ULTRAPURE GASES

Beta testing an in-line monitor for moisture measurement in inert gas lines

(Second of two parts)

Maura I. Toth, Millipore; David A. Zatko and David F. Yesenofski, Air Products and Chemicals; and Manfred Schneegans and Johann Helneder, Siemens

A significant stage in the development of new products, beta-site testing enables manufacturers and potential users to collaborate on studies that complement the manufacturers' laboratory and in-house evaluations. Beta-site participants have an opportunity to learn about technological advances and have an impact on how a new system is finally configured for production, while the manufacturer gains insights into the ways the tool performs in the field.

As shrinking device sizes and increasing numbers of process steps increase the risks associated with moisture contamination, the semiconductor industry needs new techniques for controlling gas purity. Extensive beta-site collaboration is needed to develop products that can minimize contamination-related downtime, provide more knowledge about the moisture levels that are acceptable for a given process, and allow purge cycles to be determined with precision.

This article reports on the beta tests performed at two sites for an in-line moisture detection device based on a new monitoring technology developed by Millipore (Bedford, MA), which was described in an earlier issue of MICRO.¹ Designed to measure moisture levels in ultra-high-purity (UHP) nitrogen and argon gases, the monitor uses a piezoelectric quartz crystal microbalance (QCM) sensor. The quartz crystal is coated with a thin film of barium metal, with which moisture reacts irreversibly:

Ba + 2H₂O — Ba(OH)₂ + H₂

The speed of this chemical reaction enables the sensor to respond rapidly.² As moisture reacts with the barium, the mass of the coating increases, causing a decrease in the frequency of crystal oscillation. The rate of change in oscillation frequency is correlated to moisture concentration through a one-time factory calibration. A calibration constant is programmed into the monitor's onboard electronics for use in calculating the moisture level (in parts per billion by volume, or ppbv) from the frequency change per unit of time.

The gas flow rate is also needed to calculate the moisture concentration. It can either be input manually as a flow default or via connection with the analog signal from a mass-flow controller (MFC) or meter. The sensor's operating flow range is 10—1000 std cm³/min and it can provide continuous moisture readings to within ±5 ppbv at moisture levels between 0 and 14 ppbv and within ±30% of the indicated reading at levels between 15 and 50 ppbv.

The in-line monitor was originally intended for installation at points of use (POUs), between an MFC and a chamber isolation valve, in gas lines for UHP nitrogen and argon, which typically have moisture levels below 50 ppbv. The small footprint and internal volume of the device make such installation feasible. The monitor can also be installed on a gas stick with flow and pressure control to monitor bulk gas quality in a slipstream application. The use of a gas stick avoids the bulk gas line properties that are incompatible with the monitor: flow rates higher than 1000 std cm³/min, unstable gas pressure resulting from changes in gas load, and pressures that may be higher than 7 bar (100 psig), which is the maximum recommended operating pressure of the sensor.

Beta-Site Test Programs

Before the start of full production of the monitor, beta-site evaluations were conducted at 15 locations. These studies included comprehensive testing on calibration stands at analytical laboratories, POU installations on metal-deposition tools, and conventional monitoring in bulk gas applications. The research described below represents the wide range of tests that were conducted in the course of the overall beta-site program for the in-line monitoring technology.

In the analytical technology laboratory of the electronics division of Air Products and Chemicals (Allentown, PA), tests on a multiple-analyzer test bench have examined the performance of the monitor in carefully controlled conditions. The testing began in 1993 and has involved both alpha and beta versions of the monitor, which were evaluated for accuracy, speed of response, statistical limit of detection (LOD), and correlation with an atmospheric pressure ionization mass spectrometer (APIMS) analyzer. The goal of Air Products' analytical technology group is to assist manufacturers with evaluations of new devices from the alpha stage through pilot production, ultimately providing end-users with products that require no rework in the first year, fit the specification requirements for the gas, and are backed with documented statistical evaluations for the LOD.^3—5 Air Products does not specifically endorse any instruments that it beta tests in collaboration with manufacturers. Data are presented only as results of controlled testing in the context of the development of an instrument and are not offered as an indication of results achievable under other conditions.

At Siemens's advanced research center in Neuperlach, Germany, the in-line monitor was installed in March 1996 on the UHP argon gas line supplying a metal-deposition sputtering chamber. The ongoing testing has provided insight into the sensor's ability to provide continuous monitoring of process gas quality for a physical vapor deposition (PVD) process. The advanced research center at Siemens evaluates new technology and optimizes process parameters for new semiconductor processes in aluminum and barrier (titanium/titanium nitride) metallizations.

Response Time Studies

The strong chemical reactivity of the active part of the QCM sensor enables the in-line monitor to detect moisture upsets quickly. Consequently, response time was one of the features examined extensively during beta testing. The response time to detect a moisture upset in a gas line is a combination of the distribution system, sampling-line, and analyzer response times. In turn, the analyzer's response time is the sum of its internal sampling-line response time and sensor response time. The distribution system and sampling-line response times include moisture transport and equilibration of components and tubing. Two evaluations of the monitor's response time are presented here: a test conducted by the manufacturer on a very short sampling line (3 cm downstream of the mixing point for the "dry" and "wet" gas streams) in order to quantify the response time of the device as close as possible to the moisture source, and a test conducted by Air Products on a moderately long (3-m) sampling line.

Figure 1: The in-line monitor's response to a 6.5-ppbv moisture challenge using a 3-cm sampling line and a nitrogen flow of 1 std L/min.

In the first test, a moisture challenge was generated using a calibrated moisture diffusion tube with a known emission rate. Purified nitrogen flowing at a known rate over the diffusion tube generated a stream of wet nitrogen. Dilution of this wet gas stream using purified nitrogen and MFCs gave the moisture concentrations desired. As seen in Figure 1, the monitor achieved a 60-second response to a 6.5-ppbv moisture challenge at a flow rate of 1 std L/min. The average moisture level detected was 6.9 ppbv with a standard deviation of ±0.4 ppbv. When the sample gas was switched to "zero"-ppbv nitrogen, purified using the manufacturer's Waferpure in-line system, the analyzer reading returned to an average of 1 ppbv within 60 seconds (see Figure 2), demonstrating a fast recovery from the moisture challenge.

Figure 2: The in-line monitor's response to the introduction of purified "zero"-ppbv nitrogen at 1 std L/min following a 6.5-ppbv moisture challenge.

At Air Products, the response time testing was made more severe by using a moisture challenge of only 0.5 ppbv with the monitor located 3 m downstream from the gas mixing point. Sampling-line response time was measured separately so that its contribution to the total response time could be differentiated from the response time of the analyzer. The monitor was incorporated in a test bench as shown in Figure 3. Moisture challenges were mixed using purified nitrogen as the diluent and wet gas generated by a Span Pac 261 with a calibrated, extended-life (~5 years) permeation tube (Kin-Tek Laboratories, La Marque, TX). Further dilution of the wet nitrogen with purified, dry nitrogen was made with an MFC calibrated with a bubble meter. The 1/4-in. electropolished 316-stainless-steel lines were heat-traced to a controlled temperature of 50°C to minimize moisture adsorption and maximize moisture transport. The test system connections are either welded or metal face—sealed and valves are diaphragm type with zero dead volume.

Figure 3: Schematic of the test bench used for evaluation of the in-line monitor (ILM) at the Air Products laboratory.

After the analyzer equilibrated with purified nitrogen gas (0.04-ppbv moisture detected), a 0.5-ppbv moisture challenge was made. The moisture breakthrough appeared at the in-line monitor after approximately 72 minutes, at which point the detected moisture level rose to an average of 0.4 ppbv (standard deviation = 0.06 ppbv), stabilizing within 2 hours of the start of the experiment (see Figure 4).

Figure 4: The in-line monitor's response to a 0.5-ppbv challenge, optimized for low values, using the apparatus in Figure 3 and a sampling rate of two data points per minute.

The total response time was then broken down by determining the breakthrough time attributable to the sampling line alone, which was calculated using a model based on tubing surface area and an empirical fit to an experimental database of breakthrough determinations. The estimated breakthrough time for moisture at 0.5 ppbv (at 1 std L/min) was estimated to be 14 hours at 21°C. Application of a temperature-effect algorithm, to take into account the heat-tracing at 50°C, reduced the estimated breakthrough time to 60—90 minutes. Thus, the experimental breakthrough result of 72 minutes for the monitor to detect the moisture in the system was due almost entirely to the sampling line response time. This result is consistent with others obtained by the analyzer manufacturer in tests at higher moisture levels. Together, the two response time evaluations described indicate that the in-line monitor's response time is faster than that of the sampling system.

Moisture Linearity Testing

Using the same experimental setup (Figure 3), Air Products also conducted a series of moisture challenge tests over the range of 0—10 ppbv to evaluate the accuracy and linearity of the analyzer's measurements. A regression analysis was performed to determine the linearity of the moisture level detected versus the moisture level generated (see Figure 5). Each data point in the figure represents the mean of 120 measurements (taken at a rate of two data points per minute), and the error bars on each point represent the minimum and maximum values seen. Linear regression indicates that the correlation is 1:1, with a slope of 1.001, an intercept of 0.39 ppbv, and a correlation coefficient R² = 0.99. The maximum deviation from the generated moisture value was 0.5 ppbv for the 1-ppbv moisture challenge. The statistical LOD, determined by the propagation-of-error method,^3—5 was 0.3 ppbv with a lowest challenged value of 1 ppbv. Figure 5: Linearity correlation between moisture detected and moisture generated over 0—10 ppbv. Each data point shows the average, minimum, and maximum over a 1-hour period. (Instrument calibrated at Air Products.)

APIMS Correlation

As seen in Figure 3, the test bench is set up to carry out simultaneous challenges to several analyzers. In another test series, step-down moisture challenges ranging from 10.5 to "zero" ppbv were directed to the in-line monitor and an APIMS analyzer connected in parallel. Following equilibration of the two instruments at a 10.5-ppbv challenge, step-down dilutions to one-half the previous value produced challenges of 5.2-, 2.6-, 1.3-, and 0.75-ppbv moisture followed by "zero" gas. As Figure 6 indicates, results from the two analyzers correlated well; the maximum deviation of the in-line monitor from the APIMS output was 1.2 ppbv at the 5.2-ppbv challenge. The spikes in the monitor data correlate exactly with temperature swings in the laboratory. The averaged numerical data are given in Table I; these should be considered as representing the monitor's capabilities under well-controlled conditions within a laboratory environment.

Generated Values (ppbv)	APIMS (ppbv)a	ILM (ppbv)a
10.5	10.4	10.6
5.2	5.3	4.1
2.6	2.5	2.2
1.3	1.2	1.1
0.75	0.72	0.6
0.	0.04	0.18
a Values are averages of at least 10-minute periods near the end of the challenges. APIMS has background correction.

Table I: Averaged numerical data from the step-down moisture challenge with the in-line monitor (calibrated at Air Products) and an APIMS analyzer connected in parallel. (Results are also shown in Figure 6.)

Figure 6: Results for a series of step-down challenges, starting from 10.5 ppbv, with an in-line monitor (ILM) and an APIMS analyzer in parallel on the test bench.

Continuous In-Line Monitoring of an Argon Supply

The objective of the testing at Siemens was to evaluate the new analyzer as a continuous, real-time monitor of the moisture level in process argon. To do so, a gas stick was installed in a supply line approximately 3 m from the gas point of use in an aluminum-copper sputtering chamber and nearly 100 m from the bulk argon supply (see Figure 7). The stick was constructed with two monitors in series separated by a bypass with an in-line gas purifier. This configuration provided continuous monitoring of the supply argon or, following the bypass, argon purified to remove moisture, carbon dioxide, carbon monoxide, and oxygen to levels of <1 ppbv (quantified via APIMS in nitrogen and argon).⁶ In the event of a moisture upset, the bypass line could be opened, allowing the argon to be purified to an acceptable (<20 ppbv) moisture level.

Figure 7: Schematic of the gas line configuration used to evaluate the in-line monitor (ILM) at Siemens.

Figure 8: Upstream and downstream monitor responses indicating base moisture levels following line evacuation and argon purge at a 140-std cm³/min flow. (The downstream monitor follows the purifier bypass.)

The gas lines were prepared by evacuation to a chamber base pressure of 5 x 10^—8 torr followed by purging with argon. As seen in Figure 8, the base moisture level measured by the upstream monitor averaged 13 ppbv with a 140-std cm³/min gas flow, which was consistent with the 20-ppbv value given by a capacitive sensor used to measure bulk gas quality. The downstream analyzer, used to monitor the gas flowing through the purifier, indicated moisture values ranging from 0.8 to 3.0 ppbv, with an average of 1.7 ppbv after 1000 seconds. During one iteration of the PVD process, an elevated moisture level was detected and the gas stick was used to help control the contamination. When the upstream monitor reading increased to approximately 69 ppbv, well above the acceptable level of 20 ppbv, the argon was diverted through the purifier line. The downstream analyzer then detected an average moisture level of only 2.8 ppbv in the purified argon with a 60-std cm³/min stable flow (see Figure 9). This is an example of the advantage of using two analyzers: The upstream monitor provided a warning that a high moisture level was present, while the downstream one verified that the purified gas reaching the process chamber was within the moisture specs.

Figure 9: Upstream and downstream monitor responses when the gas was diverted to the purifier bypass following detection of an elevated moisture level during PVD processing.

This beta test also revealed the need to compensate for the effects of pressure. Because the gas stick was upstream from the chamber MFC in this installation, pressure effects resulting from changes in gas load caused momentary instabilities in monitor performance, as seen in Figure 9. After the pressure stabilized, the sensor oscillation frequency also became stable. (Pressure stability within ±0.06 bar [1 psig] is recommended for in-line monitor installations, and POU installation after the MFC and directly before the chamber isolation valve is suggested to minimize the destabilizing effect of pressure variations.) Figure 10 illustrates one technique for providing pressure-compensated output from the monitor, and current production units incorporate pressure compensation capability.

A 7-ppbv moisture challenge was generated on the manufacturer's test bench to conduct comparisons of a prototype unit and a pressure-compensated unit. Nitrogen was cycled from 100 to 0 std cm³/min at 30-second intervals to simulate gas flow during wafer processing and transferring in a PVD process. As indicated in Figure 10, the overall pressure effect observed as a moisture increase was reduced from a 13-ppbv change to output within 3 ppbv of the generated moisture challenge, providing an accurate moisture reading at start-up.

Figure 10: Responses of a prototype monitor and a pressure-compensated production model to a 7-ppbv moisture challenge.

Leak Detection

In another installation at Siemens, when the monitor detected moisture levels up to 100 ppbv following evacuation and purging of the gas lines, the gas flow rate was varied to determine if the elevated moisture level was dependent on flow rate, which would be indicative of a leak or a virtual (moisture desorption) leak. A series of flow rate changes indicated that the moisture level was inversely proportional to the flow, which can be attributed to the concentration and dilution of moisture as flow rate is decreased and increased, respectively. Figure 11 represents the series of flow changes and the corresponding moisture levels detected by the in-line monitor. Decreasing the flow rate from 60 to 30 std cm³/min nearly doubled the moisture concentration detected, while subsequently increasing the flow from 30 to 60 std cm³/min caused the moisture level to decrease by approximately 20%. This flow dependence suggested the presence of a leak in the system. If the actual moisture level in the gas was >100 ppbv, the concentration detected by the in-line monitor would not have varied with flow.

Figure 11: Moisture levels detected by the in-line monitor as the gas flow rate was changed. The flow dependence of the moisture shown here suggested the presence of a leak.

To determine if there was a leak, the moisture level in the supply gas was measured with a capacitive sensor at two locations (indicated on Figure 7 as in-house measurement sites) and both were found to be 20 ppbv. The lines were then evacuated through the in-line monitor gas stick to the chamber for 50 hours, during which time elevated moisture levels were again detected by the monitor. Next, the integrity of the line fittings was probed using in-board helium leak testing with a quadrupole mass spectrometer operating as a residual gas analyzer (RGA) installed at the chamber. The RGA was unable to detect the increase in moisture, but it did detect a helium leak into the system. A leak was then found in a face-seal fitting, and after it was repaired the moisture detected by the monitor returned to acceptable levels.

Conclusion

The beta-site testing of the in-line monitor illustrates the benefits of collaboration between a manufacturer and potential users during product development. The complementary response time analyses performed by the manufacturer and at Air Products' laboratory confirmed the monitor's rapid response to and recovery from moisture challenges. The sensor response time is faster than that of the gas sampling lines. The 0.5-ppbv challenge testing also illustrated the sensitivity of the device to moisture levels below 1 ppbv. Additional tests at Air Products addressed the monitor's accuracy. Moisture challenges over a 0—10-ppbv range showed there was a 1:1 correlation between moisture generated and moisture detected, and a series of moisture challenges fed to the monitor and an APIMS showed that monitor readings compare well with the highly sensitive mass spectrometric analyzer.

The beta testing at Siemens research center involved installing a two-monitor gas stick in an argon line for a PVD application. Results showed that the devices could be used to track differences between gas with moisture levels of 20 ppbv (near the threshold of process acceptability) and purified to moisture levels below 1 ppbv. Monitor data also led to the discovery of a leak that otherwise was not detected. In addition, the test data gathered at Siemens indicated that the monitor should be installed in pressure-stable locations to minimize the effects of pressure variations on measurement stability. These results led to modifications of the monitor that minimize the effects of pressure variations.

The extensive testing performed at Air Products and Siemens, along with additional beta testing elsewhere, provided valuable insight into the new monitoring technology and how it can be used in detecting moisture contamination in-line in UHP argon and nitrogen applications before such moisture causes particle formation and electrical defects on wafers. The analyzer's accuracy, fast response time, and continuous measurement capability demonstrated in these beta tests complement other system attributes, such as a small footprint, ease of use, and relatively low cost. Further multisite evaluations will be performed for forthcoming monitor models for the measurement of moisture in helium, hydrogen, oxygen, and corrosive gases to achieve increased control over moisture levels in process gas lines.

References

1. Wei J, Pillion JE, King SM, et al., "Using an In-Line Monitor to Obtain Real-Time Moisture Measurements," MICRO, 15(2):31—36, 1997.

2. Wei J, Pillion JE, and Hoang C, "In-Line Moisture Monitoring in Semiconductor Process Gases by a Reactive-Metal-Coated Quartz Crystal Microbalance," Journal of the Institute of Environmental Sciences, March/April, pp 43—48, 1997.

3. Bzik TJ, Smudde GH Jr., Zatko DA, et al., "Limit of Detection," in Specialty Gas Analysis: A Practical Guidebook, New York, Wiley, chap 8, 1997.

4. Bzik TJ, Smudde GH Jr., and Martinez de Pinillos JV, "How Good Is Your Limit of Detection?," in Proceedings of the 1994 Microcontamination Conference, Santa Monica, CA, Canon Communications, pp 653—671, 1994.

5. Ridgeway RG, Ketkar SN, Zatko DA, et al., "Determining Limits of Detection for Analytical Methods Used for the Determination of Trace Impurities in Gases," in Proceedings of the 1992 Microcontamination Conference, Santa Monica, CA, Canon Communications, pp 293—302, 1992.

6. Waferpure Micro Integrated Filter/Purifier, data sheet no. PF023, Bedford, MA, Millipore, 1993.

Maura I. Toth is an applications engineer at Millipore in Bedford, MA, where she has been involved in the development of the in-line monitor and is responsible for providing technical support for the European monitor market. She has a BS in chemical engineering from the University of New Hampshire, Durham. (Toth can be reached at 617/533-2143.)

David A. Zatko, PhD, is a lead research chemist for Air Products in Allentown, PA, with responsibility for evaluation and development of new trace-level analyzers for semiconductor applications. He is also responsible for the development of multimedia-based training in use of the new analyzers.

David F. Yesenofski is principal research technician in the analytical technology group of the electronics division in the gases group of Air Products and Chemicals in Allentown, PA. He works on development of new trace analyzers for moisture and oxygen in bulk and specialty gases.

Manfred Schneegans, PhD, is on the technical staff at Siemens's advanced research center in Neuperlach, Germany, working in PVD metallization for DRAM and logic devices. He studied physics at the Technical University of Braunschweig in Germany, and investigated metallurgical alloys melted in UHV-vacuum systems in his PhD thesis.

Johann Helneder is an engineer in semiconductor development at Siemens Munich, where he is responsible for the development of advanced semiconductor metallization techniques.

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