Mix-and-Match: A Necessary Choice
Ruth DeJule, Associate Editor -- Semiconductor International, 2/1/2000
Five years
ago, the capability of 365 nm, step and repeat lithography tools easily met the
200 nm overlay requirements. The sequence of tools used to expose successive
layers made little difference. One group of steppers of the same brand and model
number with the best lens match among them accommodated all levels. DUV step and
scan systems (scanners) had been introduced, though initially they were used at
only two or three critical layers — not enough to impact the fab’s overall
productivity. Today, as design rules dip below 0.18 µm, DUV scanners are used at
all critical levels with overlay requirements approaching 50 nm. On a typical
device, one-third of the levels are truly critical, one-third moderately
critical and one-third non-critical. While there is some argument for using all
DUV scanners, it’s too expensive for most applications because the exposure
tools capable of handling these critical levels are far more costly than tools
designed for less critical levels. DUV scanners are approaching $10 M and i-line
steppers about half that. The consensus is that mix-and-match is needed to
divert the expense of using critical level lithography for all device layers.
Tighter overlay
Critical levels generally refer to resolution. For example, for a 150 nm process, critical levels contain 150 to 180 nm feature sizes, moderately critical levels 80 to 250 nm features, and the rest are non-critical. However, non-critical in terms of feature size does not necessarily mean non-critical in terms of overlay specifications. Overlay specifications are typically 30% of critical dimension (CD), 45 nm maximum error for 150 nm geometries. But even non-critical layers at a quarter micron might still need to maintain a 45 nm overlay for optimum device packing densities. So in addition to all critical feature size layers, most of the mid-critical feature size layers and perhaps even a couple of the non-critical ones may be critical overlay levels. Therefore, the arbitrary use of an old stepper for non-critical layers may no longer be acceptable for mix-and-match to new tools, according to Chris Mack, president of FINLE (Austin, Texas).
|
| A 4X overlay reticle, used for mix-and-match of steppers and scanners, is clamped on the reticle table of an advanced exposure tool. (Source: ASML) |
Three fundamental components of overlay are the alignment target detection capability, the stage positioning accuracy and precision, and the difference in lens distortion between two tools used to expose overlaying process layers (see Lithography News, p. 52). Alignment target detection became a show-stopper for many exposure tools with the proliferation of CMP levels, where very planarized metal layers present considerable challenges to finding and measuring a target’s position. The performance of stages has advanced such that the moving standard deviation (MSD) is nearing specifications of 15 nm. While high performance in each of these two categories is prerequisite to achieving current overlay performance requirements, optical exposure field distortion is fast becoming the most critical of the three components, according to Steve Slonaker, senior technology engineer at Nikon Precision (Belmont, Calif.). With the prerequisites in place, lens distortion essentially determines the ability or inability to mix and match within a given overlay requirement.
Lens distortion
Each lens has a characteristic distortion map, a placement error as a function of exposure field position. If every layer of a device is processed using the same lens, that error becomes unimportant because only the placement error relative to the next level matters. If one distortion map has a 40 nm error pointing radially away from the center of the lens and a second lens system has a 35 nm distortion error pointing in roughly the same direction, the overlay error is 5 nm. Another tool with a 40 nm error pointing in the opposite direction will result in 80 nm combined error. Due mainly to this issue of distortion mismatch, total machine-to-machine overlay errors may be 50% to 100% more than machine-to-itself overlay errors. Tool manufacturers have managed to get non-correctable distortions on scanners down to the 20 nm range even as lens complexity increases.
In a stepper, a square field, typically 22 x 22 mm, is fully illuminated and stepped across the wafer.
|
| 1. A single matching reticle containing 4X and 5X features can be used for non-concentric matching. The 4X tool exposes the entire pattern in one pass while the 5X tool only a quarter of the pattern, stepping four times. (Source: Benchmark Technologies) |
The characteristic lens distortion patterns are circularly symmetric placement errors, with the magnitude of the error being some function of the radial position in the exposure field. Toward the corners of the square field where the radial distance from center is greatest, the magnitude of distortion error often increases dramatically. A vector plot of such an error distribution yields the well-known pincushion or barrel pattern. In contrast, the scanner field, typically 26 x 33 mm, has a linear distortion pattern. In this case, a slit, typically ~8 mm high and the width of the exposure field, may have vectors running up and down to varying degrees. The slit and optics are fixed, and the stage moves the wafer through the illuminated slit region, averaging out the distortions in the direction of the scan. So placement distortion may appear along the length of the slit but not in the direction of the scan, according to Harry Sewell, director of technology for SVG Lithography (Wilton, Conn.). With the small optical field, optical assembly that typically involves 20 to 25 pieces of glass becomes easier, so it’s easier to minimize optical distortion, he added.
Matching aids
|
| 2. When a scanner attempts to align to the linear overlay error pattern laid down by a stepper (l), what appears to be a lens distortion pattern (r) is actually the interference of the non-concentric configuration and the linear error on the earlier level. (Source: NVS) |
To meet tight overlay budgets, careful matching of tool combinations from one level to the next is critical. For 10 exposure tools, there are n(n-1)/2 or 45 possible tool combinations. For 40 tools, there are 780 combinations. In addition, as you change the numerical aperture (NA) or other lens characteristics like annular illumination or partial coherence, the lens distortion may change, further adding to the complexity. To optimize the matching of exposure tools, matching reticles with specific overlay patterns are used to calibrate each system, and algorithms have been developed that analyze overlay data. Nikon and ASML systems have self-metrology capability, where the tools measure their own distortion patterns within the equipment. To calibrate each system, a matching reticle is used to fabricate a reference wafer (also called golden or zero wafer). This wafer, typically an etched wafer with a very dense array of overlay patterns across the wafer, is used as a yardstick and undergoes lithography and exposure on each tool. Overlay is then measured. Reference wafer and overlay data are input into software packages such as New Vision Systems’ LEMSYS, which compares and optimizes every combination of lens/stepper pairs in the fab.
Field matching
Steppers typically are 5X reduction tools with a 22 x 22 mm field size, while scanners are 4X tools with 26 x 33 mm fields. Field matching can be achieved concentrically, non-concentrically rotated and non-concentrically non-rotated. In concentric matching, only the common field area between the stepper and scanner is available for lithography. To improve overall matching performance, placing both 4X and 5X features on the same reticle can eliminate reticle-to-reticle errors (Fig. 1). Though concentric matching is a straightforward approach, “a field usage efficiency as low as 40% for the scanner reduces the scanner throughput well below its potential,” stated John Cossins, U.S. strategic marketing manager at ASML (Tempe, Ariz.).
Better throughput can be achieved with non-concentric matching; however, it requires a more complex evaluation. Because the centers of the fields are not aligned, a coupling of systematic stage and intrafield errors on the scanners and steppers occurs (Fig. 2). Under normal conditions, a stage error on a stepper, such as scaling or orthogonality, is independent of intrafield error on the scanner, such as magnification and reticle rotation, and vice versa. A stage orthogonality error on one tool normally can be corrected on the next tool by making an equal and opposite correction. But with field centers in different places (no common reference), the systematic errors laid down by one system cannot be corrected perfectly by the next system.
Non-concentric matching takes advantage of the larger 4X scanner field size. Some device manufacturers match a single 4X field with two 5X fields. This can be performed in two different ways. One method restricts the 5X field to 16.5 mm in the y direction, placing two 22 x 16.5 5X fields in one 4X field. The other method rotates the wafer 90° between the 5X and 4X exposures, completely filling the larger 4X scanner field with two 16.5 mm x 26 mm 5X fields.
Benchmark Technologies (Lynnfield, Mass.) has designed a matching reticle that introduces a 90° rotation to maximize the amount of area filled on the scanner field. On the wafer patterned with the matching reticle, the 5X stepper prints patterns 16.5 mm in the x direction and 26 mm in the y direction. The 4X scanner prints two of these patterns simultaneously, with the wafer rotated 90° so the resulting pattern is 26 mm in the x direction and 33 mm in the y direction. The pattern is set up to interlock the two stepper fields within the scanner field. Structures are added to ensure proper stitching of the two fields. The matching wafer provides a common reference, so mechanical adjustments can correct for placement or stitching errors. According to Patrick Reynolds, vice president of Benchmark, non-concentric matching with 90° rotation maximizes scanner throughput. Before running process, the reticle is used to characterize the tools, and adjustments are made to match to the closest grid. “To optimize the process for any mix-and-match lithography set, whether all-optical or even optical to NGL, matching-reticles are available or under development,” stated Craig Sager, president of Benchmark.
Alternative philosophies
|
| 3. Using a SEMATECH COO model, a 40 mJ/cm2 DUV scanner was compared with a 200 mJ/cm2 I-line scanner. A 20 x 30 mm field and 49 shots/wafer were assumed. (Source: ASML) |
In terms of cost of ownership, a larger field size means higher throughput, and tighter overlay means high packing densities. As overlay becomes a very critical part of the overall process budget, some fabs may choose to dedicate a critical level to a given tool or tool clusters. Within those clusters, the best match-ups between i-line steppers to DUV scanners are made. But other choices are becoming available (Table). One is an all-scanner approach, i-line scanner to DUV scanner, a scheme chosen by both ASML and Nikon. According to Cossins, the motivation is cost of ownership. I-line scanners take full advantage of the larger scanner field size, but with a low-cost illumination source, easier optics due to relaxed specifications, lower running cost compared to laser systems and less expensive i-line resists. An i-line scanner also eliminates the need for non-concentric matching. Though there are arguments in terms of the potential need for resolution enhancement technologies (RET) such as OPC and phase shift masks, ASML’s COO figures (Fig. 3) indicate i-line scanners are a good investment, roughly 75% of the cost of DUV scanners.
Others see all-DUV scanner fabs as a growing trend using low- and high-end DUV scanners. According to Sewell, all-DUV eliminates costly RET masks and, with the cost of DUV resists progressively decreasing, may have the best COO figures. “Whether to go to i-line scanners or all-DUV is a business decision that is based in part to how you choose to manage your factories and how many different products you’re running,” noted Bernie Roman, of the advanced optical lithography development group for Motorola’s Advanced Products Research and Development Laboratory. Motorola tends to run many technology generations, so it may not necessarily be advantageous to go to all-DUV. However, for fabs producing essentially high volumes of one-generation products, an all-DUV factory may be viable.
|
| 4. Beam positioning finally adjusts to match a distortion signature put down on a previous layer. (Source: Lucent Technologies) |
Companies such as Canon and Nikon have chosen to stick with i-line steppers, bypassing synchronization and vibration issues of scanners. Through a reengineering program in 1997, Canon analyzed all the possible causes of aberrations and distortions due to lens manufacturing and was able to reduce distortion characteristics from 50-70 nm 3 to 5 years ago to 20 and 10 nm today, thus directly benefiting the matching process, according to Phil Ware of Canon. To maintain high throughput between stepper and scanner, Canon is introducing a large field, 26 x 33 mm, 4X i-line stepper, the FPA5500iZ, that is easily converted between 200 and 300 mm wafers.
Matching to NGLs
At the conclusion of International SEMATECH’s three-day NGL workshop, participants indicated a preference for 157 nm, electron projection lithography (EPL) and extreme ultraviolet (EUV) as technologies for the 70 nm technology node. Among optical systems, whether i-line to DUV or DUV to 193 nm or 157 nm, the primary issue is lens distortion and characterization of the distortion profile, noted Doug Anberg, director of product marketing at Ultratech Stepper (San Jose, Calif.). He said no additional mix-and-match issues are expected with Ultratech’s new 157 nm exposure tool, a 10X, small field (2 mm diameter) system for research use. However, if 157 nm production tools are 6X reduction systems, as some speculate, the concern is compatible field sizes, noted Anberg. He questioned whether the 6X field size will be able to be the same as the 4X. For 6X, the size of the reticle may be prohibitive.
|
| 5. Electrostatic multipole fields up to 4 m for x/y direction and up to 200 rad rotation have been demonstrated in pattern locking 8 mm exposure fields. (Source: Infineon) |
EUV and EPL tools are scanning systems that may offer low inherent lens distortion, but they may be subject to intrafield image placement errors due to exposure-induced reticle or wafer heating, according to Roman. Though SCALPEL technology champion Lloyd Harriott, from Lucent Technology’s Bell Labs, acknowledges heating issues, initial studies indicate they may be manageable. And in terms of mix-and-match, SCALPEL may have unexpected advantages. SCALPEL is a sub-field scanner with a relatively small beam, a quarter of a millimeter square. Similar to optical scanners, the electron beam is scanned back and forth electronically and swept through the chip by moving the stage. But because of the size of the beam, it can be finely adjusted to match a distortion signature put down on a previous layer (Fig. 4), noted Harriott. And SCALPEL’s “field size” is determined by the mask. Because of the stripe stitching and sub-field scanning, the size of the chip printable on a wafer is dictated by the size of the mask you can make. Masks made from a 200 mm diameter wafer provide a nominal square field of 25 x 25 mm, said Harriott. Ion projection lithography (IPL) exposes a 12.5 mm square field in one shot and stitches fields together for larger chips. Mix-and-matching may be facilitated with fine electronic steering of the projected ion image of stencil mask patterns, thus eliminating the need for precise wafer positioning during field stitching. To lock onto the pattern, electrostatic multipole fields run current through a solenoid surrounding multipoles integrated into aluminum lens electrodes (Fig. 5). Field composable lens (FCL) electrodes can potentially reduce distortion at the wafer level to virtually zero, noted Rainer Kaesmaier of Infineon, technology champion of the IPL project. He commented that IPL has unique advantages in overlay and mix-and-match over photon-based lithographies. With the continuing device shrinks, tighter overlay bugets and huge tools costs, mix-and-match is an economic necessity, clearly here to stay. The number of available mix-and-match strategies will meet current needs, but beyond optical systems, matching to NGL systems is a future necessity that holds many unknowns. •
ASML
Benchmark Technologies
Canon
FINLE Technologies
New Vision Systems
Nikon
SVGL
Ultratech Stepper
