Four Steps to Healthier Vacuum system
03/01/1997
Four steps to a healthier vacuum system
Herbert W. Gatti, Mass-Vac, Billerica, Massachusetts
Lisa C.H. Laurin, ClearTech, Newton, New Hampshire
The costs of maintaining vacuum processes in a semiconductor manufacturing facility exceed most other maintenance costs. Pump failures, clogged or sticking valves, and clogged exhaust lines are some of the catastrophic failures that cause lost yield, high downtimes, and high labor costs. Routine maintenance can reduce the frequency of catastrophic failures, but requires significant downtime and labor. In addition, many vacuum processes cause stress on the downstream effluent treatment facilities, further increasing equipment downtime and maintenance costs. To reduce these costs, many vacuum experts have turned to multiple-step trapping. Efficient trapping of nongaseous effluent keeps the vacuum system free from residues, reducing or eliminating pump failures and clogged lines. The cost savings that result include increased throughput (capacity), reduced personnel costs, and reduced service costs.
Traditionally, trapping uses a single particle-elimination step, either a "knock-down" step, where baffling and gravity eliminate large solid particles from the gas stream, or a water cooling step, which solidifies and removes high vapor pressure materials. While these methods reduce the periodic maintenance required, they do not eliminate catastrophic failures because they leave significant quantities of nongaseous effluent in the exhaust stream. Multiple-step traps (also called multitraps), however, remove the majority of failure-causing effluent. The steps include a "knock-down" step, and an optional water-cooling step, followed by one or two steps of mechanical and chemical filtration (see "The MV MULTI-TRAP - Multiple trapping steps in an efficient package"). Results from ion implant, LPCVD, PECVD, and plasma etch show that this multiple-step approach reduces unscheduled downtime by over 70% and scheduled maintenance by over 50%, and increases equipment capacity by 1-2%. No catastrophic failures have been reported after installation of the appropriate multitrap. By eliminating solids from the exhaust stream, multiple-step trapping also reduces stress on the scrubbing system.
Cleaning up an LPCVD silicon nitride vacuum system
The deposition of silicon nitride films from the thermal reaction of dichlorosilane and ammonia creates large amounts of solid matter in the exhaust line, which, if not taken care of, cause significant downtime. The solid effluent from an LPCVD silicon nitride process consists of ammonium chloride, which sublimes at 340?C [1], silicon nitride, and silicon dioxide [2, 3]. Ammonium chloride is the most difficult constituent to deal with, since it condenses in active components such as the isolation valve and the pressure control valve, as well as in the active areas of the pump. Ammonium chloride build-up causes frequent pump failures - as often as once every two months. The problems caused by the effluent from this process appear to be universal, with failures occurring on both horizontal and vertical furnaces, from various manufacturers (ASM, BTI, SVG to name a few), and with both wet and dry pumps.
The dry pump on one ASM vertical furnace running silicon nitride failed, on average, every two months. This system deposits a 7500-? thick film, over the course of 5 hr, using a 3:1 ammonia to dichlorosilane ratio. The user performed routine maintenance every 50 runs, or every 375,000 ? of deposition. To reduce maintenance and catastrophic pump failures, the user installed a three-step trap, consisting of knock-down, water-cooling, and filtration with long stainless steel mesh filter elements (Fig. 1). The trap proved to be efficient, doubling the time between routine maintenance to 100 runs, or every 750,000 ?. The trap has successfully protected the dry pump from failure for well over a year.
Upon disassembly of the three-step trap during routine maintenance, the user found heavy accumulations of ammonium chloride in the knock-down and water cooling steps. Apparently, the ammonium chloride remained in the vapor phase and did not condense until reaching the sharp pressure drop provided by the knock-down step. Additional ammonium chloride condensed in the water cooling step and was subsequently removed as the exhaust vapor cooled. A fine dust, probably consisting of silicon nitride particulates, deposited on the stainless steel gauze elements as well.
Figure 1. A multitrap in the foreline of an LPCVD system protects from catastrophic pump failure.
High payback
A few rough calculations show that improving this vacuum system provides a huge payback:
Loss of a production run due to a pump failure could occur every two months before installation of the trap. After installation, no losses have occurred in more than a year. Assuming that this is one of the first process steps on the wafer, we can look at the loss of one production run simply as the cost of the wafers themselves:
1 run/2 months ? 12 months/year ? 100 wafers/load ? $150/wafer [4] = $90,000/year savings
Pump rebuild frequency drops from six times/year to under once/year. At $5000
ebuild [5], the cost reduction per year is 6 ? $5000 = $30,000/year savings
Frequency of preventive maintenance drops in half, increasing the furnace throughput. Assuming 100% equipment utilization and four hours downtime each time preventive maintenance is performed on the vacuum system, the minimum effect on throughput can be calculated:
Downtime before installation: 4 hours/250 hours operation
Downtime after installation: 4 hours/500 hours operation
Minimum throughput increase = 4 hours additional operating time/500 hours operation = 0.8%
(Note: Using a capital cost of $1,000,000 for the vertical furnace [6], this 0.8% capacity increase can be evaluated at 0.8% ? $1,000,000 = $8000, more than twice the cost of the trap.)
Figure 2. Multiple trapping steps remove nongaseous effluent, reducing or eliminating vacuum system failures.
Similar results on other systems
When installed on a new SVG vertical furnace, a similar trap configuration shows similar results, with 980,000 ? of deposition between preventive maintenance and no pump failures to date. In this case, the exhaust line between the furnace and the trap is heated to 150?C, creating the same effect as the water cooling step in the trap.
In another example, an SVG horizontal furnace, depositing silicon nitride at a 5:1 ammonia to dichlorosilane ratio, incorporated a four-step trap. This trap included the knock-down step and water cooling, followed by a stainless steel gauze filtration step and a 20-?m polypropylene filtration step (Fig. 2). The combination of the stainless steel gauze and the polypropylene filtration creates the same net result as the long stainless steel filter elements. Preventive maintenance requirements dropped from one every 30 runs to one every 52 runs. The user has reported no pump failures after installation of the multitrap.
Significant throughput increases for CVD silicon dioxide
Silicon dioxide dust wears at a vacuum system, reducing pumping efficiency. Many silicon dioxide sources, such as tetraethylorthosilicate (TEOS), can recondense in the vacuum lines [7], causing valves to stick and eventual production losses. With most silicon dioxide sources, multiple-step trapping reduces these problems, saving the user time and money. SEMATECH installed a four-step trap [8] in an Applied Materials Precision 5000 TEOS reactor. The trap reduced the preventive maintenance frequency from one/shift to one every three months. SEMATECH estimates that the trap added 250 hours of operation/year to the system.
A BTI horizontal furnace running undoped TEOS averaged a pump failure every six months. To minimize these failures, the user performed preventive maintenance every 47 runs. To reduce the occurrence of pump failures, the user installed a three-step trap, consisting of a knock-down step, a stainless steel gauze filter, and a 20-?m polypropylene filter. After installation of the trap, the user reported no pump failures for over one year, and had increased the time between preventive maintenance sessions to one every 88 runs. In this example, the savings include reduced line yield losses due to pump failure, reduced costs for pump rebuilding, and a slight capacity increase (Table 1).
A silane-based boro-phosphosilicate glass (BPSG) process provides similar results. One user had frequent line clogging and pump failures, despite the presence of a single-step trap in the vacuum line. This trap used only a knock-down step and required cleaning every 60,000 ? of deposition, or every two days. Once installed, the multitrap requires cleaning every 350,000 ?, which occurs only every 10 days. Inspection during cleaning shows that heavy silicon dioxide powder accumulates in the knock-down step, followed by a light powder coating of the filter elements. No accumulation appears in the vacuum line outside the trap, giving a good indication that clogged lines have been eliminated as a source of vacuum line failure. Savings for this process are also summarized in Table 1.
Plasma processes also benefit from effluent trapping
Plasma processes produce the same type of effluent as LPCVD systems, and plasma vacuum systems fail for the same reasons. In a Plasma Therm process for silicon nitride, pump failures occurred every two to three months. Installation of a three-step trap cleaned up the vacuum system. This trap consists of the knock-down step and a stainless steel gauze filtration step, followed by a 5-?m polypropylene step. It allowed the user to reduce maintenance frequency from once every 250,000 ? to once every 500,000 ? of deposition, giving up to 30 days between maintenance services.
Another user did not perform any preventive maintenance on the vacuum system of an ASM PECVD system, resulting in pump failures every two months. To eliminate these failures, the company installed a four-step trap consisting of knock-down, water cooling, a stainless steel gauze filter, and a 5-?m polypropylene filter. The trap requires cleaning after 60 days of service. The result is a scheduled down period of about four hours every two months, in place of a catastrophic failure at about the same frequency. Table 2 shows the minimum payback for these two processes.
Trapping after the pump
In the case of a Novellus PECVD system, a multitrap installed downstream of the pump solved clogging problems in the lower stage of the dry pump. Initial installation consisted of a knock-down step and a stainless steel gauze filtration step, followed by a 5-?m polypropylene filter. This set up required cleaning every six to eight days. To reduce the cleaning frequency, the user replaced the 5-?m polypropylene filters with 20-?m filters. While there is too little data to calculate a payback for this user, particles no longer clog the lower stage of the pump, and there is a significant reduction in required maintenance.
Figure 3. With metal etch processes, a multitrap installed after the pump can prevent clogged exhaust lines.
Figure 4. Multiple-step trapping after the vacuum pump of a metal etcher precipitated large quantities of solid material that would otherwise have clogged the exhaust lines. When placed after the vacuum pump in a metal etch process, the cooling and filter steps are more effective than the knock-down step, as shown in this cooling section from a trap installed on an Applied Materials etcher.
Trapping prevents exhaust line clogging in metal etch processes
When using a dry pump with a metal etch process, pump failures are infrequent. However, the exhaust line tends to clog, causing process failures. Once the exhaust line clogs, it needs replacement, a costly and time-consuming task.
Users of Applied Materials, Electro Tech, and LAM etchers have installed three-step traps in the exhaust line immediately fol
lowing the pump (Fig. 3). The traps are similar to the one used in the SVG horizontal furnace for silicon nitride deposition. In all cases, they have avoided exhaust line replacement with a simple filter change. In the Electro Tech installation, the user replaced 20 feet of exhaust line every two months. Estimating the cost for new exhaust line at $100/linear foot, this operation cost the user at least $2000 every two months. Installation of the multi-trap saved over $12,000/year.
Table 3 shows savings achieved on the Electro Tech and LAM metal etch processes. Figure 4 shows the actual effluent trapped from the Applied Materials etcher. The trapping in these cases is different from trapping achieved in the foreline, upstream of the pump. With the trap after the pump, much less accumulation occurs in the knock-down step. A fine powder collects in the water-cooling step, followed by heavy powder in the filter elements.
Reduced maintenance on ion implant
An Eaton Ion Implanter using phosphine and arsine required changing the exhaust line every two weeks, due to build-up. By installing a simple two-step trap, the user reduced maintenance frequency from once every two weeks to once every six months. This user estimates that the trap collected 95% of the phosphorus coming out of the process. After six months, the exhaust line had only a slight coating. Assuming that it took four hours every two weeks to change the exhaust line before installation of the trap, and four hours to replace the filter every six months after installation of the trap, the throughput increase equates to $23,000 (see Table 4), nearly 40 times the cost of this trap.
Multiple steps are required to keep a clean vacuum system - and it pays. The cost savings achieved by installation of a multitrap on CVD, etch, and ion implant processes far exceeds the cost of the trap and its consumables. Even when replacing a single-step trap, the multitrap keeps the vacuum system cleaner, reducing maintenance downtime, and apparently eliminating catastrophic failures of pumps. The capacity increase afforded by these improvements is small, typically 1-2%. With the high cost of vacuum-related processes, however, this small increase is worth many times the cost of the trap.
Acknowledgments
MV MULTI-TRAP is a trademark of Mass-Vac Inc. Sodasorb is a registered trademark of W.R. Grace & Co.
References
1. CRC Handbook of Chemistry and Physics, 61st edition, CRC Press, p. B-75, 1980.
2. John F. O`Hanlon, David B. Fraser, Journal of Vacuum Science Technology, A6 (3), pp. 1246-1247, May/June 1988.
3. Raul A. Abreu, A. Troup, M.K. Sahm, Journal of Vacuum Science Technology, Volume B 12 (4), pp. 2763-2767, July/August 1994.
4. Industry average of an n- or p-type 200-mm wafer without epitaxy.
5. Source: Mass-Vac Inc. Typical rebuild costs for a dry pump are between $5000 and $6000. In addition to manufacturing traps, Mass-Vac also provides pump rebuild services.
6. SEMATECH, "0.25 Micron Process Equipment Requirements," Oct. 31, 1994.
7. Peter Burggraaf, Semiconductor International, pp. 83-86, Sept. 1995.
8. Dan Babbitt, SEMATECH Technology Transfer #95072916A-TR, Aug. 31, 1995.
9. SEMATECH, "0.25-micron process equipment requirements," Oct. 31, 1994.
HERBERT W. GATTI holds an associate`s degree in chemical engineering and a BBA in management and engineering from Northeastern University. He is president and acting director of Research & Development for Mass-Vac Inc., a rebuilder of mechanical vacuum pumps and distributor of vacuum equipment, which he founded in 1971. Mass-Vac Inc., 247 Rangeway Rd., P.O. Box 359, N. Billerica, MA 01862; ph 508/667-2393, fax 508/671-0014, e-mail [email protected].
LISE LAURIN received her BS degree in physics from Yale University. She started her career as a process engineer at Intel Corp, and after 15 years of technical and marketing experience in the semiconductor industry, she founded Clear Tech, a marketing services business serving the semiconductor and other technology industries.