This paper describes Digital Equipment Corporation Semiconductor Division s

WHITEPAPER By Edd Hanson and Heather Benson-Woodward of Digital Semiconductor Michael Bonner of Advanced Energy Industries, Inc. This paper describes Digital Equipment Corporation Semiconductor Division s application of real-time process control, excursion detection, and clean optimization on a PECVD (Plasma Enhanced Chemical Vapor Deposition) wafer processing tool. The purpose of this effort is to reduce test wafers, spare part usage, and tool downtime, while increasing product yield and tool throughput. This tool optimization effort is made possible by the addition of process control technology, which monitors the RF signals at the point of use on plasma process tools. The RF monitoring system is an RF Metrology System (RFMS) supplied by Fourth State Technology of Austin, Texas. Digital Semiconductor uses a PECVD tool to deposit post metal dielectric layers during the production of their integrated circuit devices. OPTIMIZING CHEMICAL VAPOR DEPOSITION PROCESSING THROUGH RF METROLOGY BACKGROUND The Process Tool A PECVD tool is used in semiconductor manufacturing to deposit a variety of dielectric (insulator) layers. A common configuration is a multi-chamber platform consisting of a loadlock chamber and up to four additional process chambers. Each process chamber processes one wafer at a time, and there are a variety of process chambers that can be mounted on the loadlock. The chamber type this paper deals with is a universal chamber. The Process The process used in this particular universal chamber consists of two different depositions and their corresponding chamber cleans. The first deposition is an SACVD (subatmospheric chemical vapor deposition), phosphorus doped, silicon oxide deposition followed by a corresponding chamber clean. The second deposition is a phosphorus doped, silicon oxide, plasma enhanced deposition (PECVD). This step is also followed by a chamber clean. The PECVD deposition begins with a stabilization step, where the process gases or chemicals, in this case TEOS (Tetraethylorthosilicate), TEPO (Triethylphosphate), and oxygen, begin flowing, and the chamber pressure stabilizes. In the next step, RF (radio frequency) power is supplied to the electrode, a plasma ignites, and the dielectric film is deposited on the wafer. The stability and reproducibility of the plasma is critical to dielectric film properties and subsequent product yield. As the dielectric film is deposited on the wafer, it is also deposited on the walls of the chamber and the process kit hardware. This film should be removed from the chamber walls and hardware before the next wafer is processed. If this film is not removed, it can jeopardize the on-wafer uniformity and thickness, as well as the on-wafer particle levels. This film removal is accomplished with a chamber clean. The first step of the chamber clean is again a stabilization step, where the process gases are introduced, in this case, C 2 F 6, O 2 and NF 3, and the chamber pressure stabilizes. In the next step, the RF power is applied, igniting a plasma that etches away the dielectric film deposited on the walls and process kit. The SACVD process differs from the PECVD process in that it does not use RF power during the deposition phase, and it uses O 3 instead of O 2. The corresponding chamber clean differs from the PECVD process chamber clean in that it has an additional throttle valve clean step, and its clean time is slightly different. The RF Monitoring System Fourth State Technology s RFMS employs an RF sensor, installed inline between the impedance matching network and powered electrode. This sensor contains current and voltage transducers and is designed to survive the harsh conditions at the powered electrode without inducing process shifts in key process parameters. Shielded cables carry the voltage and current signals from the RF sensor to a base unit that houses the RF electronics designed to filter and analyze the RF signal. Measurements are made of the first five harmonics of voltage and current and the phase angle at the fundamental frequency. The RFMS was designed at SEMATECH for application to the PECVD tool, and as such is four chamber capable. 1

FST s unique process control software makes use of the RF voltage and current spectral information to evaluate process performance. Process control is achieved using three main modules: trend analysis, excursion detection, and endpoint detection. All three modules can be employed in real time for maximum performance improvement. These will now be discussed in detail. The first software module is FST s endpoint detection module. Endpoint refers to the point at which a process chamber is completely clean and the clean process can be terminated. During a chamber clean, the chamber impedance changes as the process byproducts are removed from the chamber walls and surfaces. Depending on the composition of the byproducts and the process used to remove them, the chamber impedance either increases or decreases until all of the byproducts are removed. Once all of the byproducts are removed from the chamber, the impedance (and thus voltage, current, and phase angle), stabilizes. Consequently, a chamber clean endpoint can be detected using the voltage, current, or phase angle signals. With the FST endpoint package, the user can develop such an endpoint algorithm using one of the 11 signals collected, to provide real-time chamber clean end-pointing. The second software module, designed for excursion detection, is FST s "Go-No-Go" package. In this module, the user enters reference values and warning and alarm limits for the mean, range, and standard deviation of the ten voltage and current signals. The "Go-No-Go" module will monitor these ten signals and provide warnings and alarms in real-time to the process tool if these limits are exceeded. These warnings and alarms are displayed on the PECVD tool s human interface. Process shifts and mechanical failures in the process tool will often cause the voltage and current signals to shift. By establishing warning and alarm limits for the different signals, the user can receive warnings and alarms on the process tool when process shifts and mechanical failures occur. FST has developed a simple implementation strategy for determining these warning and alarm limits. The third software module is a trend analysis package useful in understanding long term transients in the process tool performance such as wet clean cycles and transitions in film stress. FST s SPC module makes use of standard Western Electric control chart rules as well as customized versions to provide real-time, run-by-run process control. THE PECVD OPTIMIZATION APPLICATION This application focused on the Endpoint and "Go-No- Go" capabilities of the RFMS unit. The endpoint capability was used to compare single-step and two-step chamber cleans, and provide information on clean time optimization. The "Go-No-Go" capability was used to investigate test wafer reduction, excursion detection, and equipment troubleshooting. Endpoint Application Initially, an RFMS unit was installed on two process chambers for passive endpoint monitoring. One chamber used a single-step, timed chamber clean. The other chamber used a two-step, timed chamber clean. Several thousand chamber clean endpoint traces were collected from both chambers with the RFMS unit. Sample endpoint traces of the single-step PECVD and SACVD chamber cleans are shown in Figures 1 and 2, respectively. Sample endpoint traces of the inner and outer cleans of the two-step PECVD chamber clean are shown in Figures 3a and 3b. Figures 1 and 2 show that the single-step chamber cleans had distinct, repeatable endpoints. Figures 3a and 3b show that the inner cleans (first step) of the two-step cleans, had distinct, repeatable endpoints. However, the outer cleans (second step) show no visible endpoint. The chamber clean reached endpoint during the inner clean step of the two-step clean. The entire chamber and kit hardware was clean before the end of the inner step. The outer clean was therefore not necessary, and was consuming substantial amounts of clean gases and causing undue wear on hardware, a known cause of yield limiting particles. Figures 1 through 3b also show the differences between RF based endpoint times versus the duration of the timed cleans. It was determined that the chamber clean times could be greatly reduced, cutting process gas consumption and wear on process kits. Reducing chamber clean time also reduces RF hours on the chamber, which should result in extending time periods between wet cleans. Fewer wet cleans translates to less tool downtime and increased tool throughput. Table 1 shows the time savings to be gained using an RF endpoint system as opposed to the standard timed cleans. Based on the reduced clean times and RF hours resulting from RF endpoint, yearly savings were calculated. The savings calculations determined the amount of gas saved due to reduced clean times and the number of process kits saved by the reduced RF hours. Figure 4 shows the results of these calculations. 2

Note: All specific process parameters have been removed. 70.0 60.0 Phase Angle 43.3 26.7 Phase Angle 43.3 16.7 10.0 0 10 20 30 40 50 60 70 80 90 100 A7613080.914 A7613083.506 A7613093.011 A7613095.711 A7613102.413 A7613105.055 A7613111.712 A7613114.331 A7613120.921 A7613123.512 Figure 1. Single-step, PECVD chamber clean endpoint traces. Traces show repeatable, distinct endpoints for PECVD single-step cleans. 60.0-10.0 A7324023.736 A7324025.129 A7324030.521 A7324031.930 A7324033.325 A7324034.711 A7324040.342 A7324041.740 A7324043.314 A7324044.714 Figure 3a. PECVD inner chamber clean endpoint traces. Traces show repeatable, distinct endpoints for PECVD inner clean steps. 50.0 0 10 20 30 40 50 60 70 Phase Angle 43.3 26.7 Phase Angle 33.3 16.7 10.0 0 5 10 15 20 25 30 35 40 45 50 55 60 A7620185.135 A7620190.426 A7620191.711 A7620193.004 A7620194.255 A7620195.604 A7620200.853 A7620215.352 A7621225.401 A7621230.646 Figure 2. SACVD inner chamber clean endpoint traces. Traces show repeatable, distinct endpoints for SACVD inner clean steps. 0 0 10 20 30 40 50 60 70 A7324023.859 A7324025.251 A7324030.645 A7324032.052 A7324033.447 A7324034.834 A7324040.504 A7324041.902 A7324043.436 A7324044.837 Figure 3b. PECVD outer chamber clean endpoint traces. Traces show no distinct, repeatable endpoint for PECVD outer clean steps. Average FST FF Endpoint Time Saved % Savings Timed Endpoint Plus 20% Due to Due to Duration Overetch Endpoint Endpoint SACVD Single-Step 57.17 s 31.50 s 37.80 s 19.37 s 33.88% Clean PECVD Single-Step 95.12 s 57.51 s 69.01 s 26.11 s 27.45% Clean SACVD + PECVD 45.48 s 29.86% Single-Step PECVD Two- Step Clean 270.69 s 121.57 s 145.88 s 124.81 s 53.89% Table 1. 3

1560.0 1493.3 1426.7 1360.0 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 Figure 4. "Go-No-Go" Application The RFMS unit collected data on the deposition steps during the same time period that it was collecting the endpoint data on the chamber cleans. Over 800 data files from each chamber were collected from the depositions. The mean, range, and standard deviation was calculated in real-time for each of the ten voltage and current signals automatically by the FST process control software. These means, ranges, and standard deviations were used to establish reference values and warning and alarm limits for each of the signals. During the startup phase for the "Go-No-Go" application, data was reviewed to determine the potential to detect process shifts or equipment failures leading to undesirable wafer results. During this application, two equipment problems resulting in process excursions occurred. One chamber experienced a helium (TEOS) regulator failure. This regulator failure was detected by the RFMS unit. The regulator failure triggered range alarm limits for the V0 and V2 signals. Figure 5 shows the V0 output trace for one of the files that alarmed for the regulator failure. The figure shows a large drop in the voltage signal that takes approximately ten seconds to recover. This type of signature is typical of flow or pressure fluctuations. It should be noted that the actual flow fault was a short transient. However, the resulting plasma relaxation time was substantial. Figure 5. HE (TEOS) regulator fault. Trace of VO prior to repair. Excursion takes approximately 10 seconds to recover. After the regulator was replaced, there were no further alarms. Figure 6 shows a sample of the V0 output traces without regulator problems. The figure also shows no large voltage drops throughout the traces. Similar results were seen with the V2 output traces. While running chamber characterization monitors, the second chamber monitored experienced an episode of extremely high particle counts. This problem was also detected and documented by the RFMS. It triggered range alarm limits for the V1 signal. Figure 7 shows the V1 output trace for one of the files that alarmed for the particle problem. Figure 7 shows several large drops in the voltage signal, which take two to three seconds to recover. This type of signature is very typical of micro-arcing in the chamber. Microarcing in a chamber often generates particles. The chamber was wet cleaned, eliminating the arcing problem. V0 1560.0 1493.3 1426.7 1360.0 0 10 20 30 40 50 60 70 80 90 100 110 120130140 04 A7702145.250 A7702150.610 A7702151.854 A7702153.148 A7702154.439 A7702155.730 A7702161.021 A7702162.313 A7702163.614 A7702164.9 Figure 6. HE (TEOS) regulator fault. Traces of VO after regulator was repaired. No large signal drops. 4

Figure 8 shows V1 output traces after the chamber was wet cleaned. The figure shows no large voltage drops in the V1 output traces. 250.0 216.7 183.3 Response Model Terms Adj R-square Film Stress V0, V3, I0 0.920 Film Thickness V2, I0, I2 0.961 Uniformity V2, I0, I2 0.738 RF Power V1, I0, Phase 0.986 Electrode Spacing V2, V3, I0, I3 0.932 Pressure V0, I0, I2 0.959 Table 2. Designed experiment model summary 0 150.0 0 10 20 30 40 50 60 70 80 90 100110120 130140 E l fv1 E i k i l d Figure 7. Micro-arcing, particle problem. Example of V1 trace. Excursion takes approximately 5 seconds to recover. 250.0 Predicted Stress -5-10 V1 216.7 183.3 150.0 0 10 20 30 40 50 60 70 80 90 100 110 120130140 A7613091.332 A7613092.658 A7613094.027 A7613095.357 A7613100.745 A7613102.059 A7613103.402 A7613104.740 A7613110.053 A7613111.358 Figure 8. V1 output trace after wet clean. No large voltage drops. A more severe arcing problem will exhibit itself as much larger drops in all five voltage signals. Severe arcing has the potential to damage devices and cause serious yield loss. The final part of the "Go-No-Go" analysis was a designed experiment (DOE). The DOE was used to determine the correlation between the eleven signals collected by the RFMS unit and the process tool parameters and wafer results. The DOE was a Box-Behnken design using eighteen wafers. The adjusted variables were pressure, power, and gap spacing. The adjusted range of variables was 10%. The RFMS unit collected RF data during the DOE and all wafers were measured for stress and thickness. Upon completion of the DOE, the data was analyzed to create models for predicting wafer results and improving process parameters. -15-15 -10-5 0 Measured Stress Figure 9. Measured film stress vs FST predicted film stress Table 2 shows the DOE model summary. These represent linear models using FST parameters as predictors. The table shows that the film stress and film thickness wafer results can be predicted at a high confidence level. The process parameters of RF power, electrode spacing, and pressure can also be predicted at a high confidence level. Figure 9 is a plot of the predicted film stress vs. the actual measured film stress. One can see close agreement between the model prediction and the measured values. Figure 10 is a plot of the predicted film thickness vs. the actual measured film thickness. As in the case of the stress model, the thickness model was quite close to actual measurement across the design space. This presents an opportunity to use in-situ RF measurements to confirm or replace expensive wafer tests. 5

RESULTS AND CONCLUSIONS In this evaluation, Digital Equipment Corporation successfully optimized their PECVD and SACVD processes in terms of reducing manufacturing cost, throughput, and yield. Using a Fourth State Technology RFMS unit, it was verified that a single-step chamber clean could successfully replace the two-step chamber clean. The RFMS unit showed that the process chamber and hardware were completely clean, reaching endpoint before the end of the inner step of the two-step chamber clean. Predicted Thickness 11000 10000 9000 8000 7000 7000 8000 9000 10000 11000 Measured Thickness Figure 10. Measured thickness vs FST predicted thickness It was also determined that the single-step clean time could be greatly reduced. Moving from the two-step chamber cleans to single-step chamber cleans greatly reduces clean times. This time reduction results in reduced gas consumption of costly clean gases. It also reduces the RF hours the chamber experiences, extending kit life and tool throughput, and reducing preventive maintenance events and tool downtime. The times were also reduced on the singlestep chamber cleans, further reducing gas consumption and tool downtime and increasing kit life and tool throughput. The "Go-No-Go" portion of this evaluation revealed several substantial applications of the RFMS unit for process control and measurement wafer reduction. Digital Semiconductor was able to detect two different processing faults with the RFMS unit: a Helium regulator failure, and a gross particle failure during chamber characterization runs. With the aid of a simple designed experiment, Digital Semiconductor was also able to identify the RF parameters that correlated to excursions in RF power, electrode spacing, and pressure, as well as stress and thickness wafer measurements. These correlations provide not only detection capability, but also valuable troubleshooting information. In conclusion, all goals of process optimization, process control, and excursion detection established for this application were met. As a result of this evaluation, Digital Semiconductor has moved all post-deposition cleans to single step cleans for both doped and undoped films. Also, all doped film chamber cleans have been evaluated and post-deposition clean times optimized using FST RFMS equipment. Overall, "On Product" particle counts have seen a measurable improvement, and "Plasma On" time during chamber cleans has been reduced by 30%. A further FST evaluation has been planned for Digital Semiconductor s undoped chambers with the expectation of additional process optimization. 6

7

Advanced Energy Industries, Inc. 1999 All rights reserved. Printed in USA SL-WHITE11-270-01 1M 01/01 Advanced Energy Industries, Inc. 1625 Sharp Point Drive Fort Collins, Colorado 80525 800.446.9167 970.221.4670 970.221.5583 (fax) support@aei.com www.advanced-energy.com California T: 408.263.8784 F: 408.263.8992 New Jersey T: 856.627.6100 F: 856.627.6159 United Kingdom T: 44.1869.320022 F: 44.1869.325004 Germany T: 49.711.779270 F: 49.711.7778700 Korea T: 82.31.705.2100 F: 82.31.705.2766 Japan T: 81.3.32351511 F: 81.3.32353580 Taiwan T: 886.2.82215599 F: 886.2.82215050 China T: 86.755.3867986 F: 86.755.3867984 8