In Situ Gas Analysis Improves Cluster Tool Contamination Control

		At a Glance
	High levels of moisture and oxygen contamination in the process chamber can be a direct result of the purge procedures used in the loadlock and the choice of cassette materials. By evaluating several different purge procedures and cassette materials, an optimized loadlock process was developed.

Single-wafer cluster tools are used in many critical applications in today's semiconductor manufacturing process. The cluster tool provides ambient control during wafer processing that is not attainable using batch processing. Clustering sequential process chambers to match the IC process flow results in control of interfacial properties that cannot be achieved if the wafers are exposed to cleanroom ambient between processing steps.

The loadlock of the cluster tool is the key isolation gate that prevents contamination from the cleanroom ambient from entering the process chambers. In this paper we report the results of a comprehensive evaluation of cluster tool contamination using a residual gas analyzer (RGA). In situ analysis has shown loadlock oxygen and moisture contamination levels are significantly impacted by system operating parameters and the choice of wafer cassette material. We have also shown that contamination in the loadlock chamber is transported to the central wafer handler and subsequently to the process chamber during wafer processing. Using the data from in situ analysis, equipment operating procedures were developed to minimize contamination transport from the loadlock during wafer processing.

Experimental

The loadlock, central wafer handler and process module of a single-wafer cluster tool were analyzed for moisture and oxygen contamination using a differentially pumped quadrupole mass spectrometer (commonly referred to as an RGA). The supply gas streams for each of the chambers were fitted with purifiers rated to reduce the moisture and oxygen impurities to <1 ppb. The RGA system was connected to the chambers under test using a custom-designed sampling system and inlet. The sampling system included a temperature-controlled sampling manifold, pressure reduction, pressure control, sample switching and a calibration gas source. To improve the accuracy of the analysis, sensitivity factors were obtained using calibration gas sources containing known concentrations of moisture and oxygen. During this investigation, the RGA was always calibrated at the same pressure as the sample measurement in order to minimize the pressure dependence effects on sensitivity factors. Typical operating pressure for the RGA was 2x10^-6 Torr. Typical detection limits were <200 ppb for moisture and <40 ppb for oxygen.

Initially, each of the test chambers were evaluated to determine the moisture and oxygen contamination levels using the standard operating procedure. The standard operating procedure after loading a cassette of wafers into the vacuum loadlock was to pump the loadlock to a pressure of 1 Torr, and then backfill with nitrogen to the central handler pressure. When the central handler pressure was reached, the nitrogen flow to the loadlock was shut off, and a valve was opened to equalize the pressure between the loadlock and the central wafer handler. The central wafer handler was maintained at 10 Torr with a continuous nitrogen purge.

Results

The initial evaluation of the loadlock is shown in Figure 1. A cassette of wafers was loaded and gas sampling was done from LLA at the standard conditions. The oxygen and moisture levels measured in the loadlock for three time intervals after pumpdown are shown. Ten minutes after pumpdown/backfill, the oxygen level in the loadlock was 20,000 ppm, and the moisture level was 2000 ppm. After six hours the oxygen had decreased to <5000 ppm, but the moisture level had increased to >60,000 ppm. Although outside the calibration range of the QMS (maximum error estimated at 25%), the presence of these impurities at percentage levels was quite alarming. There was immediate concern that the surfaces of wafers that were in the vacuum loadlock for an appreciable time period would be subject to significant contamination potential before being processed through the clustered thermal deposition or oxidation process.

The oxygen levels could be explained by simple dilution calculations. The ambient room air that entered the loadlock during loading of the cassette contained ~21% oxygen (210,000 ppm). The pressure in the loadlock was reduced to 1 Torr, then backfilled to 10 Torr with nitrogen. This would give a dilution factor of 10:1 and an initial oxygen concentration of 21,000 ppm after backfill. With the central wafer handler under nitrogen purge, the oxygen concentration was <1 ppm. Because of this concentration gradient, the oxygen would diffuse from the loadlock to the central handler chamber via the open equalization valve, thus explaining the reduction in oxygen over time. The source of the moisture was not as readily understood. Assuming cleanroom conditions of 70ˆand 50% relative humidity, the moisture concentration in the cleanroom ambient would be ~13,000 ppm. The likely sources for moisture was outgassing from the cassette of wafers or the loadlock chamber walls. Since there is no nitrogen purge in the loadlock, the moisture level would likely increase with time, assuming that the outgassing rate is greater than the diffusion rate through the open equalization line.

1. In the initial evaluation of a loadlock, oxygen and moisture levels were monitored. Note the high level of moisture even after 6 hrs.

2. Cycle-purging can significantly reduce oxygen levels. Three different pressure cycles are shown.

3. Oxygen and moisture levels for various time intervals after the Mod 2 sequence, where the chamber was pumped to base pressure (about 25 mTorr), backfilled to 700 Torr, pumped to base pressure and then backfilled to 10 Torr.

To achieve further dilution, two cycle-purge pumpdown/backfill sequences were evaluated. In the first sequence (Mod 1), the loadlock was pumped to 1 Torr, backfilled to 700 Torr, pumped to 1 Torr and then backfilled to 10 Torr. The second sequence (Mod 2) was pump to base pressure (~25 mTorr), backfill to 700 Torr, pump to base pressure and backfill to 10 Torr. During the initial cycle-purge tests, no wafers were loaded into the loadlocks because of the lengthy time required for sample manifold drydown after sampling very high levels of moisture.

Figure 2 shows that cycle-purging (C-P) significantly reduced the oxygen level as expected. Ten minutes after the Mod 1 sequence, the oxygen level was 377 ppm; 10 min after Mod 2, the oxygen level was 1.41 ppm. Figure 3 shows the oxygen and moisture levels for various time intervals after the Mod 2 sequence (again with no wafers loaded). The initial oxygen and moisture levels were 4 ppm and 8 ppm, respectively. The oxygen level increased to ~37 ppm after 2 hrs and then remained stable for the remainder of the 5 hr test.

The moisture level increased to 850 ppm after 3 hrs and then stabilized. The increase in oxygen and eventual stabilization was attributed to the intrinsic leak rate of the loadlock chamber and reaching a steady-state concentration gradient between the chamber leak and the nitrogen purged central handler chamber. The increase in moisture was assumed to be related to the leak rate and desorption from the chamber walls.

After completing the initial loadlock sequence optimization work, the impact of the loadlock C-P on contaminant migration into the central handler was evaluated. For this experiment, a full cassette of 25 wafers was placed in the loadlock. After the loadlock sequence was complete, an automated wafer handling program was executed, which removed a wafer from the loadlock and placed it in the wafer cooldown chamber. After 8 min, the wafer was returned to the cassette, and the next wafer in the cassette was removed and transferred in the same manner. Figure 4 shows the resulting change in moisture and oxygen concentration in the loadlock for the standard loadlock sequence (atm/1 Torr/10 Torr) and the Mod 2 sequence (atm/0.025 Torr/700 Torr/0.25 Torr/10 Torr).

A sharp increase in oxygen was seen when the slit valve was opened to extract the first wafer following the standard loadlock sequence. This result is consistent with the high oxygen levels measured in the loadlock after the standard preparation sequence. The oxygen would enter the central handler each time the slit valve was opened, increasing the concentration to about 100 ppm, and then decrease back to about 10 ppm before the wafer was returned to the loadlock and the next one removed. (Note that the initial oxygen concentration in the central handler was ~0.1 ppm.)

4. The resulting change in moisture and oxygen concentration in the loadlock for the standard loadlock sequence and the Mod 2 sequence.

Using the Mod 2 sequence, the initial oxygen contamination of the central handler oxygen was minimal but increased as wafer handling continued, eventually peaking at about 5 ppm when the slit valve was opened and dropping to about 0.5 ppm between wafer movements. This increase over time suggested that most of the oxygen contamination of the central handler when the Mod 2 sequence was used was not from air residual but rather from outgassing of oxygen from the loadlock chamber, the cassette of wafers and/or a chamber leak.

Figure 4 also shows the results for moisture during the same experiment. Except for the first few wafers, the loadlock sequence had little impact on the moisture contamination. For both sequences, the significant increase in measured moisture in the central handler chamber did not occur until after several wafer exchanges had taken place. This was attributed to the adsorption of moisture by the central handler chamber and sampling line after increasing from the starting moisture content of ~1 ppm. After the walls and sample line were wetted, the moisture content peaked at about 250 ppm when the slit valve was opened and would drop to 150 ppm between wafer movements. The similarity of the moisture trace for both loadlock sequences led to the belief that the most significant moisture source was not from the cleanroom air directly, but again from desorption from the chamber walls and the cassette of wafers. These results illustrated that C-P alone would not adequately reduce oxygen and moisture contamination of the central handler chamber.

5. Cassettes are shown to be the main source of moisture and oxygen in the loadlock.

6. Five cassette materials were tested to determine their outgassing characteristics.

Following the initial evaluation of the central wafer handler, a series of experiments were completed to better understand the source of moisture in the loadlock chamber. Four experiments were run: 13 wafers single-spaced, 13 wafers double-spaced, cassette only and no cassette/no wafers. It was originally believed that the increase in contamination with wafers present might have been due to insufficient purging of impurities between wafers. Thirteen wafers were used so that the overall surface area for both the single- and double-spacing experiments would be the same. The optimized loadlock sequence (760/1/700/1/700/1/10 Torr) was used prior to each experiment. The results of the experiments for moisture are shown in Figure 5.

Although the experiments with single-spaced wafers gave slightly higher concentrations of moisture, the most important result was that the cassette was the main source of moisture in the loadlock. Similar results for oxygen also indicated that the cassette was a major contributor of oxygen into the loadlock. The outgassing of these impurities from the cassettes explained the gradual increase in impurities measured in the central handler chamber during earlier experiments.

Subsequent experimentation revealed that the amount of moisture and oxygen outgassing was dependent on the cassette material. Five cassette materials were tested to determine their outgassing characteristics. The standard cassette material for this cluster tool process was black static dissipative polyetheretherkeytone (PEEK). This cassette was tested along with black static protective polypropylene (PP), blue polypropylene (PP), translucent perfluoroalkoxy polytetrafluoroethylene (PFA) and black static protective PFA. The results are shown in Figure 6. Again, the optimized loadlock sequence (760/1/700/1/700/1/10 Torr) was used prior to each experiment. Data were taken 15 min after the loadlock sequence. From these results, it may be deduced that the addition of carbon powder or fibers for anti-static properties makes the cassette material either more porous or more adsorbent of moisture. The loadlock moisture contamination was an order of magnitude higher for black Teflon (PFA) and polypropylene compared to standard cassettes without the anti-static properties. The anti-static PEEK displayed the most moisture outgassing of all cassettes tested. The materials displayed quite different behavior for oxygen. The anti-static additives did not affect the oxygen outgassing for PFA and polypropylene. PFA cassettes had the highest level of oxygen outgassing and PEEK had the lowest.

7. Oxygen and moisture concentration in the loadlock after an optimized loadlock sequence.

8. No signature related to wafer movement was present for either moisture or oxygen in the central handler chamber.

As described earlier, after the central handler pressure was reached at the end of the loadlock sequence, the nitrogen flow to the loadlock was shut off and the a valve was opened on a 3/8 in. line to equalize the pressure between the loadlock and the central handler chamber.

Thus, except for diffusion through the equalization line and passage through the slit valve during load/unload, impurities that desorbed from the cassette after completion of the loadlock sequence as well as impurities from leakage remained in the vacuum loadlock chamber. This has been evident throughout the figures by the increase in oxygen and moisture concentration with time. It was apparent that the loadlock sequence needed to be modified to allow the impurities to be continually removed from the loadlock as they were desorbed.

A series of experiments were conducted to determine the effect of adding a nitrogen purge to prevent desorbed contaminants from accumulating in the loadlock. Figure 7 shows the oxygen and moisture concentration after an optimized loadlock sequence (760/1/700/1/700/1/20 Torr) for no continuous purge flow and with continuous purge flow. A full cassette (PEEK) of wafers was placed in the loadlock for the experiments. The purge flow condition was achieved by manually opening the slow pump valve (1/4 in. line) and the high-vac valve (1 in. line), while at the same time flowing backfill nitrogen to maintain the loadlock pressure at 20 Torr. Based upon chamber volume and pressure rate of rise, the flow rate was estimated at 10 slm. These results indicated that continuous loadlock purge could reduce the contamination levels by more than two orders of magnitude.

The central handler chamber was re-monitored during wafer movement from the loadlock to determine the impact of the culmination of all improvements to loadlock/central handler operation. The results of oxygen and moisture are shown in Figure 8. No signature related to wafer movement was present for either moisture or oxygen at these final operating conditions.

Characterization of the process chamber showed similar improvements in the moisture and oxygen contamination levels.

9. The optimized loadlock sequence showed improvements in the moisture and oxygen contamination levels in the process chamber.

The comparative results for the process chamber moisture levels are shown in Figure 9. Similar improvements were seen for oxygen contamination where the peak oxygen spikes during wafer transport were reduced from 1 ppm to <40 ppb.

Conclusion

It has been shown that significant contamination of single-wafer cluster tool chambers can occur during normal process operations. This contamination can be caused by the operating parameters of the cluster tool or the choice of cassette material. RGA can be used to identify the sources of this contamination and to develop procedures for minimizing the transport of contaminants to the process chamber. This can lead to substantial improvements in contamination control for single-wafer cluster tool processes during integrated circuit fabrication.