Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Experimental SEM images of an ArF-photoresist pattern. The images are 2000 nm long and 192 nm wide, i.e., 400 by 218 pixels with a size ΔyPIX or ΔxPIX along y or x-axis as 5.00 or nm, respectively. (a) Captured using a CD-SEM, model S-9380, from HHT with an acceleration voltage of 300 V and a probe current of 4.8 pA. 26–29,32,33,36,37 A total of 16 frames were averaged to reduce image noise. (b) or (c) Obtained by crosswise or longitudinally averaging 13 or 16 pixels of (a), respectively. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Pattern-edge images simulated using a MC method. (a)–(c) correspond to (a)–(c) in Fig. 1, respectively. The simulation was carried out using the method described in Sec. 2.2 assuming a variance and correlation length of LWR to be 6.25 nm2 and 35 nm, respectively, 26–29,32,33,36,37 here and in the following. A half scan-signal width λ, for which a scan-signal intensity drops to 1/e of its peak, and a ratio θ of an image-noise SD to the peak were 7.4 nm and 0.45, respectively. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Crosswise distributions of scan signal. The solid line in (a) or (b) represents a scan-signal distribution in Figs. 1(a) or 2(a), respectively. The dashed line in (b) represents an image-noise-free distribution assumed in the simulation. Pattern edges were detected at the maxima of scan-signal profiles, as indicated by arrows. The edge assumed in the simulation is shown using Xα. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. PSDs of photoresist LWR obtained using crosswise filtering. (a) Experimental results obtained putting a sampling interval Δy and length L of line segments as 5 and 2000 nm, respectively, here and in the following. The PSDs were obtained by averaging the results of 452 samples. The number NPIX,X of crosswise-averaged pixels ranged from 7 to 23. For reference, a result obtained without the noise filtering is also shown (NPIX,X=1). (b) The average of theoretical PSDs for 2000 trials carried out using the method of Sec Each line in (b) reproduces a corresponding result in (a). For reference, a simulated result for no image noise is also shown. This noise-free PSD is the same for all NPIX,X values except for statistical noise. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Difference in theoretical PSDs obtained crosswise averaging scan-signal distributions with and without image noise. The differential PSDs ΔIτ’s were obtained using the PSDs in Fig. 4(b). Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Image-noise-induced variance as a function of number of crosswise-averaged pixels. Closed plots in (a) and (b) represent the same experimental results. The lines in (a) represent theoretical results calculated assuming θ to be 0.22, 0.45, and 0.90 with λ fixed at 7.4 nm, whereas the lines in (b) were simulated changing λ from 3.7 to 14.8 nm with θ fixed at The number of MC trials was 500 in both (a) and (b). Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Experimental statistics of photoresist LWR estimated using crosswise filtering. The circles, triangles, and squares represent a mean of line widths, correlation length, and variance of LWR, respectively. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. PSDs of photoresist LWR obtained using longitudinal filtering; (a) and (b) show experimental and theoretical results, respectively. These PSDs were obtained in the same way as those in Fig. 4, except for the noise filtering. The number NPIX,L of longitudinally averaged pixels ranged from 2 to 16. The result for NPIX,L=1 in (a) or (b) was obtained without the noise filtering and is the same as that for NPIX,X=1 in Fig. 4(a) or 4(b), respectively. The number of samples or trials in (a) or (b) was equal to that in Fig. 4(a) or 4(b), respectively. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Image-noise-free PSDs obtained using longitudinal filtering. The number of local minima is equal to NPIX,LΔyPIX/Δy−1. 36 Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Difference between theoretical PSDs obtained longitudinally averaging scan-signal distributions with and without image noise. ΔIτ’s were obtained using the PSDs in Figs. 8(b) and 9. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. PSDs obtained using longitudinally or non-filtered images; (a) or (b) shows the results with or without the filtering, respectively, for θ ranging from 0 to NPIX,L was 2 in (a). Black or gray lines in (b) represent the PSDs of uniformly or non-smoothed LWR, respectively. The smoothing-window size in the former was equal to the filtering-window size in (a). Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Nominal and image-noise-induced variances and their sum as a function of number of longitudinally averaged pixels. The closed circles represent an experimental result on a nominal total variance, from which the variance of a longer-range-correlation component was subtracted. The solid line represents a theoretical sum of nominal and image-noise-induced variances, which are represented by the dashed and dotted lines, respectively. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Normalized ratio of maximal scan-signal intensity to standard deviation of image noise as a function of normalized crosswise- filtering-window size. The normalized ratio f(W^) was calculated using Eq. (10). Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Comparison of experimental image-noise-induced variances between previous studies and this current work. Closed and open plots represent experimental var(φ)’s from previous 26 and current studies, which used Terminal PC and new image-analysis software, respectively. Solid and dashed lines represent theoretical fits to these experimental results. The solid line was simulated assuming scan-signal profiles as shown in Fig. 15. Both the results of this study (open plots and dashed line) are the same as those for θ=0.45 in Fig. 6(a) and for λ=7.4 nm in Fig. 6(b). Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Scan-signal profile assumed in the simulation best fitted for a previous image-noise-induced variance. The integral signal intensity ASEM was the same as that in Fig. 3(b). The thin solid line represents the scan-signal profile, whereas the dashed line represents the image-noise-free distribution assumed in the simulation in a manner similar to how the dashed line in Fig. 3(b) did. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Theoretical maximum and half width of averaged scan-signal as a function of number of averaged image pixels. The solid or dashed lines represent maxima or half widths of averaged scan signal, respectively, and thick (black) or thin (gray) lines represent the results for the longitudinal or crosswise averaging, respectively. λ was assumed to be the same as that in Figs. 4(b) and 8(b), but θ was set to zero. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM

Date of download: 6/29/2016 Copyright © 2016 SPIE. All rights reserved. Theoretical variance of image-noise-induced LWR as a function of normalized image-noise intensity. Closed circles represent the results simulated assuming λ to be the same as that in Figs. 4(b) and 8(b). Image pixels were not averaged. These plots are approximated by a linear function with a zero offset, as represented by a solid line, in the range of θ from to 0.3, where var(φ) in Fig. 12 was inversely proportional to the square root of NPIX,L. Figure Legend: From: Noise filtering of scanning-electron-microscope images for accurate analysis of line- edge and line-width roughness J. Micro/Nanolith. MEMS MOEMS. 2012;11(4): doi: /1.JMM