Optimization of C2F6 Burnbox Destruction
Burnbox destruction of C2F6 high pump purges can achieve efficiencies near 80%.
John McNabb, Scott Bischke, Hewlett-Packard Co., Inkjet Business Unit, Corvallis, Ore -- Semiconductor International, 4/1/1998
 |
At a Glance | | |
|
| ||
| |||
Within HP and the semiconductor industry in general, the most emitted PFC is C2F6,2 used primarily to clean process chambers following chemical vapor deposition (CVD) operations. Control of C2F6 emissions is one of the end goals of this study. The effluent from the CVD tools examined in this study goes to H2 burning Guardian 4 burnboxes before release to the atmosphere. The primary purpose of these burnboxes is to safely combust silane, but they may also be effective in destroying C2F6. This study characterizes C2F6 destruction removal efficiency (DRE) in the burnbox to allow a more accurate estimate of past C2F6 emissions and provide data to optimize the burnbox to minimize future emissions.
Experiment
The experiment was a three-factor central composite in a cube experiment, with six center points and four test points. The test points were added to facilitate a neural net analysis. The three factors in the original design were total N2 pump purge (slm), air draw (scfm) and inlet thermocouple temperature (Å). As explained below, the outlet thermocouple temperature was also recorded at each treatment condition and was actually used in the modeling. The responses were the H2 valve setting (number of turns) and C2F6 DRE (%).
Optimally, the H2 flow would have been employed as a variable in the experimental design. This flow could not, however, be measured with the available equipment set. Additionally, H2 flow measurement is not available in the multiple burnboxes used at HP and could not be used as a variable to optimize other burnboxes running a similar C2F6 chamber clean. Inlet temperature, which correlates directly with the H2 fuel flow, was selected as a variable in the experimental design. The use of the H2 valve setting (i.e., number of full turns from closed) was considered as a design variable but then rejected as being inconsistent across multiple burnboxes.
The burnbox chosen for the characterization typically receives the effluent from two Novellus Concept I plasma-enhanced CVD (PECVD) systems. However, during the data collection, this burnbox was receiving effluent from only one PECVD system. Under these conditions, the N2 pump and exhaust purges are ~70 slm. An additional N2 source was added to simulate the effect of two PECVD systems. The N2 pump purge rate was varied via using a rotameter.
 |
| 1. Schematic of the Guardian 4 burnbox shows the experimental setup. |
The air draw was controlled with a butterfly valve in the exhaust duct downstream from the burnbox and was measured with a handheld vane-type velometer at the air inlet to the burnbox. The air draw reading was taken at the set point temperature for the given treatment condition. The inlet and outlet thermocouples were cleaned in deionized (DI) water before the start of the experiment and once during the middle of the data acquisition. Figure 1 shows a schematic cross section of the burnbox.
The C2F6 data were acquired using a gas chromograph thermal conductivity detector (GC-TCD) apparatus, sampling from the exhaust duct ~3 ft downstream from the burnbox exit. The sampling tube was several feet in length, allowing the sample temperature to equilibrate with ambient before entering the GC-TCD. The sample temperature was verified with an in-line thermocouple. A sampling pump continuously drew exhaust from the duct, with a valve routing a sample to the GC-TCD on demand. Calibration standards containing C2F6 and CF4 were employed to determine peak retention time.
For each of the treatment conditions, C2F6 abundance was measured at the set point temperature and compared with a baseline measurement of C2F6 abundance taken with no H2 flow after the burnbox was allowed to cool. The ratio of these abundances determined the DRE for that treatment condition. The C2F6 flow was kept constant at 1.6 slm, with no plasma in the CVD tool.
The C2F6 DRE was assumed to have very little dependence on actual C2F6 flow from the processing tool. Since this flow is small compared with the pump purges, the flow dynamics and thermal profiles are likely unaffected by changes in C2F6 flow. Also, the fuel and airflows greatly exceed that of C2F6, so an abundance of combustion species likely exists, suggesting that the DRE is independent of the C2F6 flow.
Results
Table 1 shows the experimental design and the measured responses at each treatment condition. The estimated uncertainty3 of the C2F6 DRE data is 5%.
Table 1. Experimental Data
| Trial | N2 flow (slm) | Air draw @ temp. (scfm) | Inlet temp. (Å) | Outlet temp. (Å) | H2 valve setting (turns) | C2F6 DRE (%) |
|---|---|---|---|---|---|---|
| 1 | 104 | 209 | 700 | 61 | 3.8 | 28.4 |
| 2 | 104 | 166 | 797 (a) | 232 | 9.2 | 33.3 |
| 3 | 104 | 209 | 690 | 64 | 3.8 | 36.3 |
| 4 | 57 | 201 | 703 | 59 | 3.9 | 30.6 |
| 5 | 104 | 205 | 701 | 61 | 3.9 | 31.7 |
| 6 | 152 | 288 | 500 | 40 | 3.2 | 0 |
| 7 | 152 | 244 | 783 (a) | 235 | 12 | 40.9 |
| 8 | 57 | 288 | 505 | 41 | 3.2 | 24.6 |
| 9 | 57 | 240 | 800 (a) | 230 | 12 | 74.8 |
| 10 | 130 | 262 | 800 (a) | 180 | 10 | 52 |
| 11 | 105 | 279 | 692 | 54 | 4.4 | 48.6 |
| 12 | 74 | 214 | 645 | 55 | 4.5 | 42.1 |
| 13 | 104 | 209 | 716 | 57 | 3.7 | 28.9 |
| 14 | 104 | 209 | 465 | 49 | 3.3 | 25.6 |
| 15 | 104 | 140 | 716 | 71 | 3.7 | 22.5 |
| 16 | 152 | 137 | 508 | 43 | 2.3 | 6.4 |
| 17 | 57 | 118 | 832 (a) | 237 | 7.4 | 43 |
| 18 | 57 | 140 | 490 | 50 | 2.3 | 11 |
| 19 | 83 | 140 | 568 | 46 | 2.5 | 6.2 |
| 20 | 152 | 118 | 830 (a) | 220 | 7.1 | 39.7 |
| 21 | 152 | 205 | 700 | 60 | 4 | 33.7 |
| 22 | 104 | 201 | 694 | 59 | 4 | 39 |
| 23 | 119 | 196 | 755 | 64 | 4.3 | 24.7 |
| 24 | 104 | 201 | 697 | 64 | 4.2 | 34.3 |
| (a) Inlet temperature saturated at this value before reaching set point. | ||||||
The inlet temperature was found to saturate at high H2 flows. Further increases in fuel flow affected no increase in inlet temperature. However, the outlet temperature continued to rise and varied at all fuel flows.
The conversion of C2F6 to CF4 in the burnbox was also examined. No CF4 peak could be resolved, and based on our sensitivity to this gas, the maximum amount of CF4 produced was estimated at 10% of the C2F6 concentration on a molar or volume basis.3
The data were analyzed using a neural net solver, NNAPER.4,5 This analysis gave superior fits for C2F6 DRE (adjusted R2 = 0.95) compared with response surface analyses of the same data (best adjusted R2 = 0.79). Figures 2-5 show predicted contours of C2F6 DRE at two different N2 purges and outlet temperatures. The predicted C2F6 DRE is a strong function of outlet temperature up to about 100Å, increasing rapidly with rising temperature. Under almost all conditions, DRE improves as the air draw increases. The N2 purge dependence is more complicated, but generally DRE is higher at lower N2 purges with the dependence getting stronger as the air draw increases.
 |
 |
| 2. Predicted C2F6 DRE at N2 purge of 70 slm. | 3. Predicted C2F6 DRE at N2 purge of 140 slm. |
 |
 |
| 4. Predicted C2F6 DRE at outlet temperature of 60 Å. | 5. Predicted C2F6 DRE at outlet temperature of 142 Å. |
Discussion
To achieve the desired high C2F6 DRE, outlet temperature and air draw are relatively easy to vary. In contrast, N2 purge is usually set by the number and type of pumps whose effluent goes to the burnbox and is less easily varied.
The burnbox in this study is configured to take the effluent from two CVD tools, with about 140 slm N2 purge. The neural net model was optimized at two fixed N2 purges. These results are summarized in Table 2, along with the predicted DRE at our current operating conditions for comparison. Without changing the N2 purge, the model predicts DRE can be doubled from ~25% to 53% by increasing the outlet temperature to ~115Å. By lowering the N2 purge, DREs approaching 80% can be achieved. Though desirable for minimizing pump maintenance, the use of high purge pumps degrades the PFC destruction efficiency of the burnbox.
Table 2. Predicted C2F6 DRE at Various Burnbox Settings and N2 Purges
| Configuration | Outlet temp. (Å) | N2 purge (slm) | Air draw (scfm) | Model DRE (%) |
|---|---|---|---|---|
| Current setup (N2 = 140 slm) | 50-60 | 140 | 250 | 18-31 |
| Optimized (N2 = 140 slm) | 116 | 140 | 288 | 53 |
| Optimized (N2 = 70 slm) | 142 | 70 | 288 | 78 |
Two factors that are expected to strongly influence C2F6 destruction are temperature and residence time in the combustion zone. The combustion zone temperature is only indirectly measured by the thermocouples. At low fuel flows, the combustion zone is near the inlet. In this regime, the inlet thermocouple is a good indicator of combustion conditions. However, at high fuel flows, the combustion zone shifts down the burnbox, away from the inlet and toward the outlet thermocouple. At high fuel flows, the H2 flame extends far below the deflector plate, and the outlet thermocouple becomes more sensitive to the combustion conditions as the combustion zone temperature rises.
This effect is evidenced by the data in Table 1. At low fuel flows, as the inlet temperature rises from 500Å to 700Å, the outlet temperature increases only about 20 In this regime, DREs are relatively low. However, as fuel flow increases further, the inlet thermocouple temperature saturates, while the outlet temperature rises to more than 200Å. Higher DREs are also observed. Since the highest DREs occurred in a regime in which the outlet thermocouple is the best indicator of combustion zone conditions, we chose to base the model on the output thermocouple temperature, contrary to the experimental design.
In Figures 2-5, DRE generally increases with outlet temperature but levels off as outlet temperature rises above about 100Å. This leveling off is likely an artifact of the nonlinearcoupling between the combustion zone and the outlet thermocouple. The relationship between DRE and outlet temperature would be quite nonlinear, even if the DRE was a linear function of the combustion zone temperature. Figures 2 and 3 suggest that at high air draw, DRE is not a monotonic function of outlet temperature, decreasing slightly at highest fuel flows. This inconsistency is not well understood.
The increase in DRE as air draw increases may result from greater turbulence at higher airflow. The greater turbulence should increase residence time, increasing the chance of the reaction between C2F6 and an active combustion species. Another possibility is that higher air draw simply increases the flux of oxygen, increasing combustion efficiency. The N2 purge dependence shown in Figures 4 and 5 is complicated, but may also be related to turbulence-induced changes in residence time.
These optimized results only apply in the high pump purge, low air draw regime imposed by our experimental setup. The maximum measured air draw on this tool is 288 scfm, while the lowest pump purge investigated was 57 slm. Without these restrictions, optimization will likely lead to higher DREs. Indeed, considerably higher DREs have been demonstrated at low pump purge6 and by increasing air draw with the addition of a blower.7
Conclusion
We have characterized a hydrogen burning burnbox for its C2F6 DRE in a high pump purge, low air draw regime. We find that DRE generally increases at higher outlet temperature, higher air draw and lower N2 purge, in rough agreement with basic considerations of thermodynamics and residence time. We predict that with our current pump packages and air draw restrictions, we achieve about 25% DRE, and without changing the N2 purge, we could increase this destruction to 53%. The model predicts a maximum DRE of almost 80% within the parameter space studied. This level of DRE, however, could only be achieved with N2 purge rate modification by a change in the number of tools serviced by a burnbox or a decrease in pump purge requirements.
Acknowledgments
We would like to thank Mike Mocella and Chuck Allgood of DuPont for performing the data collection and providing the neural net analysis software. Many thanks also go to Hewlett-Packard fab design, facilities and maintenance personnel for the preparatory work for the testing.
References
- Mike Mocella, SRC Technology Transfer Course: PFC and CFC Emissions and Abatement for Plasma Processes, Dec. 1995.
- Walter Worth, SRC Technology Transfer Course: PFC and CFC Emissions and Abatement for Plasma Processes, Dec. 1995.
- Personal communication, Chuck Allgood, DuPont.
- M.T. Mocella, J.A. Bondur and T.R. Turner, Process Module Metrology, Control and Clustering, SPIE Proceedings, 1594.
- A.J. Owens and M.T. Mocella, Proceedings of the IEEE International Workshop on Artificial Neural Networks, (1991).
- Report available upon request from Mike Mocella, DuPont.
- Report available upon request from Micheal Hayes, Ecosys.
John McNabb is
a manufacturing development engineer with Hewlett-Packard, focusing on dry
aluminum etch processing, global warming gas reduction, process control and
technical information archiving.
Phone: (541) 715-6797
Fax:
(541) 715-0378
E-mail: john_mcnabb@hp.com
Scott Bischke
joined Hewlett-Packard's Inkjet Business Unit in 1989. He has worked as a
manufacturing engineer focused solely on environmental projects since 1991.
Phone: (541) 715-3801
Fax:
(541) 715-0378
E-mail: scott_bischke@hp.com
