In Situ Particle Detection for Pre-Metal Sputter Etch

Particle control during wafer processing continues to be a critical factor in the successful production of semiconductors. Although cleanroom particle control has matured with the implementation of fully automated wafer handling systems, the major source of yield-limiting particle events — the actual wafer processing environment — continues to create particle control issues. The use of in situ particle monitors (ISPMs) has long been identified as one method to monitor and control the actual wafer processing environments, and to drive wafer yield and equipment productivity improvements.¹ Successful implementation of ISPM systems has been hampered by two key issues: the inability to correlate sensor results with chamber condition and wafer results, and the seamless integration of the sensor into the manufacturing environment.

With the advent of advanced process control (APC) systems, there has been a renewed interest in obtaining sensor-based control of process conditions, including particle counts. LSI Logic's 200 mm fab in Gresham, Ore., has been a long-time advocate of advancing the role of sensors in its award-winning facility.² LSI has implemented a sensor integration and analysis system that allows engineers to access sensor and equipment information in real time to enable statistical control of key parameters, including particle levels detected by ISPM systems installed throughout the factory.³

Control of advanced sensors like residual gas analyzers and particle sensors requires tight integration to both the host tool and the manufacturing environment. To achieve this integration, LSI Logic has implemented Inficon's FabGuard Sensor Integration and Analysis System. The FabGuard system communicates with the manufacturing tool through standard SECS protocol, via the primary or secondary SECS ports available on the tool. Sensor data is merged with tool status variable IDs (SVIDs) and served through the web to engineers anywhere in the factory.

The system uses local data collection peers, the IPM controllers, which provide rapid response for real-time applications such as endpoint detection, critical fault, and tool interdiction, similar to a traditional peer-to-peer architecture. Centralized servers — executive and database — provide management of acquisition, analysis, modeling, reporting, data archiving, database and web services, similar to a client-server architecture. This integration system is scaleable, and allows LSI Logic engineers to monitor multiple sensors on various process tools through a single user interface. Sensor configuration, data collection plans, data analysis and statistical process control (SPC) can all be administered remotely from the fab through the web client. If an excursion occurs, the FabGuard system can notify engineering via e-mail, the manufacturing operator through the tool screen, and/or halt the tool with various alarms that can be set up by LSI Engineering.

The sputter etch clean process removes native oxide from the bottom of contact and via holes prior to the metal physical vapor deposition (PVD) process typically used in semiconductor manufacturing. This oxide removal enables good electrical conductivity between the metal and polysilicon levels or between successive metal levels.

Oxide is sputtered from the underlying surface using a highly biased argon plasma. Throughout the process the oxide is redeposited on the surface of the chamber dome, creating the potential for particle excursions. As the redeposited oxide grows thicker, it will eventually flake off and land on the wafer being processed. Once the chamber dome begins to degrade, it is replaced (kit change), and a manual cleaning of the process chamber is performed to return the chamber to a baseline particle-free condition. By monitoring the chamber condition in real time, it is possible to identify variations in chamber kit performance and improve process yields and productivity through the control of particle levels.

Inficon's Stiletto ISPM is installed in the pump line of the sputter etch clean chamber, between the chamber and the turbo-molecular pump (Fig. 1 ). It is important that the ISPM system is installed in close proximity to the process chamber to ensure good particle transport and improve correlation to both chamber events and wafer defect levels.⁴

1. The physical location of the ISPM in the pump line between the chamber and the turbo pump.

The Stiletto in situ particle monitor uses basic light scattering to detect particles as they pass through a laser beam that is projected in front of the turbo inlet. The sensor employs a scanning laser that covers a large percentage of the pump line with laser light, and captures particles down to 0.2 µm in size. This sensitivity provides the necessary count rate during wafer processing to enable tight control of chamber particle levels.

Real-time particle counting is a unique process control issue, since particle levels vary based on recipe chemistry and incoming film properties. Also, since particle events occur without warning and have significant impact on the product yield, it is critical that the control methodology provides maximum sensitivity to both normal chamber wear and spontaneous excursions. The key to implementing effective control of in situ particle counts is to establish the particle baseline of the process chamber, and correlate excursions from this baseline to actual chamber degradation and/or wafer-level particles.

Particle levels inside the tool are affected by the incoming condition of the wafer and the process conditions in the tool. Since the integration system collects logistic information like recipe ID and lot ID, it is possible to implement control limits based on product mix and process type. Separate limits were created for multiple recipe types on the chamber. Each recipe type has unique particle behavior and unique control limits.

Once a baseline particle level is observed, it is important to monitor the chamber condition over time to determine control methods. Based on historical data collection and analysis, it was determined that two separate control methods were required for optimal protection against particle excursions:

1. A trend rule was implemented for process control using an M of N limit. If any five out of nine wafers run consecutively have total particle counts more than the upper specification limit (USL), an alarm is generated.
2. A single point limit was also implemented to guard against catastrophic particle events. An alarm was generated whenever the counts from a single wafer broke the high-count limit.

During the initial implementation of Stiletto on the sputter etch chambers, several faults were detected and subsequently classified through maintenance and engineering activities. Preventive maintenance (PM) for particle control usually consists of a kit change to replace internal components of the chamber, including the chamber dome. For chambers being monitored with the Stiletto sensor, the PM cycles are performed after the kit has 30 kWh of plasma-on for wafer processing.

Case 1. Normal PM Cycle — Figure 2 shows the Stiletto count trend during a PM cycle that lasted to the 30 kWh limit. Notice the low particle count baseline, which is stable over the course of the cycle.

2. The in situ particle trend for a normal PM cycle with a dome in good condition. A normal PM cycle lasts 30 kWh.

Case 2. Misprocessed Wafers — Figure 3 shows the Stiletto count trend during a PM cycle that only lasted 24 kWh. Particle levels are elevated, and there are three failures of the run rule control limit. After the third failure, engineering took the chamber offline and inspected the kit. The dark flaking condition found is caused by wafers being processed in the sputter etch chamber without an oxide film. When wafers without oxide are processed, the adhesion of the sputtered oxide on the chamber components becomes poor and tends to flake from the chamber walls back onto the wafer.

3. The in situ particle trend during a PM cycle with a contaminated dome. SPC failures indicated poor dome condition and triggered premature shut down of the chamber after 24 kWh.

Case 3. Contaminated Dome — Figure 4 shows a premature failure of particles only 12 kWh into processing on a kit. Multiple SPC failures led engineering to take the chamber offline and inspect the kit condition. When the chamber was open, contamination was found on the dome and in the chamber. A kit change was completed before returning the chamber back to production. Notice that the count baseline is much lower after the kit change, indicating that the particle problem was resolved.

4. The in situ particle trend indicating a contaminated chamber. SPC failures shut down the chamber after only 12 kWh. The chamber was cleaned and the system returned to a baseline count level.

This is an optimal use of ISPM, since it detects premature failures and provides a quick way to qualify a chamber after maintenance activity.

Case 4. Elevated Counts for Single Lot — Particle levels can increase rapidly if the adhesion of the sputtered oxide film begins to degrade. Often there will be short-lived flaking of the walls until additional wafer processing stabilizes the film. Figure 5 shows an example of this phenomenon, where one lot (#37081) is processed while the chamber is flaking particles. The baseline count level is elevated during several wafers in this lot.

5. This control chart shows how one lot failed because it was processed while the chamber was flaking particles. The baseline count level is elevated during several wafers in this lot.

To make a proper disposition of the wafers processed during this excursion, it is common to measure the wafer(s) using an ex situ particle measurement tool. Figure 6 shows the ex situ and in situ results for two wafers processed during the excursion seen in Figure 5 . The elevated counts on the ex situ scans validate the excursion detected by Stiletto.

6. In situ and ex situ counts from two wafers processed during the particle excursion seen in Figure 5 .

Extending kit change cycles

To extend the time between kit changes, it is important to be able to detect a shift in the normal operating condition of the chamber and to halt manufacturing when the chamber begins operating outside of its normal region. Since the baseline defect levels of the chamber have been established and an acceptable M of N rule is in place to stop processing and send an alarm to manufacturing, production wafers are protected from a defect excursion and the kit can be extended to any desired lifetime, or until the Stiletto sensor has determined end-of-life by alarming the chamber.

When an alarm occurs, it is important to determine whether the alarm is caused by changing chamber conditions or by the incoming wafers. It is uneconomical to replace the kit each time an alarm occurs. At the same time, it is a risk to production if the alarm is cleared and wafers continue to process. To alleviate this issue, a troubleshooting flow diagram was created by LSI engineers for the manufacturing team to use as a guide when reacting to alarms. The troubleshooting guide leads the user through steps that eventually determine whether the chamber is operating out of control or if the incoming material is the root cause of the elevated defect levels. By reviewing the failing SPC chart, a 30-day lot average chart, inline particle monitor data, and inspection data of the current wafers leading to the alarm, it can be determined whether the issue is chamber-based or based on incoming wafers.

Through the implementation of in situ particle monitoring and control, LSI Logic was able to increase kit lifetime by 25%.

Implementation of in situ particle monitors into mainstream semiconductor manufacturing was recognized by the National Technology Roadmap for Semiconductors as early as 1997 as "the only way to provide the resolution and timeliness necessary to detect equipment and process deviations." Issues with capture efficiency, noise rejection and integration of these particle sensors has delayed the large-scale proliferation of this technology. With the advent of more advanced sensors and easy-to-use manufacturing integration systems, chipmakers are now able to realize the benefits of improved process control and productivity.

The ability to capture particle excursions and correlate them to wafer-level defects and chamber condition has enabled a 25% improvement in the mean time between kit changes, and has provided an unprecedented level of particle control within the LSI Logic fab in Gresham, Ore.