Ultrapure Gases

Implementing on-lineand in situ moisture monitoring in reactive gas environments

James McAndrew, Melanie Bartolomey, and Jean-Marc Girard, Air Liquide; and Gerhard Göltz and Jean-Michel Flan, STMicroelectronics

TDLAS performs trace moisture measurements in process chambers and pure gases, achieving sensitivity in the low-parts-per-billion range and fast response times.

As device dimensions shrink, manufacturers must improve the control of molecular contamination in semiconductor process environments. While it has long been assumed that there is a correlation between shrinking design rules and the need to control contamination, concrete evidence has begun to emerge that bolsters the validity of this assumption. For example, the effects of trace moisture on the resistivity of titanium silicide serpentine structures was observed to increase as structure width was reduced from 2.88 to 0.72 µm.¹ Improved control of molecular contamination is also necessary for reducing the cost of ownership (COO) of wafer fabrication facilities. The major practical obstacle to achieving a lowered cost of ownership is the lack of easy-to-use sensors that are accepted by the manufacturing community and are compatible with reactive process gases and atmospheres.

This article describes tunable diode laser absorption spectroscopy (TDLAS), a technique developed by Air Liquide (Paris, France/Chicago) for measuring moisture in process gases.^1­11 This technique meets the increasingly stringent requirements of contamination control in the production of devices with small design rules and promotes a lower cost of ownership. The application of TDLAS to rapid thermal processing, where the only process gas is nitrogen at atmospheric pressure, has been described elsewhere.^1,2 Based on studies conducted at STMicroelectronics (Crolles, France), this article examines the capability of TDLAS in more aggressive atmospheres. Table I summarizes the principal beta tests of the in situ diode laser technology at various customer sites.

Developing State-of-the-Art Process Gas Monitoring Methods

Through atmospheric pressure ionization mass spectroscopy (APIMS), contamination levels in carrier gases such as nitrogen, argon, helium, and hydrogen can be detected to the single-parts-per-trillion level.¹² This is sufficient to meet the most stringent requirements for atmospheric contaminants in inert gases envisioned in the 1998 version of The National Technology Roadmap for Semiconductors and the 1999 revision, The International Technology Roadmap for Semiconductors. In addition, APIMS has been used in the continuous monitoring of real-world distribution systems and is available in a user-friendly package requiring only moderate operator expertise. While APIMS is expensive, a single system with multigas, multisampling capabilities can replace multiple conventional analyzers, making it the option of choice for ultra-high-purity systems.

Tool (Process)	Gases	Process Pressure	Installed
AMAT Centura (TiSi2   rapid thermal processing)	N2	Atmospheric	April 1996
SVG furnace (silicon nitride   low-pressure chemical   vapor deposition)	NH3, SiH2Cl2	200 mTorr	January 1998
AMAT Endura (TiN   chemical vapor deposition)	TDMAT N2, H2	10 Torr	June 1998
AMAT Centura (SiGe   epitaxy)	Hydrides, H2	20 Torr	July 1998
Lam TCP 9600   (aluminum etch)	Cl2, BCl3, N2	30 mTorr	January 1999

Table I: Summary of the principal beta tests of the in situ diode laser technology at various customer sites.

Existing options for monitoring specialty gases, however, are much less favorable. Fourier transform infrared (FTIR) spectroscopy can deliver the sensitivity levels (50 ppb) prescribed by the roadmap, but it frequently requires several minutes of acquisition time. In addition, this system is often bulky and difficult to use. Mass spectroscopy (MS) is an alternative to FTIR, but when used for gas delivery systems at high pressure it must work in conjunction with large and expensive vacuum systems. Moreover, limited filament lifetime is frequently a problem with MS.

To meet the challenge of monitoring reactive gases, techniques are required that are capable of performing measurements at or below projected roadmap specifications. In order to achieve reliable measurements at the 50-ppb level, a low-parts-per-billion detection limit is necessary. Likewise, fast response times (i.e., seconds) and excellent reproducibility are essential for improving statistical process control in gas production and cylinder-filling facilities. Technology designed to accomplish these tasks can also be used to improve handling techniques to ensure the purity of gas delivery systems.

In addition to being able to rapidly monitor reactive gases at low-parts-per-billion levels, such a system must be able to measure reactive gases on-line in order to ensure significant cost savings. Currently, reactive gas distribution systems are qualified in carrier gases only. While qualifying in carrier gases can aid in appraising the condition of the distribution system, it does not tell the whole story. For example, introducing hydrogen chloride (HCl) to a delivery system can result in a significant moisture release, even if the system was previously thoroughly dried in nitrogen. It is reasonable to assume that releases of this type can have a negative impact on the qualification of new fabs. Qualifying reactive gas distribution systems in actual process gases reduces the number of nonproduct wafers consumed during qualification. Similarly, on-line monitoring accelerates yield learning by certifying the purity of process gases.

While on-line gas purity measurements provide the best qualification of the process gas supply, in situ measurements in the process chamber are sometimes preferable. The chamber, wafers, and the process itself are important sources of contamination, which are neglected by upstream measurements. Current process gases and distribution systems are usually relatively minor contamination sources. Because they can detect contamination originating in the process itself, measurements of contamination just downstream of the process chamber are considered in situ monitoring. This usage has been widely accepted for in situ particle monitors. While MS is usually proposed for in situ contamination measurements, its use has gained wide acceptance only for ultra-high-vacuum processes such as ion implantation and sputtering, where an additional vacuum system is often unnecessary. For higher-pressure processes, however, which involve a dedicated vacuum system, filament lifetime limitations, and calibration difficulties, performing in situ measurements through MS has gained only limited acceptance.

For practical on-line, in situ measurements, it is important to restrict the quantity of data generated to manageable levels. For this reason, this study has chosen to focus on water vapor monitoring. Water vapor is a good indicator of ambient contamination, because it is much more difficult to eliminate than oxygen. In addition, water vapor has an important impact on process performance and causes corrosion of chambers and gas delivery systems.

TDLAS in Reactive Gas Environments

TDLAS addresses the problems associated with reactive gases. It is an absorption spectroscopic technique that uses a diode laser as its light source. Like many spectroscopic techniques, it can be used in most environments because the only elements to make contact with the sample, besides light itself, are mirrors and windows, which can be made from very robust materials. TDLAS has an advantage over other spectroscopic methods in that the diode laser can be operated in a relatively high-frequency modulation regime, where its noise contribution is very low. In addition, it has a very narrow linewidth (10^­3 cm^­1), which greatly facilitates the resolution of the absorption signal of interest, even in the presence of interference at nearby wavelengths.

The sensors developed for this study use a custom diode laser emitting at 1.368 µm. The emission wavelength may be tuned by varying either the laser temperature or current. The temperature is kept constant while the current is varied to tune the emission wavelength over the region where water vapor absorbs. The fraction of the laser emission absorbed by water molecules in the light path is then related to the moisture concentration.

The study examined the use of TDLAS in a variety of processes, including chemical vapor deposition (CVD), low-pressure chemical vapor deposition (LPCVD), and on-line monitoring of trace water in pure reactive gases.

In Situ Monitoring of Titanium Nitride (TiN) CVD. In the TiN CVD process, the process gases used are nitrogen, hydrogen, and tetrakis dimethylaminotitanium (TDMAT). The sensor configuration is illustrated in Figure 1. The diode laser and detector are mounted in a 12 x 12 x 12-cm optical head, which communicates with a multipass cell incorporated into the exhaust of the process chamber via a fused silica window. The sample region, incorporated into the exhaust line, consists of a Herriott-type multipass cell containing a pair of curved mirrors that are arranged so that the exhaust gas flows through the region between the mirrors. Light from the diode laser enters the exhaust line through the window, passes though a hole in one mirror, is reflected 20 times between the mirrors, and finally exits through the same hole and window through which it entered, eventually striking the detector in the optical head. The mirrors are 22 cm apart from each other and the total path length is 4.4 m. The mirrors and window of the multipass cell are heated to minimize deposition.

Figure 1: Schematic of in situ moisture sensor as configured for monitoring titanium nitride CVD. A similar geometry can be used for other single-wafer processes.

Second-harmonic detection is used to improve measurement sensitivity. High-frequency modulation is applied to the laser current (and hence emission wavelength), and the signal-processing electronics extract that portion of the detector signal that is in phase with the modulation and varying at twice its frequency. The second-harmonic signal is normalized to the total light intensity, which makes the calibration independent of such factors as mirror or window degradation and variations in alignment.

The diode's laser current and temperature are controlled by electronics in a separate chassis, which is mounted up to 1 m away in order to minimize space requirements on the exhaust line. Data can be provided to the fab control system via analog or serial port outputs or stored directly on a removable hard drive. A monitor connected directly to the electronics chassis provides local real-time display of moisture versus time, if required. Moisture concentration is measured every 2 seconds while the local video display is updated approximately every 6 seconds.

Figure 2 shows a moisture trace collected during wafer processing. Each moisture peak corresponds to the processing of a wafer. The origin of the moisture is believed to be wafer outgassing. In the trace shown, the initial wafers show much higher moisture peaks than later wafers, which is unusual. Because wafers were not tracked individually in this study, it is not possible to say with certainty whether the electrical properties of the initial wafers differed significantly from those of other wafers in the lot. The system offers sufficient time resolution to follow rapid events occurring in a matter of seconds and can discern steps in the processing of each wafer. While the large moisture spikes associated with the processing of the initial wafers coincided with the presence of plasma, later wafers showed very little moisture evolution. It can also be seen in Figure 2 that the standard deviation of the background moisture trace (after wafer processing) is 3 µTorr, indicating that the system has a sensitivity of approximately 10 µTorr of water.

 

Figure 2: Moisture trace collected during a titanium nitride CVD run. Insets show details for first wafers processed (high moisture level), later wafers (low moisture level), and after processing is completed (background).

In Situ Monitoring of Silicon Nitride (Si^₃N^₄) LPCVD. Si^₃N^₄ LPCVD has been the subject of several in situ monitoring studies, but generally they have been confined to pilot lines or laboratory studies. In this investigation, TDLAS was implemented on an LPCVD system under manufacturing conditions. The process gases used were dichlorosilane (SiH^₂Cl^₂) and ammonia (NH^₃), with a nitrogen carrier gas. HCl is generated as a reaction product. These gases presented a significant challenge to the instrument. However, the greatest challenge was the deposition of ammonium chloride (NH^₄Cl) in the exhaust line, a result of the reaction of HCl with NH^₃.

The sensor integration into the exhaust line is shown in Figure 3. Its placement downstream of the throttle valve was expedient; it would have been preferable to place it upstream in order for it to be physically closer to the process and to avoid pressure drop. As in the case of TiN, heating of the optics is critical. When the sensor was first installed, the mirrors, but not the window, were heated. Within 24 hours, deposition on the window was so severe that no light was transmitted. The sensor was removed during preventive maintenance and modified to include window heating. Figure 4 shows the power incident on the detector monitored over several months after modification and reinstallation. As a result of poor alignment, the power declined by about 25% over the first month. The system was realigned to ensure that the laser light was centered on the detector and other optics, making it less sensitive to small drifts. After this adjustment, the power remained constant within a few percentage points over three months. The system remained operational with as little as 50% of the original power reaching the detector, demonstrating that it is stable enough to operate for an extended period without a significant performance loss caused by deposition (assuming that there is no loss of power to the heating circuit during processing).

 

Figure 3: Schematic showing the integration of an in situ moisture (ISM) sensor into a batch furnace exhaust. The optical path is perpendicular to gas flow.

Figure 4: Variation in total light intensity reaching the detector for Si3N4 LPCVD. Initial drift was corrected by realigning without cleaning optics; the lack of subsequent decay in light intensity indicates minimal deposition on optics.

Figure 5 depicts a typical trace observed during processing. Large moisture peaks ("A" and "E") were observed during wafer loading and unloading. Spike "B" appeared when pumping speed was increased. A small peak "C," caused by a decrease in gas flow, was observed when the throttle valve was closed for a leak check. Peak "D" occurred when the throttle valve was reopened and gas with accumulated moisture reached the sensor. Figure 6 illustrates that "C" does not vary greatly over time, while "D" varies significantly, especially during periodic maintenance. This shows that the moisture level in the exhaust line remains more or less constant (or decreases slowly) while the moisture level representative of the furnace itself is highly variable.

 

Figure 5: Moisture and gas temperature traces during Si3N4 LPCVD.

Figure 6: Variation in height of moisture peaks over several months. High moisture levels were detected during preventive maintenance.

On-Line Monitoring of Trace Water in Pure Reactive Gases. While the first two examples describe the implementation of TDLAS for in situ monitoring of semiconductor processing, this example discusses its application to monitoring pure gas upstream of the process chamber.¹³ This important application can help ensure the quality of the reactive gas supply.

The challenges posed in monitoring pure gas upstream of the process chamber differ from the challenges of in situ monitoring. On one hand, very few problems caused by deposition arise in monitoring pure gas, since the sample gases are clean and dry. On the other, meaningful measurements must be significantly more sensitive than most measurements involved in in situ applications. In response to these different challenges, a somewhat different solution was implemented. The greatest degree of sensitivity was achieved using a dual-beam system, in which the laser beam is divided into two components, one of which passes through the sample and the other is reserved as a reference. Each portion of the laser beam impinges on a separate detector, and the difference between these two detector outputs gives the desired signal. While beam subtraction can be combined with second harmonic detection, good sensitivity was achieved in this investigation without doing so.

The use of a dual-beam system is not recommended at present for most in situ applications, because deposition on the optics makes it difficult to maintain the necessary balance between the two beams. (In situ monitoring of very clean processes might be an exception.) While deposition is not a concern in pure gas monitoring, the cell is kept at 60°C in order to minimize the interaction of its wall surfaces with moisture. It is also helpful to operate at reduced pressure (~ 60 Torr) in order to have as sharp an absorption feature as possible (by minimizing pressure broadening) and to minimize the residence time of a gas sample in the cell.

Figure 7 presents a series of absorption spectra and the corresponding calibration curve for trace moisture in HCl. The system's detection limit is 10 ppb of water in HCl and hydrogen bromide (HBr) and 5 ppb in nitrogen, carbon tetrafluoride (CF^₄), nitrogen trifluoride (NF^₃), and similar gases. NH^₃ is of particular importance for gallium nitride (GaN) processes and the production of blue lasers and light-emitting diodes. NH^₃ poses a particular difficulty because it interferes with the water absorption line selected. Therefore, it is necessary to subtract the NH^₃ spectrum from the data before analyzing them. With this procedure, a detection limit of 50 ppb of water in NH^₃ has been achieved. A sample spectrum illustrating background subtraction and a calibration curve are shown in Figure 8.

 

Figure 7: Observed spectra (moisture peaks) for parts-per-billion-level moisture additions to HCl (left); calibration curve for water in HCl (right).

Figure 8: Observed spectra (moisture and ammonia peaks) for pure ammonia and ammonia with 1.6 ppm of water and result of spectral subtraction (left); calibration curve of water in NH3 based on subtracted spectra (right).

TDLAS has been used successfully to monitor the initial moisture release from a dry electropolished stainless-steel pipe in HCl. Figure 9 depicts the setup used for this experiment. The tubing sample under investigation was 4 m long and made of electropolished stainless steel that had been dried under dry nitrogen down to a moisture level of <10 ppb. Then HCl from a cylinder was introduced and the moisture concentration versus time was recorded, as depicted in Figure 10. Fresh and dry electropolished tubes cause temporary (5-minute-long) contamination, which can be very detrimental if delivered to the process chamber. The source of released water is believed to be a combination of adsorbed water displaced from the surface by HCl and water generated by reaction with surface oxides, as in the following formula:

Fe^₂O^₃ + HCl * FeCl^₃ + H^₂O

As can also be seen in Figure 10, the diode laser system also responds quickly to a decrease in the contamination level, such as when the gas flow is switched to pass through a purifier. Fast response is achieved despite a modest flow of only 200 std cm³/min of HCl.

 

Figure 9: Schematic of laboratory apparatus for investigating the effects of the distribution system on moisture in reactive gases.

Figure 10: Evolution of moisture level as HCl was introduced to a tubing sample previously purged with nitrogen; after 58 minutes, the HCl stream was switched to pass through a purifier.

Reducing the Cost of Ownership

One of the major reasons for implementing in situ, on-line moisture monitoring of reactive gases is to reduce wafer fabs' cost of ownership. Several factors contribute to lowering the cost of ownership:

• Improving Yield: If contamination events are detected in real time, corrective action can be taken to prevent or minimize yield loss. To implement a real-time strategy, it is necessary to have baseline data for typical contamination levels and to know the contamination level at which corrective action is required. Real-time monitoring provides these data.

• Accelerating Yield Learning: Because contamination is often a suspected factor in yield loss, knowledge of contamination levels will accelerate the determination of actual yield detractors (even if they turn out not to be contamination related). With increasing process complexity and the use of new chemistries, the importance of this factor can be expected to increase in the future.

• Improving Overall Equipment Effectiveness: The time devoted to the recovery of process chambers after maintenance is strongly affected by the need to adequately remove the atmospheric contamination that enters while the chamber is open. Purging time is often set by experience and padded with a large safety margin. By measuring actual contamination levels, technicians can resume processing sooner and with reduced risk. This consideration is particularly important for tools that present bottlenecks. Contamination monitoring can also improve overall equipment effectiveness by enabling improved scheduling and making it possible to avoid unplanned interruptions. For example, knowledge of rising contamination levels can be used as an indicator that maintenance is due.

• Reducing Nonproduct Wafers and Consumables: Apart from its impact on overall equipment effectiveness, maintenance also usually requires the use of consumables and test wafers. Optimizing the frequency of maintenance periods can reduce both. Furthermore, if recovery after maintenance does not go smoothly, the number of required nonproduct wafers can increase greatly. Contamination monitoring detects infiltration and can accelerate the solution of other problems by removing uncertainty about contamination issues.

• Improving Throughput: Just as contamination monitoring can reduce the time spent on purge procedures following maintenance, it can also reduce the time spent on purge procedures during processing itself. This application demands that analytical equipment respond rapidly to rising contamination levels, but, as has been shown, fast-responding sensors are available.

• Increasing Tool Lifetime and Accelerating Fab Start-Up: Process tool lifetime can be increased by accelerating fab start-up. Because the useful lifetime of most process tools is determined by obsolescence, their effective lifetime is increased when they are brought on-line quickly, saving time at start-up. It can be expected that fab start-up will be accelerated by qualifying process gas lines in reactive gases. Current practice is to qualify all gas lines in nitrogen only. But unknown levels of contamination are present in new piping systems, which can contribute to start-up delays. As has been demonstrated, when lines are first exposed to reactive gases, a contamination spike can result. This can occur even though the lines have been thoroughly purged with nitrogen and are as clean as possible.

An improvement in the cost of ownership is calculated by considering the cost of ownership of a manufacturing system with and without monitoring. The cost-of-ownership reduction is given by

where CF = fixed cost, CR = recurring cost, CY = cost of yield loss, L = tool lifetime, TPT = throughput, Y = yield, and U = utilization. Yield improvement affects both Y and CY. An acceleration in yield learning and an improvement in overall equipment effectiveness affect U. A reduction in nonproduct wafer and consumables reduces CR.

Decisions regarding the implementation of process and gas monitoring are in principle a matter of simply comparing the increase in CF due to the analytical system with the decrease as a result of other factors. In practice, however, decision making is more complicated than this because the impact of contamination on yield is often not known precisely.

Conclusion

TDLAS has been shown to be capable of performing trace moisture measurements in semiconductor process chambers and pure gases. Sensitivity in the low-parts-per-billion range and fast response times have been achieved. By heating the optics, reliable operation has been demonstrated in extremely challenging environments. Implementation of this technique can improve yields in moisture-sensitive processes and is expected to facilitate the extension of current processes to smaller device dimensions. Significant improvements in overall equipment yield, leading to a reduced cost of ownership, are achievable. The on-line monitoring of reactive gases enables fabs to start up more rapidly than previously, leading to an effective increase in tool lifetime during the critical early stage of product supply. Finally, yield learning can be accelerated by ensuring the purity of the process environment.

Acknowledgments

The authors would like to thank Dmitry Znamensky and Ronald Inman (Air Liquide, Chicago) and Patrick Mauvais (Air Liquide, Paris) for their contributions to the experimental aspects of this article. They also wish to thank Alain Cannizarro, Didier Dutartre, and Dominic Malgouyres (STMicroelectronics) for their critical support.

References

1. J McAndrew et al., "Increasing Equipment Up-Time through In Situ Moisture Monitoring," Solid State Technology 41, no. 8 (1998): 61­71.

2. J McAndrew, "Progress in In Situ Contamination Control," Semiconductor International 21, no. 5 (1998): 71­78.

3. B Jurcik et al., Method for the design of spectroscopic sample cells and cell based on this method, U.S. Pat. 5,351,120, 1994.

4. C Ronge and P Mauvais, Process and device for analyzing traces of impurities in a gas sample by means of a diode laser, U.S. Pat. 5,705,816, 1998.

5. J McAndrew and R Inman, Method for stabilizing the wavelength in a laser spectrometer system, U.S. Pat. 5,742,399, 1998.

6. R Inman and J McAndrew, Polygonal planar multipass cell, system, and apparatus including same, and method of use, U.S. Pat. 5,815,578, 1998.

7. J McAndrew and R Inman, Method for calibration of a spectroscopic sensor, U.S. Pat. 5,835,230, 1998.

8. J McAndrew and R Inman, Method and system for sensitive detection of molecular species in a vacuum by harmonic detection, U.S. Pat. 5,880,850, 1999.

9. J McAndrew and R Inman, In-line cell for absorption spectroscopy, U.S. Pat. 5,949,537, 1999.

10. J McAndrew, H-C Wang, and B Jurcik, Chamber effluent monitoring system and semiconductor processing system comprising absorption spectroscopy, U.S. Pat. 5,963,336, 1999.

11. J McAndrew, Method for intelligent data acquisition in a measurement system, U.S. Pat. 5,991,696, 1999.

12. JM Girard and Y Marot, "PPT-Level Analysis of UHP Hydrogen," European Semiconductor (April 1996).

13. J-M Girard and P Mauvais, "PPB-Level Hygrometry in Nitrogen and ESGs using Tunable Diode Laser Absorption Spectroscopy," in Proceedings of the International Symposium on Semiconductor Manufacturing '96 (Piscataway, NJ: IEEE, 1996), 325­328.

James McAndrew, PhD, leads the electronics R&D group at the Chicago Research Center of Air Liquide. His research interests lie in high-purity gas analysis and in situ monitoring of semiconductor process chambers. He has published many papers and holds many patents on various aspects of semiconductor gases. He received his PhD in chemical physics from Yale University (New Haven, CT) in 1987. (McAndrew can be reached at 708/579-7780 or james.mcandrew@airliquide.com.)

Melanie Bartolomey is an R&D engineer at the Air Liquide research center in Les Loges en Josas, France, where she works in the R&D group dealing with the applications of gases in electronics and laboratories. She has worked on the development of analytical methods for trace measurements in process gases used in the electronics industry. She received her MS in chemistry from Ecole Superieure de Physique et Chimie Industrielle (Paris, France) in 1996. (Bartolomey can be reached at +33 1 39076353 or melanie.bartolomey@airliquide.com.)

Jean-Marc Girard, PhD, is a senior scientist at Air Liquide Laboratories in Tokyo. Previously he was an R&D engineer at the Air Liquide research center in Les Loges en Josas, France, where he worked in the R&D group dealing with the applications of gases in electronics and laboratories. Girard has worked on the development of analytical methods for trace measurements in process gases used in the electronics industry. He received his PhD in chemistry from Ecole Normale Superieure (Lyon, France) in 1993. (Girard can be reached at +81 298 790050 or jean-marc.girard@airliquide.com.)

Gerhard Göltz, PhD, is a process development manager at STMicroelectronics in Crolles, France. In 1981 he joined the electronics center of the Centre National d'Etudes des Télécommunications in Grenoble, where he worked with silicide and metallization processes. Göltz joined the new STMicroelectronics facility in Crolles in 1992, working in different management positions in the R&D area. In 1976 he obtained his PhD in Physics from the University of Stuttgart in Germany. (Göltz can be reached at +33 4 76926000 or gerhard.goltz@st.com.)

Jean-Michel Flan manages an equipment engineering group at the STMicroelectronics R&D facility in Crolles, France, where he is responsible for driving a variety of equipment performance programs. Previously he was involved in equipment reliability and quality engineering for STMicroelectronics and other international semiconductor manufacturers. He has a BS in electrical engineering. (Flan can be reached at +33 4 76426263 or jean-michel.flan@st.com.)