Authored By:
Esra Daniel Stoll
Basler AG
Germany
Summary
The design recommendations for the latest chips often contain overprints to handle the small land patterns with common type 4 pastes and 100 μm (4 mil) stencils. Such workarounds make careful data preparation necessary. A standardized data processing for the stencil apertures would be preferred.
To match the best practice recommendations for stencil printing, type 5 solder paste, and thinner stencils foils are necessary. So - what does make more sense: Fiddling or Switching? To get out of the conflict, a Design of Experiments (DoE) was proceeded to find answers. The experiment should find the optimum printer parameters, compare type 4 vs. 4.5 vs. 5 solder paste and 4 different stencils.
Conclusions
We asked many experts about the different stencil technologies used by our production. There were many different opinions, but statistically robust evidence was absent. Thanks to the DoE we have now answers and know that the extra effort of introducing a type 5 will be worth it. And that the fiddling can end, too. The typical AR=0.66 recommendations can already be critical for a capable printing process. Using an AR~0.7, the printing process is by far better under control. The main DoE created 890,000 data points and the confirmation run a another 475,000 data points.
With this massive amount of data, more research and prediction can be done. By example: The result of the printing quality depends not only on the area and the foil thickness, but also on the shape and – for a rectangular – on the aspect ratio of the edges. Hence, linear regression is not the best prediction tool. Neural networks are better in finding and predicting “spots”. A simple example is shown in Figure 19. In this neural neutral with five hidden nodes, you can – carefully using domain knowledge – predict the percentage of misprinted pads.
Figure
Initially Published in the SMTA Proceedings
|
