Using partial pressure analysis to monitor wet clean recovery
08/01/2000
Thomas P. Schneider,* Craig H. Huffman, Kathryn Morse,
Texas Instruments Inc., Silicon Technology Development Center, Dallas, Texas**
Rick Van Meurs, Lam Research Corp., Fremont, California**
Christopher A. Tripp, Ferran Scientific Inc., San Diego, California**
*Currently with Cree Inc., Durham, North Carolina
**Additional authors are listed in the Acknowledgments.
overview
 The main processing chamber of Lam Research's 9600 TCP metal etch system. |
In general, when a fault in a vacuum system occurs, particularly after wet cleaning, it is usually the troubleshooting that takes the largest amount of time. Equipping the troubleshooter with a partial pressure analyzer, however, provides an effective tool for accelerating the troubleshooting process and, hence, accelerating corrective action. This tool can be an effective means for productivity enhancement of vacuum process tools.
The very nature of vacuum processing involves the science of carefully cleaning internal chamber surfaces and new components. In general, the cleaning process includes exposing the chamber to atmosphere or high-purity N2, cleaning the internal surfaces and adding new components, and evacuating the chamber for processing. Even if high-purity N2 is used for venting the chamber, the internal surfaces experience some atmospheric exposure.
Re-establishing a vacuum following a high-pressure exposure can be quite challenging. Most vacuum technicians and engineers have a good sense for the rate of change of pressure values observed on the pressure sensors during the pump-down from atmosphere. If a problem with the vacuum quality occurs, the pressure sensor will detect it in some cases. The pressure sensor is not useful for diagnosis, however. Examples include:
- Atmospheric leaks — they are common, and typically a result of a poor air-to-vacuum seal.
- Virtual leaks — they are a result of trapped gases inside the vacuum slowly diffusing out of a component or seal and into the vacuum system.
- Outgassing components — they are as simple as a fingerprint on a newly installed component, or as complicated as an o-ring seal cleaned with a solvent.
- Leaky flow control systems — they are the result of poor seals on valves for mass flow controllers (MFCs) or helium leaks from electrostatic chucks (ESCs).
Fortunately, partial pressure analysis (PPA) is a metrology technique that is easily applied to all vacuum-processing systems. Each problem listed can be detected and diagnosed rapidly with PPA where, by comparison, a simple pressure manometer is insufficient.
 |
Most processing systems are without PPAs. The current practice in the semiconductor industry is to initially watch the total pressure gauge for monitoring the process chamber during the pump-down from atmosphere to the required base pressure. The process chamber is pumped for 15-60 min, or a pump-purge sequence is used to cycle dry gas into the chamber. Finally, the process chamber is isolated from the pump and monitored for rate of rise (or leak-back). If the base pressure and rate-of-rise test are acceptable, the process chamber is considered for process qualification.
If the vacuum quality is poor following an atmospheric exposure, the total pressure will occasionally reflect a problem. Typically, diagnosis of the symptoms ranges from using a portable leak detector to dripping solvents on air-to-vacuum seals. Both methods can be time consuming and clumsy; neither method is sufficient for diagnosis.
 |
Application of PPAs is more productive. There are reported successes using PPA for process gas purity [1] and fault detection and classification (FDC) on ion implantation systems [2], chemical vapor deposition (CVD) tools [3], and sputter deposition tools [4]. These past successes point the way to the clear need for PPA use in next-generation tools [5]. Indeed, PPAs can be applied to catch process-tool problems before they result in yield loss.
We have developed a rather elegant approach to monitoring vacuum quality during recovery from atmosphere. Measuring and trending key species during the pump-down sequence yields information on the quality of:
- vacuum seals (e.g., o-rings or Conflat),
- the presence of virtual leaks,
- outgassing of newly installed components (e.g., ceramic inserts for a plasma processing system or targets and shields in a sputter deposition system), and
- the flow control system (e.g., gas manifold valves and MFCs).
In addition, monitoring the pump-down sequence provides information on the evolution solvents used to wipe down the internal chamber surfaces.
The variables listed above should not go unmeasured before or during manufacturing. Using the breadth of information monitored during the pump-down sequence, one can optimize protocols used to return process chambers to a productive status and maintain on-board fault detection systems.
Our development work involved using PPA to find key gas species during the pump-down from atmosphere for a plasma metal etcher [6, 7]. This technique can be generalized to any process chamber including ion implanters, furnaces, sputter deposition systems, and high-temperature annealers, thus optimizing the wet clean recovery procedure.
PPA configuration
We found that the two most useful data display modes from the PPA were analog spectra, initially used for species identification, and partial pressure trend, used to monitor changes in gas species' partial pressure levels. Our mass resolution was <1amu. In addition, our PPA enabled a fast scan mode that captured the maximum number of data points for trend curves. This was especially important during the initial pump-down from atmosphere.
We networked two PPA systems to monitor two metal etchers from the quality control (QC) ports on the endpoint window. We added a pneumatic valve to this setup so that we could isolate the PPA from the vacuum system during regular processing; the PPA control unit automatically controlled this valve by opening it at 50mtorr during the chamber pump-down and collecting data when the pressure reached 10mtorr. If the pressure went above 10mtorr, the system stopped data collection. If the pressure went above 50mtorr, the pneumatic valve closed.
Automatic operation of the PPA provided a data-management advantage. The system's Windows-based software provided automatic data storage and the capability to recall wet clean recovery data for offline analysis. This software also supports dynamic data exchange (DDE), where the PPA data can be called by the factory automation system for display in the factory system software.
This enhancement makes it possible for factory personnel to view, analyze, and develop statistical process control (SPC) with the PPA data in a familiar environment.
Standard wet-clean recovery
We can generalize most post-maintenance pump-down processes in terms of the partial pressures of the constituent gases. For purposes of simplicity, our focus here is on moisture (H2O), nitrogen (N2) and oxygen (O2), and expected trends of these atmospheric constituents during the pump-down of a vacuum chamber to the chamber's base pressure (Fig. 1).
Initially, these trends are extremely steep as the pumping system removes the bulk of the N2 and O2. After these are predominantly exhausted from the chamber (see crossover points in Fig. 1), H2O becomes the dominant species contributing to the total pressure; after an atmospheric exposure, moisture desorption from internal chamber surfaces is the largest contributor to the total pressure.
 Figure 3. Partial pressure trends from a recovering plasma processing system, showing what is typical for a normal process chamber pump-down. |
The crossover points indicate two key aspects of the vacuum chamber: the volume of the chamber and the efficiency of the pump. Following the crossover points, moisture desorption from the internal chamber surfaces determines the rate at which the system reaches base pressure. If a significant atmospheric leak is present, no crossover points would exist and the N2:O2 ratio would be close to 5:1 [8]. PPA data can be collected shortly after the crossover points (for non-differentially pumped PPAs). The uniqueness of the PPA used here allows for data collection beginning at 10mTorr.
A typical analog spectrum (Fig. 2) collected shortly after the crossover points shows dominant spectral features at 18amu (H2O+), then 28amu (a combination of N2+ and CO+). These data also show peaks at 2amu (H2+), 32amu (O2+), and 44amu (CO2+) [8]. Notice that the residual partial pressures of atmospheric species contribute to the background.
Plasma system recovery
We used the PPA to monitor the standard procedure for a wet clean recovery for a plasma processing system, noting the partial pressure trend for the 18amu, 28amu, and 32amu species shortly after the crossover points (Fig. 3). The data in Fig. 3 include the initial pump-down, pump/purge cycling time, a rate-of-rise test, and the background partial pressures after the vacuum system was deemed ready for process qualification. The overall trends are quite similar to those shown in Fig. 1 (see region labeled "MPA data" in Fig. 1).
 Figure 4. PPA data from an unsuccessful wet clean recovery that led to a diagnosis of two different problems. |
In Fig. 3, pump-purge cycling data were not included because the pressure reached 125 mtorr during cycling. Therefore, the horizontal axis has a gap of ~70 minutes. Although the purpose of the pump-purge cycling is to dry down the vacuum system, the data in Fig. 3 show that the pump-purge cycling was not effective. With such knowledge from a PPA, one can develop an optimum dry-down procedure.
We used the PPA rate-of-rise test to determine that the vacuum chamber was, indeed, leak-tight. In contrast to waiting 100 min for the conventional rate-of-rise testing, it was clear from the PPA data that the process chamber was leak-tight ~2 minutes after data collection began. The gap in the data at 110 min, shortly after the rate-of-rise test, shows when the PPA system was temporarily turned off.
Collectively, the PPA data that we gathered provide insight into the standard recovery process and enable a more rapid return of a tool to production status.
Fault detection and classification
In addition to monitoring and optimizing a vacuum recovery protocol, a PPA can be used for fault detection and classification. Classification is imperative during troubleshooting or diagnosis. We define fault detection as significant variation from normal and classification as the determination of the root cause for variation.
For example, consider an attempted wet clean recovery of a plasma processing system, which was subjected to a brief exposure to atmosphere and a quick wipe-down of the internal chamber surfaces, that is inhibited by both desorption of a cleaning solvent and a process gas manifold leak (Fig. 4). For this problem, data collection was initiated ~20 minutes after the pump-down was started because the chamber had not passed a conventional rate-of-rise test, so the equipment technician decided to diagnose the problem with the PPA. Figure 4 shows two distinct regions that help diagnose the two components of the overall situation.
 Figure 5. Analog spectra comparing a normal (from Fig. 2) and an unsuccessful wet clean recovery. |
The detection of cleaning solvents (Fig. 4, region 1). When we compared analog spectra data from a normal wet clean recovery and the region 1 data (Fig. 5), it was clear that the region 1 data contained additional spectral features. We identified these as gas-phase isopropyl alcohol (IPA) in the process chamber. The source of the high partial pressure of IPA was determined to be the polymer on the process chamber walls (i.e., aluminum-etch by-products and photoresist). More specifically, when the internal chamber surfaces were quickly wiped down, the polymer absorbed the IPA. Following the pump-down the IPA was desorbing from the polymer and contributing to a significant partial pressure background. This example shows how PPA data are useful for both fault detection and classification of solvents in the process chamber during an attempted wet clean recovery.
 Figure 6. Analog spectra comparing a normal and an unsuccessful wet clean recovery showing a unique spectral feature at 36amu. |
The detection of a process gas leak (Fig. 4, region 2). Prior to the use of PPA in this situation, the equipment technician thought that the rate-of-rise test failed as a result of moisture slowly desorbing from the chamber walls. So, he decided to strike an inert gas plasma to accelerate the desorption of species from the chamber walls. With the proper gases flowing, however, a proper RF match was not established, and no on-board system diagnostics were available to help diagnose the problem. Here again, the PPA analog spectra data were very useful. Comparing the Region 2 spectrum data to a normal wet clean recovery (Fig. 6), the technician saw an additional spectral feature at 36amu, representing a collection of Cl+ and HCl+. It was then quickly determined that the root cause of the high partial pressure of HCl was a result of Cl2 in the gas manifold, likely due to a leaky MFC valve.
Cost avoidance analysis
Early detection of process chamber problems is critical to preventing misprocessing of product wafers that typically results in scrap. The ratio of the cost to develop and maintain PPA applications to the cost of lost revenue from scrapped product approaches zero within a short time (weeks to months). Consider, for example, the cost-effectiveness of gas purity monitoring:
Currently, 300mm diameter silicon wafers cost $1190 [9]. Wafer value increases with each process step, but for simplicity in this example we ignore processing costs. We also assume that each wafer takes an average of 3 minutes to process and includes set-up time and chucking and dechucking. The PPA used in this study costs $9000 for the PPA system, a pneumatic ISO-valve, and engineering time for installation and training.
Using the scenario, 3 min/wafer x 1 wafer/$1190 x $9000/system = 22.6 min/system.
The PPA pays for itself in 22.6 min of scrap prevention. In other words, the PPA system cost is nearly the same as seven wafers, and the PPA pays for itself by saving seven wafers that would have otherwise become scrap. If an end-product profit per device of $100 is assigned and there are 250 devices/wafer, the scrap produced in 23 min impacts ~$189,000 of gross revenue (23 min x 1 wafer/3 min x 250 devices/wafer x $100/device = $188,332).
We can only conclude that the development of PPA applications in high-volume manufacturing will result in increased profits.
Conclusion
We used a PPA to monitor a process chamber recovery from atmospheric exposure after wet cleaning maintenance. The PPA data were useful for determining the extremely low efficacy of the standard dry-down process. It also eliminated extensive rate-of-rise testing and other unnecessary time wasted during vacuum recovery. The rapid fault detection and classification led to accelerated corrective action that quickly returned the process tool to a productive state; the PPA was used as a tool for productivity enhancement. A cost avoidance analysis clearly suggests the potential of PPA for increasing profit from wafer processing operations.
As a general guideline, one should apply the PPA to monitor wet clean recovery by trending atmospheric residual gases, the solvents used for cleaning, and process gasses. The wet clean recovery work presented here can easily be applied to most vacuum-processing systems.
Acknowledgments
Additional authors include Gregory H. Leggett with Texas Instruments Silicon Technology Development Center; Dave M. Bullock with Lam Research Corp.; Said Boumsellek and Robert J. Ferran with Ferran Scientific Inc.; and Brad Van Eck at International Sematech. The authors appreciate the support they received from Rebecca Gale, Stephanie Butler, Gabe Barna, and Lam Research Customer Service. We also appreciate the efforts of Tammy Fletcher, A. Dean Springfield, and Dennis Shipley for assisting in the installation of the MPA data collection systems.
References
- T.P. Schneider, et al., "Using partial-pressure analysis to detect contamination in an oxygen gas supply," Micro, January 1999.
- T.P. Schneider, P. Krocak, B. Van Eck, "Real-Time In Situ Residual Gas Monitoring During Ion Implantation in High Volume Semiconductor Manufacturing," Future Fab International 4, pp. 237-239, 1997.
- A.M. Haider, T.T.H. Fu, R.W. Rosenberg, "Investigating an In Situ RGA for Process Monitoring and Diagnostics," Micro, p. 35, October 1995.
- R.J. Markle, et al., "Volatile Contamination Monitoring for Titanium and Titanium Nitride Sputter Systems," Semiconductor Fabtech 4, p. 237, 1997.
- L. Peters, "Residual Gas Analysis," Semiconductor International, pp. 94-101, October 1997.
- A commercially available miniature PPA, a MicroPole Analyzer (MPA) available from Ferran Scientific Inc., was used to monitor the main chamber of a Lam 9600 TCP metal etcher.
- R.J. Ferran, S. Boumsellek, "High-Pressure Effects in Miniature Arrays of Quadrupole Analyzers for Residual Gas Analysis from 10-9 to 10-2 Torr," J. Vac. Sci. Technol. A 14, p. 1258, 1996.
- M.J. Drinkwine, D. Lichtman, "Partial Pressure Analyzers and Analysis," American Vacuum Society Monograph Series, eds. N.R. Whetten and R. Long, Jr., 1978.
- We have used the 300mm dia. wafer example from the Si-based microelectronics industry for a reasonable comparison to the compound materials (II-VI and III-V) and batch vacuum-processing industries.
Thomas P. Schneider has a BS in physics from Moravian College, an MS in physics in applied quantum theory from Saint Bonaventure University, and a PhD in physics from North Carolina State University. Schneider works on process engineering and defect reduction for Cree Inc., 4600 Silicon Drive., Durham, NC 27703; ph 919/ 313-5300, fax 919/313-5452, e-mail [email protected].
Craig H. Huffman is a member of the technical staff and etch equipment module manager at Texas Instruments' Dallas facilities.
Kathy Morse is a plasma etch equipment engineer in Texas Instruments' Kilby Fab.
Rick Van Meurs is senior technology manager in the etch division at Lam Research Corp.
Christopher A. Tripp is VP of Sales at Ferran Scientific Inc., San Diego, CA.