Maximizing process latitude by specifying via/contact layer reticles
07/01/2003
Overview
Shrinking k1 factors are making via and contact layers more difficult to print with acceptable latitude and low defectivity. A typical method for improving the common process window is to use embedded attenuated phase shifting masks (EAPSM). Even with the improved resolution offered by this technology, small deviations in reticle contact size are producing increasingly severe patterning problems and, at the extreme, missing contacts. An inspection algorithm that measures reticle contact energy was examined.
This technique detected slightly undersized contacts directly corresponding to the coordinates of the repeating wafer defects, yet these same defects had not been detected by standard reticle inspection. A test reticle with programmed over- and undersized contacts was produced and a new reticle specification to detect defective contacts before they are shipped to the wafer fab was created.
Leading-edge technologies must have low defect density if they are to produce final die at acceptable cost. Completely missing contacts or vias are a particularly severe variety of defect since their die kill rate is 100%, but they are easy to detect during die failure analysis. Figure 1 shows a defect scan from an early test vehicle contact layer utilized by Motorola's Dan Noble Center during development of a 90nm node contact process. There are three defect locations: two appear in every field scanned (hard repeater) while one appears with <100% frequency (soft repeater). The hard repeaters are a classic signal of a reticle level defect. Soft repeaters, though, can easily be misclassified and attributed to a random cause. They are particularly insidious because great time and effort can be spent working on nonroot cause troubleshooting and fixes, wasting resources and time.
The mask (Fig. 1) was shipped to Motorola with no defect detections using the most sensitive die-to-die algorithm available at the mask shop. Return and re-inspection failed to find any of the known hard or soft repeater locations being seen on wafer prints. Since wafer locations of the defects were known, it was possible to backtrack bad printed locations to specific reticle locations. Figure 2 shows the result of just such an exercise. The left image is a reticle image of a bitcell that passed inspection and printed well. The center image is a reticle image of a bitcell that passed inspection but, as the right image shows, printed poorly.
 Figure 1. Inspection wafer map showing repeating defects. |
null
The reticle was inspected using the TeraFlux algorithm that measures the total transmitted light energy, or flux, of contacts in a die-to-die mode and flags variances above a user-defined pre-determined threshold. In this case, there were 56 defect detections on the reticle, including the three locations highlighted above (Fig. 1). All 56 detections were true reticle defects, even though their impact on wafer printing was apparently below the ability of the wafer inspection tool to detect. Since the new algorithm was capable of detecting defects that caused such subtle printed defects that were not readily detectable on the wafer, the next step was establishing an appropriate specification for it. The specification needed to be tight enough so that reticles are delivered that will not cause print defects, but not so tight as to cause unnecessary rejects in the mask shop, which will raise cost and increase cycle time.
 Figure 2. Slightly undersized reticle contacts causing repeating wafer defects. |
null
Reticle flux inspection
Traditional die-to-die reticle inspection algorithms compare the shape of each feature to that of an adjacent die. Traditional reticle pattern inspection methods are good at detecting defects on pattern edges. These methods have evolved together with the industry and are very sophisticated. Currently, XPA (eXtreme Performance Algorithm) is an advanced pattern inspection algorithm that enables the TeraStar tool to inspect advanced reticles, achieving an overall 100nm defect detection sensitivity. In contrast, the TeraFlux algorithm sums the pixel energy, or flux, of small, clear features for comparison to a like feature on an adjacent die.
The primary goal of the TeraFlux development program was to create an inspection algorithm to meet the new high MEEF (mask error enhancement factor) sensitivity requirements for contact and via layers. To achieve this, an energy flux approach was chosen. The algorithm was targeted to find all defect types that cause a change in energy flux, including CD variations, semitransparent defects, classical corner, intrusion and extrusion defects, and any combination of them. High productivity was achieved by designing the algorithm to operate in a concurrent inspection mode. Therefore, traditional die-to-die inspection (XPA), contamination inspection (STARlight) and flux energy measurement (TeraFlux) can all be accomplished simultaneously in one inspection pass. To detect defects, the energy flux difference between test and reference contacts is evaluated using the following equation:
 |
null
When inspecting a reticle in die-to-die mode, it is not always known by the inspection system which die is the defective die; therefore, the denominator is the average of the energy intensity of the test and reference die.
Test mask
A new test mask was designed in order to fully characterize the performance of the new flux measurement algorithm [1]. Traditionally, defect magnitudes have been measured in linear units (µm or nm). Showing a contact defect size in linear units, however, does not tell much about what kind of effect the defect will have on the total energy (flux) transmitted through the contact. Same-size defects may produce very different results on wafers depending on the defect type, location, transmissivity, etc. It makes much more sense to express the magnitude of contact defects as a total transmission variation. In so doing, how the defect affects the lithography process is measured more directly. The approach is also very practical since all the possible defects affecting the transmission of energy on contacts get reduced to only one type: transmission defects.
 |
null
To characterize the performance of TeraFlux, a test plate (called Cetus) with programmed defects on contacts was designed and fabricated. Different from previous sensitivity test masks, the Cetus- programmed defects are designed to change the defective contact's area as a percentage of the nominal contact size. The plate is composed of test patterns containing 600 and 800nm contacts to be used with the TeraStar's 150 and 186nm pixels, respectively. A programmed defect die has 13 columns of programmed defects and 20 rows of defect sizes. Starting at 1% transmission difference (defect 1), defects 1–10 change incrementally by 1% at a time. Starting at 11% transmission difference (defect 11), defects 11–20 change incrementally by 2% at a time. Table 1 shows details about defect types and sizes; defect types included in the plate are shown in Fig. 3.
 Figure 3. Programmed defect types on the Cetus plate. |
null
Experimental
The initial focus of this evaluation was to determine if the new inspection algorithm was capable of detecting the defective reticle features. Since the initial inspection results demonstrated that the algorithm could find the repeating defects, as well as many more real defects, it was necessary to quantify the threshold of flux loss that could be tolerated. The goal was to determine what was named the critical flux tolerance (CFT). CFT is defined as the percent transmission loss of a feature that corresponds to the percentage process window loss that can be tolerated. The key point now was to correlate loss of overlapping process window with flux transmission variation. Once it is determined how much loss of process window to assign to the reticle, the maximum amount of flux variation — or the CFT — can be set.
A 193nm 6% attenuated PSM (EAPSM) reticle with the Cetus pattern was manufactured so that the CFT could be determined. Past work has shown that the area of the contact is most important for determining how the feature prints, which makes the Cetus pattern an ideal test vehicle for the study [2]. The wafer target CD was set to 140nm for the 150nm (1x) nominal contacts. An ASML 5500/1100 193nm scanner system was used in conjunction with Motorola's baseline 90nm node contact resist process. Three groups of focus-exposure wafers were exposed at different numerical aperture (NA) settings and the ADI CD process window responses were measured. The NA settings were referred to as high, medium, and low. Varying the NA is well known to affect the process window and, as a result, it was expected to have an impact on the printability of the programmed defects. Since the printable area of the reticle was fairly small, it was possible to expose 17 steps of focus and exposure. Multiple wafers were exposed in each group on wafers with a dielectric film stack appropriate for subsequent post-etch defect analysis.
Once the wafer exposure was complete, the next step was to measure the ADI CD process windows. Attention turned to the uniformly undersized contact (type "B") at the densest pitch available (600nm wafer scale) as it was expected to be the most sensitive. Coarse inspection of the wafers established the extent of focus, exposure and defect severity necessary for measurement. Recipes were then written on a CD SEM to measure this parameter. Since it is well known that SEMs cause 193nm resist shrinkage, great care was taken in the recipe setup to minimize this effect on the resulting measurements.
Wafer results
Process window analysis of the ADI CD SEM data was performed in KlarityCD. The windows were analyzed using an elliptical (as opposed to rectangular) process window and the focus was fixed at "best focus" for all of the analyses. Exposure was also fixed at a value that gave the target CD of 140nm. In order to truly gauge the impact on overlapping process windows, fixing the focus and exposure is an important step. Allowing them to float can result in the reporting of an overly optimistic overlapping process window as the algorithm continually changes best dose as defocus is increased. More importantly, this is not how real lithography processes are run — one typically fixes focus and at most allows dose to vary from lot to lot to hit the target CD. Also, since the best dose and focus is chosen based upon nondefective contacts, it was deemed unrealistic to allow the programmed defects to affect the choice of best dose and focus during the ADI CD process window analysis.
Figure 4 displays the process window results for the three cases of NA examined. These plots show how the percent exposure latitude rolls off as the depth of focus (DOF) requirement increases for a given feature. The focus value at 0% exposure latitude is typically what would be reported as the DOF. Examination of this data shows the expected behavior for this kind of nearly isolated pitch: for the nominal contact, as NA decreases, exposure latitude declines and DOF increases. As NA decreases, the impact of large programmed defects is increasingly severe. For example, at high NA there is some small overlapping window for the 6% defect. The 6% defect does not show up at all on the medium and low NA cases. From this data, it is clear that the worst defect severity that could be tolerated under any circumstance is a 6% defect. Even for this case, allowing a 6% defect implies that nearly 100% of the CD tolerance window is effectively allocated to the reticle — clearly not realistic. For this reason, it was decided to call the 6% defect the point at which the overlapping process window collapses.
 Figure 4. Effects of NA and defect type on process window. |
null
Figure 5 is another plot of the ADI CD. In this case, the indicated curves enclose all values of dose and focus for a given feature that produce in-spec CDs for the high NA case. The heavy solid line is the combination of dose and focus values that are common to all features and which produce in-spec CDs. The dashed horizontal and vertical lines indicate where best dose and focus have been fixed to center in order to produce the required 140nm CD. Examination of this plot indicates that for increasingly severe defects (i.e., increasingly undersized) the process windows shift to higher and higher doses. This leads to effectively eliminating any underexposure latitude — indicated by the small ellipse centered at best dose and focus.
 Figure 5. Overlapping process windows for the high NA case. |
null
Reticle inspection results
The Cetus test mask was inspected 20 consecutive times on a KLA-Tencor SLF87 with both XPA (traditional pattern inspection) and TeraFlux. Figure 6 shows the resulting sensitivity analysis. The shaded areas display the 100% detection region of each defect detection algorithm; TeraFlux displayed superior detection capability vs. the standard XPA algorithm. Wafer print data of defect B (discussed above) shows a dramatic loss of process window (±10% CD tolerance) at the 6% programmed error (actual error size has not been SEM-measured). Because some overlapping process window exists, it was decided to set the process window collapse point at <6% of the programmed error as indicated by the solid red line in Fig. 6. More defect types will be evaluated in the future; however, in order to emphasize the gap in traditional defect detection capability, the >6% programmed error line for all defect types was extrapolated. Defect type M is totally missing contacts, which are captured 100% of the time by both detection algorithms.
 Figure 6. Cetus defect sensitivity comparison of TeraFlux vs. XPA (traditional inspection). |
null
Conclusion
Standard die-to-die inspection techniques will not detect some contact/via reticle defects that can cause wafer print defects for a 90nm node contact process. The impact on the wafer fab of lost yield, delayed time-to-market and additional expense is unacceptable. An inspection algorithm based upon flux variance is capable of detecting the locations causing hard and soft repeating wafer defects — those that the standard inspection missed. Additionally, the flux-measuring algorithm is capable of detecting reticle defects that may cause wafer defects that impact product performance, but are below the ability of wafer inspection tools to find. A procedure to determine how much of a variance in flux can be tolerated before there is an unacceptable impact on the process window was also demonstrated. This technique provides a means to set a reticle flux deviation specification that ensures detection of defects that may cause wafer printing problems, while minimizing unnecessary mask rejects and associated costs at the mask shop.
Kirk Strozewski, Joe Perez, Motorola Dan Noble Center/APRDL, Austin, Texas
Anthony Vacca, Art Klaum, KLA-Tencor, Austin, Texas, Keith Brankner, G2Concepts LLC, Wimberley, Texas
Acknowledgments
The authors would like to acknowledge the critical support given to this project by John Fretwell, Rick Fossum, and Loraine Villegas of Motorola, and also Lantian Wang, Larry Zurbrick, Charika Becker, and Scott Pomeroy of KLA-Tencor. TeraFlux, TeraStar, and STARlight are trademarks of KLA-Tencor.
References
1. Volk, et al., "A New Energy Flux Method for Inspection of Contact Layer Reticles," Proc. SPIE 4754, pp. 499–510, 2002.
2. C. Mack et al., "Lithography Performance of Contact Holes Part II: Simulation of the Effects of Reticle Corner Rounding on Wafer Print Performance," Proc. SPIE 4066, pp. 172–179, 2000.
For more information, contact Kirk Strozewski, 3501 Ed Bluestein Blvd., MD K10, Austin TX, 78721; ph 512/933-8556, fax 512/933-5304, [email protected].