The roughness of a sample is an indicator of the roughness or smoothness of the sample surface. Roughness is specified as Rq, in this paper.
The result of measuring the right sidewall of the pattern is shown as an automatically 3D rendered image in Fig. 5 (a). The 3D rendered image clearly shows the difference in the roughness of each region through the topographic image, dividing the structure into top, sidewall (right sidewall), and bottom. In Fig. 5 (b), line profile measurement was performed after defining each line of region. In the profile of measured line, the Rq of the sidewall line was 1.376 nm, which was rougher than the top line and bottom line. In addition, the Rq of the top line is 0.522 nm and the Rq of the bottom line is 0.218 nm, indicating that the surface is gradually smoothing.
Fig. 5 (c), a 2D image of Fig. 5 (a), was achieved by a post-processing technique that first order flattens the measured 3D image. Then, the 2D image is divided into three regions, top, sidewall, and bottom. Since the image of the sidewall is relatively rough compared to other parts, sidewall can also be distinguished by the difference in image contrast.
As in Fig. 6, to demonstrate the reproducibility of SWR measurements using 3D-AFM, 15 repetitive measurements were performed per identical location on point 1, point 2, and point 3, randomly selected from three locations on the wafer dies.
As a result, in the point 1, mean roughness (Rq) was measured as 2.26 nm, with standard deviation of 0.051 nm, and RMS value with 2.26%. For the point 2, the mean roughness (Rq) was 1.479 nm, with standard deviation of 0.023 nm, and RMS value with 1.56%. For the point 3, the mean roughness (Rq) was measured as 1.124 nm, with standard deviation of 0.016 nm, and RMS value with 1.42%.
From the measurement results, it was found that the RMS value was close to 2% at all measurement positions in the wafer level, and the reproducibility of the measurement was proved based on these results. Accordingly, the Z scanner tilting technique for 3D measurement has high reproducibility, and it has been demonstrated that the measurement method and the structural stability of the equipment are powerful .
Above all, even if in the same pattern in single wafer, not only the topography but also the roughness is different depending on the die. The SWR analysis results of measurement points 1, 2 and 3 are distributed as 2.37 nm, 1.67 nm, and 1.23 nm.
After confirming the reproducibility, total of 13 points was measured to compare the overall measurement results of the 300 mm wafer as shown in Fig. 7 (a). Including the center and edge of the wafer, 13 dies were selected, and we measured the same pattern 1 time per each die. As a result, the SWR, Rq, was distributed in 1.124 ~ 2.260 nm. The measured Rq for the entire wafer is distributed as Fig. 7 (b).
Fig. 8 shows the actual analysis result of the developed SWR automatic analysis program. When a raw image is selected, it automatically detects the sidewall with a process developed by the user, and it calculates a roughness value. As shown in Fig. 8 (a), either left or right sidewall of the pattern analysis is possible. Also, even in a repeated trench structure, multi-sides can be detected as shown in Fig. 8 (b).
Furthermore, as in Fig. 8, SWR automatic analysis program defines the maximum value (top of structure) and minimum value (bottom of structure) of the structure height which is used to define the sidewall to be analyzed according to the amount of data desired. After the sidewall is defined, the roughness of the region can be calculated.
The devised algorithm also presents a methodology for analyzing SWR by introducing the basic idea of detecting an inflection point to exclude variations in the measurement sample as shown in Fig. 9. The measurement result shown in Fig. 9 (a) is suitable for use in the existing method of detecting sidewall by obtaining the max and min values of the sample height. In contrast, if the top and bottom of the structure are not flat as shown in Fig. 9 (b), the existing method cannot accurately detect the sidewall. Therefore, the improved algorithm enables accurate sidewall detection even in difficult-to-analyze samples or environments, so an improved algorithm that can obtain the same analysis results in both Fig. 9 (a) and (b).