 |
 |
 |
 |
|
Comments? Suggestions? Send us your feedback.
MicroMagazine.com |
PROCESS CLEANLINESSImproving tungsten CVD performance with in situ particle monitoringJenny Asbell and Zhijiang Sun, Intel; and Brian Bosch, HYT/Pacific Scientific Advanced, real-time particle detection methods are essential to preventing die yield loss during such critical operations in semiconductor device manufacturing as tungsten chemical vapor deposition (WCVD). Both the process- and hardware-related failures that can affect film quality are accompanied by elevated particle levels, which can signal the nature of the problem. For example, a sudden increase in particle levels during the bulk deposition step can indicate a gas line leak. In situ particle monitoring can provide immediate feedback on such particle excursions, enabling corrective actions to be taken before yield is significantly affected.1,2 This article reports on the evaluation of an in situ monitor installed on WCVD systems at Intel's Fab 9 in Rio Rancho, NM. After a brief description of the equipment and data collection setup, the particle distribution analysis and equipment particle performance are discussed. The accuracy of the in situ monitor was evaluated by applying statistical distribution analysis, steady-state particle modeling, and particle transport calculations to both equipment and process data. Finally, a number of practical uses for the monitor in engineering and production environments are reviewed. Equipment Setup The Model 70 in situ monitor from High Yield Technology (Sunnyvale, CA) selected for this study was evaluated on Precision 5000 WCVD systems (Applied Materials, Santa Clara, CA). This cluster tool can accommodate as many as four cold-wall reactors and one loadlock chamber. Each chamber independently deposits a blanket tungsten film on the wafer and then etches the wafer backside to remove excess tungsten. Two isolated gas lines deliver reactants through the chamber lid, and a perforated pumping plate distributes the reactants through six evenly distributed holes. This configuration ensures even gas flow distribution in the chamber. After each wafer is processed, a chamber clean recipe is executed to remove particles from the chamber walls. As depicted in Figure 1, the monitor's sensor assembly was installed on a weldment housing positioned above an isolation valve and directly beneath the deposition chamber pumping port. The circled area in the figure depicts the cross section where particles intersect with the sensor beam, which is generated by a GaAlAs/GaAs semiconductor laser and focused through a cylindrical lens to increase signal area. A detector oriented perpendicular to the beam path measures the amplitude of the Mie-scattered light that results from the collision of a particle with the near-infrared laser beam. A band-pass filter centered at 776 ± 5 nm filters ambient light, increasing the signal-to-noise ratio.
The tested systems were equipped with SECS-based data collection software installed on a host PC. The software package can provide real-time feedback on key CVD process and chamber parameters, as well as such monitor-related data as particle counts, laser current, and stray light. The monitor sensor reports particle counts to a controller, where the information is digitized and sent via computer bus to the data collection software program. The monitor can store up to 256 data points, providing a buffer to the host computer communication system. Data points are transferred to a raw-data file on the host PC every 1.1 minutes. The software program allows the monitor's sampling frequency to be synchronized with each step in the process recipe, so each particle count represents a step-specific, wafer-partitioned data point. The monitor controller and data collection software can also divide the particle counts into bin numbers based on particle size. For this study, the in situ instrument was used to monitor nucleation, bulk deposition, backside etch, and clean process steps on a wafer-level basis, and the monitor controller was programmed via the host computer to detect the particle size ranges listed in Table I. Thus, particles >=0.2 µm were monitored throughout the entire process.
Table I: Size ranges used to group particle count data into bins, which represent an amplitude range of the Mie-scattered light values. In addition, the monitor controller collects and communicates laser diagnostic values to the data collection software. Laser current data indicate the remaining useful life of the laser, while laser stray values indicate the cleanliness and accuracy of the detector. Observation of these diagnostic values is critical to ensuring accurate data collection. As products from the deposition process coat up the laser window, the stray light value increases proportionally with the level of scattered laser light. If the detector window is not cleaned regularly, it will coat up as well, in which case reflected light from passing particles will not be detected and the stray light value will drop to zero. The software recognizes when these diagnostic values are out of control, triggering an audible alarm. However, during the study preventive maintenance procedures were put in place to reduce unscheduled monitor downtime. Particle Distribution Analysis During scheduled maintenance cycles, a statistical analysis was performed on the particle data collected by the monitor, revealing the relationship between the count rate and the associated counting frequency. A histogram created from the sample of collected data, shown in Figure 2, closely resembles a Poisson distribution, which is representative of steady-state particulation. Because particle data from in situ monitors and data from the wafer surface scanners exhibit different count rate distributions (Poisson versus polynomial-exponential), a correlation study between the monitor and wafer scanners was not conducted.
In order to assess the accuracy of reported bin-level counts, a chart that compares the actual particle distribution with a calculated steady-state distribution was created (see Figure 3). The steady-state distribution was derived from a form of the Gates-Gaudin-Schumann distribution.3 As the figure indicates, the profile of the particle count data follows an inverse third-order distribution between particle frequency and size. The actual bin-level distribution conforms to the calculated distribution within the region of sensor calibration (i.e., 0.3 to 0.5 µm); however, the profiles begin to deviate from one another for particles >0.5 µm. This discrepancy may be the consequence of interpolating the calibrated detection area at a particular size of 0.5 µm with the calculated detection area of a 1.5-µm-diam particle. A third calibration point may further enhance the accuracy of the detection area for particles >=0.5 µm. In addition, the sensor location and bulk flow properties of the effluent sample may have introduced additional bias within this particle size range.
To determine if collisions between molecules were frequent within the monitor's sampling area, an analysis of bulk flow properties was done, revealing that the sensor samples in a molecular diffusion environment. This determination was important since molecular interactions greatly influence the distribution of particles in the effluent stream. Table II summarizes the calculated mean free path data for nucleation, bulk deposition, backside etch, and chamber clean steps. The mean free path of effluent stream molecules in all process steps was much smaller than the 50-mm diameter of the pump foreline. Since Fickian diffusion governs molecular transport in this regime, molecules undergo several collisions with each other before encountering the boundary of the pump-line wall.4 Although these mechanics influence particle distributions, further studies are required to determine the exact nature of particle transport in the sampling region.
Table II: Bulk flow properties of the effluent stream for various process steps. SPC Implementation Because the monitor will detect particles contributed by previous processes as well as particles originating from the WCVD process, a method of discriminating between such particles was required. Statistical process control (SPC) limits were calculated for nucleation, bulk deposition, and backside etch steps from a 3-month population of particle data from multiple chambers, where two out of three out-of-control data points defined the SPC limit. Thereafter, when two out of three consecutively processed wafers exceeded an upper SPC limit, an audible particle alarm was triggered. The bulk deposition step is executed for a substantially longer period of time than the nucleation, backside etch, and clean steps. This longer processing time results in substantially higher particle counts for this step, and masks subtle particle trends. In contrast, the in situ particle counts during nucleation, backside etch, and clean steps typically generate easily recognizable particle trends. In Situ Monitor Applications Failure Mode Analysis. One important application of in situ monitors is to achieve real-time process control. Measuring particles on test wafers using surface scanners is still the most common monitoring method for tungsten processing equipment. In addition to requiring costly test wafers, this postprocessing inspection procedure decreases tool utilization and increases the operators' workload. More importantly, if a problem is not identified until after a postprocessing inspection, additional production wafers may be subjected to improper processing. In some cases, severe particle excursions result in the loss of many wafers. The following three particle excursions were detected during the evaluation of the in situ monitor. Wafer Lift Mechanism Misalignment. As mentioned earlier, the tungsten deposition process includes a final etch step to remove residual tungsten from the backside of the wafer. To initiate this step, four ceramic fingers lift the wafer from the susceptor to the backside etch position. The step motor used to perform this operation has a flag that aligns with a reference laser to indicate the home position. The ideal etch position is reached when the fingers move a specified distance from this home position. In one instance, a deposition chamber was failing to achieve film-thickness uniformity during the tungsten etch. Although the problem was not detected by wafer surface scans, the in situ monitor indicated an increase in particles had occurred. The in situ particle counts at the time of this incident are displayed in Figure 4. The solid squares are total particle counts for the entire WCVD period, while the open diamonds are particle counts reported during the backside etch step. As the figure indicates, counts increased suddenly following the completion of a scheduled preventive maintenance (PM1), with the etch step contributing most of the total particles. Soon afterward, ending point failures were detected and the tool was then brought down for unscheduled maintenance (PM2), at which time the root cause was determined and corrected. Examination revealed the reference laser was moving out of position as it rubbed against the home flag, which, in turn, moved the home position of the ceramic fingers each time the step motor actuated, ultimately modifying the backside etch position. When the home flag was realigned, film uniformity and particle counts returned to normal.
Contaminated Gas Line. Tungsten hexafluoride (WF6) is one of the gas species used in WCVD. In another particle excursion instance, the WF6 gas line became contaminated. This problem was not detected by wafer scans, but the in situ monitor data, shown in Figure 5, revealed the particle baseline had increased and what is known as the first-wafer effect was occurring. (The first-wafer effect is characterized by elevated particle counts immediately following an idle period.) As the figure insert indicates, the longer the chamber was idle, the higher the associated in situ particle counts were for the first wafer. The counts declined during consecutively processed wafers, reaching a steady baseline; however, this baseline was 20x higher than previous ones.
An investigation of the WF6 gas line was initiated since most of the particles were observed during the bulk deposition step. When argon was substituted for WF6, the counts dropped immediately (see circled area of Figure 5); then, when wafers were again processed with WF6, counts returned to the elevated baseline (see the last two data points in the figure). This comparison confirmed the WF6 gas line was the origin of the particulation. A root-cause investigation was subsequently performed, revealing there was a leak in the line that allowed WF6 to react with oxidants in the gas flow, producing WO3 particles. These small particles are ideal low-energy sites for surface-catalyzed reactions, such as gas-phase nucleation. This problem probably was not detected in postprocess inspections with wafer scanners because small particles deposit early in the process and become masked by the thick films deposited during the latter process steps. In other words, postprocess scanning is biased toward measuring particles generated at later stages of the CVD process. In situ monitoring does not have such a drawback, since particle flux is measured during every step of the process in real time. Foreign Material on Susceptor. In a third instance, film uniformity abnormalities were detected downstream of the tungsten operation by a wafer surface scanner. The problem was traced back to the WCVD process step and the failing chamber was identified. During the investigation, the chamber was opened and foreign material was discovered on the susceptor. Because this material had prevented wafers from sitting flat on the susceptor during processing, the deposited film was distributed unevenly over the wafer surface. The in situ particle counts shown in Figure 6 reveal there had been a sharp increase in contaminants during processing, which apparently occurred when the material fell onto the susceptor.
Process Characterization. Unlike wafer surface scanners that measure particles after all processing steps are finished, in situ monitors measure particles continuously. The resulting data are thus synchronized with various process events, such as wafer movement and changes in gas flow and pressure, susceptor position, and plasma power and ramp. Because of this real-time, equipment event—synchronized particle measurement capability, such monitors are very useful in process characterization. For example, Figure 7 reveals that two distinct particle baselines resulted when alternating lots were processed through a WCVD system. Further investigation revealed the difference was due to the incoming materials. This process characterization data eventually contributed to overall process optimization.
Preventive Maintenance Prediction. Process tool maintenance is scheduled based on the number of wafers processed through a chamber. As seen in Figure 8, the in situ monitor clearly indicates particle trends during such preventive maintenance cycles. Particle counts are very low immediately following each scheduled maintenance period and gradually increase thereafter until the cycle is repeated. This trending leads to the possibility of scheduling preventive maintenance based on monitor data, not on wafer counts. Contamination-related problems would be reduced if maintenance procedures were performed as soon as a chamber showed a trend toward high particle levels. On the other hand, if particle counts remained low when the current wafer count trigger was reached, the maintenance cycle could be extended, saving money and reducing equipment downtime.
Conclusion The evaluation of an in situ particle monitor in WCVD systems indicated that the instrument accurately resolves particle size and counting frequency within a steady-state particulation regime. The monitor detects particles within a molecular diffusion environment, exhibiting an inverse cubic relationship between counting frequency and particle size. Furthermore, the frequency distribution profile resembles a Poisson distribution, a statistical characteristic of steady-state particulation. Because the monitor reports particle flux in a uniform flow regime, resolving steady-state particulation conditions of the process chamber, this technology is very different from wafer surface scanners, which report particle counts postprocess. The particle excursions reported by the monitor during the study revealed the tool can be useful in both engineering and production applications. By detecting higher particle counts associated with the flow of a specific gas, the monitor facilitated identification of a leak in the delivery line. In addition, the monitor data were useful in solving problems involving poor film uniformity. The monitor also can be used during characterization of new processes or optimization of existing ones by synchronizing data collection with mechanical and process states of the tool. Finally, particle trending between maintenance periods can be discerned by the monitor, enabling prediction of the need for repairs or part replacement. Based on these considerations, in situ particle monitoring offers an unprecedented capability to reduce scheduled maintenance, optimize tool performance, and prevent die loss by accurately monitoring steady-state particulation. Acknowledgments The authors would like to thank Dean Robinette of High Yield Technology for his support during the monitor evaluation period. We would like to recognize James Dean of Intel manufacturing, who provided valuable insight into the production applications of the monitor, and Kyle Ramsey and Tom Wash of the Intel management team for their continuous support and valuable advice throughout this project. Finally, we would like to thank Dave Nelsen of Intel for his support and direction during the evaluation. References 1. Borden PG, "Meeting the Challenges of Monitoring Particles in a Tungsten CVD System," Microcontamination, 9(3):39—43, 65, 1991. 2. Bosch BP, Hess D, Smith C, et al., "In Situ Particle Monitoring in a Poly LPCVD Diffusion Furnace," in Proceedings of the 43rd Annual Technical Meeting of the Institute of Environmental Sciences, Mount Prospect, IL, IES, pp 315—320, 1997. 3. Perry RH, Chemical Engineer's Handbook, 6th ed, New York, McGraw-Hill, pp 8-5—8-6, 1984. 4. Geankoplis, CJ, Transport Process and Unit Operations, 2nd ed, Englewood Cliffs, NJ, Prentice-Hall, pp 451—453, 1983. Jenny Asbell is a graduate rotation engineer at Intel in Rio Rancho, NM. During her first rotation there, she worked with the tungsten Applied Materials operations group on qualifying the in situ monitor for use in manufacturing. Asbell has a BS in chemical engineering (1996) from the University of New Mexico. (Asbell may be reached at 505/893-6431.) Zhijiang Sun, PhD, is a senior CVD process engineer at Intel in Rio Rancho, NM. The recipient of a BS in solid-state physics (1985) from the University of Science and Technology of China, an MS in semiconductor physics (1988) from the Chinese Academy of Sciences, and a PhD in chemistry (1994) from the University of Texas at Austin, he is the author of more than 20 technical publications and review papers and a member of the American Chemical Society, American Physical Society, and the American Vacuum Society. (Sun can be reached at 505/893-1893.)
Brian Bosch is a senior applications engineer with the High Yield Technology division of Pacific Scientific (Sunnyvale, CA). Previously, he worked for Applied Materials, holding positions in the global microcontamination division and service engineering field support division. He received his BS in chemical engineering (1994) from the University of Minnesota, where his studies focused on fluid dynamics. (Bosch can be reached at 505/792-8595.)
MicroHome |
Search | Current Issue | MicroArchives Questions/comments about MICRO Magazine? E-mail us at cheynman@gmail.com. © 2007 Tom Cheyney |
Figure 2: Histogram showing that the monitor's count rate distribution resembles a Poisson distribution.
Figure 3: Particle distribution chart showing that particle frequency conforms to an inverse cubic function of particle diameter.
Figure 6: In situ particle counts during an incident involving film uniformity; the problem was traced to foreign material on the deposition process chamber susceptor, which apparently appeared at the time of the particle excursion.
Figure 7: An example of in situ particle counts for use in process characterization. The two apparent baselines were synchronized with the processing of alternating lots. 
