3.1.

Comparison of Defect Detectability Between Electron-Beam and Deep Ultraviolet Inspection Systems

Figure 2 shows an EUV aerial image, EB (SEM) image, and EB (PEM) image of the hp 64-nm L/S patterned EUV mask. The contrasts of EB images are reversed compared with the aerial image. The contrast of the DUV image is similar to that of the aerial image because DUV is also reflected by the EUV reflective multilayer (ML) and is also absorbed by the EUV absorber layer as shown in Fig. 3.

Fig. 2

(a) EUV (aerial), (b) SEM, and (c) PEM images of hp 64-nm L/S patterned EUV mask.

Fig. 3

Schematic explanation of image contrast of EUV mask obtained by (a) DUV and (b) EB inspection tools.

As a result, bright and dark areas in the DUV and EUV images directly correspond to the ML and absorber parts.^1,27 On the other hand, in the EB inspection system, the secondary electron emission coefficient (SEEC) determines the material contrast on the SE image.^9,19,21,23 The SEEC of the absorber is larger than that of the ML. Therefore, the bright and dark areas in the EB image correspond to the absorber and ML parts as shown in Fig. 2.^{9,19,21,23,27} The defect image contrast is also influenced by the defect size and shape. In the cases of pinhole and intrusion defects, as the defects become smaller, the signals from the ML become lower in both cases of DUV and EB images. Hence, the contrast of those kinds of small defect signals becomes low in the case of DUV image, but in the case of EB image, it becomes high as shown in Fig. 4.^9,21,23

Fig. 4

Experimentally obtained SEM image of a 30-nm-sized pinhole defect on hp 88-nm L/S pattern and its signal intensity profile (superimposed).

The intensity profile of a pinhole defect is much weaker than that of the ML level. This is because the SEs from the bottom of the pinhole are blocked by the side walls of the defect.^9,21,23 We had already reported that the PEM detectability of intrusion defects was better than that of extrusion defects as the defect size became small.^9,21,23 On the other hand, the printability does not coincide with the detectability in the case of EB inspection due to the reversed contrast.

Figure 5 shows simulated printability of the intrusion and extrusion defects on the hp 64-nm L/S pattern. The impact of an extrusion defect on printability is larger than that of an intrusion one.¹⁴ Therefore, the inspection condition should focus on the detection of an extrusion defect due to its better printability. We had also reported that a landing energy with 250 eV created the most sensitive condition to detect the 16-nm-sized intrusion defect, whereas in the case of same size of an extrusion defect, 1000 eV created the most sensitive condition.⁹ Because the detectability of an extrusion defect degrades under the condition of 250 eV, the landing energy should be set to 1000 eV to detect more printable defects by taking into account the printability.

Fig. 5

Simulated printability of intrusion and extrusion defects on hp 64-nm L/S pattern.

3.2.

Impact of Line Edge Roughness on Defect Detectability of Projection Electron Microscope Inspection System

Figure 6 shows the defect detection mechanism of the PEM. The defect signal is determined by the summation of gray-level difference between defect and base pattern, similar to a DUV inspection system.¹⁴ Defect signal intensity is defined as the ratio between the intensity peak and the standard deviation of the background noise in the difference image. Therefore, as the LER becomes large, the background noise of the base pattern becomes large in the difference image as shown in Fig. 7(a).

Fig. 6

Schematic explanation of defect detection mechanism of PEM: (a) defect signal in base pattern and (b) defect signal intensity in difference image.

Fig. 7

Dependence of LER on (a) background noise of difference image and (b) defect detectability.

As a result, the defect signal intensity degrades with LER as shown in Fig. 7(b). However, this result also indicates that the detectability of extrusion and intrusion defects keeps its 10 sigma below 2 nm LER.

3.3.

Defect Detectability of Scanning Electron Microscope Inspection System

In order to investigate the nature of the pattern inspection with EB, the detectability of the SEM inspection system was evaluated. If the electron dose per unit area and the landing energy are the same, the gray level of the defect signal intensity in the difference image is the same between PEM and SEM when the same algorithm as PEM is used. The strong edge effect on an SEM image can enhance the defect signal intensity. However, the background noise derived from the LER is also enhanced. When the defect size is significantly larger than LER, a binarization process can take advantage of a higher resolution of the SEM. In this process, if the signal from a pixel is larger than a threshold, the absolute value of the signal intensity from the pixel becomes unity. On the other hand, if the signal is smaller than the threshold, the signal intensity becomes zero.

Figure 8 shows the simulated SEM images with resolutions of 1.5 nm, and the difference image using the image processing algorithm with the binarization process. As shown in Figs. 8(a2) and 8(b2), the binarized defect signals and the background noise in a difference image become 1 or $-1$. Pixel-sized signals, such as the background noise derived from the LER, are not printable defects. Moreover, because the signal area of the background noise is much smaller than that of the defect signal, the noise can be removed by a noise canceling process as shown in Figs. 8(a3) and 8(b3). This result shows that the high resolution leads to a robust detectability. However, in order to improve the throughput of the SEM inspection system, a lower resolution with a shorter dwell time is required. The resolution and dwell time correspond to the pixel size and number of electrons per pixel of the SEM image, respectively.

Fig. 8

Simulated SEM image with 16-nm-sized defects on hp 64-nm L/S pattern and the difference images with image processing algorithm for SEM using binarization process before and after canceling the noises, with [(a1) through (a3)] 1 nm LER and [(b1) through (b3)] 2 nm LER, respectively. The resolution (pixel size of the image) is 1.5 nm, and the electron dose per unit area is the same as in PEM.

Figure 9 shows the dependence of the number of electrons per pixel on the detectability for the case of 6 nm resolution. A pixel size of 6 nm was the largest resolution to detect the 16-nm defect on the hp 64-nm L/S pattern with 1 nm LER. In the case with more than 100 electrons per pixel, all defects are detected. However, in the case of 50 electrons per pixel, pin-dot and intrusion defects are not detected, and a false defect is observed.

Fig. 9

Difference images with the resolution (pixel size of the image) of 6 nm before and after noise canceling process: number of electrons per pixel are [(a1) through (a2)] 200, [(b1) through (b2)] 100, and [(c1) through (c2)] 50, respectively. Sample target of the simulation is same as Fig. 8(a1) (1 nm LER).

Figure 10 shows the standard deviation of the background noise in a difference image before the noise canceling process [as shown in Figs. 9(a1), 9(b1), and 9(c1)] as a function of the number of electrons per pixel. As the number of electrons becomes large, the background noise becomes low and shows asymptotic behavior. The asymptotic value is determined by the LER. If the LER is reduced to zero, the value approaches zero. This result shows that 100 electrons per pixel is a required condition for detection, but 200 electrons per pixel is a more robust condition. In the same manner, a pixel size of 3 nm was found to be the largest resolution to detect the 11-nm-sized defect on the hp 44-nm L/S pattern with 1 nm LER. Moreover, it was also found that 100 electrons per pixel is a required condition for detection, but 200 electrons per pixel is a more robust condition. The required pixel size to detect the defect can be approximated by the following equation:

Fig. 10

Standard deviation of difference image before noise canceling process [as shown in Figs. 9(a1), 9(b1), and 9(c1)] as a function of number of electrons per pixel.

In real applications, there exist L/S patterns with LER aligned with 45 deg angle to a horizontal line as shown in Fig. 11. Moreover, when the center of a defect is located just on the corner of the pixels, this is the worst-case situation for defect detection. If at least one pixel is covered by the defect, the defect detection capability becomes more robust under the condition of 200 electrons per pixel.

Fig. 11

Relationship between defect size and pixel size in the binarization process: (a) not detectable case and (b) detectable case. The L/S patterns are aligned with 45 deg angle to a horizontal line.

However, if a part of the pixel is not covered by the defect as shown in Fig. 11(a), the probability that the signals from the pixel exceed the threshold level reduces. This leads to a degradation in the defect detection capability. Hence, in this situation, all of the nearest four pixels need to be covered by the defect as shown in Fig. 11(b). This condition is expressed as Eq. (1). According to the condition, the required pixel sizes for 16- and 11-nm-sized defects are estimated as $\sim 6$ and 4 nm; these values are almost in agreement with the simulation results of 6- and 3-nm-sized pixels, respectively. In the case of an 11-nm defect, the simulation result is smaller than the estimated value. This result shows that the impact of LER becomes large as the defect size becomes small. As a result, a smaller-sized pixel (higher resolution) is required to detect a smaller-sized defect. Malloy et al. have reported that the high-throughput image acquisition of EUV mask can be demonstrated using a multibeam SEM with 69 beams with currents of 672 pA in high-resolution mode.⁷ By using this beam current, we can estimate the throughput to inspect the EUV mask with $100\times 100\text{-}\mathrm{mm}$-sized patterned area. The dwell time of 48 ns on a 6- and 3-nm pixel is needed to accumulate 200 electrons per these pixels. As a result, the throughput using a 69-beam SEM is found to be 58 and 232 h to scan the whole pattern area of EUV mask and to detect the 16- and 11-nm-sized defect on the hp 64-nm and hp-44 nm L/S pattern (16- and 11-nm node), respectively. On the other hand, in the PEM for the 16-nm node, because the TDI sensor has 2048 pixels, a $32.8\text{-}\mu \mathrm{m}$-wide image can be simultaneously acquired with a processing rate of 600 MPPS. Therefore, a throughput of 19 h can be achieved. However, in order to detect the 16-nm-sized defect on the hp 64-nm L/S pattern, 3000 electrons per pixel (16 nm) are required in PEM.⁹ These results show that the required number of electrons per pixel for SEM to detect a same sized defect is an order of magnitude lower than that for a PEM due to the SEM’s higher resolution.

Table 1 shows the performance comparison between PEM and multibeam SEM inspection system. In the PEM for the 11-nm node, which is now being developed, a throughput of 8 h with a pixel size of 11 nm is designed. In order to achieve the same throughput as PEM, 200 to 1850 beams are required in the multibeam SEM.

Table 1

Comparison between multibeam SEM and PEM.