Effect of Reticle CD Uniformity on Wafer CD Uniformity in the Presence of Scattering Bar Optical Proximity Correction

Transcription

1 Effect of Reticle CD Uniformity on Wafer CD Uniformity in the Presence of Scattering Bar Optical Proximity Correction Konstantinos Adam*, Robert Socha**, Mircea Dusa**, and Andrew Neureuther* *University of California at Berkeley, 231 Cory Hall, Berkeley, CA **National Semiconductor Corp., 2900 Semiconductor Dr., Santa Clara, CA Abstract Amplification of reticle linewidth variations in imaging is examined through direct measurements of the mask error factor (MEF), which typically is used to describe this undesirable effect. The error observed in the aerial image linewidth is decoupled from the error in the resulting resist profile linewidth with the introduction of two separate mask error factors, namely the aerial image MEF aerial and the resist MEF resist. These MEF s are evaluated from systematic aerial image measurements and resist profile measurements on printed wafers respectively. In many cases the noise in the metrology tools used in the experiment, combined with the very high quality of the test reticle used, limit the statistical confidence of our results. However useful insight is gained on the role of the non-linearity of the resist in reducing the error observed at the wafer (MEF resist ) in comparison to the error observed in the aerial image (MEF aerial ). It is found that 180nm lines (k1=0.38) have a MEF aerial 1.5 and a MEF resist 1. The effect of scattering bars OPC on the MEF for features sizes 180nm and 220nm (k 1 factors of 0.38 and 0.47 respectively) is generally small and on the order of <10%. 1. Introduction In order to print fine features carefully controlled linewidths are needed and this control must be engineered in the presence of nonlinearities associated with imaging and the resist reaction. Aerial images are well understood and linewidths typically increase more rapidly than the mask linewidth by a factor called the "mask error factor" (or MEF). For small features in the regime of 0.4λ/NA this factor is as much as 1.5 or 2. In printing on wafers the nonlinear response of the resist material which helps convert a sloped image into a more desirable square resist profile may also play a moderating role. This paper investigates the extent to which the resist behavior may actually reduce the dependence of magnification of the linewidth. The overall effective MEF is important as it plays a key role in determining tolerances on photomasks. Developing this understanding is a challenge because other sources of variation such as across mask variations, across field variations, across wafer variation and measurement noise confound the MEF. The effect of optical proximity correction (OPC) such as scattering bars (SB) on MEF also remains an open question. A number of authors have looked at image formation effects on MEF and there are also some indications of the role of the resist [1]-[3], [6], [7]. These publications indicated that reticle variations result in magnified linewidth variations even when adjusted for the reduction ratio of the exposure tool and that this effect rises rapidly with small features. Magnification of mask dimensional error has been shown by simulation to be a concern in the k 1 <0.5 regime and is more severe for dense rather than for isolated features [1]-[3]. However it has been suggested that optical proximity corrections can alleviate mask induced CD errors [2]. Attention should also be paid when assigning reticle linewidth errors to resist linewidth errors because of the various nonlinearities of the lithography step [4]. This paper systematically compares mask, image and wafer resist profile measurements to track mask error factors for isolated and dense feature sizes of 0.4<k 1 <0.5. The merit of using scattering bars at the reticle for optical proximity correction (SB OPC) is examined for a potential reduction in the MEF. In the following section we introduce the definitions for MEF aerial and MEF resist and also describe the experimental approach in measuring these mask error factors. In the third section aerial image measurements are presented and the correlation of aerial image linewidth variations with mask linewidth variations is assessed, which leads to the MEF aerial values. Following the aerial image measurements, results from printed wafers are presented and the MEF resist is also evaluated. A large signal estimate of the MEF is then attempted as an alternative approach.

2 2. Description of the Experiment First some definitions and conventions for the mask error factor (MEF) are introduced. The mask error factor of an optical system is by convention the ratio of the change of the printed linewidth to a change in the linewidth of the mask, scaled by the demagnification of the optical system. To distinguish effects on the aerial image and the resulting resist profile we make separate definitions for the aerial image MEF (MEF aerial ) and the resist MEF (MEF resist ) as follows: LW aerial = MEF aerial LW LW mask resist = MEF resist LW mask where LW mask is the linewidth of a line on the mask, LW aerial is the linewidth of the aerial image at 25% of the maximum intensity and LW resist is the resist linewidth of the line printed on the wafer. The symbol denotes variation of the respective quantity. Note that the demagnification factor M of the imaging system is suppressed throughout our analysis and all mask dimensions are given in wafer (1X) scale. In our experiment M=4, therefore any mask dimension given in 1X scale should be multiplied by 4 to give the actual dimension on the mask. Obviously what matters most is the resist mask error factor, since the resist linewidth of the printed feature is the end result of the lithography step. The mask error factor (both MEF aerial and MEF resist ) is constant for large features with k 1 >~0.6, where k 1 =CD/ λ. However, for k 1 <~0.6 the nonlinearities become prominent and both MEF aerial and MEF resist are functions of k 1 and can be larger than unity. NA Initially the linewidth variations of 8 selected features on a binary test reticle were measured. Isolated lines of sizes 0.18µm (1X) and 0.22µm (1X) with and without scattering bar OPC were considered. Also semi-dense lines of 0.18µm with 0.27µm space and lines of 0.22µm with 0.33µm space with and without bias OPC were examined. The reticle CD measurements were done with a Zygo-Technical Instruments AMS400 system. Next, using an area on the center of the test reticle where linewidths varied by ~10nm (1X) from the nominal linewidth, the aerial images were acquired repeatedly at 25 different locations for each one of the 8 selected features. The aerial image measurements were performed with a Zeiss MSM100 (AIMS tool). Following the mask linewidth and aerial image measurements, 6 wafers were printed in order to understand the effect of reticle CD uniformity on wafer CD uniformity and on resist profile. The resist linewidths were measured in a topdown CD SEM. Details on this part of the experiment are given in section 4. The exposure tool used was a ASML/300 stepper and the CD SEM was a KLA The resist used was Shipley UV5-HS on top of BARC AR2. The illumination wavelength was λ=248nm and the partial coherence factor was σ=0.67. The AIMS measurements were performed with NA=0.53. However, the test reticle was designed for NA=0.57. Therefore wafers were printed and measured at both NA=0.53 (k 1 factors of 0.38 and 0.47 for 0.18µm (1X) and 0.22µm (1X) lines respectively) for direct comparison with the AIMS data and NA=0.57 (k 1 factors of 0.41 and 0.51 for 0.18µm (1X) and 0.22µm (1X) lines respectively). In every case we plot either the aerial image linewidth or the resist linewidth versus the demagnified mask linewidth and the slope of the best fitted line to the data corresponds to the MEF. The following table summarizes the experiment conditions: Table 1: Experimental conditions Resist/type Stepper NA AIMS tool NA Illumination λ Partial coherence Mask reduction Shipley UV5/positive 0.53 or nm X A schematic of the test reticle is given in Figure 1. It is a 6 inch binary COG clear field reticle with a cell design that is repeated 17(rows) by 13(columns) times. Each cell consists of 2 dies, the top one where no OPC was used (Control die) and the bottom one where a combination of scattering bar OPC and mask biasing was used in the design (OPC die). Note that both the Control die and the OPC die contain the same test patterns and the only difference between the two is the use of nonprinting assist features in the OPC die. Various test patterns are laid out in each die but we will only be concerned in this study with the elbows shown enlarged in Figure 1. Specifically we will examine isolated and dense y-oriented lines. Our measurements were repeated in a 5 by 5 array of identical die-pairs (25 different measurement sites) located at the center of the reticle.

3 col 1 } col 13 } line space (line/space)=(1/1.5) row 1 { LW mask,dens Control Die LW mask,iso row 17{ scattering bars (SB OPC) LW mask,dens OPC Die LW mask,iso Figure 1. Schematic of the test reticle. The basic cell which is repeated 17 (rows) by 13 (columns) times consists of a pair of dies shown enlarged at the inset. The Control die and the OPC die contain the same test designs but the OPC die uses scattering bar optical proximity correction and biasing. The inset only shows an enlargement of the elbows test patterns that were examined in this study. The test reticle was of higher quality than we anticipated and it posed certain difficulties in our study. The demagnified mask LW was always within the 5% spec and in some cases the LW variation was only 3%! The standard deviation of each 25-member population was at most 3nm. We had planned to use the linewidth variations in the mask to give a sufficiently broad spread of feature sizes which was not the case. In retrospect, it would have been better to specify programmed mask variations. 3. Aerial Image MEF from Aerial Image Measurements Aerial image measurements of the test reticle were performed in the 5 by 5 array of sites. At each site 8 measurements were made: the 0.18µm and 0.22µm lines, isolated (without OPC and with SB OPC) and dense (without OPC and with bias OPC). A complete aerial image measurement of a single site consisted of 9 through focus images of the two-dimensional intensity variation, where the focus step size was 0.25µm (1X). An instructive way to view the results of each measurement is to plot the aerial image linewidth (LW aerial ) for different intensity thresholds versus defocus. Then, the LW aerial at the 25% threshold level at best focus and at 0.3µm defocus can be used to determine the variation LW aerial due to correlated mask variation LW mask. Consequently, the optical mask error factor (MEF aerial ) can be estimated at the corresponding k 1 factor, for best focus and 0.3µm defocus conditions. Figure 2 shows an example of these Bossung-like plots (LW aerial vs. defocus) obtained from the AIMS measurement of a 0.18µm (1X) isolated line without OPC and with scattering bar OPC (this is from a single pair of Control and OPC features in Figure 1). Each of the two plots in Figure 2 shows the dependence of LW aerial on defocus for intensity thresholds from 15% to 45% with a step of 5%. Notice the improved iso-focal response of the isolated line at all threshold levels in the presence of scattering bars OPC. From these plots LW aerial at best focus and at 0.3µm (1X) defocus is immediately extractable at the 25% intensity threshold. This will appear as a single point (LW mask, LW aerial ) in subsequent plots of LW aerial vs. LW mask.

4 Control Die OPC Die Figure 2. Linewidth vs. Defocus (Bossung-like) plots for various intensity threshold levels acquired from aerial image measurements. To the left you can see the optical response of a 0.18µm isolated line without OPC (control die), whereas to the right you can see the optical response of the same line with the use of scattering bars assist features. Figure 3 depicts a plot of the aerial image LW vs. the demagnified mask LW at best focus for each of the 25 measured sites of the 0.18µm isolated lines. Two series of data are displayed in the same plot, where the first corresponds to the control lines without OPC and the second to the lines with scattering bar assist features. The solid line is a linear fit to the data from the control features and the broken line to the data with scattering bar corrected features. The resulting line equations are displayed together with the correlation coefficient R 2 of the data, for each set of data. The slope of each line is the respective MEF aerial, and MEF aerial 1.5 for the control lines without OPC and it slightly decreases to MEF aerial 1.4 for the lines with scattering bar assist features. Figure 3. LW aerial vs. LW mask at best focus for 0.18µm isolated lines, with and without OPC. The slope of the best fitted lines is the estimate of the MEF aerial at k 1 =0.38. Figure 4 shows a plot of the aerial image LW vs. the demagnified mask LW at 0.3µm defocus for each of the 25 measured sites of the 0.18µm isolated lines. Again, two series of data are displayed in the same plot, the one corresponding to the control lines without OPC and the other to the lines with scattering bar assist features. A linear fit to the data is plotted in each

5 case and the resulting line equations are also displayed together with the correlation coefficient R 2 of the data, for each set of data. The mask error factor of the corrected with SB OPC lines increases to MEF aerial 1.8 from 1.4, but no conclusion can be made for the lines of the control die because of the very poor correlation of the data in this case. The data from the aerial image measurements were noisy. This is reflected in Figures 3 and 4 in the poor linear correlation of the points (LW aerial, LW mask ), which leads to low values of R 2. The source of the noise was the CCD camera of the AIMS tool and appeared to be associated with the resolution of 25nm/pixel. 4. Resist (Wafer) MEF Figure 4. LW aerial vs. LW mask at 0.3µm defocus for 0.18µm isolated lines, with and without OPC. The slope of the best fitted lines is the estimate of the MEF aerial at k 1 =0.38. In taking wafer measurements a particular methodology was applied to reduce track coat and develop effects and average CD SEM measurements. First we exposed a calibration wafer using a FEM (focus-exposure-matrix) where we varied both focus and dose in order to determine best focus and best dose. Then, a second wafer was printed using a FEM where focus was set at best focus, as determined from the first wafer, and the exposure dose was varied with steps of 0.5mJ/cm 2 around the center dose, which was set to be equal to the best exposure dose, again determined from the first wafer. The resulting wafer contained a rectangular array of fields, where all the fields were exposed at best focus and all the fields in the same column were exposed with the same dose. In this way the field at best focus and best dose was repeated 7 times on the wafer. A third wafer was printed using a FEM with a defocus of 0.3µm from best focus and dose was varied the same way as before. These wafers were exposed with NA=0.53, which was the same as that in the AIMS measurements. The above procedure was also repeated with another set of wafers (wafers 4, 5, 6) exposed with NA=0.57. This was the NA value for which the test reticle was optimized. CD SEM measurements of the resist linewidths of every wafer (except for the calibration wafers 1 and 4) and for all 7 fields with best dose were performed at all of the 25 different sites on the reticle. At each site 8 measurements were made: the 0.18µm and 0.22µm lines, isolated (without OPC and with SB OPC) and dense (without OPC and with bias OPC). For each column of 7 measurements the outfliers were first discarded to remove coat and track effects. The remaining values (in all cases at least 5) were then used to compute the average and standard deviation to reduce the large variation in SEM measurements. Estimates of the remaining errors are depicted in the plots of this paragraph with the error bars along the y-axis, that are one standard deviation long in each direction and denote the spread of the LW resist values. Figure 5 shows plots of the resist linewidth vs. mask linewidth for the 0.18µm isolated lines, with NA=0.53. Figure 5(a) is the best focus exposure case (second wafer) whereas Figure 5(b) is the defocused by 0.3µm from best focus case (third wafer). Remember that each point is the average of the linewidths of at least 5 different fields that correspond to the same demagnified mask linewidth. Therefore the spread along the y-axis is to be interpreted as noise from the process and SEM measurements. From plot (a) observe that MEF resist 1 for the control case, and the SB OPC case only slightly decreases the MEF resist. From plot (b) observe that the MEF resist 1.3 for the SB OPC case. Although low values of coefficient R 2 are indicative of poor correlation of the data due to noise, there is still a definite trend indicating that the MEF resist 1 at best focus regardless of the presence of SB OPC and that it increases by as much as 30% as we move 0.3µm out of

6 focus. (a) (b) Figure 5. LW resist vs. LW mask for 0.18µm isolated lines, with and without SB OPC (NA=0.53). (a) best focus, (b) 0.3µm defocus. The slope of the best fitted lines is the estimate of the MEF resist at k 1 =0.38. Figure 6 shows plots of the resist linewidth vs. mask linewidth for the 0.18µm isolated lines, with NA=0.57. Figure 6(a) is the best focus exposure case (fifth wafer) whereas Figure 6(b) is the defocused by 0.3µm from best focus case (sixth wafer). The same observations apply to these two plots as for the plots in Figure 5, which strengthen the fact that MEF resist 1 for 0.38<k 1 <0.41 for best focus exposure and MEF resist 1.3 for 0.3µm defocused exposure. Note that in (b) the data series for the control die is to be discarded as very noisy and with very poor correlation. (a) (b) Figure 6. LW resist vs. LW mask for 0.18µm isolated lines, with and without SB OPC (NA=0.57). (a) best focus, (b) 0.3µm defocus. The slope of the best fitted lines is the estimate of the MEF resist at k 1 =0.41.

7 Figure 7 depicts a plot of resist linewidth vs. mask linewidth for the 0.22µm isolated lines (with and without SB OPC) at best focus and NA=0.57 (from wafer 6). In this case the R 2 is again low and nothing can be safely inferred about the MEF resist. Figure 7. LW resist vs. LW mask for 0.22µm isolated lines, with and without SB OPC (NA=0.57) at best focus. The slope of the best fitted lines is the estimate of the MEF resist at k 1 =0.51. Figure 8 depicts a plot of resist linewidth vs. mask linewidth for 0.18µm dense lines (with and without bias OPC) at best focus and NA=0.57 (from wafer 3). It is interesting that for dense lines of k 1 =0.41 there is quite strong indication that the MEF resist 1, which is also much lower than MEF values for dense lines reported by other authors. The separation on the graph of the two sets of data series takes place because the mask biasing OPC was not optimized for k 1 =0.41. Figure 8. LW resist vs. LW mask for 0.18µm dense lines, with and without bias OPC (NA=0.57) at best focus. The slope of the best fitted lines is the estimate of the MEF resist at k 1 =0.51. From all the previous plots (Figures 5-8) it is obvious that noise from the metrology tool is very dominant making it very difficult to draw definitive conclusions with high statistical confidence. The data for scattering bar OPC shows both increases and decreases in MEF resist, but the data has a low degree of statistical confidence. The noise of the experiment can, for example, be estimated from the plot in Figure 7, where the slope due to MEF resist is expected to be near unity. A total mask LW range of 8nm is seen (with σ 3nm) as compared to a total resist LW range of over 20nm (with σ 8nm). The statistical confidence increases as the correlation coefficient R 2 increases (0<R 2 <1). 5. Large Signal Estimate of the MEF In an attempt to overcome the noise limitations in our initial approach we introduced an alternative method, which we call "large signal estimate of the MEF". In this method isolated lines with mask linewidths of 0.22µm and 0.18µm were

8 treated as large signal variations (±10%) from 0.20µm linewidths. Obviously we do not expect this approach to give an accurate estimate of the MEF resist at k 1 =0.46 (NA=0.57)or the MEF aerial at k 1 =0.43 (NA=0.53), since with 10% variation we cannot claim the linear approximation to be accurate. Nevertheless, the outcome will be instructive and can reinforce many of our previous observations. Figure 9 shows a plot of LW resist and LW aerial vs. LW mask, where each x-coordinate is evaluated from subtracting linewidth values for 0.18µm lines from 0.22µm lines for all 25 measured locations and each corresponding y-coordinate comes from subtracting the average resist linewidth values or the aerial image linewidth values for the respective lines. The standard deviation along y-axis for the data from resist measurements is now the square root of the sum of the two squared standard deviations, which corresponds to uncorrelated errors. The best fitted lines have to pass through the origin, by their definition. The slope of these lines is a large signal estimate of MEF resist 1 and MEF aerial 1.3. While half of the MEF s are expected to be lower at k 1 =0.43 instead of 0.38, this data is in very good agreement with our conclusions from sections 3 and 4. As expected the MEF resist is significantly lower than the MEF aerial. Figure 9. LW resist and LW aerial vs. LW mask for isolated lines, with and without SB OPC at best focus. The slope of the best fitted lines is a large signal estimate of the MEF resist and MEF aerial for 0.2µm lines. Note that each line has to cross exactly from the origin. 6. Conclusions The mask error factor from resist linewidth measurements (MEF resist ) for 0.38<k 1 <0.51 was found to be approximately 1 for both isolated and dense lines, which is smaller than that observed in AIMS data (MEF aerial 1.5 for isolated lines) or reported from simulation (>2). The observation that the mask error magnification is nearer to unity was further reinforced by the "large signal estimate" of the MEF resist that was presented. There was also some indication that scattering bars on isolated lines (SB OPC) reduce the MEF aerial by 5%. This however is not statistically significant. The low value of MEF observed experimentally is believed to be due to the tendency of the resist to respond to the toe area of the image (region below 10% of the clear field). The low statistical confidence in determining the MEF was due to noise inherent in both aerial image and resist linewidth measurements, combined with the high quality of the mask, which did not have a sufficient spread of values to exceed the noise. For accurate evaluation of the MEF more accurate metrology tools or methods are needed. Still the resist linewidths will need to be averaged over a big population of printed fields of the same focus and dose on the wafer. Finally, a test reticle with programmed linewidth variations is required, whose magnitude exceeds the noise of the experiment. Acknowledgments The authors would like to thank Xuelong Shi and Karthik Rammohan for their support in the wafer processing and resist profile measurements at the National Semiconductor R&D facility in Santa Clara. This work was sponsored in part by the California SMART Program, Grant No. S97-01.

9 References [1] Alfred Wong, et. al., Lithographic Effects of Mask Critical Dimension Error, Proceedings of the SPIE, vol. 3334, The International Society for Optical Engineering, (Optical Microlithography XI, Santa Clara, CA, USA, February 1998.) SPIE-Int. Soc. Opt. Eng, 1998, p [2] John Randall, et. al., Reduction of Mask Induced CD Errors by Optical Proximity Correction, Proceedings of the SPIE, vol. 3334, The International Society for Optical Engineering, (Optical Microlithography XI, Santa Clara, CA, USA, February 1998.) SPIE-Int. Soc. Opt. Eng, 1998, p [3] Pei-yang Yan, J. Langston, Mask CD Control Requirement at 0.18 µm Design Rules for 193 nm Lithography, Proceedings of the SPIE, vol. 3051, The International Society for Optical Engineering, (Optical Microlithography X, Santa Clara, CA, USA, March 1997.) SPIE-Int. Soc. Opt. Eng, 1997, p [4] R. E. Gleason, Hua-Yu Liu, Impact of Photomasks on Linewidth Variation, Proceedings of the SPIE, vol. 3236, (17th Annual Symposium on Photomask Technology and Management, Redwood City, CA, USA, Sept ) SPIE-Int. Soc. Opt. Eng, 1998, p [5] Pei-Yang Yan, et. al., Mask Defect Printability and Wafer Process Critical Dimension Control at 0.25 µm Design Rules, Japanese Journal of Applied Physics, Part 1, vol. 34, no. 12B, (8th International MicroProcess Conference (MPC'95), Sendai, Japan, July 1995.), Dec. 1995, p [6] A. Vacca, et. al., 100 nm Defect Detection Using an Existing Image Acquisition System, Proceedings of the SPIE, vol. 3236, (17th Annual Symposium on Photomask Technology and Management, Redwood City, CA, USA, Sept ) SPIE-Int. Soc. Opt. Eng, 1998, p [7] W. Maurer, Mask Specifications for 193 nm Lithography, Proceedings of the SPIE, vol. 2884, (16th Annual Symposium on Photomask Technology and Management, Redwood City, CA, USA, Sept ) SPIE-Int. Soc. Opt. Eng, 1996, p