Modified pump, trap system cuts down PECVD maintenance
07/01/2000
Peter Fancourt, Motorola MOS9, East Kilbride, Scotland
Sajid Ishaq, Mike Czerniak, BOC Edwards, Crawley, England
overview
PECVD deposition of oxy-nitride films is well-known as a source of large quantities of fine particulate by-product that can cause maintenance issues for vacuum pumps and downstream exhaust gas treatment devices. Selection of an appropriate pump and modifications to the gas abatement system can significantly improve the maintenance lifetime of equipment downstream of the PECVD process tool. These straightforward modifications have eliminated associated process downtime for more than two years at a Motorola lab.
The basic PECVD-oxide deposition reaction chemistry has the following form:
SiH4 + 2N2O --> SiO2 + 2H2 + 2N2
It is possible for silica (SiO2) to form homogeneously in the gas phase, resulting in the formation of fine powder in the vacuum foreline of a deposition system. The quantities involved can amount to grams per wafer and can result in maintenance issues for vacuum pumps and exhaust gas-abatement systems. At Motorola's MOS9 Fab in Scotland, this meant that vacuum pumps needed servicing as frequently as monthly, and the thermal processor unit (TPU) on the exhaust line needed daily attention.
 TPU combustion unit at operating temperature. |
Motorola uses the TPU as part of its commitment to reduce global warming, having signed the EPA's memorandum of understanding to reduce emissions of perfluorocompounds (PFCs) used in chamber cleaning. When using C2F6, the BOC Edwards TPU achieves >99% destruction removal efficiency (DRE) [1], and safely abates leftover pyrophoric deposition gases.
A traditional approach to dealing with solids in downstream lines from a process chamber is to use trace heating that encourages volatile species to remain in the vapor phase. Unfortunately, this approach does not work for nonvolatile silica, as evidenced by thermogravimetric analysis (TGA) (i.e., weighing the sample during controlled heating; see Fig. 1 on p. 148).
Our problem amounted to being able to successfully handle the large quantities of by-product silica powder without compromising wafer throughput or uptime on a Novellus Concept 1 process tool, or compromising PFC destruction removal efficiency on the TPU. Our goal was to extend vacuum pump and TPU service intervals to match those of the tool, making the former effectively "invisible" from a downtime point of view, which for the process tool was typically 12,000 wafers, or about four to five weeks.
Separating the issues
Our first priority involved separating the various elements in our systems that could contribute to the problem.
 Figure 1. Thermogravimetric analysis of foreline powder. |
Two TPUs, each serving two PECVD chambers, were originally connected in parallel so that either could back up the other. For our test, we isolated these to prevent any crossover effects. Not all our deposition tools run the same recipe; other processes included silicon nitride and a TEOS-based oxide.
We installed catchpots and gate valves at the inlets to process pumps. The former collected some of the powder generated as a by-product of the process, thus protecting the pump and TPU from blockage. The latter allowed the pump to run continuously, avoiding possible problems associated with cold restarts.
We fitted a cyclone trap, which we expected to be effective for particles >1µm [2], to the pump exhaust as an additional level of particle protection for the TPU. This trap included pressure monitoring and a bypass assembly. It also served as an expansion volume to condense saturated TEOS vapors from the oxide process. (A cyclone trap is a cylindrical device with a tangential inlet and coaxial outlet that separates powders from a gas stream by virtue of their greater mass than those of the gas molecules. The powder collects at the base of the trap.)
Vacuum pump issues
We assessed the condition of the process pump—80m3/hr with a 1200m3/hr booster—by monitoring motor current. Such monitoring showed us that during normal pump operation (Fig. 2), no current transients or "spikes" occurred that would be characteristic of the ingestion of significant quantities of solids.
Because the pump running current remained stable, even over several months, we concluded that there was no significant long-term accumulation of solids in the pump.
 Figure 2. Typical vacuum pump motor current characteristics. |
Although we noted some powder still appeared downstream of the pump, the catchpot did not fill to a depth greater than ~1.5cm (a circle of compacted powder was evident on the transparent catchpot lid). Observation during foreline pumpdown confirmed that these effects were due to rapid depressurization of the collected powder while opening the gate valve (an effect that was simulated in the BOC Edwards laboratory). Our solution was to fit an antisurge valve (ASV, a spring-loaded baffle plate) to throttle the gas flow during rapid pumpdown. We sized the ASV to have minimal effect on conductance during normal operation.
The ASV has been in use for the last two years and has enhanced the catchpot's efficiency to the extent that approximately 0.51kg of powder is collected between tool services. Estimated efficiency is ~50%; this probably consists of particles >10µm [2]. Due to its low density, this accounts for a volume of ~1 liter.
 TPUs in production at BOC Edwards' new facility at Clevedon, UK. |
We concluded that the foreline catchpot with baffle valve was effective at trapping foreline powders that could reduce pump lifetime. Maintenance of the catchpot and filters has been phased to coincide with regular scheduled maintenance of our deposition tools. We have not seen any pump failures due to solid ingestion throughout our tests. Furthermore, analysis of the pump motor current indicates that the pumps are not under stress, and therefore, powder ingestion is unlikely to be a long-term cause of failure.
At this point in our tests, we replaced the original pump with a newer model [3] that features a roots-claw mechanism, a cantilevered shaft design, an inverter-powered direct-drive booster, and specific improvements to combat process by-products. Such improvements include improved powder handling, which is very important for PECVD applications in particular, and the elimination of high-vacuum bearings that can be vulnerable to process by-products and require regular maintenance.
Importantly, when we replaced the pump we also achieved reduced cost-of-ownership benefits, including:
- interface compatibility with earlier pump models to ensure simple swap-out,
- reduced water consumption because the pumping mechanism is not water-cooled,
- reduced nitrogen consumption, without compromising dilution for toxic or pyrophoric gases, because the pump runs hotter, and
- a smaller footprint, which is becoming a critical factor in new fab designs.
With our original pump and vacuum configuration, the interval between pump servicing could be as short as four weeks. Since installing the catchpot, ASV, and new pump in April 1998, the pump has not required servicing (as of April 2000) and plans are to leave it running until it stops.
Exhaust gas management issues
Initially, although we decoupled both TPUs and installed a cyclone trap in the inlet line, the TPU combustion liner (i.e., where the thermal decomposition of the incoming gas takes place) still experienced considerable accumulation of solids. Inspection revealed:
- Heavy deposits of nonvolatile, fine white powder at the top of the combustion liner, identified as SiO2(TGA analysis).
- Hard, white deposits on the TPU inlet that reduced the effective inlet diameter for process gases, thereby increasing process gas exit velocity into the combustion chamber.
- Resultant hardening of the surface of the ceramic liner of the combustion chamber, which is usually quite soft to the touch, reducing its service life to a matter of weeks.
- A high, 60-80°C plenum-chamber temperature, where 30-50°C is normal; this chamber acts as a manifold for the methane-air fuel mixture.
- A low, 550°C combustion temperature, where 850°C is normal, measured at the inlet to the combustion chamber.
(The function of the combustion chamber liner is to support uniform incandescent combustion using an inward-flowing gas-air mixture, similar to a gas mantle. Operation is detailed in Ref. 4.)
 Figure 3. A 16-inlet head TPU liner after burning >6000 liters of silane in "high-fire" mode, revealing the same symptoms seen at MOS9. |
We sent a used ceramic liner to the manufacturer (Alzeta Corp.) for analysis. The surface showed signs of crystallization, which reduced the pore size, restricting the flow of the methane-air mixture, especially at the top inlet-end of the liner. We believe this was the underlying cause of our TPU problems. Low liner porosity results in less volumetric flow through the liner surface per unit time than standard. This means that heat build-up in the liner from the combustion processes is not as efficiently removed. This heat build-up can lead to further degradation of the liner material.
With a replacement ceramic liner in place, we increased the pressure difference (DP) across the venturi mixing tee by 40%. This had the effect of increasing the methane-air flow through the liner, which we hoped would help keep solid deposits away from the surface. The DRE of C2F6 was measured at this stage and found to be complete, within the >99% resolution of the mass spectrometer.
Much to our surprise, far from resolving the solids problem, the new set-up actually made things worse, as it reduced the liner life to two weeks. The increased volumetric flow through the liner was not sufficient to stop deposition gases from impacting the liner, and the higher running temperature inside the combustion chamber, resulting from the increased use of fuel and air, accelerated the densification of the liner behind the deposited material. At this point we decided to mimic the process and abatement set-up at the UK BOC Edwards Research Laboratories.
Laboratory simulation
In the lab, we simulated the Motorola set-up by running silane (SiH4) into a TPU continuously set to "high-fire" mode (i.e., the set-up for PFC destruction); this TPU had a 16-nozzle inlet [4]. Our test reproduced the 30-min.-long cleaning step used on Novellus Concept 1 tools after numerous wafers are processed. (Single-wafer PECVD tools typically clean for approximately 90 sec after each wafer.) With this simulation, we observed the exact same symptoms we had previously seen at MOS9 (see Fig. 3).
 Figure 4. Relative gas velocities of the two types of inlet heads versus combustion gas velocity. |
At the time of these tests, a new design of the TPU was available featuring a "larger" four-nozzle inlet system. Detailed mathematical analysis of the relative process gas exit velocities from the 16- and 4-nozzle TPU heads was performed in the laboratory. The results suggested that for the 16-nozzle head the expansion velocity of the inlet gas exceeds the inward velocity of the methane-air mixture through the TPU liner. This allows process gases to impinge on the hot liner surface and decompose to form silica in the form of a hard "glaze" that blocks the pores of the ceramic matrix that makes up the liner. The hard glaze reduces fuel-air throughput, and therefore, there is very little gas to transport heat build-up in the liner out into the combustion section. This effect accelerates the wear already started by the deposited materials.
Use of larger-diameter inlet nozzles reduces process gas velocity and, hence, the expansion velocity of these gases from the nozzles. This should have the effect of reducing the amount of process gas impingement on the liner surface (see Fig. 4).
To examine this theory, we tested a four-inlet head under similar conditions; the results are shown in Fig. 5.
 Figure 5. TPU liner with four-inlet head after >6000 liters of silane in high-fire mode, showing minimal hardening and minimal deposits. |
Our laboratory work showed us that, as process gases enter the hot combustion section from the relatively cool inlet nozzles, they expand outward toward the liner. The rate of expansion is proportional to the process gas velocity. If the velocity is high, the rate of expansion will be high and there will be a higher probability of process gases impacting the liner material. Using a 4-nozzle head in place of a 16-nozzle head reduces the process gas velocity and, hence, decreases the probability of process gases impacting the liner. The expansion due to heat is the same for both 4- and 16-nozzle TPU heads. The difference between the heads lies solely in the expansion due to nozzle velocity.
Thus, the TPU 16-nozzle process gas velocities appeared to be the root cause of liner deposition in this instance. Our conclusion from these tests was that reducing the expansion velocity of the inlet gas by using a four-nozzle inlet head was effective at minimizing liner degradation. This resulted in negligible solids accumulation on the liner. The total solids collected after 6000 liters of SiH4 was <0.5g; this represents >99.997% removal of the 16kg of SiO2 created by the oxidation of this gas. This is removed by the combined action of the combustion chamber and subsequent wet scrubber stage of the TPU.
Field tests
We repeated the laboratory results at Motorola MOS9 by fitting a new four-nozzle inlet head to TPU No. 1 and reducing dP back to its original value. This resulted in minimal deposition on and no hardening of the ceramic liner, normal combustion chamber and plenum temperatures, and no reduction in the effective diameter of the inlet nozzles. In addition, our service interval extended from a few days to four to five weeks (i.e., equal to the service interval of the PECVD tool). Further, liner life increased from several weeks to six months.
Conclusion
Several accessories and changes can result in a transformation in the reliability of pumping and gas abatement downstream of a PECVD oxy-nitride tool. These changes have now resulted in several years of vacuum pump lifetime (i.e., no maintenance). In addition, servicing of the TPU abatement equipment is matched to that of the PECVD process chamber, resulting in zero impact on wafer throughput, while maintaining the highest levels of PFC removal (>99%).
In summary, the modifications involved:
- fitting a catchpot to the vacuum foreline with an ASV to trap and hold by-product powder,
- installing a gate valve at the pump inlet to allow the pump to run continuously and, therefore, eliminate possible restart seizures,
- using a process-hardened dry vacuum pump,
- fitting cyclone powder traps in exhaust lines between the vacuum pump and the TPU to remove residual particles before they enter the abatement device, and
- using a four-nozzle inlet TPU.
The set-up at Motorola MOS9 has been running without issues since the middle of 1998.
Acknowledgments
The authors thank Ken Aitchison of Novellus, Santa Clara, CA.
References
- S69 Evaluation of an Edwards/Alzeta Thermal Processing Unit (TPU) Designed to Abate Perfluorocompounds (PFCs), Sematech Technology Transfer 95113010B-ENG, 22 December 1995.
- C.E. Lapple, Stanford Research Institute Journal Vol. 5, p. 95, (Third quarter, 1961).
- BOC Edwards iH1000.
- P. Chandler, et al., Fabtech International, 10th Edition, p. 103, 1999.
Peter Fancourt started with Motorola's MOS9 Fab as a shift technician in 1989. Since 1996, he has worked on equipment enhancement projects involving Novellus Concept One, Watkins Johnson 999, AMAT P5000 systems, and Edwards pumps and gas abatement systems. Fancourt is a CVD equipment engineer at Motorola MOS9, East Kilbride, Scotland; ph 01355-356201, fax 01355-265460.
Sajid Ishaq received his BSc in applied chemistry and a PhD from the University of Strathclyde, UK. Ishaq is an applications technical specialist covering Europe for BOC Edwards.
Mike Czerniak received his BSc, MSc, and PhD in physics from Manchester University, UK. He is the applications engineering manager for BOC Edwards' exhaust gas management facility in Clevedon, UK.