REACTIVE ION ETCHING PROFILE AND DEPTH CHARACTERIZATION USING STATISTICAL AND NEURAL-NETWORK ANALYSIS OF LIGHT-SCATTERING DATA

As device design rules continue to shrink for the manufacturing of integrated circuits, unprecedented challenges for process inspection appear. No longer is optical microscopy adequate for determining if process results meet specifications. On the other hand, the alternative-scanning electron microscopy-is time consuming, destructive, and costly. Another approach is to measure scattered light intensity as a function of scattering angle, as opposed to imaging, to obtain distinct signatures for submicron structures. In this work, a set of Si wafers with photolithographically defined lines and spaces are reactively ion etched. By varying process conditions, a range of depths and sidewall profiles is generated and then inspected by detecting visible scattered laser light over 180-degrees. The resultant scattergrams are then analyzed both by using discriminant analysis and by training a neural network to catalog the microstructures according to depth and profile. We find that this approach is a viable alternative to destructive sampling and off-line inspection by scanning electron microscopy: depth and profile are correctly classified with better than 95% accuracy.