Progress in In Situ Contamination Control

The drive of semiconductor manufacturing to reduce device dimensions has required major improvements in process gas purity. Contamination levels have been reduced by four orders of magnitude or more over two decades. Along with straightforward improvement in gas specifications has come an increased emphasis on maintaining purity to the point-of-use. In the late 1980s, gas distribution systems became the focus of a major effort based upon component testing,¹ numerical modeling² and sophisticated analytical techniques (notably atmospheric pressure ion mass spectroscopy ³). As a result, sub-ppb purity of nitrogen is now routinely achieved at the point of connection to the process tool. Although it was widely predicted that all other gases would rapidly follow the same trend (the 1994 National Technology Roadmap for Semiconductors predicted that contamination levels in specialty gases would be in the 1-100 ppb range by 1998 and the 1-10 ppb range by 2001), the actual evolution in specifications fell short of these expectations. In fact, the 1997 Roadmap⁴ indicates that 500 ppb is sufficient for today's production and 100 ppb should be sufficient until 2006.

This surprising change in purity requirements primarily resulted from a shift in emphasis from upstream purity specifications to purity in the process atmosphere itself. In addition, the technical basis for purity specifications is be-ing questioned more carefully today than in the past. The process chamber, the wafer and the process chemistry are now recognized as major contributors to contamination in processing. Process environment purity is governed by not only gas purity, but also chamber materials of construction; operating, maintenance and wafer handling procedures; and even device materials.

In situ sensors can be used to reduce yield loss and improve equipment effectiveness by optimizing purge and pumpdown procedures, while providing diagnostic information to ensure process chamber purity prior to processing. The most logical point to measure contamination is in situ in the process chamber, as close to the wafer as possible. If in-chamber sensing proves impractical, sensor placement immediately downstream is an alternative. In both cases, a single sensor is designed to monitor contamination from all possible sources. Both locations are particularly demanding because of the aggressive nature of the process atmosphere, noise sources, etc. In situ analysis of the chamber environment requires techniques that can withstand exposure to reactive gases, proximity to plasma, elevation to high temperatures, etc. A brief review of in situ particle monitors, RGAs and other sensors shows the advantages and limitations of using these sensors in production semiconductor process tools.

Particle monitors

As particles have the greatest impact of any contaminant on device performance and yield, it is not surprising that in situ particle monitors (ISPMs) have formed the spearhead of in situ contamination measurement. ISPMs are usually placed on the exhaust line of the process tool, relying on the transport of particles from the process chamber to the sensor. They were introduced in the early 1990s, ably supported by many articles by Peter Borden and others.

• Particle transport mechanisms carry different populations to the sensor from those carried to the wafer
• Particles are not uniformly distributed through the small fraction of the exhaust volume sampled⁶
• High exhaust velocities result in short-lived signal bursts that are often rejected by detection electronics
• Minimum detectable particle size may be too large to detect particles that are still detrimental to the process
• Background signals in process chambers (caused by RF and magnetic fields, optical emission from plasma, etc.) are difficult to distinguish from true particle counts

ISPMs have been particularly successful in ion implant applications, where the sensor can be installed in the process chamber itself. Modeling techniques should lead to improved correlation with wafer particle counts and improved sensor placement. Although placement in the process chamber appears to be the most promising approach for the future, it is also the most challenging because of the sensor's vulnerability to process noise.

RGAs

RGAs are mass spectrometers used to measure residual gas composition in vacuum chambers. Despite a long history of use and various papers describing their successful implementation,^{7, 8} RGAs have had limited success in the manufacturing environment. They have been most successful in ion implantation, an ultrahigh-vacuum process that enables the RGA to be used without differential pumping for substantial advantage in simplicity of operation and accuracy of sampling. Recently developed compact RGAs⁹ are inexpensive and operate at relatively high pressures (several mTorr), enabling some sputtering processes to be monitored without differential pumping. Currently, compact RGAs are most commonly used to monitor evacuation procedures.

In order to sample low and moderate vacuum environments, an RGA must be differentially pumped, and gas samples must be drawn into the RGA via a pressure reduction system. When operated in this fashion, the RGA is not truly an in situ sensor. For water vapor measurement in particular, adsorption/desorption effects in the sampling system and the sensor itself will reduce measurement sensitivity and increase response time.

RGAs provide a wealth of information on the process, yet the data are often complex and difficult to interpret. Calibration of RGAs is particularly problematic as it is sensitive to the condition of the filament and ion optics as well as to the gas pressure (i.e., calibration at low pressure is not sufficient for quantitative measurements at higher pressures).

Optical sensors

Optical emission spectroscopy is widely used for process control (endpoint detection) in semiconductor processing. Although its use for contamination measurement has been reported, it does not appear to be successful, presumably because of the difficulty of interpreting the data and to the obvious restriction to operation during plasma processing. Other optical spectroscopic techniques, widely implemented for laboratory studies, do not apply as well to manufacturing because optical access to the process environment requires modification of the process chamber.

In situ humidity measurement

Air Liquide's research program in in situ monitoring was initiated in response to an invitation from Texas Instruments (Dallas, Texas) to propose a system that would be suitable for use in development and production fabs and would be focused on the measurement of molecular contamination. We chose water vapor as a diagnostic of ambient contamination because it is usually the most difficult atmospheric component to eliminate from any environment. Therefore, if water vapor concentration is low, it can usually be assumed that other contaminants are also present at low levels. Water is also the most difficult species to measure at a low concentration using an RGA.

The main drawback of measuring water vapor only is that one cannot detect contamination events because of the presence of inappropriate process gases (as can be caused by errors in construction and some types of component failure). However, the atmosphere is the most common source of chamber contamination, and the advantages of simplicity obtained  by monitoring a single species compensate for some loss of completeness.

Humidity can enter the process environment accidentally, caused by leaks, insufficient purge procedures, excess moisture on incoming wafers, etc. Such conditions can lead to undesired oxide formation or corrosion of metal layers on the wafers -- eventually resulting in yield loss. Unavoidable presence of humidity arises, for example, after periodic maintenance or whenever wafers enter the loadlock from the cleanroom environment. While purging can remove this moisture, it is a time-consuming process, often extended as a safety measure. In all these cases, an in situ humidity sensor can be used to eliminate or shorten such procedures.

We have selected tunable diode laser-absorption spectroscopy (TDLAS) as a measurement technique because it is compatible with a wide range of environments and is based on an absolute principle. We overcame the difficulty of directly accessing the process environment by placing the sensor on the exhaust line of the process chamber (lead photo and Fig. 1), in a manner analogous to that used for particle monitors. Control electronics are mounted remotely from the sensor to minimize space requirements near the process tool. The sensor (Fig. 2) uses a tunable diode laser emitting at 1.368 µm as the light source. The emission wavelength may be tuned by varying either the laser temperature or the laser current. In our case, 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. Second harmonic detection, used to improve measurement sensitivity, applies high-frequency modulation to the laser current (and hence the wavelength of its emission) and uses only the portion of the detected signal that is in phase with the modulation, varying at twice its frequency. The second harmonic signal is normalized to the total light intensity -- an extremely important step that removes the effect of mirror or window degradation, variations in alignment, etc. If the light intensity decreases because of these effects, the sensitivity will decrease, but the average value of the reading will be unaffected (unless it is at or below the detection limit).

In comparison to particle monitors, the moisture sensor is not subject to complex effects arising from differences in transport mechanisms for particles of different sizes. The TDLAS sensor samples most of the exhaust stream, and because of the modulation techniques used to optimize sensitivity, it is relatively insensitive to noise. Fortunately, high exhaust velocity has no effect on the measurement. Importantly, the sensor must be placed close enough to the process chamber so that upstream adsorption/desorption of water on surfaces does not corrupt the data.

The performance of the TDLAS sensor in different gas matrices and at different pressures is summarized in Table 1. It can detect water vapor in almost any matrix gas. Obvious exceptions are gases that react rapidly with water vapor such as BCl₃ and WF ₆. Silane (SiH₄) reacts with water vapor in the presence of steel surfaces, but the reaction is sufficiently slow that water vapor is detectable in SiH₄. Spectroscopy of SiH₄ near 1.368 µm is not well characterized, and given the abundance of absorption features of SiH₄ throughout the infrared region, it is reasonable to assume that SiH₄ will interfere somewhat with H₂O detection by TDLAS. It has been shown that NH₃ interferes with water vapor detection using TDLAS.

In an extremely low-pressure environment(<10^-6 Torr), the sheer lack of molecules severely limits the usefulness of TDLAS instruments. Fortunately, the RGA is most easily used in this pressure regime, making the two techniques complementary. At high pressures, pressure-broadening reduces sensitivity somewhat, and pressure-broadening coefficients are usually larger in specialty gases than in carrier gases. For this reason, operation at atmospheric pressure in nitrogen provides greater sensitivity than operation at atmospheric pressure in HCl.

Texas Instruments decided to evaluate the diode-laser moisture sensor on a rapid thermal processing (RTP) chamber, because the process is known to be sensitive to ambient contamination. The in situ moisture sensor, which can operate at atmospheric pressure as easily as at vacuum, is quite convenient. In this test, the exhaust line was modified to include a fused silica window for the entry and exit of light, and a pair of mirrors form a multipass cell, the details of which are published elsewhere.¹⁰ Figure 3 shows the sensor and control electronics. A real-time display of water vapor concentration is provided. Water vapor concentration and other diagnostic information is written to an ASCII file, updated every 2 sec.

Click to see larger image. 4. Each peak of the spectrum corresponds to moisture released from the wafer.

Figure 4 shows a moisture trace during TiSi₂ formation. Each peak corresponds to moisture released from a wafer. Note that moisture events occur on a relatively fast time-scale and are yet easily detected. The standard deviation of the background signal is 25 ppb, consistent with the claimed detection limit of <100 ppb. Moisture traces observed during different processes (TiSi₂ formation vs. annealing) and even, in some cases, of the same process carried out on different devices, turn out to be quite different and characteristic. Detailed interpretation of the moisture traces and of the impact of moisture on device characteristics has been presented elsewhere. ¹¹

The sensor was first used to optimize chamber purging following maintenance procedures. Previously, the purging time was set at 12 hours without knowledge of the typical moisture level during processing or purging speed. Once the typical chamber baseline level was determined (about 0.5 ppm), it was easily determined that four hours of purging sufficiently removed moisture in the chamber -- enabling a saving of eight hours. In a similar manner, the purge procedure following loading of a batch of wafers into the loadlock was reduced from 2.5 purge cycles (i.e., two cycles from atmospheric pressure to vacuum and one from 0.5 atm to vacuum) to 1 purge cycle, for a savings on the order of 5 min per wafer batch. Once the in situ sensor was installed, it identified an inadequate purging cycle caused by human error. It also effectively found an error in setting of the process gas flow. The sensor further detected a major leak in the exhaust line of a second RTP chamber.

Conclusions

The benefits of implementing in situ moisture sensing technology include yield loss reduction and improvement in OEE. To enjoy these benefits, the sensor output must be integrated into the fab or process tool control system, and the sensors must be properly maintained. Eventually, in situ contamination sensors may be integrated with new semiconductor manufacturing equipment, but this approach alone is insufficient as it neglects the users of existing equipment. We believe that certain in situ monitoring services should be included as part of an overall gas and chemical management package, enabling the semiconductor manufacturer to benefit from in situ contamination control while concentrating on the company's core business.

Acknowledgments

James McAndrew is the senior scientist responsible for analytical techni-ques research at Air Liquide's Chicago Research Center. He holds several patents and has published more than 30 papers on various aspects of gas-phase chemical physics, gas analysis, ultrahigh-purity gas delivery and in situ analysis in semiconductor process tools. He was awarded the Ph.D. in chemical physics by Yale University in 1987. Phone: (708) 579-7780 Fax: (708) 579-7833 E-mail: james.mcandrew@airliquide.com

Texas Instruments' Contamination Free Manufacturing Group, under the leadership of Venu Menon, was the driver in initiating our work in Dallas. The contributions of Michael Grobelny, Asad Haider and Edwin Arrant were vital. No less important was the cooperation and contribution of TI manufacturing personnel, especially Neal Murphy, Jody Brookshire and Paul Gillespie. Ronald Inman of Air Liquide's R&D group provided on-site sensor implementation, with assistance from Richard Kuan of Air Liquide's Electronics Chemicals and Services.

References

1. J.J.F. McAndrew, M.D. Brandt and G. Kasper, "Establishing Moisture Test Methods for Process Gas Distribution Systems," Microcontamination, May 1991, p. 33.

2. J.J.F. McAndrew, B. Jurcik and D. Znamensky, "Using Simulation to Optimize Gas Distribution System Cost and Performance," IES Journal, Vol. 37, No. 3, 1994, p. 30.

3. C. Ronge, et al., "APIMS Technology: Sub-ppb Measurement of Contaminants in Nitrogen with APIMS Trace," Microcontamination Conference Proceedings, 1991, p. 153.

4. National Technology Roadmap for Semiconductors, Semiconductor In-dustry Association, Nov. 1997.

5. K.M. Takahashi and J.E. Daughtery, "Current Capabilities and Limitations of In Situ Particle Monitors in Silicon processing Equipment," J. Vac. Science and Tech., A, Vol. 14, No. 6, 1996, p. 40.

6. P.D. Kinney, et al, "Particle Behavior in Vacuum Systems," IES Journal, Vol. 39, No. 6, 1996, p. 40.

7. C. DiCouto, Micro, Oct. 1996, p. 103.

8. B. Bradshaw, et al., "Process Monitoring of Tungsten Chemical Vapor Deposition and Etchback with Residual Gas Analysis," IES Conf. Proceedings, 1996, p. 382.

9. S. Boumsellek and R.J. Ferran, "Detection of 1 ppm Contamination Levels Using the Micropole RGA," IES Conf. Proceedings, 1996, p. 1.

10. Submitted to J. Vac. Science and Technology

11. A. Haider, J.J.F. McAndrew and R.S. Inman, Proceedings of Symposium on Crystalline Defects and Contamination, Electrochemical Society, 1997, Vol. 97-22, p. 484.