Expanding capabilities in existing fabs with lithography tool-matching
06/01/2000
Frank Goodwin, Infineon Technologies, White Oak Semiconductor, Sandston, Virginia,
Ted Bodine,* ASM Lithography, Tempe, Arizona,
Craig Sager,* Benchmark Technologies, Lynnfield, Massachusetts
*Additional authors are listed in the Acknowledgments.
overview
Rotated nonconcentric matching of advanced lithography tools to an installed base of older tools is emerging as a viable technique for production wafer fabs. The technique involves use of common matching reticles and golden standard wafers. When properly applied, this technique enables improved repeatability of a process from tool-to-tool and from day-to-day. As lithography technology evolves, the matching process facilitates the integration of tools with different capabilities into the fabrication process, optimizing process flow and in turn, increasing process yield, without the high capital outlay required to outfit an entire fab with the most advanced tool.
The high cost of lithography has made mixing and mathing tools with different capabilities a necessary strategy in today's wafer fabrication lines [1, 2]. To use less expensive, lower resolution steppers for some layers and more sophisticated steppers and scanners for others requires that the images the tools produce overlay within a known, specified limit. For the 140nm lithography node, for example, the total allowable overlay budget will be <45nm [3]. Determining where the overlay fits within the limit is called matching.
To match a new lithography tool into an installed base of lower resolution steppers, a dedicated matching pattern is printed on a wafer using one of the lower resolution steppers. The higher resolution tool then prints the pattern onto the same wafer. The difference between the two patterns is a combination of stage errors, lens errors, column errors, and reticle errors. Detailed measurements of the reticle allow reticle errors to be subtracted from the total, leaving only the difference between the two tools. As much as possible, the tools are then adjusted to minimize correctable errors.
 |
The topography of a wafer requires that certain layers match well with the previous layers, while other layers can be printed with a much looser match. To achieve maximum yield using a mix of tools, lithography engineers characterize their tools through the matching process. Table 1 shows an example of matching data used to optimize a process. The matching data in each column of Table 1 represents the maximum x and y translations, and the maximum vector data. This format allows a process engineer to establish a tool path that matches product requirements. If a process layer requires tighter x registration, then the tool that has the best registration in x and meets the minimum specifications for y and maximum vectors would be chosen over a tool with a better overall matching.
 Figure 1. Traditionally, only the center portion of the available 4X field can be actively matched to a 5X-stepper field. |
In the process flow shown in Table 1, the first layer is processed using Tool 4. Of the tools that match well enough to Tool 4 to be used for the second layer, Tool 6 shows the better match, and is the preferred tool for this layer. Tool 13 is also acceptable for the second layer. Four tools match well enough to be considered for the third layer, Tool 13, however, is preferred because on average it matches better to both tools 6 and 13 used on the previous layer. The other three tools are still within specification and may be used if the primary or secondary tools are unavailable. For the fourth layer, where the y registration component is the critical matching parameter, Tool 15 is considered the primary tool to Tools 8, 12, and 15, and Tool 1 the better match to Tool 13.
Why is matching needed?
The ability to match DUV 4X step and scan systems to 5X i-line steppers gives existing fabs the flexibility to move below 300nm processing without replacing their entire line of lithography tools. Many fabs use newly-purchased high-end scanners or steppers for their critical layers and their original, lower-resolution steppers for the less critical layers. The use of existing steppers for less critical layers saves significantly on capital investment.
Another use for matching is to maximize throughput of lithography tools, especially 4X scanners, through an optimized use of the field size. The 5X tools in use today have a maximum square field size of 22mm by 22mm and a maximum rectangular field size of 16.5mm by 26.0mm, half the field size of the 4X scanners. Traditionally, using concentric matching, the optimum portion of the 5X lens, used for device imaging, restricts the exposure field used to match the 5X tool into the 4X scanner. The useable area is generally limited to within a 22mm by 22mm center region of the 5X stepper lens, reducing the 4X tool's potential exposure field (Fig. 1).
Using the entire 4X field instead of just the center 22mm by 22mm portion nearly doubles the throughput of a 4X machine. To do this, however, requires a different matching technique called 90° rotated nonconcentric matching. With this new technique, the 5X tool prints a field that is 16.5mm in the x direction and 26mm in the y direction. Then the 4X scanner prints a field that is exactly twice the size of the 5X field, but rotated 90°, or 26mm in the x direction and 33mm in the y direction. Once the tools are matched so the residual registration errors are known and product overlay can be maintained to within specification, the scanner is capable of imaging a field that is twice as large as the 5X stepper in any single exposure, optimizing the throughput and the utilization of the 4X tool.
The "golden standard"
Matching techniques allow a facility to improve lithography process repeatability from tool to tool and from day to day. Some facilities use matching techniques on a daily basis to match all their tools into a "golden standard." This idea can also work for matching new tools into an existing line. Since reticles exist for nearly all lithography tools in use in today's fabs, there are few limits to tool matching.
 |
A golden standard is a wafer that is created using the tool that produces results closest to the average of all the tools in the fab. Matching to this tool minimizes the difference between any specific tool and the standard. Prior to exposing the standard, critical tool parameters are verified to ensure that the tool is operating at peak performance. The tool parameters include pre-aligner accuracy and repeatability, stage accuracy, and lens distortion. The wafer is then exposed using an appropriate matching reticle and the pattern etched into bare silicon or an oxide layer. Once prepared, the golden wafer can be used over and over to match to this optimized tool or to new tools.
Addressing implementation problems
 |
Traditionally, matching uses a different reticle for each lithography tool. Undocumented, differences between two reticles can appear to be an overlay mismatch between the tools, contributing to errors in the calculated results. To minimize the difference between the reticles, they are written sequentially on the same maskmaker's tool whenever possible. Once made, a registration measurement system thoroughly inspects each reticle and allows the user to document any differences between the measurement marks. Table 2 shows a portion of data generated from a comparison of two matching reticles. In this case, the position of each feature on the first reticle is compared with the corresponding point on the second reticle. The summary data at the bottom shows how closely the two reticles match. This data must then be included in the calculation of the matching error.
Alternatively, some companies are using a single reticle (Fig. 2 on p. 103) in both tools, significantly reducing the error contributed by the reticle.
Examples of matching
It is instructive to consider an example where the practical implementation of matching allowed the use of both 4X and 5X lithography tools in subsequent layers.
 |
For example, ASM Lithography had placed two 4X step-and-scan systems into a fab with existing 5X steppers manufactured by two different toolmakers. The first set of measurements matched one of the ASM 4X scanners to a 5X stepper. In this case, two reticles were used: a single matching reticle for the ASML tools and a second reticle for the 5X tools. The 5X stepper printed the golden wafer, and the ASML scanner printed the same pattern in resist over the golden standard. In this case, the ASML scanner was used as the overlay measurement tool and inspected 25 sites/field and 20 fields/wafer.
 |
Figure 2. A reticle designed specifically for matching 4X to 5X tools. This reticle has a complex matching pattern with many sites that can be measured. Complex reticles offer the opportunity for multiple correlations. One of the challenges for a lithography engineer is to optimize the data to produce the best results in the least amount of time. Usually, a number of sites will be measured at several locations or fields on the wafer. The errors are then averaged for a group of these sites. In some cases, depending upon the results of the statistical analysis, a small number of sites will provide sufficient information for matching.
The data in Table 3 on p.103 show match before any corrections were made to the tools. Directly below that data, are the residuals from this process (see "What are residuals?"). The ASML tools then printed the first critical process onto a set of wafers, and the 5X steppers printed the second layer onto the wafers. The user measured both grid matching error (intrafield) and field matching error (interfield) on the product wafers. The data in Table 4 show how well the matching process corrected both interfield and intrafield errors between the two tools.
The two tools matched to each other — ASML Tool 1 and 5X Tool A3 — have the lowest maximum overlay error. Once the other tools are matched, they should have similar overlay error results. It is clear from Table 4 that all the correctable errors on these two tools (both interfield and intrafield errors) were corrected to produce this small overlay error. From this type of data, the lithography engineer can evaluate the limits on the process generated by the use of this tool set and select the best tools to use for a critical process.
The future of matching
The techniques of rotated nonconcentric matching and using the same reticle to match different tool sets are just now evolving to maximize the matching capability. (Nonconcentric matching has been around for a while. Rotated nonconcentric is new.) In the future, we expect to see these two technologies accepted across a wider base. The most interesting future trend may be to combine these two technologies, providing nonconcentric matching capability on a single reticle.
 |
As we move beyond traditional optical lithography, the capital cost of the tools will almost certainly drive the continued use of mix and match. The benefits derived from tool matching will also grow, so that matching becomes standard practice for most types of device manufacturing. This may create the need to match tools that use different lithography techniques (e.g., matching an x-ray tool to a DUV tool). The sophisticated matching patterns are not bound by lithography technique, however, and we expect they will continue to be used for the foreseeable future.
Conclusion
The matching process makes it possible to take advantage of the most advanced lithography tools where they are needed, while keeping costs low by using an installed base of lower resolution steppers on less critical layers. Matching can also improve the repeatability of a process from tool-to-tool and from day-to-day. Matching tools such as single, common matching reticles and golden standard wafers let lithography engineers match tools accurately and efficiently. As lithography technology evolves, the matching process facilitates the integration of tools with different capabilities into the fabrication process, optimizing process flow and in turn increasing process yield, without the high capital outlay required to outfit an entire fab with the most advanced tool.
Acknowledgments
Additional authors include Patrick Reynolds of Benchmark Technologies and Doris Chin of ASM Lithography.
References
- T. Shafer, M. Bigelow, J. Greeneich, "Optimizing Stepper Mix to Combat Rising Fab Costs," Semiconductor Fabtech, Edition 2, 5-1995.
- F. Goodwin, J. Pellegrini, "Characterizing Overlay Registration of Concentric 5X and 1X Stepper Exposure Fields using Interfield Data," SPIE Vol. 3050, p. 407-417.
- C. J. Gould, F. Goodwin, W. Roberts, "Overlay Measurement: Hidden Error," presented at the SPIE Microlithography 2000 Meeting, 3/2000.
- This analysis technique uses 99.7% of the data, throwing away 0.3% at the high end. If the data conform to a Gaussian distribution, this technique is equivalent to a 3s calculation. In matching, however, the distribution is often non-gaussian, making a 3s calculation less representative of the actual process. ASML uses the 99.7% technique to represent the process under both Gaussian and non-Gaussian conditions.
Frank Goodwin received his BSEE from the University of Massachusetts and his MSEENG from Cornell University. He is the lithography process tool group leader for White Oak Semiconductor, 6000 Technology Blvd., Sandston, VA 23150; ph 804/952-7685, fax 804/952-7683, e-mail [email protected].
Ted Bodine is an applications engineer for ASM Lithography, 8555 South River Parkway, Tempe, AZ 85284; ph 480/383-4356, fax 480/383-3995, e-mail [email protected].
Craig Sager earned a BS in photographic sciences from the Rochester Institute of Technology. Sager is president and co-founder of Benchmark Technologies, 7E Kimball Lane, Lynnfield, MA 01940; ph 781/246-3303, fax 781/246-0308, e-mail [email protected].
The Challenges of Overlay Control
Overlay has always been an underappreciated area in the lithography world. The ability to place images properly and consistently on a wafer just does not have the headline grabbing power of resolution and critical dimension (CD) control. In reality, the ability to control image placement is just as important to yield as CD control and has fewer "tricks," such as wave front engineering to call upon. In fact, many techniques used to improve resolution are detrimental to overlay control. Although the general algorithms and models for calculating unusual mix-and-match implementations have been long known and understood, the art is application of these concepts to practice in a real production environment.
Lloyd C. Litt, Advanced Optical Lithography, Motorola, Semiconductor Products Sector, Advanced Product Research and Development Lab
What are residuals?
The matching process will uncover two different types of errors: correctables and residuals. Correctable errors are errors that can be corrected through machine adjustments. Residual errors are errors that cannot be corrected.
Intrafield (or field) errors are errors that occur within a single exposure field. Interfield (or wafer grid) errors are errors that occur when the tool steps from one field to the next.
Intrafield and interfield errors can be either correctable or residual. Correctable interfield errors include translational errors, non-orthogonality, mirror curvature, and wafer scaling — errors that can be adjusted for by correcting the machine constants. Correctable intrafield errors include magnification, rotation, and trapezoidal or tilt errors.
 A vector map showing residual errors between two different lithography tools. (Source: White Oak Semiconductor) |
Advanced lithography tools, such as the ASML PAS 5500 step-and-scan systems, also include correction techniques for third-order distortion. The most advanced ASML scanners also include correction techniques for asymmetric rotation and asymmetric scaling. Noncorrectable lens errors and a few interfield errors make up the residuals.
Vector maps make it easy to visualize residual errors (see figure). Each arrow indicates the translational error in that particular location within the field. The arrow points in the direction of the error, and the larger the arrow, the larger the error. In this case, the vectors indicate residuals between an ASM Lithography 4X scanner and another manufacturer's 5X stepper. By accurately identifying correctable errors, an exposure tool can be adjusted to compensate. On the other hand, identifying and quantifying the re-sidual errors defines the limitations on the process and allows a lithography engineer to choose the tools that match most closely.