Determining printable defects with MEEF-based mask inspection
09/01/2005
The aim of mask error-enhancement factor (MEEF)-based inspection is to increase photomask yields by adjusting defect specifications to exclude harmless aberrations and small defects that have no impact on printed devices at the wafer level. Using a defect printability score based on MEEF, design software and layout-to-silicon verification technology demonstrate how mask inspection can be optimized and tailored for specific features.
The main difference between a state-of-the-art photomask and a state-of-the-art IC is that the mask has to be absolutely free of defects that will print on the wafer, whereas a certain number of defects are allowable at any process step in IC fabrication. This situation has generated an entire industry of mask defect inspection, repair, and cleaning tools, and has led to the invention and application of the pellicle, which preserves a mask in its pristine state. In spite of heroic efforts in this area, mask yield has been drastically decreasing in recent years, with mask defects being one of the main reasons [1].
For almost every combination of design rule generation and lithographic strategy, there is an experimental or simulation-based evaluation of the size and type of the smallest critical defect. These defect specification techniques have become among the most critical parameters in determining the maskmaking yield and, therefore, mask cost. However, it is still a human call to use the capabilities of mask metrology to find and then eliminate a defect based on its size. And more complex mask processes with their potential for new defect types (e.g., quartz defects in alternating phase-shift masks) make this pass/fail decision increasingly difficult. Lithography simulation tools have been deployed to simulate the effect of a defect identified by a mask inspection tool to the printed image on the wafer [2]. This approach helps in the case of very small defects, which may have no effect at all on the printed image; but in many cases this just shifts the question “does this defect hurt?” to a different level of abstraction.
Managing mask defects
Since their beginnings, mask inspection tools have had the option to designate certain areas on a mask as “Do Not Inspect” regions (DNIR). This technique saves inspection time and eliminates potential issues with defect deposition in DNIRs on masks. It has been particularly useful for dealing with test layouts that violate design rules and generate a large number of so-called nuisance defects. These defects are actually visible on the mask, but are known not to affect the printed image.
In 2003, the concept of “smart inspection” was proposed as an extension to the fully sensitive/zero-sensitive inspection strategy of DNIRs to create multisensitivity-level mask inspection [3]. Under the concept, a smart inspection tool accepts the layout information and compares it to the mask at different data levels, each of which can be assigned a different inspection sensitivity. Using a design rule checking (DRC)/layout manipulation tool to generate the different data levels, this strategy allows, for example, inspection of active gates (defined as features on the gate level overlapping diffusion areas) with higher sensitivity than all other features on the same level. Another possibility is to keep the fill (“dummy”) pattern, which smoothes the feature density for more effective processes, on a different level than the rest of the features, and then inspect all fill features with reduced sensitivity, unless a certain fill feature gets close enough to a main feature.
Taking this concept further, this article outlines a strategy to automatically assign a score of defect printability to each feature element in a mask layout, such that the higher the score the greater the likelihood that a defect will print on the wafer. This score is compiled from the simulation of MEEF of each feature element, using a software tool and all the necessary input data that are used for optical proximity correction (OPC). Applying this strategy to a mask inspection tool with smart inspection capability generates a score specifically tailored to each feature element of the input data.
How MEEF-based mask inspection works
To demonstrate this concept in MEEF-based inspection, the current production and beta capabilities of Calibre software from Mentor Graphics is used. The specific target of this work is to generate marker shapes that identify high-sensitivity and low-sensitivity areas for the defect inspection process. All results are for a hypothetical 65nm process, which uses a 0.75 numerical-aperture (NA) exposure tool with 193nm wavelength illuminated by an annular source to expose features on an attenuated phase-shift mask into a resist/antireflection coating stack on top of silicon. The layout was modified via the insertion of subresolution assist features (SRAF) with subsequent OPC. The OPC tool has also been used to perform the following simulations.
OPC tools optimize the printed image of a given layout by varying the layout piece by piece, and checking the result using a model of the pattern transfer process. The first step of this process is to cut up the features within the layout into so-called fragments. This fragmentation is usually performed according to a complex rule set, but can also be done based on a model. During OPC, each fragment gets assigned certain properties, called tags. The most important tag for OPC is the position of the fragment in the printed image relative to the nominal position; this is edge placement error (EPE). Tags can be the result of a logical operation (“all fragments of layer A, which overlap features of layer B”), a topological operation (“all fragments, which are part of a concave corner”), the optical imaging process (“all fragments, whose slope of the aerial image is smaller than x”), the model (like EPE), or any user-defined logical combination of other tags.
One of the tags is MEEF, a dimensionless number, which compares the change in the width of a printed feature on the wafer as a function of an incremental width change of that feature on the mask [4]:
 |
null
Besides the process window given as depth-of-focus (DOF) vs. exposure latitude, MEEF has become the most important parameter to characterize a lithographic pattern-transfer process. The OPC tool is capable of sorting fragments according to their tag values and performing logical operations such as generating a data output layer in OASIS or GDSII, which contains all fragments within a certain bin.
Figure 1 shows simulation results of mask defect printability based on the width and fragment length of a defect for the hypothetical 65nm process technology. This plot shows the maximum change in linewidth, when a long line on a mask has a fragment of given width but varying length. As long as the fragment length is large, the linewidth change in the printed image is proportional to the width change on the mask, with a MEEF in this case of ~0.7. When the fragment length gets below the characteristic optical length, however, the width change decreases as well.
 |
A small defect on a mask feature prints proportional to MEEF (at least an incremental small defect). So the result of Fig. 1 suggests adding a correction factor to the MEEF reported by the OPC tool, which is constant for large fragment lengths, but increases below - in this case - 200nm. Table 1 provides such a correction factor, derived from Fig. 1. Three classes of feature segments to mask inspection are provided, sorted according to their tendency to print defects. Another binning step is necessary to discriminate between the different MEEF levels. This process is described in Table 2.
 |
This binning strategy has been applied to a real 65nm layout, and a small cutout of the result is shown in Fig. 2. It shows that a potential mask defect between the two wide lines on the left side of this figure (location a) will have a much higher score for printability as a defect on the wafer than a mask defect in the adjoining space to the right of that area (location b). Defect printability factors are influenced by a combination of SRAF elements, linewidths, spaces, and environment of the mask area. Here, the left-side space contains an additional SRAF between the two wide lines, while the other space has no SRAF. By the same token, a mask defect at or near the narrow line in the center (location c) will be less likely to print on the wafer than a mask defect located at or near the wide line (location d). The SRAF/OPC strategy was not optimized to amplify the defect sensitivity of the layout for illustration purposes.
null
MEEF tagging for contact levels
The well-known printing issues of contacts can be summarized as “contacts are four line-ends close together.” Regardless of which edge of the contact is changed, the effect on the printed image is the same.
Based on similar experimental and simulation results, Sturtevant et al. have proposed to define an area MEEF to characterize the printing of contacts [5]:
 |
null
This definition implies that virtually all that counts in printing a contact is its area on the mask. Using this definition, a comparison of MEEF and defect printability is straightforward, as Fig. 3 shows. As Sturtevant et al. [5] show, area MEEF is twice as large as the conventional width MEEF. This could be confirmed using simulation numbers, which yielded the results in Fig. 3. However, this does not change the potential to bin contacts according to their MEEF using the MEEF-tagging capability of the OPC tool. A small cutout of an actual layout treated with this described strategy is depicted in Fig. 4. The difference in MEEF for some contacts in Fig. 4 can be attributed to their different neighborhoods, which are mainly defined by SRAF. Statistical results characterizing the entire design were derived. Figure 5 shows a histogram of MEEF for all contacts in the evaluated layout. The distribution of contacts by MEEF categories can be used to set specifications for mask inspection and repairs.
An actual application of the strategy demonstrated here needs also to take into account the potential spread of feature width on a real mask, as allowed by the mask specifications. This means that three runs need to be performed - one with all contacts at nominal size, and two with all contacts at the upper and lower limits of the mask specification. The output then must be a summary of all three runs, and a contact with a high score in any run has to be placed into the high category for the output.
 Figure 4. Binning of contacts, according to their MEEF. All contacts shown have a MEEF <3.5, except for those in blue. |
null
Summary
MEEF-based mask inspection can help to increase mask yield by adjusting the defect specifications to the actual requirements of the pattern transfer process. This method allows photomask operations to reduce the number of defects to be repaired on a critical mask and thus the number of reworks because of irreparable defects. At least this strategy successfully replaces human judgment with a combination of number-driven tools.
 Figure 5. Histogram of MEEF for contacts in a given layout. |
All the necessary information for this MEEF-based mask inspection is available at tape-out of the design, where resolution-enhancement technology and OPC are applied to the layout data. At that stage, it can be combined with additional information for other mask inspection strategies with variable-sensitivity bins. MEEF-based mask inspection is an important and useful step toward design-for-manufacturability, which, in its widest definition, includes anything that can be done in the design space to ease the manufacturing of ICs.
Acknowledgment
Calibre is a registered trademark of Mentor Graphics Corp.
References
- K. Kimmel, “Mask Industry Assessment 2003,” Proc. 23rd Annual BACUS Symp. on Photomask Technol., Vol. 5256, p. 331, SPIE, 2003.
- Y. Maenaka, N. Takatsu, I. Kagami, D. Kakuta, “Defect Printability Analysis of Attenuated PSM Using PASS,” Proc. 21st Annual BACUS Symp. on Photomask Technol., Vol. 4562, p. 468, 2002.
- W. Volk, C. Hess, W. Ruch, Z. Yu, W. Ma, “Investigation of Smart Inspection of Critical Layer Reticles using Additional Designer Data to Determine Defect Significance,” Proc. 23rd Annual BACUS Symp. on Photomask Technol., Vol. 5256, p. 489, 2003.
- W. Maurer, K. Satoh, D. Samuels, T. Fischer, “Pattern Transfer at k1 = 0.5: Get 0.25mm Lithography Ready for Production,” Proc. Optical Microlithography IX, Vol. 2726, p. 113, 1996.
- J. Sturtevant, J. Opitz, J. Word, “Contact-hole MEEF Comparison between ALTA and 50-KeV Written Masks,” Proc. Optical Microlithography XVI, Vol. 5040, p. 1055, 2003
Wilhelm Maurer is principal engineer of lithography at Infineon Technologies AG, St. Martinstrasse 76, D81609 Munich, Germany; ph 49/89-234-54070, e-mail [email protected].
Steffen Schulze is product marketing manager for Calibre mask data preparation products at Mentor Graphics Corp., 8005 S.W. Boeckman Rd., Wilsonville, OR 97070; ph 503/685-7000, e-mail [email protected].
James Word and Shumay Shang are technical marketing engineers for Calibre RETs at Mentor Graphics.