LEVELS OF ATMOSPHERIC CONTAMINANTS IN INERT
SEMICONDUCTOR PROCESS GASES
J. Scott McWilliams, Alan Johnson, Charlene Long, Jeremiah D. Hogan*
Texas Instruments, Inc.
Dallas, TX 75265
Low ppb detection limits for contamination in cylinder gases have been achieved with gas chromatography using a modified gas injection system and megabore capillary columns with a variety of detectors including mass spectrometry, atomic emissions, and discharge ionization. Components of interest are separated in a single run via a unique valve switching scheme. A computer-controlled gas blending unit provides multiple levels of dilution of standard gases allowing for the generation of comprehensive calibration curves with a high degree of precision and accuracy. Reproducible detection limits below 50 ppbv are routinely achieved with the equipment and methods described herein.
The extreme requirements for chemical purity enforced by the growing demands of large scale integration in the semiconductor manufacturing industry have driven the technology for increased sensitivity in the field of gas analysis. Those laboratories tasked with performing analysis on cylinder gases have traditionally depended on dedicated analyzers for their work, but have more recently turned to gas chromatography (GC) with a variety of detectors. Ionization detectors such as flame (FID), discharge (DID), helium (HID), and photo (PID), along with ultrasonic detectors (USD), and thermal conductivity detectors (TCD) have been employed to characterize gas samples. Most labs taxed with these types of analyses have turned to GC with one or more of the detectors mentioned above. While the sensitivity provided by instruments such as reduction gas analyzers, electrolytic oxygen analyzers and similar dedicated analyzers is impressive, the information they provide gives but a narrow glimpse of the gas under scrutiny. Also, purchasing an array of dedicated analyzers such as this in order to get a broader view can be quite expensive. Performing full-scale analysis, such as that offered by chromatography, can provide much more insight into the contamination present in the gases of interest. Virtually all chromatographs, however, are designed for the analysis of liquid samples. This design characteristic causes problems for the gas analyst, as instrumental sensitivities are typically below that required for trace-level analytical work on gaseous samples, not to mention the modification of hardware which is required before gas phase samples can be successfully introduced into the instrument. This work details a chromatographic method and instrument modifications which have led to the development of capabilities for routine trace-level analytical testing of cylinder gases
A Hewlett Packard 5988A GC/MS was used in all of this work. The stock sample inlet of the 5890 GC was replaced with a Valco ten-port gas injection valve with a 250 ml sample loop. Figure 1 shows the valving scheme utilized for all work presented here. This arrangement allows for baseline separation of the major components of air in one 16-minute run. Separations are accomplished with the use of two columns, in series. The first column is a Chrompak PoraPLOT Q. It is followed by a Molsieve 5A column from the same manufacturer. Both are 0.53 mm OD x 0.33 mm ID x 25m fused silica 'Megabore' capillary columns.
The stock data system and software of the 5988A were replaced with a third party system from Teknivent, as the reproducibility and accuracy of the Teknivent Vector/2 software integration routines proved to be more accurate and reproducible at the levels of interest in this work. Since the spectra here acquired are not subjected to library searches or otherwise used for comparison, electron energy is used as a tuning variable. For most of this work an electron energy of 58 eV was used. This value is well above the ionization potential of all species of interest in these analyses. Also, since the contaminants of interest are below m/z 69, the m/z 18, 28, 32, and 44 peaks of air were utilized in conjunction with perfluorotributlyamine (PFTBA) for tuning/calibration purposes. Normal system leakage provides sufficient air for this operation. Figure 1 shows a calibration report which lists all parameters for a typical tune file used in routine analyses performed in this lab. Also shown in this figure are three of the peaks used for tuning and calibration, namely water (m/z = 18), nitrogen (m/z = 28), and carbon dioxide (m/z = 44).
RESULTS AND DISCUSSION:
The prime contaminants involved in most areas of gas analysis are the components of air. Nitrogen, oxygen, carbon dioxide, carbon monoxide, methane, and several others at lower concentrations, are the focus of most of the attention when analyzing cylinder gases to the sub- ppm level. The act of filling a cylinder with a product gas while surrounded by a 'sea' of air is difficult for obvious reasons. These same pitfalls are again encountered when trying to retrieve a representative sample from a cylinder and deliver it to an instrument for analysis. System mechanical integrity must be of the highest level in order to avoid violating the sample with the omni-present contaminant, air. Toward this end, tedious schemes for minimizing sampling- system leakage have been developed. The word 'minimize' is used here to emphasize the fact that there is essentially no way to completely avoid system leakage. All systems leak at some level. It will therefore not be considered possible to produce a truly 'leak-free' system when performing gas analysis at the trace levels.
Since there are few analytical instruments on the market which arrive from the manufacturer in the appropriate configuration for performing trace-level gas analysis, virtually all equipment used for this purpose requires modification prior to utilization. While the majority of the modifications required are performed on the inlet system, several other variations from convention are employed in this work. As the 5890 GC arrives from the factory, it is setup for liquid injection. The liquid injector must first be removed and replaced with suitable gas injection system. The immediate consideration in this modification is the design of the chromatographic separation. The chromatography involved in the separation of the components of air from an inert balance gas such as argon or nitrogen is not particularly difficult. In this work, the PoraPLOT Q (PLOT) column is used to separate methane and carbon monoxide from the other major components of air. As shown in figure 2, the eluents from a sample injected onto a PLOT column include a single peak for nitrogen, neon, oxygen, carbon monoxide and argon. This 'air' peak is followed by methane, and then by carbon monoxide. The Molsieve column will further separate the 'air' peak, but will irreversibly retain carbon dioxide. It is therefore necessary to arrange the separation such that the carbon dioxide is not exposed to the Molsieve column. This is accomplished with the injection valve arrangement shown in figure 3. Using this configuration, the sample is introduced onto the PLOT column when the valve is switched from the Load position to the Inject position. This performs the initial separation described above. The valve is returned to the Load position when the methane has cleared the PLOT column, and before the carbon dioxide has entered the Molsieve column. The timing of this event is determined by performing a pretest injection on an 'as necessary' basis, typically immediately after tuning or calibration, (figure 4). This valve switch forces the carbon dioxide to be 'short circuited' directly into the mass spectrometer, avoiding the Molsieve column altogether. The remainder of the constituents are directed onto the Molsieve column, where the additional separation takes place. Note that the flow direction through the columns is never reversed, as doing so tends to degrade chromatographic resolution. After elution from the Molsieve, the separated components again pass through the PLOT column. This additional trip is unavoidable without flow-reversal, but it does not cause any deleterious effects. Figure 5 shows the completed separation of a 100 ppb standard multicomponent blend utilizing the method described above. In practice, detection limits below 50 ppb are normally obtained.
As in all trace level gas analysis, system leakage is paramount in achieving sufficiently low background levels to facilitated low detection limits. Toward this end, several measures have been taken which either minimize system leakage, or at least attenuate the effects of such leakage. For work of this nature, it is not reasonable to believe that a truly 'leak free' system can exist. At the limit of reality, virtually everything leaks, to some degree. While some types of fittings are proven to seal better than others, all fittings are considered to be sources of leakage. Given this fact, and since it is impossible to entirely eliminate all connections from a system of this sort, all fittings in the chromatographic stream are located in a purged housing on top of the chromatograph which is kept under pressure with a few psi of ultrapure helium in a custom made, clear plastic housing. The only fittings in the flow stream of the chromatograph which are not located in this housing are those at the purifier for incoming carrier gas, and the connection at the interface between the GC and the mass spectrometer. In addition to this measure, the number of fittings has been further minimized by connecting all chromatographic columns and the GS/MS transfer line directly into the ports in the injection valve, without adapting from capillary column to l/16' stainless steel tubing. This has facilitated the removal of eight fittings from the flow stream. The use of glass-filled Teflon ferrules (Scientific Instrument Service) has helped to facilitate this interconnection. These (green) ferrules allow for this direct connection more readily than ferrules made of other materials. Care must be taken when making these interconnections, as it is possible to push the end of the capillary column into the valve head too far, causing a scoring of the valve rotor or crushing of the end of the column. Removing the preload assembly and rotor during the column installation process allows the operator to verify proper column placement in the valve assembly by looking down into the rotor seating area during and observing column placement.
The interface between the GC and the mass spectrometer is also non-traditional, in that there is no form of separator or other eluent splitting involved. The transfer line consists of ten meters of 0.2 mm I.D. fused silica capillary column which is connected to a port on the injection valve and routed directly into the mass spectrometer. All eluent from the chromatograph is swept into the mass spectrometer, yielding base pressures which are typically on the order of 2 x 10-6 torr.
Calibration is accomplished with the aid of a microprocessor-based Environics Gas Blender. This particular unit is the result of several years of collaboration with Environics, Inc. (Tolland, CT). The original commercial unit consisted of Swagelok connections, standard-grade stainless steel tubing, mass flow controllers with polymeric seals, and industrial-grade solenoid valves. These materials of construction were unacceptable for calibration gas generation in the region of interest for this work. Refinements include a fitting minimization scheme, and replacement of all unavoidable fittings with metal-seal VCR connectors. All tubing in this system is chemically cleaned and electropolished, and joints are orbitally welded. All mass flow controllers in the system are of all-metal seal construction, and the solenoid valves are zero dead volume bakeable T-solenoids. In addition, the unit is designed such that each individual gas circuit is automatically and continuously purged with high purity helium when it is not actually involved in the dilution scheme. This feature minimizes memory effects and allows the user to more quickly achieve desired concentrations. These refinements have dramatically improved the ability of this unit to produce reliable dilutions of a multicomponent calibration gas over many orders of magnitude with a high level of accuracy and reproducibility. In these applications, a calibration cylinder consisting of fifty ppm each of eight components in a balance of helium are dynamically diluted in five steps over a range from 1 ppm to 50 ppb in order to generate calibration curves which span the area of interest. Calibration acceptance is based on obtaining a correlation coefficient for a linear least-squares-fit line through the six data points (forcing zero) of at least 0.995. The analytical sample is then introduced into the injection valve sample loop through the same pathway as the calibration standard, via the gas blender, in order to ensure consistency of sampling.
This work has provided the ability to perform routine quality control testing of cylinder-based gas phase products with the necessary precision, accuracy, and flexibility required for screening a variety of gases utilized by the semiconductor industry. The purity specifications for gases analyzed in this lab range from several ppm to below 100 ppb for the contaminants listed above. The modified instrumentation and chromatographic methods described herein provides the ability to perform routine QC screening for a variety of grades of argon, nitrogen, and helium cylinder gases. The dynamic gas blending system has been paramount to the success of this work. Reliance on common 'zero-and-span', two-point calibrations has proven to be unacceptable for this type of analysis at these levels. Plans are to extend these methods to the analysis of non-inert gases including those which are corrosive, pyrophoric, and/or toxic.