Mask-Level Measurements Predict Imaging Performance for Flash Designs
Flash memory is accelerating the push for continued scaling, and hyper-NA immersion lithography has brought 45 nm and below imaging within reach. Several mask parameters are important for imaging performance, and mask-level aerial imaging is useful for characterizing that performance before exposure.
E. van Setten, O. Wismans, K. Grim and J. Finders, ASML Netherlands BV, Veldhoven, Netherlands; M. Dusa, ASML TDC, Santa Clara, Calif., www.asml.com; R. Birkner, R. Richter and T. Scherübl, Carl Zeiss SMS GmbH, Jena, Germany -- Semiconductor International, 9/1/2008
Flash memory has become one of the most important segments of the semiconductor industry in recent years, driving the lithography roadmap with its dramatic acceleration in dimensional shrink. Meanwhile, the introduction of hyper-NA immersion lithography has brought the 45 nm node and below within reach for memory makers.
| An operator prepares a test mask for AIMS qualification. |
At these feature sizes, mask topology and the material properties of the film stack on the mask play an important role in imaging performance. Furthermore, the breakup of the array pitch regularity in the NAND-type flash memory cell by two thick word lines and a central space leads to feature-center placement (overlay) errors that are inherent to the design. Mask-level aerial image measurements have been performed to characterize the imaging performance of a NAND-flash memory gate layer.1
The NAND-type flash memory cell (Fig. 1) is organized in separate modules (pages) containing 32 word line transistors (word line poly gates) with two select transistors (select poly gates), and with one source/drain (S/D) contact for the entire 32-page module. The array-core zone consisting of 32 equal line/spaces (L/S) allows for aggressive low-k1 imaging with extreme off-axis illumination. However, the presence of the two mirrored select gate (SG) lines and the SG-SG gap break up the array pitch regularity and create a discontinuity in layout topology. Dusa et al.2 showed that this discontinuity introduces an asymmetry in the near-field diffracted intensity under the left and right edges of the SG line, as well as under the next adjacent four to five word lines (Fig. 2). The lithographic performance of the mask layout is affected by this cross-coupling effect, which induces feature-center placement errors through dose and focus.
| 1. In a flash memory core layout for critical poly word line and interconnect bit line and contact layers (left), layout simplicity with unidirectional regularity is noticeable. However, two mirrored select gate (SG) lines and an SG-SG gap (right) create a discontinuity in layout topology. Design rule defines SG and SG-SG gap dimensions as multiple integers of half-pitch. |
| 2. This scattered near-field diffracted intensity for TE-polarized illumination from a flash word line mask pattern shows an asymmetry under the left and right edges of the SG line and adjacent word lines (e.g., compare the intensity at the black arrows). |
Because the mask is the dominant contributor to CD uniformity via the mask error enhancement factor (MEEF) and feature-center placement errors described above, we measured a test mask using AIMS 45-193i. This enables the separation between mask and scanner when looking at the lithographic performance of the NAND-flash gate layer.
Exposures were performed with a Twinscan XT:1900Gi hyper-NA exposure tool. The test mask contains NAND-type flash memory clips at multiple feature sizes patterned in molybdenum silicon (MoSi), and MoSi covered with a chrome (Cr) layer. The results reported here are from the binary (MoSi+Cr) features with a 40 nm design rule minimum half-pitch. We will focus on CD uniformity, Bossung curves and feature-center placement errors.
Experimental conditions
All exposures and measurements were done with a test mask from the Advanced Mask Technology Center (AMTC, Dresden, Germany). The film stack of the mask consists of a 680 Å MoSi layer covered by a 580 Å chrome layer.
The mask has been measured extensively with an AIMS 45-193i aerial image measurement system from Zeiss. The system has a maximum scanner equivalent NA of 1.4, which enables the emulation of the latest generation of hyper-NA immersion scanners. Furthermore, it is possible to use polarization and multiple off-axis illumination modes, equivalent to scanner illumination modes. The aerial image measurements were done at NA=1.35 and =0.78/0.98 dipole X illumination with 35° opening angle and linearly polarized light (Y-polarized) in scanner mode3 to capture the vector effects associated with hyper-NA imaging.
In scanner mode, the aerial image in resist will be generated as a result of a combination of actual image measurements in the scalar mode (or AIMS mode) and a Zeiss proprietary algorithm to emulate mask effects. Additionally, the user must input the refractive index of the resist. In this work, the refractive index was set to 1.7, which is equivalent to the refractive index of the photoresist used during the exposures.
ASML's Twinscan XT:1900Gi exposure tool has a maximum NA of 1.35. The illumination conditions were optimized for 40 nm half-pitch using NA=1.35 and =0.807/0.967 dipole X illumination with 35° opening angle and Y-polarized light. The exposures were done on bare silicon wafers coated with 93 nm Brewer Science ARC93SR bottom antireflective coating (BARC), 95 nm TOK TARFPi6001 resist and a 90 nm JSR TCX041 topcoat. All wafer measurements were done by CD-SEM (Hitachi CG-4000).
The difference in sigma settings between the AIMS measurements and exposures has been chosen purposely to closely match the intensity profile of the scanner illumination source. The AIMS source can be regarded as top-hat illumination, whereas the scanner source has a more Gaussian intensity distribution. Furthermore, sigma-inner/-outer on the scanner are defined as the radius where 10/90% of the total energy is encircled, whereas the top-hat sigma settings are defined at the edges (0/100%) of the encircled energy.
Flash word line mask pattern
Figure 3 shows the features from the flash word line mask pattern that were evaluated and their design CDs: The central space (SG-SG); SG; word line 1, 2 and 7; and the spaces between SG and WL1 (SP0) and between WL1 and WL2 (SP1). All seven features were measured with AIMS at mask level and with CD-SEM at wafer level in resist on both sides of the central space (only the right side is shown). WL7 is regarded as fully dense and is used as reference to print the gate layer on target. A basic optical proximity correction (OPC) treatment has been applied using the thin mask approximation (Kirchhoff) and a lumped parameter resist model. The feature-center placement errors of the SGs, WL1 and WL2 have been determined from the CD-SEM and AIMS measurements through focus. The center of the central space has been chosen as reference. A positive shift indicates that the line moves away from the center (i.e., in the positive X direction for features on the right side of the central space and in the negative X direction for features on the left side of the central space). Conversely, a negative shift indicates that the line moves toward the central space.
| 3. AIMS and CD-SEM evaluated several features from the flash word line mask pattern (upper left): central space (SG-SG), select gates (SG), space between select gate and first word line (SP0), word line 1 (WL1), first space (SP1), second word line (WL2) and the seventh word line (WL7). Definition of feature-center shift (upper right): A positive shift moves the line away from the center; a negative shift toward the center. The evaluated features are SG, WL1 and WL2. The table details the design CDs at wafer level for each of the features. |
Figure 4 shows the aerial image contour of the flash word line mask pattern that was measured with AIMS on top of a CD-SEM image of the identical module. This picture exemplifies that the AIMS 45-193i can accurately capture the optical effects of a complex 2-D structure, closely matching the final image in resist.
| 4. For an aerial image contour of an SP0 1.5 design, AIMS measurements are shown by the red lines on top of a CD-SEM image. The landing pads are also shown. |
Mask CDU measurements
The qualification of a mask is typically done using mask-level CD-SEM measurements. This gives a good indication of the quality of the mask in terms of linewidth variations and mean-to-nominal CD. However, additional contributors to the CD on the wafer, such as sidewall angle variations, transmission variations and phase-shift errors, remain invisible and must be qualified with other techniques.
With AIMS, it is possible to measure the CD variation on the mask level for a specific illumination mode, including all variations and material imperfections that are not picked up by CD-SEM.4 Furthermore, the AIMS measurement includes the MEEF, which is dependent on the illumination setting and feature size. The effect of resist contrast, resulting in an increased MEEF at the wafer level, cannot be included, although this may become possible in future software releases.
Figure 5 shows across-field CD fingerprints for the binary pattern measured at both mask and wafer levels. To determine the aerial image CDs of the features in the gate layer, first WL7 is evaluated. A threshold value is determined to print WL7 to size (i.e., 40 nm) for the center position on the mask. This threshold is used to calculate the CDs of the other features of interest, both within the word line pattern and across the mask. For all evaluated features, the fingerprints match well, indicating that the CD variation on the mask dominates the CD uniformity on the wafer. This not only holds for the dense word lines and spaces like WL7, WL2 and SP1, but also for semi-dense or semi-isolated features such as WL1, SP0 and the SGs, which are much more sensitive to focus errors on the wafer.
| 5. Across-field CD fingerprints for a 40 nm half-pitch flash word line pattern were measured on the wafer with CD-SEM and at the mask level with AIMS. The X and Y axis of the field fingerprints are indicated in millimeters; the color axis represents the CD in nanometers. |
Figure 6 shows the mean CD and across-field CD uniformity corresponding to the features evaluated in Figure 5. The feature sizes in the gate layer measured by CD-SEM and AIMS correspond well to each other. The maximum deviation was found to be 4.4 nm, which is surprisingly small given that no effort has been made to calibrate the metrology tools, the CD-SEM measurements are done in resist while the AIMS measurements are done in air, and the measurements were not performed on exactly the same location in the flash structure (same feature but not the exact same X-Y coordinate). The across-field CD uniformity of the word line pattern on the wafer is larger than the mask CD uniformity, but this is as expected because the wafer measurements include across-field dose and focus errors from the scanner and a higher MEEF caused by the photoresist layer. However, the graph clearly shows that the mask CD uniformity is the dominant contributor to the final CD uniformity on the wafer. CD uniformity differences between the features in resist closely follow the CD uniformity differences at the mask level.
| 6. Mean CD and across-field CD uniformity of binary 40 nm half-pitch flash word lines were measured with AIMS at the mask level (numbers are at 1×) and with CD-SEM in resist on the wafer. |
Figure 7 shows the Bossung plots of all evaluated features in the gate layer. To convert the focus values at the mask level to focus values on the wafer level, we used an XT:1900Gi-specific mask-to-wafer conversion factor, which deviates ~10% from the non-paraxial formula5 used in the AIMS software. The focus range over which the select gates show a good image integrity can be measured accurately by AIMS at the mask level. Both the AIMS and CD-SEM measurements report a maximum depth of focus (DoF) of 130 nm for this focus-critical feature. The focus range of the other features is slightly overestimated by AIMS because the aerial image measurements cannot capture resist effects such as pattern collapse, which is one of the DoF-limiting factors on the wafer for the minimum resolution structures. However, the Bossung curvature of the aerial image Bossungs correspond well to the resist Bossungs, although the focus sensitivities are slightly underestimated.
| 7. Bossung plots for all evaluated features show good correlation between mask-level AIMS measurements (left) and wafer-level CD-SEM measurements (right). |
Figure 8 shows the feature-center placement error measured with both techniques. The SGs have a positional shift of 2 nm in best focus, which increases to ~9 nm when going 60 nm out of focus. WL1 goes in the opposite direction of the SGs and shows a shift of -7 nm in best focus and approximately -11 nm when going 60 nm out of focus. WL2 shows the smallest shift — only -4 nm in best focus — which remains almost stable through focus (approximately -5 nm when going 60 nm out of focus).
| 8. This graph shows feature-center placement error measured for SG, WL1 and WL2 by AIMS (left) and CD-SEM (center), as well as the correlation plot between the two (right). |
If we assume that the overlay budget is 20% of the CD node (8 nm in this case), of which a 4 nm error can be tolerated for these intra-module overlay errors, then it becomes clear that the positional shift consumes almost the total overlay budget, leaving no room for additional overlay errors from the scanner and reticle. But if we allocate the full 8 nm overlay budget to the placement errors, then the focus range should still be controlled within ±40 nm (WL1 being the limiting feature). This means that the required focus control is driven by the placement errors through focus rather than CD variations.
Besides the lithographic implications that can be derived from Figure 8, a remarkable agreement between the AIMS measurements on mask level and CD-SEM measurements on wafer can be observed. The correlation plot in Figure 8 shows that the AIMS and CD-SEM data sets for both modules are statistically almost identical, assuming a measurement error of 1.0 nm 1 for CD-SEM and 0.4 nm 1 for AIMS. These measurement errors include local CD variations caused by wafer non-flatness and processing (CD-SEM), local mask CD variations (AIMS) and cumulative measurement errors from the calculation method of the placement error (adding L/S measurements). To arrive at these results, we fitted the original measurement curves with a second-order polynomial fit to get rid of most of the measurement noise, centered best focus for 1:1 comparison, and calibrated the central space (SG-SG) size, which is the reference for the feature-center calculation.
Conclusions
We have shown that the AIMS 45-193i aerial image measurement system is useful for characterizing the imaging performance of flash memory masks. Important lithographic parameters such as CD uniformity, optical proximity effects (OPE), DoF and placement errors can be quantified at the mask level before the mask is actually exposed on a scanner.
By comparing AIMS measurements with wafer prints using an XT:1900Gi scanner, it has been experimentally proven that feature-center placement errors in flash memory designs are optically induced effects, which are inherent to the feature design. AIMS measurements can accurately capture these optically induced mask effects and, therefore, help to distinguish between mask effects and scanner or resist effects in lithography.
The final imaging performance on the wafer is also heavily dependent on the photoresist stack used, which is not taken into account in AIMS measurements. However, the AIMS measurements of MEEF, OPE, exposure latitude (EL) and DoF predict the correct trends and can be used as indications.
Feature-center placement errors induced by the design on the mask can be accurately measured by AIMS and mapped almost perfectly on wafer measurements with CD-SEM. DoF-limiting factors, such as image reversal, can also be accurately measured. Feature-center placement errors have been shown to be inherent to the design of the flash word line mask pattern, consuming a large part of the overlay budget. To prevent impact on the final device performance, this effect should be taken into account in the design of the flash memory layout.
In a related study,6,7 we found that the flash word line mask pattern can be desensitized for placement errors by increasing the space between the SGs and WL1. This also results in a larger EL of the dose-critical features for the attenuated modules and a reduced Bossung curvature of SP0 and WL1 for both the attenuated and binary modules.
| Acknowledgements | ||
| The authors would like to thank Carl Zeiss SMS for supporting the AIMS measurements; and Brid Connolly from AMTC-Toppan for the support, technical suggestions and the effort to get the test mask delivered on very short notice. We furthermore would like to acknowledge the "1900 apps team" for carrying out the exposures on the Twinscan XT:1900Gi, and a special thanks to Mariette Berende-Hoogendijk and other members of the CD-SEM metrology group for the CD-SEM measurements and pictures. | ||
How CD-SEMs Complement Scatterometry
06/01/2009CD Metrology Confidently Looks Beyond 32 nm
03/17/2010CD-SEM: Precision or Accuracy?
03/17/2010Designing Tracks for Better CD Control
08/31/2003
VIEW ALL PODCASTS
