Siemens' white paper on curvilinear design in semiconductor manufacturing introduces a machine learning approach to classify verification error markers. Traditional geometry-based methods struggle with curvilinear shapes, common in silicon photonics and metasurface optics, due to grid-snapping inconsistencies and high edge complexity. Siemens’ OPCVerify ML Classify addresses these challenges by analyzing error features in the vector space rather than relying solely on geometrical outlines. In testing, the model reduced 837,072 raw errors into just 51 classes, streamlining review without compromising accuracy. Further subclassification with user-controlled parameters increased flexibility for design-specific classification.

Join for free to read