Sub-resolution assist feature printability prediction using machine learning

Abstract

Sub-resolution assist feature (SRAF) is a mask pattern nearby main feature to promote pattern fidelity of main feature but should not be printed on wafer. SRAFs are sometimes unintentionally printed and the printed SRAFs are critical defects in semiconductor manufacturing. To prevent the accident, the SRAF printabiltiy check is essential before mask tapeout. A conventional SRAF printability check method has large false alarm error because the method does not consider surrounding mask patterns, which effects on SRAF printability. Another conventional SRAF printability check is accurate but time-consuming so it is used only in small layout. We propose new SRAF printability check using machine learning and achieve 12%false alarm error and 69% runtime reduction.