Abstract

In semiconductor manufacturing, submicrometer linewidth measurements are essential because they can be used to characterize VLSI processing steps; for example, in photolithography, the measured width is directly related to critical process parameters such as focus distances and exposure periods. Optical metrology is preferred because it is non-destructive and has a high throughput. Optical confocal microscopes¹ and correlation microscopes² are especially useful for metrology because they have better transverse and range definition than standard microscopes. However, as the dimensions of circuits shrink to below 0.5 μm (~ one wavelength of light), it is difficult to interpret the optical images obtained by these improved microscopes. Traditional measurement techniques based on thresholding the intensity of an optical image are not accurate enough for submicrometer measurements.³ We describe here a pattern recognition technique (based on data clustering) for recognizing line features which are otherwise difficult to measure.

© 1992 Optical Society of America