efficient reconstruction.

The DeepRecon module uses custom-trained neural networks to enable higher throughput and higher success rates for FA and structural analysis applications requiring repetitive analysis of samples that are the same or similar to each other.

Zeiss offers customized neural networks for specific sample classes that can be optimized to meet customer needs. Compared to FDK, DeepRecon enables scans up to four times faster with similar or better image quality for designated sample classes, as well as superior image quality with low noise when using the same scan time. Minimal effort is required to apply the desired DeepRecon network model for the workflow. The tool operator simply selects one of the Zeiss-developed network models from a drop-down menu.

“Since its introduction last year, the Xradia 600-series Versa has seen strong momentum in the electronics and semiconductor packaging industry due to its superior resolution, image quality and throughput for package failure analysis,” said Dr Stefan Preuss, head of Zeiss Process Control Solutions.

“As our customers continue to face new challenges in advanced package failure analysis, Zeiss in turn is continually innovating to bring new capabilities and higher levels of performance to our products in order to meet those challenges head on. A case in point is our Advanced Reconstruction Toolbox with OptiRecon and DeepRecon modules, which provides a significant boost to throughput and image quality for these world-class imaging solutions — enabling our customers to accelerate their time to results and achieve higher package yields even more than before.”

The Advanced Reconstruction Toolbox with OptiRecon and DeepRecon modules is now available for upgrade on Xradia Versa and Context microCT systems, as well as future Versa and Context microCT systems.

Related articles 

Other articles on eeNews Europe


Linked Articles
eeNews Europe