Curvilinear mask metrology: toward standardizing contour-based metrics for EUV mask qualification

Abstract

Curvilinear EUV mask layouts challenge conventional qualification because local geometry changes continuously along the contour. Sparse CD-style sampling can therefore miss or average out the regions that drive the measured response. We present a contour-based evaluation workflow using PMsite outputs and mask SEM data from a curvilinear logic-cell use case acquired on an Advantest E3650 system. The workflow combines signed edge placement error (signed EPE), curvature, and CurviCD to describe complementary aspects of contour placement, local bend severity, and local dimensional behavior. A simple curvilinear test feature is first used to show how the metrics should be interpreted together. The same workflow is then applied to a 28 nm wafer-scale pitch (P28) curvilinear logic standard cell, with repeated SEM acquisitions used as a qualitative contour-extraction consistency check before metric interpretation. Global metric distributions provide a useful overview, but they combine geometrically different contour regions into a single population. By separating the contour into geometry-aware classes, including straight-line, curved-line, line-end, and junction-like regions, the mixed response becomes more interpretable and the contour classes contributing to the observed behavior can be identified more clearly. These results support class-aware contour analysis as a practical step toward more actionable curvilinear mask qualification and focused hotspot review, without relying on a single global summary metric.