Process Equipment Control

Modeling and controlling the effects of base contamination in DUV lithography resists

Devon A. Kinkead and William Goodwin, Extraction Systems; and Karen Turnquest, Advanced Micro Devices

Now that the effects and probability of base contamination have been quantified, fabs can use real-time monitoring to help achieve the stringent CD control budgets of the SIA roadmap.

The airborne base contamination of fast, chemically amplified photoresists is a yield-limiting factor in atmospheric pressure, deep-UV lithography and will remain so as device features continue to shrink. Lithography-induced critical dimension (CD) variations will have a particularly acute effect on device characteristics in the 150-nm technology node and beyond.¹ The standard deviation of propagation delay times for CMOS-based ring oscillators, for example, will increase from 1% for 300-nm devices to 20% for 250-nm devices as a result of variations in gate oxide, impurity, and gate length. The impact of gate length variation will account for a remarkable 80% of the effect on devices below 200 nm. To contain this impact, the 1999 edition of The International Technology Roadmap for Semiconductors (ITRS) stipulates a postetch budget value of 15 nm for CD variation in 150-nm devices.² Recent quantitative studies on the effect of airborne molecular base contamination on CD variation found that exposure to contaminant levels of 15 ppb caused a CD change of 6 nm/ min.^3,4 Thus, an intratool 15-ppb concentration spike in an interfaced lithography cluster could lead to a postdevelop CD growth of 6 nm (assuming a throughput of 60 wafers/hr), which would consume 40% of the available postetch CD budget. This level of budget consumption may not cause killer defects, but it will cause a change in the mean process CD and an increase in CD variation.

The work presented in this article carried the research effort in this area to the next logical step by modeling how contamination can add to intra- and interwafer variation as well as lot-to-lot variation. The article also analyzes the economics of a monitoring strategy designed to reduce the CD variation caused by imperfect control of molecular base levels in DUV lithography clusters.

Photoresist Sensitivity

A major difference between the photoresists used in DUV lithography and longer-wavelength resists is the chemical amplification mechanism used to bleach the resists--that is, to make them soluble or insoluble.⁵ Generally associated with a characteristic feature of i-line resists whereby the total absorbance/transmission of the resist layer changes upon exposure, the term bleaching may become less descriptive when it is applied to DUV resists that use a photoacid-based development mechanism. In DUV lithography, throughput is controlled by the activation energy and concentration of the photo-acid generator and the activation energy of the protective group on the polymer resin. Unfortunately, this bleaching mechanism, which involves acid generation, makes the resists susceptible to base contamination, effectively neutralizing the acid catalyst.

The least-controlled sources of base contamination in coat, develop, and exposure tools are airborne molecular emissions from people and processes.^3,6 These parts-per-billion-level fugitive gases, which may include ammonia, n-methyl pyrrolidone, and tetramethyl ammonium hydroxide, permeate the resist polymer according to Henry's law (surface concentration equals solubility of base in the polymer multiplied by the base concentration in air) and change the quantum efficiency of the reaction at the resist's surface/air interface, resulting in a widening of the top of the resist profile, called a "T-top."¹ Chemical filters are employed in DUV lithography equipment to control this kind of contamination effect, but CD changes caused by acid diffusion and substrate poisoning remain an issue.

Documented sources of molecular base contamination in the fab environment include ambient air, other manufacturing process chemicals containing volatile molecular bases, paint, ceiling tiles, and volatile humidifier system boiler additives such as morpholine, cyclohexylamine, and dimethylaminoethanol.⁷ One study reported that additional "hidden," but significant, pollution sources were people, gloves, and caulking.⁸ Of these sources, people were the most significant, generating between 70 and 3800 ppb of continuous molecular base contamination.

Recent postexposure delay studies at IMEC (Leuven, Belgium) using a state-of-the-art 248-nm DUV resist put the CD variation cost of contamination at 6 nm per minute of exposure to a 15-ppb total molecular base concentration.³ Research results reported by International Sematech on 193-nm chemically amplified photoresists revealed similar sensitivities to molecular contamination in both isolated and dense (375-nm-pitch) features.⁴ Examples of thick (490-nm) and thin (153-nm) profiles of K98, manufactured by Shipley (Marlborough, MA), are given in Figures 1 and 2, respectively. Figure 3 shows examples of even thinner (60-nm) profiles of PAR101, a photoresist manufactured by the Sumitomo Chemical fine chemicals division (Osaka, Japan). The tops and bottoms of the profiles shown in Figure 3 increase at about the same rate. After 20 minutes of exposure to 5 ppb of molecular contamination in a chamber, there was an ~17% increase in CD for each measurement, which equals a CD growth of approximately 1.2 nm/min.

 

a	b
Figure 1: Examples of thick (490-nm) K98 profiles: (a) dense and (b) isolated. Dense features have a 375-nm pitch.

a	b
Figure 2: Examples of thin (153-nm) K98 profiles: (a) dense and (b) isolated. Dense features have a 375-nm pitch.

a	b
Figure 3: Examples of 60-nm-thick PAR101 profiles: (a) dense and (b) isolated.

Graphs of CD growth versus time for the thick K98 profiles with dense and isolated 150-nm features are given in Figures 4 and 5. Assuming that the CD growth rates are linear with increasing concentration (Henry's law), these results are similar to those (within the error bars) obtained in the research at IMEC. International Sematech, however, also found that the thinner (153-nm) resists were about three times more sensitive to contamination than the thicker (490-nm) resists. The ITRS projects that resist thickness will drop from the present range of 540­720 nm for the 180-nm technology node to 390­520 nm for the 130-nm technology node.⁹ This 28% reduction in resist thickness, independent of other factors, will increase contamination sensitivity, which, when placed against a tightening CD variation tolerance, will require better management of resist contamination.

Figure 4: CD growth of dense 150-nm features versus time in a chamber contaminated with 5-ppb of NH3 for thick K98 profiles.

Figure 5: CD growth of isolated 150-nm features versus time in a contaminated chamber for thick K98 profiles.

Contamination Event Probability

The development of an intelligent strategy to control the effects of resist contamination necessarily involves some discussion of event probability. Almost 100% of DUV lithography clusters are chemically filtered to preclude contamination-related problems. The industry assumes that these filters will maintain contamination levels below 1 ppb; thus, if a contamination event does occur, it is attributed to filter failure. Because less than 5% of the total installed base of lithography clusters include monitoring systems, little data to the contrary have been available. However, data recently collected for both R&D and production tools reveal that faulty filters may not be the only source of contaminant spikes. Figure 6 shows a 51-hour contamination event in a 248-nm R&D lithography cluster. The transient nature of the event itself and the measured output of the filter over time indicate that the problem was not caused by a degraded or spent filter. If this event had been caused by a spent filter, the tool's contamination levels would not have returned to normal.

 

Figure 6: Total molecular base concentration at various times in a 248-nm lithography cleanroom and process tools, including a 51-hour period during which the stepper showed contamination levels equal to those of the cleanroom.

In a larger study of production tools, real-time total molecular base (TMB) monitoring of multiple DUV clusters in a production fab yielded a contamination frequency of 0.8%; that is, 0.8% of the 13,118 samples collected from scanners, steppers, and tracks had contamination levels of more than 5 ppb. The DUV lithography bays had an overall mean total TMB concentration of 21 ppb (with a standard deviation of 16) and a maximum value of 61 ppb, based on 966 samples. The intratool TMB levels in production steppers and step-and-scan tools were quite low, with a maximum of 13.2 ppb and a mean of 0.6 ppb, indicating that the exposure tools were generally under control. However, 0.7% of the time (in 48 of 6929 samples), the TMB concentration inside a stepper or scanner exceeded 5 ppb. Concentration levels in coat and develop tracks were higher than those in the exposure tools: the mean was 1 ppb, the maximum was 24 ppb, and the probability of a >5-ppb event was 6%. The sample total of 7299 for these tools included 367 data points taken during a filter failure. If the filter-failure counts are excluded, the probability of a >5-ppb event drops to 1%, or about the same as the probability for steppers and scanners. These tool results are summarized in Table I.

 

Result type Steppers/ Scanners Tracks (total) Tracks (without filter failure data) Mean (ppb total molecular base) 0.6 1 1 Standard deviation () 1.7 2 2 3 5.0 7 5 Maximum (ppb) 13.2 24 19 Contamination events > 5 ppb (no.) 48 426 72 Sample count 6929 7299 6932 Probability of a > 5-ppb event (%) 0.7 6 1

Table I: Results of real-time monitoring of intratool total molecular base contamination levels in a production environment.

This TMB monitoring experiment revealed that, on average, about one wafer in 100 was exposed to contamination levels greater than 5 ppb and these events were not caused by filter failure. This conclusion raises several questions: What is the yield implication of having one wafer out of every 100 experiencing a 1­6-nm/min CD growth caused by contaminant exposure during DUV processing? What is the impact of processing wafers with typical TMB exposure levels of 2 ppb? And what can happen if wafer exposure time to a given contaminant concentration is not constant?

The Impact of Contamination on Variation and of Variation on Yield

Measurements on production lots have shown that resist TMB levels as low as 1­2 ppb affect CD control. It is therefore essential to immediately catch an event that increases these levels. Although many factors can affect the length of time wafers are exposed to TMB contamination and the point at which exposure occurs, the timing of the wafers' passage through the DUV lithography cluster has been shown to have the most significant effect on the resulting CD signature.

The timing through the lithography cell itself can be affected by several factors. One of the most important is the time that elapses between the last flash and the initiation of the postexposure bake, because the resulting CD corresponds directly to the amount of postexposure delay, as seen in Figure 7. Depending on the type of stepper and track used and how track programs are set up, the delay time may vary significantly. Track engineers have taken great pains to reduce this variability, but it is still not uncommon for the delay to vary between 30 seconds and 2 minutes. Such fluctuations can result in as much as a 20-nm range in wafer-to-wafer mean variation, or 6­7 nm, 1 mean variation in a standard 25-wafer lot. A variation of 20 nm represents a significant part of the ITRS CD control budget for 280- or 300-nm features. Less CD variation has been detected when more environmentally stable resists are used to process 150­180-nm features, but the variability still represents a significant fraction of the CD control budget. Sometimes intralot, mean wafer-to-wafer variation can exceed 4 nm, 1 variation in production, which is truly significant when processing images of this size.

 

Figure 7: A typical relationship between postexposure bake (PEB) delay and resulting CD variation based on wafer position.

Other factors besides the delay between the last flash and postexposure bake can affect DUV process timing, making it important to understand contamination control as a systemwide concern. For example, a lot processed on a tool that is fully loaded and running in steady state can easily have a different intralot CD range than a lot that was run individually. This variation is largely due to the fact that lithography tools prioritize their moves, depending on whether they are fully loaded or not. Thus, in some cases, a wafer may be held in a "wait" state before, or right after, exposure. Very often, plots of CD as a function of wafer processing order show repetitive patterns, which represent how the tool consistently prioritizes its moves and show where delays between certain operations can be expected, based on relative wafer positions. Each fab must characterize its own tools and determine how to reduce the module-to-module time variation.

Assists (unplanned interruptions that require human intervention) also have an impact. During such events, in-process wafers remain in the lithography tool, where they are exposed to the contamination levels in their respective process areas. Depending on where the wafers are in the process flow, wafer-to-wafer CD variability can be dramatic. For example, a 10-minute pause can easily cause a spike in CD variation for a wafer that has just finished exposure and is not yet baked. If the pause occurs when the wafer is in an area where the contamination levels are as high as 5 ppb, a CD shift of as much as 10­20 nm may occur. Then, if that wafer is not chosen as part of the CD sample, it will continue on to the next process step and ultimately produce die with notably different performance characteristics than the rest of the lot. On the other hand, if the wafer does become part of the sample, an engineer who was not aware that an assist caused the deviation may make an improper assessment of whether the process is running in control.

The studies under discussion here detected very little across-wafer CD variation following exposure to 2-ppb TMB contamination. Most variability occurred in the wafer-to-wafer mean CD and, consequently, lot-to-lot CD, and depended on whether the lots were processed on fully loaded tools or individually.

Tight distributions in lot performance will always be one of the major objectives in lithography. While variations may be related to several processing issues, CD variation caused by TMB contamination exposure can have an impact on such distributions. If the mean times between failures and assists are low, if spikes in contamination levels occur frequently, and if the tool is not always loaded and in steady state, it is quite possible that the wafers sampled after photolithography and after etch will not represent the lot as a whole, leading to inaccurate assessments on whether the process is in control. Such variable results can also cause deviations from the expected device performance at the end of the line. Overall, variations in linewidth can affect distributions in maximum frequency (F^_max), drive currents, threshold voltages, leakage, and other important device parameters.

TMB Monitoring

Recently introduced metrology equipment, such as the monitor from Extraction Systems (Franklin, MA) shown in Figure 8, can measure intratool TMB concentrations in real time without interference and with parts-per-billion-level sensitivity. These tools permit fabs to quantify and control the intraprocess variation of molecular base contamination in DUV resists and thus reduce its effects on CD variation. Because real-time monitoring of TMB levels is a relatively new technology, its cost-effectiveness requires scrutiny. In general terms, the solution (i.e., monitoring) cannot be more costly than the problem (defined as the probability that the contamination will occur multiplied by the cost of an event when it does occur). The effect of a 15-ppb contamination event on CD variation is the product of the actual effect—6 nm per minute of exposure—and the probability that an event will occur—about 1%— which equals 0.06 nm in added variation. Estimates of the dollar cost of such variation have been reported elsewhere and are based on proportional relationships between the gate CD and average die selling price.^10–12 The cost of monitoring is $0.12/wafer.

 

Figure 8: Monitoring system designed for in situ real-time measurement of total molecular base concentrations in DUV lithography resists.

Figure 9: The relationship between the penalty for CD variation, the number of die per wafer, and the ratio of the cost of monitoring to the cost of variation. In situ monitoring only makes economic sense when the ratio of metrology cost to CD variation cost is less than 1.000.

Using these data, the economic justification for monitoring can be assessed by considering the relationship between three parameters: the economic penalty for increased CD variation, the number of die per wafer, and the ratio of the cost of monitoring to the cost of variation. This relationship is shown graphically in Figure 9. Three conclusions can be drawn from the figure:

• Monitoring does not make economic sense when the penalty for an increase in CD variation is less than $0.20/nm/die and there are less than 20 die per wafer.
• Assuming that measuring a process-limiting parameter would enable a reduction in process variation of at least 10%, an investment in monitoring equipment would yield a return whenever the ratio of metrology cost to variation cost is less than 0.10.
• An investment in monitoring would also be justified when both the CD variation penalty and the number of die per wafer are high.

Figure 10 compares the monitoring cost per die for three die densities with the penalty for variation. The vertical distances between the penalty data points and the monitoring-cost data points represent the return on investment under those conditions. Because the ratio of metrology cost to CD variation cost in all cases is less than 0.001, the investment in a monitoring system would be justified. Indeed, a 10 percentage point reduction in CD variation would be possible with a monitoring cost of only 1/100 that amount. In sum, in situ real-time monitoring of molecular bases makes economic sense under real-world conditions.³

 

Figure 10: Monitoring cost per die for various die densities compared with the penalty for CD variation. The return on investment in monitoring equipment improves as the penalty for CD variation and the number of die per wafer increase.

Conclusion

Research at International Sematech has shown that thinner resists are equally or more contamination sensitive than today's 248-nm resists, which means that airborne molecular base contamination will remain a yield-limiting factor into the foreseeable future. The CD variation cost of a typical 15-ppb intratool contamination spike with a 248-nm resist is 6 nm per minute of exposure, which is equal to the total ITRS postetch gate CD control budget forecast for 2005. Consequently, if that budget is to be met, improvements will be required. Because both the probability and effect of base contamination events have now been reasonably quantified, real-time TMB monitoring can be used to help achieve those improvements. Indeed, it seems safe to predict that the combination of constricting CD budgets and shrinking time allowances to trend and source problems will drive in situ real-time monitoring of molecular bases onto the production floor. The economics of this strategy are compelling.

References

1. F Matsuoka, "Concerns Respecting Lithography for 150-nm and Below Devices," in Proceedings of SEMI Technology Symposium '99 (Mountain View, CA: SEMI, 1999), 6-3­6-9.
2. The International Technology Roadmap for Semiconductors (San Jose: SIA, 1999), 147.
3. DA Kinkead and M Ercken, "Progress in Qualifying and Quantifying the Airborne Base Sensitivity of Modern Chemically Amplified DUV Photoresists" (paper presented at SPIE 2000, Santa Clara, CA, February 28­March 3, 2000; also to be published in European Semiconductor).
4. O Kishkovich and K Dean, "Environmental Stability of Chemically Amplified Resists: Proposing an Industry Standard Methodology for Testing" (paper presented at SPIE 2000, Santa Clara, CA, February 28­March 3, 2000).
5. CG Willson, "Resists for DUV Lithography" (short course presented at Microlithography 99, Santa Clara, CA, March 18, 1999). The discussion that follows is also based on this source.
6. DA Kinkead, "The Value of Airborne Base Contamination Measurement in DUV Lithography," Microlithography World 8, no, 4 (1999): 22­25.
7. DG Baldwin, Chemical Safety Handbook for Semiconductor/Electronics Industry, 2nd ed. (Beverly, MA: OEM Press, 1998).
8. O Kishkovich and CE Larson, "Amine Control for DUV Lithography: Identifying Hidden Sources" (paper presented at SPIE 2000, Santa Clara, CA, February 28­March 3, 2000).
9. The International Technology Roadmap for Semiconductors (San Jose: SIA, 1999), Table 40a.
10. J Sturtevant, "Manufacturing Implementation of a Feedback Controller for CD and Overlay," Microlithography World 8, no. 3 (1999): 22.
11. H de Haas, "CD Control Calculations," unpublished memorandum.
12. ME Preil and HJ Levinson, "Yield Limiting Issues in Deep UV Lithography," Microlithography World 7, no. 2 (1998): 22.

Devon A. Kinkead is a founder of Extraction Systems (Franklin, MA). He has years of experience as a project manager in the field of airborne molecular contamination (AMC) control and has managed measurement projects involving process tools and cleanrooms. He has contributed information on AMC-related issues to the SIA roadmap and was the principal author of Sematech's 1995 document "Forecast of Airborne Molecular Contamination Limits for the 0.25 Micron High Performance Logic Process." A member of SEMI, Kinkead chaired the SEMI Molecular Air Contaminants Task Force, which created SEMI standard F-21-95. He develops patents and publishes extensively in the field of AMC measurement and control. He earned a BS in biology/chemistry from the Claremont Colleges in Claremont, CA. (Kinkead can be reached at 508/553-3900, ext. 14, or dkinkead@extractionsystemsinc.com.)

William Goodwin is director of manufacturing and engineering at Extraction Systems, where he is responsible for the development and manufacture of trace gas monitoring systems. He also has experience in the design and installation of cleanroom-compatible AMC treatment equipment in wafer fabs in the United States and abroad. Before joining the company, he had nine years of experience as a process engineer and project manager for process- and facilities-related capital equipment installations. He has a BS in chemical engineering from Purdue University in West Lafayette, IN. (Goodwin can be reached at wgoodwin@extractionsystemsinc.com.)

Karen Turnquest is a member of the technical staff in the photolithography process area at Advanced Micro Devices (Austin, TX). Before joining the company she spent 10 years with IBM. She received a BS in chemical engineering from MIT in Cambridge, MA; an MS in imaging sciences from the Polytechnic Institute of New York in Brooklyn; and an MS in materials science and engineering from Rensselaer Polytechnic Institute in Troy, NY. (Turnquest can be reached at 512/933-6173 or r7276c@email.sps.mot.com.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit