Current state-of-the-art instruments for automatic visual inspection of masks, reticles, or wafers, are targeted to inspect patterns with line widths as small as 1.5 micron (gm) and defects as small as 0.5 micron. New near term targets for production inspection machines will require inspection of 1.0 micron line width patterns and 0.3 micron defects. The types of inspection machines which will adequately address these new targets will require very robust image analysis algorithms. Some current instrumentation will predictably fail in the sub-0.3 micron to sub-0.5 micron defect size range due to fundamental limitations of traditional image acquisition and processing methods (e.g., pixel to pixel comparison modes). Two studies were carried out to see how defect detection might be improved using a new approach exploiting adaptive signal processing, artificial intelligence methodologies, and CAD directed model matching. The first study investigated the current practice/ limits of defect detection; measuring the reliability with which defects could be detected, measured, and correlated with a CAD data base. We conclude that defects greater or equal to 0.5 micron can be consistently detected at a high (>90Z confidence level); with defects as small as 0.3 - 0.4 micron being detected routinely. In the second study, we explored the realistic potential of defect detectability of the new approach. In this experiment we report very encouraging experimental data extending to line widths of 0.5 micron, and defects below 0.2 micron.

Date Published: 23 July 1985
PDF: 8 pages
Proc. SPIE 0538, Optical Microlithography IV, (23 July 1985); doi: 10.1117/12.947757

Published in SPIE Proceedings Vol. 0538:
Optical Microlithography IV
Harry L. Stover, Editor(s)