Etch Profile Analysis on Taper angle using Convolution Neural Network in Narrow Gap VHF+LF driven CCP

In high aspect ratio etch process, it is challenge that maintaining profile shape and CD with "spec-in" condition. Varying equipment status and operation recipe by time also cause profile distortion e.g., tilt, bowing, tapering, twisting. Especially, in high stacked NAND process, etch profile is considered a critical parameter because small change of taper angle (<1º) causes dramatic degradation of cell characteristics and uniformity. ARDE (Aspect Ratio Dependent Etch) is representative technology for those challenges. As basis research for ARDE, correlation analysis between etch profile data (MI) with equipment status and plasma information are essential. Thus, precise measurement and parameterization of MI data are required. However, various operation condition, ambiguous etch profile and measurement error are obstacles decreasing accuracy of measurement. In this study, CNN based SEM image analysis was introduced to extract MI data with minimizing inaccuracy. Etching was performed using patterned wafer of Si3N4/SiO2 trench in narrow gap VHF+LF driven Fluorocarbon based CCP. Vertical SEM images were obtained along radial position in wafer. Extracted etch profile from SEM image using NN showed high similarity to actual etch profile with above 0.9 of SSIM and MS-SSIM value. Measured taper angles at center/edge region at LF bias power varying condition had precision below 1º angle with low error. These results show MI data from our method was precise enough to recognize variance of etch profile at high aspect ratio condition, and to be used as quantitative factor for etch profile evaluation.