This article, based on a study conduced at a West Coast IC facility with the assistance of Ultratech Stepper (San Jose), compares stepper and scanner performance. The study revealed that the manufacturer could achieve performance, accuracy, productivity, defectivity, yield, and—ultimately—return-on-investment advantages by converting to 1x stepper technology or even by implementing a mix-and-match strategy. The technology transformation enabled the chipmaker to increase throughput by 12,000 wafer starts in a given time period by lowering mask costs, boosting mask accuracy, improving total overlay, and upping speed binning.

Although the semiconductor industry is advancing toward sub-100-nm design rules, many manufacturing facilities fabricate 0.8-µm and larger devices. Facilities producing 2-µm and larger devices have traditionally used 1x projection, proximity printers, or even contact printers.

Manufacturers realized early on that contact printing creates large yield losses because that method causes masks to come into direct physical contact with the wafer, allowing residues from the lithography process to be transferred to wafers, generating killer defects. To avoid that problem, the industry introduced proximity and projection aligners. However, proximity aligners require accurate gapping, an added complexity that translates into maintenance and productivity issues.

The introduction of 1x projection scanning systems, and especially wafer step-and-repeat systems, allowed for the complete physical decoupling of the wafer and mask or reticle in the printing process, greatly improving productivity and yield. Interestingly, the optical design of both types of systems has the same roots—a unit magnification catadioptric lens. In fact, the stepper's lens design, illustrated in Figure 1, led to the optimization and simplification of the scanner's lens design by removing a secondary mirror. Moreover, because early 1x stepper systems did not have to use a scanning mask-plus-wafer carriage, they were nearly as compact as projection scanners of the same vintage. At many semiconductor manufacturing fabs, these modifications set the stage for a technology strategy in which scanner systems could be replaced with stepper systems having approximately the same footprint.

Figure 1: The stepper's 1x catadioptric lens design has only five critical elements.

The trend toward technology advances is ongoing. The increases in wafer size from 100 to 200 to 300 mm have led to dramatically higher 1x projection mask costs, especially since full-wafer masks must always be one size larger than the printed wafer size. Such large substrates are also subject to considerable mask sagging, introducing across-wafer magnification errors. While 1x step-and-repeat technology is evolving continuously, projection aligners have long since ceased to develop. Steppers have added state-of-the art laser-interferometer stages; fully automated off- and on-axis, through-the-lens alignment systems; high-powered/high-wafer-plane-irradiance illumination systems; and sophisticated multiple-size wafer handlers.

With the advent of optical proximity correction, and especially subresolution assist features, mask technology, even for reduction steppers, had to begin to focus on close to 1x scattering bars. Today, mask writers, processes, metrology, and inspection can easily deal with even <500-nm features. Meanwhile, automatic reticle handling has virtually eliminated reticle or even pellicle breakage.

But what often counts most to the end-user is the cost of ownership (COO) of the lithography setup. Productivity, tool and mask costs, and ultimately yields are primary concerns for manufacturers. The experiment described in this article examines these issues.

To evaluate and quantify the yield and performance differences between the scanner and stepper technologies, the IC manufacturer performed tests on a high-volume MEMS device resembling a regular two-layer-metal MOS device. The tests compared the performance of an SVGL 760 scanner from Silicon Valley Group Lithography (now part of ASML, Veldhoven, The Netherlands) and a Titan stepper from Ultratech Stepper. Start levels were up to 30,000 wafers per month. A 1x full-field mask set was generated for the scanner and a 1x dual-field reticle set was generated for the stepper. Consideration was given to the cleanliness and storage of both the mask and the reticle set.

A wafer lot was split so that half of the wafers were always exposed with the scanner and the other half with the stepper. Since both tools in the actual production environment are interfaced to a different wafer track system, it was decided to use the same resist, coat, and develop process for both sets of wafers; careful exposure dose adjustments were made for each tool to match afterdevelop inspection (ADI) critical dimensions (CDs). The full wafer lot was coated at the same time in a DNS8800 coating and developing track system from Dainippon Screen (Kyoto, Japan) and then split into two halves again. After being exposed, both sets of wafers were recombined for bake and develop to minimize potential CD differences resulting from processing. Mean CD-per-layer data are presented in Figure 2.

Figure 2: Mean CD data per layer from wafers processed in the scanner versus the stepper.

Subsequent processing through etch and film deposition continued according to a normal run plan. A total of 9 layers were processed in this fashion. To quantify the performance of the two exposure tools, metrology steps were performed at 16 different points to gather ADI CD data, afterclean inspection CD data, and overlay data.

After the run was completed, electrical tests were performed on the wafers belonging to each split lot. Three wafers had to be discarded because of processing errors. Histograms were generated to show the yield distribution.

Comparing 1x Masks and 1x Reticles. Although both scanners and steppers use unit magnification (1x), there are significant differences between the manufacturing performance of 1x masks and 1x reticles. Because 1x scanner projection masks have a full-wafer layout, they must be one size larger than the wafer itself. (That requirement poses difficulties for 300-mm wafers.) The study described here used 6-in. (150-mm) wafers and, consequently, 7 1/4-in. masks. The mask write area must be from 4 to 16 times larger than that of an equivalent 1x reticle.

Cost is driven mainly by write time, which on raster-scan-based mask writers scales with the area covered. Since reticles are smaller than masks, their write times are faster. Although steppers have faster write times than scanners, that benefit is partly minimized because reticles have multiple exposure fields (i.e., multiple levels) on the same mask and a redundant row for inspection purposes, which involve added expenses. Nevertheless, these features, illustrated in Figure 3, enable manufacturers to achieve better yields than masks do.

Figure 3: Comparison between (a) 1x full-field masks with one layer and (b) 1x reticles with two integrated layers and a redundant row for inspection. The dual-field layout of the stepper reticle ultimately results in a cost savings of 26%.

Reticles ultimately lower manufacturing costs because they have greater die-to-die inspection capability than masks and because manufacturers can choose to repair only the best row of die. The full 1x reticle set used to conduct this study consisted of five reticles, while the 1x scanner set had nine masks. Additional savings of close to 30% can be realized by opting for 5 x 5-in. reticles rather than 6 x 6-in. ones or even 7 1/4-in. scanner masks.

Alignment Strategies. Just as full-field masks have a full-wafer-field image while dual-field reticles have an individual die image, scanners rely on a global (full-wafer) alignment methodology while steppers rely on a local (die-by-die) methodology.

The 1x scanner aligns to two sites only per wafer. After flat finding and prealign, the wafer is transferred to the exposure chuck and clamped down via a vacuum system. The subsequent alignment sequence allows for an x-y translational alignment by using two targets (in this case, the top and bottom of the wafer). The operator brings the two alignment targets (typically crosses) under two viewing microscopes, achieving the best fit for the x and y translation as well as a rotational correction, which the system accomplishes by rotating the wafer chuck in theta. Specially designed targets are printed at the first level and replenished for all subsequent critical alignment levels (e.g., polyxisland, contactxpoly, and metalxcontact). Although this alignment methodology at best achieves a global correction for translation and a limited correction for wafer-scale magnification and rotation, it does not address rotation or orthogonality, both of which are important in a mix-and-match environment, where the scanner has to align back to a stepper level.

Figure 4: Schematic of the machine vision system, which uses two digital CCD viewing cameras to superimpose an image of a direct-referenced reticle key in through-the-lens mode onto the actual target on the wafer.

In contrast, the stepper's machine vision system (MVS) uses two digital CCD viewing cameras to superimpose an image of a direct-referenced reticle key in through-the-lens mode onto the actual target on the wafer, as depicted in the schematic in Figure 4. Any offset between the two cameras is calculated and applied as a postalignment motion correction after the completed sequence. The user can employ a minimal three-point global alignment method (x,y-translation, scaling, rotation, orthogonality), full die-by-die alignment, or a hybrid, enhanced global alignment (EGA) system, where five or more targets are chosen to enhance the statistical validity of the sites sampled. Figure 5, a comparison between global- and local-alignment strategies, highlights that EGA allows for the best grid error correction.

Figure 5: Comparison between (a) the global-alignment strategy of the scanner mask and (b) the local-alignment strategy of the stepper reticle. The stepper's enhanced global alignment capability results in the best grid-error correction.

Stepper-alignment systems allow far more flexibility than scanner-alignment systems. With MVS, the alignment targets chosen by the user can be either designed targets or virtually any sufficiently small artifact pattern out of the scribe line or even the device, such as die corners, dicing targets, or logos. Uniqueness and image contrast are the important parameters, as illustrated in Figure 6.

In the experiment discussed here, the alignment sequence and targets were kept constant between the scanner and stepper—that is, targets for the stepper were replenished at the same levels as for the scanner. Subsequently, in normal operating mode, the customer used a mix-and-match strategy, where the first level was exposed on the scanner and all critical levels were exposed on the stepper. The use of two complementary mask/reticle sets enhanced the flexibility and redundancy of the high-start-level operation.

Figure 6: Stepper alignment targets, which can be virtually any sufficiently small artifact pattern out of the scribe line or even the device, such as die corners, dicing targets, or logos.

To gauge the alignment results, an automatic overlay measurement tool read five sites per wafer on five wafers for each of the critical levels. As expected, the stepper MVS EGA alignment mode performed 47% better than the scanner's global-alignment mode (Figure 7). While system modifications were not performed prior to conducting the experiment, wafer magnification for both systems was carefully checked before production runs were begun. The optimized set-up would have resulted in an even greater improvement in total overlay than was achieved during the experiment.

Figure 7: Comparison between the registration values of the scanner and those of the stepper.

Yield Improvements. Since the mask and alignment technology of the 1x stepper performed better than that of the scanner, it was of interest to see if the stepper's technological advantages would be reflected in yield data. To determine the yield differences between the two systems, the finished split lots of wafers were probed separately at electrical testing.

Figure 8: Comparison between the wafer yields of the scanner and those of the stepper.

The customer data in Figure 8 indicate that the scanner achieved an average yield of 91.1%, while the stepper achieved an average yield of 94.7% with a tighter distribution (i.e., a higher binning level in the 94th percentile range). That 3.36% difference translates into a wafer-start savings of more than 12,000 wafers/ year, which for the device under investigation represents approximately 10 manufacturing days. Coupled with the savings that result from using a reticle set instead of a mask set, most fabs would enjoy a COO advantage with a short (6- to 9-month) return on investment by replacing scanners with steppers.

A carefully designed split-lot experiment conducted by a West Coast manufacturer of MEMS devices demonstrated that there were distinct advantages to shifting from 1x scanners to 1x steppers. Although both systems employ very similar projection technology based on a catadioptric lens (a principle that is being revisited by virtually all stepper manufacturers for their 157-nm lithography lens designs), the stepper's field-by-field reticle exposure technology has proven to be superior to the scanner's traditional full-wafer exposure and global-alignment technology. Four- to sixfold reticle write-time advantages, combined with superior alignment capabilities and ultimately higher yields, make 1x steppers with their 100-plus wafer-per-hour throughputs an attractive lithography option for high-volume manufacturing.