Investigation of interactions between metrology and lithography with a CD SEM simulator

Transcription

1 Investigation of interactions between metrology and lithography with a CD SEM simulator Mark D. Smith, Chao Fang, John J, Biafore, Alessandro Vaglio Pret, Stewart A. Robertson KLA-Tencor Corp. ABSTRACT The predictive power of computational lithography is often demonstrated by showing predicted 2D pattern shapes compared with top-down SEM images. However, image formation in a SEM is a complex process [1,2,3], and for most 3D lithography and OPC simulators, line width measurements and 2D pattern shapes are based on extracted resist polygons at a fixed height above the substrate. Generating resist polygon shapes with this method is driven by computationally efficiency instead of an attempt to describe the image formation process in an actual SEM. We present PROLITH photolithography simulations combined with simulation of the CD SEM to investigate the interactions between lithography and metrology. Our CD SEM simulator is a simplification of the complicated image formation process [4], but it captures many effects seen experimentally. For example, narrow trenches and contact holes are dark at the bottom in our simulated SEM images, while for isolated lines, the sidewall of the photoresist can clearly be observed all the way to the resist foot at the substrate. This simple result has important implications when evaluating lithographic phenomena such as LWR: for polygon-based metrology, simulated LWR is approximately constant with resist thickness; by contrast, the LWR increases with decreasing thickness when the same simulated 3D resist profiles are evaluated with the CD SEM simulator. Introduction Metrology is an integral part of lithography you cannot perform an experiment or control a manufacturing process unless you measure what has been printed on the wafer. CD SEM is the most prevalent metrology after develop, especially for OPC calibration and verification. Other metrology methods, such as scatterometry, can be used as well, and it is well-known that there is often an offset between different types of metrology [5]. An offset between CD SEM and scatterometry is not surprising when one considers that these metrology techniques are vastly different CD SEM images are formed by rastering a focused electron beam across the structures on the wafer and detecting a secondary electron signal, whereas scatterometry is based on measuring the reflectivity of a grating target through angle or wavelength and then fitting the measured spectra to a calculated spectra for a geometric model of the grating structure. Similarly, the virtual metrology performed in most lithography simulators is very different from either the physics of the CD SEM or scatterometry, so we should expect systematic offsets that could be important in interpreting the results of lithography simulations, especially when comparing with experimental results. To understand the expected differences, it is useful to review how metrology is commonly performed in a simulator. In PROLITH, the result of a lithography calculation is a 3D resist profile, and virtual metrology is often performed by slicing the 3D profile at a fixed height above the substrate. The slice gives a set of polygons representing the printed features, and CDs can be extracted by determining the width of the polygons at different locations. A similar approach is applied in OPC simulations, where the calculation is usually performed in the XY plane, and the result is either a set of edge locations or a resist polygon [6]. Advances in Patterning Materials and Processes XXXI, edited by Thomas I. Wallow, Christoph K. Hohle, Proc. of SPIE Vol. 9051, SPIE CCC code: X/14/$18 doi: / Proc. of SPIE Vol

2 Intensity Profile Line Scan SEM -> E -Beam Profile Figure 1: The simplified CD SEM image formation model in PROLITH takes the geometric shape of the resist profile, calculates the signal, and then convolves the signal with a Gaussian that represents the size of the e-beam [4]. The purpose of this paper is to compare common photolithography metrology with metrology performed by a CD SEM. In order to do this, we have built a physical-based SEM emulator in PROLITH [4], as shown in Figure 1. We call this a SEM emulator because the SEM model in PROLITH is not intended to include all of the physics required to describe how an actual SEM works (e.g., electron optics and detectors, charging, resist shrinkage, etc.). Instead, we want to produce more realistic pictures of a simulated result and to produce CD measurements that mimic the physical process of CD SEM image formation. We have included threshold edge detection algorithms that are similar to what is found on a CD SEM [7], and a few filters for noise reduction, such as Gaussian smoothing along the direction on a line scan and summing lines for averaging perpendicular to the scan direction. An example is shown in Figure 2, where we show an iso-dense 5 bar pattern with perfect profiles with a 70nm width, an 80 degree sidewall angle, and top rounding (these are geometrically defined shapes, not the result of a lithography simulation). When typical polygon-based metrology is applied to this structure, the measured CD is 70nm for all parts of the structure. By contrast, metrology on the emulated SEM image gives a CD of 63.6nm for the dense structures and a CD of 70nm for the isolated structure. This is very similar to the result reported by Hitachi for narrow trenches [8]. As demonstrated by this example, the combination of a photolithography simulator with a SEM emulator provides the ideal framework for comparing polygon-based metrology with SEM metrology. The outline of this paper is as follows. In the first section, we will show a comparison between our SEM emulator and experimental results and between our SEM emulator and a more rigorous Monte-Carlo SEM simulator. Next, we will examine the differences between metrology with the SEM emulator and polygonbased metrology for three sets of lithography simulations. First, we will show isolated trenches for positivetone develop and negative-tone develop. Second, we will show SRAF printability for contact holes. The third example will be LWR versus resist thickness for 22nm half-pitch lines printed with EUV. Finally, we conclude the paper with a summary. Proc. of SPIE Vol

3 1 Figure 2: Example 3D profile (shown on left) and corresponding PROLITH SEM image (shown on right). All features are 70nm wide on a 140nm pitch with top rounding and an 80 degree sidewall angle (see inset). Typical polygon-based metrology gives a 70nm CD for all structures, whereas metrology on the SEM image gives 64nm CD for the dense structure and 70nm CD for the isolated structure. Comparison with Experiment and with Monte Carlo Simulations We exposed 65nm lines on a 150nm pitch and collected a cross-section SEM image and a top-down SEM image for comparison with our SEM emulator. A comparison between a calibrated resist model and the cross-section SEM is shown in Figure 3, and a comparison between experimental and simulated top-down images is shown in Figure 4. This comparison demonstrates that for an accurate simulated resist profile shape, we can achieve reasonably good matching to the experimental SEM images and line scans, at least for this set of conditions. We also performed a comparison with a rigorous Monte Carlo SEM simulation, using the approach by Grella et al [2]. A comparison between the SEM emulator and the rigorous Monte Carlo simulation is shown in Figure 5. Again, the agreement is good. It should be noted that charging was not simulated in the Monte Carlo calculation or in the SEM emulator results. We should also note that the Monte Carlo is based on first-principles, so it has fewer adjustable parameters, but it also requires approximately 1000x longer simulation times compared with the SEM emulator. These successful comparisons with experiment and rigorous Monte Carlo calculations give us reasonable confidence in the SEM emulator results, so we now continue to investigate the differences between polygon-based metrology and the results from the SEM emulator. Proc. of SPIE Vol

4 SU kV 2.2mm x150k LA1(UL) 8/16/2011 ' ' ' Figure 3: Comparison between experimental cross-section SEM image (on left) and the simulated resist profiles (on right) for 65nm lines on 150nm pitch. Figure 4: Comparison between experimental top-down image and the image from the SEM emulator is shown on the left. The simulated results are shown in the red box. Comparison between the line scans is shown on the right. The data for the line scans is an average of 5 rows from each image, and the line scans are normalized to have a maximum intensity of one. Proc. of SPIE Vol

5 Figure 5: Comparison between the SEM emulator and a rigorous Monte Carlo simulation of the resist crosssection shown in Figure 3. Isolated trenches printed with positive tone develop and with negative tone develop photoresist We simulated isolated trench structures with the PROLITH Stochastic Resist Model for positive tone develop (PTD) and negative tone develop (NTD) processing [10,11]. The simulation was for an ArF immersion system with NA=1.35, XY polarization, and annular illumination. The features were nominally 65nm, with a resist thickness of 100nm. Example 3D resist profiles are shown in Figure 6, where the PTD process has prograde sidewalls (wider at the top than at the bottom) and the NTD process has retrograde sidewalls (narrower at the top). If we use polygon-based metrology on the PTD result, then we must decide which Z- height to extract the resist slice. One choice is to slice the resist at the substrate (Z=0 nm). However, if we compare this with the CD measured from a simulated SEM image, we find the average CD is much narrower: polygons predict CD failure by scumming, while the CD SEM image gives an average CD of 63.7nm. We can adjust the height of polygon slice to Z=22nm in order to match the average CD, as shown in Figure 7. It is also interesting to notice that the scumming, shown on the left side of Figure 6, is barely visible in the SEM image. Positive Tone Negative Tone Figure 6: Isolated trenches printed with positive tone develop (left) and negative tone develop (right) resist processes. Proc. of SPIE Vol

6 I 4 Figure 7: Metrology results for the PTD material shown in Figure 6. Resist polygons at the substrate (Z=0) are shown on the left, resist polygons at Z = 22nm are shown in the middle, and simulated SEM image is shown on the right. The average CD for the resist polygons sliced at Z=22nm is 64nm, which is closely matched to the SEM CD of 63.7nm using an edge detection threshold of 50%. Figure 8: Simulated SEM images through focus for the NTD isolated trench. The image on the left is underexposed at best focus, the image in the middle is at +50nm defocus, and the image on the right is for +60nm of defocus. Proc. of SPIE Vol

7 For the NTD system, matching is more difficult, because the SEM can only see the edges at the very top of the resist. Extracting polygons at the top surface of the resist is problematic for NTD systems that demonstrate resist loss because the resist thickness may vary across the simulated region on the wafer. In contrast, it is relatively straightforward to extract edges from the simulated SEM images, regardless of the cross-section shape. We show some example simulated SEM images in Figure 8 as the wafer defocus increases. The final SEM image shows pattern failure, and it is possible that a SEM CD algorithm would return a measured CD for this feature, even though measureable polygons could probably not be extracted. Such a SEM measurement should be discarded, but if it were not removed, any comparison with simulated CDs based on polygon metrology would lead to a large apparent discrepancy. SRAF printability Accurate detection of sub-resolution assist feature (SRAF) printing is a very for OPC generation larger SRAFs improve the depth-of-focus, but if they are so large that the SRAFs print, then they may lead to undesirable features in the pattern transfer step and possibly electrical failure of the device. In a paper by Li et al., a standard OPC resist model was used to predict the shape of the primary features, and a second OPC resist model was used to detect the printing of SRAFs [9]. They demonstrated their model on 100nm holes on a 300nm pitch with a single SRAF between the holes. The SRAFs varied in size from 30nm to 80nm, and they found experimentally that the 40nm SRAFs did not print and the 50nm SRAFs did print. We are able to predict this same result with a PROLITH resist model, as shown in Figure 9. PROLITH results are shown for both the continuum resist model and the stochastic resist model, and both models show SRAFs start to print at a size of 50nm. 100nm Holes w /40nm SRAF Experiment PROLITH Continuum Model PROLITH Continuum Model with SEM O 0 0 O 00 O (-) PROLITH Stochastic Model with SEM 100nm Holes w /50nm SRAF }00ß0c O o(o0c er O 0000( Figure 9: Experimental SEM images of the onset of SRAF printing (first column), the corresponding PROLITH simulation (second column), simulated SEM image of the PROLITH results (third column), and simulated SEM image of the PROLITH Stochastic Resist Model results (fourth column). Experimental images are from reference [9]. Proc. of SPIE Vol

8 "WAN Figure 10: Comparison of SEM and polygon-base metrology for onset of SRAF printing for the PROLITH stochastic resist model. As shown in the simulated SEM image (on the left), the primary feature has a CD of 128nm. We can set up polygon-base metrology to match this CD by slicing the resist at a height of Z=20nm, but this metrology height does not detect all of the printing SRAFs. We can attempt to match SEM metrology to polygon-base metrology in the same way we matched the metrology for the isolated PTD trench: we determine the CD of the contact hole in the SEM image as 128.2nm, and then we choose a height of Z=20nm to slice the 3D resist profile and generate the resist polygons, as shown in Figure 10. While this slice height gives good matching for the contact hole CD, we only detect a few of the printing SRAFs that are clearly shown in the simulated SEM image. This implies that the second resist model created by Li et al. for the SRAFs may have been required for two reasons: printing of SRAFs likely requires a 3D resist model, and metrology matching is difficult when using a polygon-based method at a single height above the substrate to evaluate the results of the lithography simulations. LWR versus Resist Thickness In our final example, we investigate line width roughness (LWR) versus resist thickness for 22nm half-pitch lines printed with EUV. We set up the simulations with NA=0.25 and an optimized dipole source, and varied the resist thickness from 35nm to 65nm with the exposure dose fixed at 26.8 mj/cm 2. Example simulated lines are shown in Figure 11, and example SEM images are shown in Figure12. For both the resist profile polygons and the SEM images, we can extract the mean CD and LWR for the 5 lines in the middle of the simulated region. We extracted data for 0.5 microns along the length of each line, and the resist polygons were sliced at a height of 10% of the resist thickness, and we repeated this process 10 times to give a total of 50 measurements. Figure 11: EUV resist profiles for 22nm HP with 35nm (left), 50nm (center), and 65nm (right) thicknesses. Proc. of SPIE Vol

9 Figure 12: Simulated SEM images for the profiles shown in Figure 11. Resist thickness of 35nm is shown on the left, 50nm thickness is shown center, and 65nm thickness is shown on the right. CD and LWR results are shown in Figure 13. We see that the polygon CD is much more sensitive to thickness than the SEM CD values, and we also seee that the LWR is approximately constant for the polygon metrology, while the LWR measured by the SEM is lower and shows a systematic trend with resist thickness. There are two reasons for these results. First, the cross-section shape of the resist changes as the resist becomes thinner. As shown in Figure 11, the 65nm thickness shows top rounding and then sloped sidewalls due to absorbance, while the 35nm resist thickness only shows top rounding because the film is so thin. This will change the shape of the signal generated by the SEM emulator. Second, the trenches are much darker for the thick resist than for the thin resist, as shown in Figure 12. For the thin resist, the SEM signal can see the bottom of the trenches, and it is likely that the SEM edge detection corresponds to a position near the bottom of the resist, which is typically rougher than the sidewall midway up the resist height. Summary and Conclusions We have demonstrated a SEM emulator that mimics the physicss of the SEM image formation process. While this simulator is not intended to rigorously model all of the physics in a real SEM, we have shown examples of good agreement with both experimental SEM images and images generated by a more rigorous Monte Carlo SEM simulator. We then applied this SEM model to examine differences compared with polygon- For based metrology for isolated trenches printed with positive tone develop and with negative tone develop. the positive tone case, it was straightforward to match the polygon and SEM metrology. Matching was more difficult for the negative tone case because the profiles were retrograde, and the SEM could only see to very Proc. of SPIE Vol

10 Figure 13: Measured CD and LWR for polygon and SEM metrology for a variety of resist thicknesses. Each result is the average of 5 lines within the field of view shown in Figure 12 repeated over 10 images (i.e., the average of 50 lines). top of the resist profile. Next, we examine SRAF printability, and demonstrated that the SEM emulator very naturally captures both the main features, which clear through the entire thickness of the film, and the onset of SRAF printing, which only appear on the top surface of the resist. Our final example showed that resist CD and resist LWR are measured differently as a function of resist thickness. For polygon-based metrology, the LWR was approximately constant with thickness, while thinner films showed greater LWR with SEM based metrology. Acknowledgements We would like to thank Luca Grella from the REBL division at KLA-Tencor for assistance and example code for Monte Carlo simulation of the SEM, and we would like to thank Alex Vaglio Pret at KLA-Tencor and our imec partners for experimental SEM image collection. References 1. Reimer, L., Scanning Electron Microscopy, 2 nd Ed., Springer, Berlin (1998). 2. Grella, L., Lorusso, G., Niemi, T., Chuang, T., and Adler, D., "Three-dimensional simulations of SEM imagingg and charging", Proc. SPIE 4344 (2001). 3. Suzuki, M., Borisov, S., Babin, S., and Ito, H. "Optimizing the detector configurationn for SEM topographic contrast by using a Monte Carlo simulation", Proc. SPIE 7729 (2010). 4. Jones, R., Byers J. and Conley W., "Top-down versus cross-sectional SEM metrology and its impact on lithography simulation calibration", Proc. SPIE 5038 (2003). 5. Allgair, J., Benoit, D., Drew, M., Hershey, R., Litt, L.C., Herrera, P., Whitney, U., Guevremont, M., Levy, A., Lakkapragada, S., Implementation of spectroscopic critical dimension (SCD) for gate CD control and stepper characterizatio on, Proc. SPIE 4244 (2001). Proc. of SPIE Vol

11 6. Cobb, N. Flexible sparse and dense OPC algorithms, Proc. SPIE 5853 (2005). 7. Yamaguchi, A., Koyanagi, H., Tanaka, J., Inoue, O., Kawada, H., Robust edge detection with considering three-dimensional sidewall feature by CD-SEM, Proc. SPIE 7971 (2011). 8. Tanaka, M., Meessen, J., Shishido, C. Watanabe, K., Minnaert-Janssen, M., and Vanoppen. P. CD bias reduction in CD-SEM linewidth measurements for advanced lithography, Proc. SPIE 6922 (2008). 9. Li, J., Gao, W., Fan, Y., Xue, J., Yan, Q., Lucas, K., De Bisschop, P., Melvin, L.S., Binary modeling method to check the sub-resolution assist features (SRAFs) printability, Proc. SPIE 8326 (2012). 10. Robertson, S. A., Reilly, M.T., Biafore, J. J., Wandell, J., Smith, M.D., Stochastic simulation of resist linewidth roughness and critical dimension uniformity for optical lithography, J. Micro/Nanolith. MEMS, MOEMS. 9 (2010). 11. Robertson, S.A., Reilly, M., Biafore, J.J., Smith, M.D., Bae, Y., Negative tone development: gaining insight through physical simulation, Proc. SPIE 7972 (2011). Proc. of SPIE Vol