5 Werkstofftechnik
Filtern
Dokumenttyp
- Posterpräsentation (2) (entfernen)
Sprache
- Englisch (2)
Referierte Publikation
- nein (2)
Schlagworte
- Machine Learning (2) (entfernen)
Organisationseinheit der BAM
The 5G technology promises real time data transmission for industrial processes, autonomous driving, virtual and augmented reality, E-health applications and many more. The Low Temperature Co-fired Ceramics (LTCC) technology is well suited for the manufacturing of microelectronic components for such applications. Still, improvement of the technology such as further miniaturization is required. This study focuses on the characterization of inner metallization of LTCC multilayer modules, especially on the vertical interconnect access (VIA). Critical considerations for this characterization are delamination, pore clustering in and at the edge of the VIA, deformation, and stacking offset. A LTCC multilayer consisting of a glassy crystalline matrix with silver based VIAs was investigated by synchrotron x-ray tomography (CT). The aim of this study is to propose a multitude of structural characteristic values to maximize the information gained from the available dataset. Data analysis has been done with the open source software ImageJ as well as several additional plugins. The high-resolution CT data was evaluated through 2D slices for accessibility reasons. The segmentation of all 2000 slices to assess the different regions e.g. pores, silver and glass ceramic was done by a supervised machine learning algorithm. A quantitative evaluation of shape, deformation, and porosity of the VIA with respect to its dimensions is presented and the suitability of the characterization approach is assessed.
The introduction of the 5G technology and automotive radar applications moving into higher frequency ranges trigger further miniaturization of LTCC technology (low temperature co-fired ceramics). To assess dimensional tolerances of inner metal structures of an industrially produced LTCC multilayer, computer tomography (CT) scans were evaluated by machine learning segmentation.
The tested multilayer consists of several layers of a glass ceramic substrate with low resistance silver-based vertical interconnect access (VIA). The VIAs are punched into the LTCC green tape and then filled with silver-based pastes before stacking and sintering. These geometries must abide by strict tolerance requirements to ensure the high frequency properties.
This poster presents a method to extract shape and size specific data from these VIAs. For this purpose, 4 measurements, each containing 3 to 4 samples, were segmented using the trainable WEKA segmentation, a non-commercial machine learning tool. The dimensional stability of the VIA can be evaluated regarding the edge-displacement as well as the cross-sectional area. Deviation from the ideal tubular shape is best measured by aspect ratio of each individual layer. The herein described method allows for a fast and semi-automatic analysis of considerable amount of structural data. This data can then be quantified by shape descriptors to illustrate 3-dimensional information in a concise manner. Inter alia, a 45 % periodical change of cross-sectional area is demonstrated.