In-Situ Particle Monitoring - FabTronics First
In-Situ Particle Monitoring in a Vertical Poly Furnace
Pete Glass, Joe Kudlacik — IBM Microelectronics
Ray Burghard — Pacific Scientific Instruments Group
INTRODUCTION
In situ particle monitoring (ISPM) is a key semiconductor initiative as wafer production facilities strive to reduce operating costs. The benefits of using ISPM include real-time process monitoring, reduced tool qualification costs, and improved product cycle time,. All of these improvements lead to increased manufacturing productivity and a lower operating cost. The following paper describes the production implementation of an in situ particle monitor on an LPCVD vertical furnace. The furnace is a high-risk, critical process to monitor in real-time due to the large number of wafers processed in each batch. Correlation between surface scanner counts, ISPM counts, test yield and process trending is discussed., along with sensor reliability and the future plan of statistical control.
SYSTEM INSTALLATION
The Model 70XE sensor from High Yield Technology (HYT) was chosen for this program (figure 1). This sensor provides sensitivity down to 0.17 microns (m m), through a dark field, light scattering detection scheme4.
Figure 1 - HYT Model 70XE ISPM
Particles in the pump line pass through the laser beam and scatter light to the photo detector. The light is converted into an electrical signal, which is analyzed at the controller. Each voltage signal greater than 30 millivolts (mV) is counted as a particle. The magnitude of the voltage signal determines the size of the particle.
The sensor was installed in the exhaust line of an ASM A300 vertical poly silicon furnace, at IBM’s production facility in Essex Junction, Vermont. The furnace is a single-wall, top-exhaust system with a 130 wafer capacity. Figure 2 is an illustration of the sensor installed in the vacuum line. It is critical to install the sensor as close to the chamber to provide good sensitivity to particles which can affect the wafers. In order to achieve the optimum sensitivity of the sensor it was necessary to relocate the throttle valve outside of the tool enclosure, which is on and to the pump side of the gate valve. This rearrangement did not necessitate any process changes.
Figure 2 - Schematic of system installation
In addition to the sensor being installed in the vacuum line, the ISPM controller was wired directly into the furnace’s digital output circuitry. This enabled the collection of particles during certain steps of the process and the ability to recognize which steps are the major particle contributors (see Table 1). All of the particle collection was done using a stand alone PC and High Yield Technology’s Particle VisionTM software.
Table 1
|
Process Step |
Description |
Particle Performance |
|
BoatIn |
cold wafers pushed into furnace |
tube life indicator, correlation |
|
Pumpdown |
pump to base pressure |
stable, no excursions |
|
Purge |
nitrogen purge, temp. stability |
sensitive to buildup in exhaust |
|
PolyDep |
polysilicon deposition |
stable, low counts |
|
PostPumpPurge |
nitrogen purge |
stable, no excursions |
|
Backfill |
tube back to ATM pressure |
stable |
|
BoatOut |
pull hot wafers out |
sensitive to leaks, correlation |
The sensor was calibrated and setup to collect data into five different bins according to desired particle size and corresponding threshold voltage value. These values can be seen in Table 2. The particle counts in bins 2-5 (or > bin 2) was used for most correlation work and for initial SPC, since the historical particle data from the furnace was recorded at 0.5m m
Table 2
|
Bin |
Particle Size (m m) |
Threshold Value (mV) |
|
1 |
0.2-0.3 |
30 |
|
2 |
0.3-0.5 |
230 |
|
3 |
0.5-0.7 |
690 |
|
4 |
0.7-1.0 |
875 |
|
5 |
>1.0 |
900 |
RELIABILITY
The HYT Model 70XE has three built in diagnostic functions to determine the health of the sensor; stray, laser and noise.
The stray light is a measure of the background light on the photo-detector when no particles are passing through the laser beam. As the optical windows of the sensor get dirty the stray light will increase. Laser is the current necessary to drive the diode laser to maintain the calibrated laser power. The laser current will gradually increase over the lifetime of the diode, until the laser can no longer reach the calibrated output. The median noise measures the background noise level seen by the photo-detector, higher noise levels can lead to "false" counting in the sensor (particle counts when none are present).
Figure 3 shows the sensor stray light increasing as the tube life increases and displays the change in stray light after cleaning the sensor optics.
Figure 3 - ISPM diagnostics
The sensor performed very reliably at stray light values below 70. Stray values above 70 resulted in "false" particle counts, which showed up under two distinct circumstances.
Initially the median noise level in the system would spike during normal operation. When the noise level would go above 12 the ISPM would start to "false" count. In Figure 3, the first increase in stray over 70 causes a corresponding increase in noise. This increase in noise occurred when the sapphire window that the laser beam passes through became coated. When this happened the laser beam reflected back into itself, causing an increase in the laser noise. To reduce the risk of false counting, the sensor is now cleaned at every tube change. HYT has also implemented a more permanent solution by angling the sapphire window to reduce the effect of this problem.
A second instance of false counting occurred when particles were left on the laser window after routine sensor cleaning. Improper removal of small particles from the window caused increase stray and noise. To make sure all debris is removed during the sensor cleaning, the stray light must be below 50 before re-installation of the sensor.
The laser has been operating in the furnace for one year and has shown no marked increase in laser current. Laser life expectancy for this sensor is over 2 years.
Table 3 summarizes the reliability information. Note that adding scheduled cleans at the tube change should increase the MTBF to well over 1 year (the expected lifetime of the laser).
Table 3
|
Reliability Parameter |
# in hours |
Description |
|
Mean Time Between Failure |
> 2000 |
dirty laser window causes high stray |
|
Mean Time Between Clean |
> 1200 |
clean windows at tube change |
|
Mean Time To Repair (Clean) |
0.5 |
clean and inspect windows, leak check |
|
Additional Scheduled Downtime |
0.5 hr. per 2 months |
one sensor clean per tube change |
|
Additional Unscheduled Downtime |
0 |
no hard sensor failures during the implementation |
CORRELATION
System Events
The most basic correlation to establish for ISPM, is between the in situ particle counts and known system maintenance events. This correlation to tool problems can be established by reviewing the in situ particle data and equipment logs to match up the particle excursions and maintenance events. This relationship can also be established if an excursion of in situ counts is observed, and tool maintenance is performed to determine the cause for increased counts.
During the early evaluation of the ISPM, an increase in particle counts was noticed in several of the process steps (Figure 4). The sharp increase in counts was most noticeable in steps where pressure was increased or the system was being evacuated. Since the tube had recently been changed, it was determined that the failure mode was not from normal film buildup on the quartz walls. Premature particle failures have occasionally been caused by buildup in the system exhaust, so the tool was taken down to clean this area. As seen in Figure 4, after the exhaust area was cleaned the in situ particle counts returned to a normal baseline, and the system was returned to production.
Figure 4 - ISPM triggers system maintenance
Tube Condition
A key benefit to using in situ particle detection is the ability to know the condition of the process chamber at all times during wafer processing. In the case of the furnace this "process chamber" is the quartz tube.
In most production facilities the lifetime of the tube is determined by how many wafer loads are processed, or how many hours of deposition have been performed since a tube was changed. The primary reason for limiting tube lifetime is particulate which is generated as the film deposited on the tube walls begins to flake off. Since the in situ particle monitor can detect this tube flaking during process runs, it should be a good gauge to indicate the need for a tube change.
Figure 5 shows how the in situ particle counts trend over multiple tube changes on the poly vertical furnace. It can be seen that after a tube change the baseline count level is very low, indicating a clean system. As more wafers are processed on the system, the baseline increases until the next tube change is performed. This signature is very repeatable, and is now used to trigger system tube changes.
Figure 5 - ISPM indicates tube condition
Wafer Measurements
In order to validate the ISPM method of particle monitoring on the furnace, a correlation to the existing wafer particle monitors was performed.
Actual product wafers were measured using a Tencor AIT surface scanner. The minimum detectable size on this system was approximately 0.5 microns (m m).
Figure 6 shows an overlay of both the ISPM data and the product wafer measurements during a three month period of data collection.
Figure 6 - Trend of ISPM and Tencor AIT
Both the in situ particle count and the product wafer particle count indicate increased levels at the end of tube life, just prior to the tube change. Just before the tube change the in situ count baseline is significantly higher than both the baseline seen at the beginning of the graph and the baseline achieved immediately after the tube change. The wafer particle count baseline never seems to change, but the standard deviation, and the magnitude of the spikes both increase prior to tube change.
The pump change also seems to have generated increased particle levels as seen by both the HYT and the wafer measurements.
Figure 7 shows the linear correlation derived between the HYT and Tencor AIT during this three month period. A correlation coefficient of R=0.73 was found, indicating that the two measurement techniques have reasonable correlation.
Figure 7 - Correlation of ISPM and Tencor AIT
Electrical Test
Following the poly deposition and subsequent poly etch step, wafers are submitted for a preliminary electrical test to determine defect density levels. This is done primarily to prevent early process steps from impacting multiple production lots.
Figure 8 shows the electrical test data for a single product type versus the ISPM particle levels in the vertical furnace. The electrical test data is displayed as a daily average of the percent of bad die per wafer. An increase in the percent of bad die per wafer indicates increased defect density and lower yield.
Prior to the installation of the ISPM, the specified maximum tube lifetime was 125 hours of deposition. Since the ISPM showed a dramatic particle increase toward the end of tube lifetime, it was decided to reduce the maximum lifetime to 100 hours of deposition. In order to determine the impact of high particle levels in the furnace on defect density, it was decided to extend one tube out to the limit of 125 hours.
Figure 8 shows consecutive tube changes on the ASM furnace, one done at 100 hours and one at 125 hours. The electrical test data from the extended tube lifetime is overlaid, showing apparent yield degradation both before and after the 100 hour mark.
CONCLUSION
The results obtained during the extensive evaluation have shown that the ISPM system is a useful monitor of processing conditions on a vertical furnace. The IBM Vermont facility has decided to populate all of the ASM A300 poly systems with the HYT ISPM system, and begin
collecting ISPM data on SVGTM vertical poly systems. In addition to installing sensors on multiple tools, the ISPM data is being fully networked into the existing fab-wide factory automation system. Real-time statistical process control of particle levels will be implemented.
The sensor’s ability to capture real-time particle contamination will enable a proactive response to potential problems, and reduce product exposure to yield-limiting process conditions.
In addition to minimizing product risk, the added benefit of reducing measurements and monitor wafers will also be realized as the response of the ISPM in relation to wafer surface particle data becomes better understood.
1 H.D. Pham, M. Elzingre, and P.G. Borden, "LSI Logic Proves Yield Gain Using In-situ Monitor," Semiconductor International, April 1995.
2 T. Scharnagl; Application of an HYT In situ particle monitor for selective nitride etching on a LAM 4420 (Rainbow); Inst. Env. Sciences ; Proceedings 42rd Annual Technical Meeting Orlando. May 12-16 1996; p317
3 R. Burghard, D. Dance and R. Markle; Reducing Ion-implant Equipment cost of ownership through in situ contamination prevention and reduction; Microcontamination, Sept. 1992, p 27
4 H.C. van de Hulst; Light scattering by small particles; John Wiley, New York (1957)
The ISPM data correlates very well with the yield information. Days when the yield is poor correspond to process runs when particle levels in the furnace are high. Based on this correlation to wafer defect levels, implementation of the ISPM for process control on the furnace was initiated.
It is important to note that the first yield hit occurs before the 100 hour mark. Just setting a 100 hour limit on the tube life will not eliminate the potential for poor yielding runs. Using the ISPM as real-time protection will ensure minimal product exposure to potential yield-limiting process conditions.
Figure 8 - Trend of ISPM and Wafer Test Yield
FabTronics Ltd |
23 Braeview Drive Paisley PA2 8PW
|
