Handling reliability improves with in situ diagnostics
10/01/2004
Wafer-handling systems in semiconductor process equipment can significantly limit overall tool reliability. Lack of objective, real-time measurements for mechanical performance and calibration of handlers can lead to unpredictable behavior and tool downtime. Applying in situ diagnostic tools to automate robot calibration and gathering meaningful performance statistics improved wafer-handling reliability and tool productivity.
Poor reliability of wafer-handling systems in IC processing tools can significantly inhibit overall tool performance and reliability in manufacturing plants. Failures due to wafer-handling systems have significant mean time to repair (MTTR), resulting in expensive tool downtime.
Fab tool data collected by a major integrated device manufacturer (IDM) show significant productivity loss due to downtime from wafer-handling failures. The IDM monitored data from several ion implantation systems over one year's time and found tools experiencing a mean time between failures (MTBF) of 155 hrs, with wafer handling the primary cause of tool downtime. The MTTR for these failures was 5.25 hrs [1]. In addition, these tools were typically down three to four times per event due to the difficulty of correctly performing diagnosis and repair.
Based on the chipmaker's data, more than 90% of failures were caused by improper placement of the wafer in the robot's end effector, resulting in broken product wafers during transfer and handling. The problems were usually addressed by the replacement of wafer-handling components or manually recalibrating the handler. Overall, <10% of the root causes for failures were clearly identified, according to the device manufacturer. The problems often were incorrectly identified as failed system components, including motors, cabling, or the robot itself.
Recent studies indicate that a number of operands in the reliability equation can be increased with deployment of in situ diagnostic tools in wafer-handling systems, resulting in higher MTBFp (mean productivity time between failures) and higher MCBF (mean cycles between failures), based on Semi E10-0701 guidelines.
Failure without warning
On the surface, current wafer-handling systems exhibit a binary behavior — they either work or they fail. In reality, these systems can successfully perform operations even while calibration and health of their critical wafer-handling devices are degrading. There are no objective measurements of how well handling components are calibrated or how consistently electromechanical elements operate compared to a "known good" baseline. As they approach critical failures, unexplainable intermittent problems may occur.
Typically, the most critical wafer-handling device in a tool is the wafer-transfer robot. Once a failure occurs, proper diagnosis and analysis frequently require the robot to be removed from the wafer-processing tool and delivered to specially designed test fixtures located at the supplier's laboratory. A great deal of cost can be incurred moving the robot between the wafer fab and the supplier's lab.
All too often, suppliers determine that there is no problem with the robot and that the cause of failure must be in another system component. Costly resources are expended because adequate diagnosis of the robot is not possible while it is in its normal production configuration.
A collection of in situ tools has been developed to improve calibration and provide active knowledge of wafer-handling health. Together, these tools are able to reduce the likelihood of failure while providing rapid and precise diagnosis of robot performance. They are utilized in situ, without removing the robot from the wafer-processing equipment.
Fast automatic teaching
One reliability factor in wafer-handling systems is the requirement to "manually teach" the wafer positions of the handler robot. The quality of this subjective manual teaching mode is directly correlated to the skills and knowledge of technicians. For years, manual teaching has been thought to be the only feasible method of setting robot positions for each wafer-handling station.
However, automatic calibration methods have gained momentum. In Berkeley Process Control's patented automatic calibration method, a controller is programmed to drive the robot's motors and set its arms to a certain position [2]. High-resolution encoders provide feedback to the controller, indicating the position of each motor. Controller software continuously compares the actual motor feedback position to the software-commanded motor position to generate appropriate drive signals. The controller's integrated drives provide the necessary motor drive current. Through this tight integration, the controller has real-time knowledge of the velocity and torque of each motor.
While manual teaching procedures may take 4–8 hrs [3] or more to complete, automatic robot calibration is typically completed in a matter of minutes — reducing downtime.
An advantage of touch-sensing calibration is that it does not require additional hardware or sensors. The controller has real-time knowledge of the velocity and torque of each robot motor and the present position of the robot's end effector. The controller also knows the approximate location of the robot's wafer handoff positions and the geometry of the end effector, based upon previous information provided by the tool application developer.
In the "touch calibration" mode, the controller commands a robot axis to slowly move the end effector into the predefined nominal location for handoff of wafers in process tools. When the end effector makes light contact, the axis slows down and the motor torque changes, indicating physical contact (Fig. 1). The controller instantly captures the encoder position as the calibration point. Since the controller is aware — in real time — of the precise torque requirements of each motor, touch calibration is achieved with very low contact forces. Sophisticated torque-data processing algorithms are used to eliminate false triggers and ensure calibration consistency despite dynamic mechanical characteristics of the robot.
Data show that automatic robot calibration cycles demonstrate a tenfold repeatability improvement vs. manual teaching methods. In a recent analysis performed on an inspection tool at a major semiconductor-equipment manufacturer, hundreds of automatic calibration cycles demonstrated repeatability within the width of a human hair. Standard deviation in a tool coordinate frame came to x = .025mm, y = .021mm, and z = .031mm.
In situ automatic calibration also provides the foundation for three additional reliability tools to monitor and diagnose the health of robotic systems while they are being used in manufacturing equipment. These new capabilities can be generally described as wafer-map tuning, calibration tracking, and mechanical-systems monitoring.
Comparing calibration approaches
For decades, semiconductor equipment OEMs have relied on manual calibration of wafer-handling systems. In a series of steps, technicians can calibrate robot end effectors by using leveling tools, turning screws, and nudging robotic arms into desired positions for wafer handling. One major semiconductor equipment OEM has estimated that highly skilled system technicians are only able to calibrate handlers to within 0.5mm repeatability using manual methods.
A number of "auto-teach" methods have been deployed in recent years to improve upon manual teach methods, but many of these approaches cannot support full in situ diagnostics of handlers. One common auto-teach method uses a combination of specially designed fixtures and sensors placed in the wafer-handling station. In some systems, fixtures detect position using mapping lasers. Others use proximity sensors to detect the location of the end effector. While reducing the time it takes to teach wafer handlers, these approaches also require special fixtures for the end effectors and robotic arms. Often, these special fixtures must be used when handlers are re-calibrated in the field. The use of sensors can also present additional reliability concerns.
Touch-sensing calibration has opened the possibility to evaluate mechanical integrity of handlers by monitoring the position, velocity, and torque of each motor in the system. Handlers with mechanically damaged robots can give the false impression that systems are working properly if positions are detected and measured only by calibration sensors. Real-time access to motor torque and other system servo control data supports the ability to provide full in situ analysis of robot performance.
Calibration quality tracking
While robot calibration may be successfully achieved when a tool is being set up, the quality of calibration during operation is rarely known. To be certain of ongoing quality, the calibration sequence must be repeated; however, there is no certainty that the new calibration data is better than the original set-up data.
One solution is to compare incoming calibration data, collected by the controller, and the set-up baseline data while robots are operating inside tools. The calibration data is compared to the baseline and significant deviations are recognized as a critical change in the wafer-handling equipment. The equipment can be recalibrated with automatic routines, without special tools and with handler devices in situ. Trends in the change of calibration data are monitored as well. The abilities to monitor the repeatability of calibration and easily perform automatic calibration routines allow the system to maintain optimum performance and wafer-handling reliability, which can be displayed on a user interface screen.
Precisely tuned mapping
Another potential source for failure is the wafer-mapping device. Robots typically utilize mapping lasers or through-beam sensors to determine the presence of wafers on each handling device and whether or not wafers are properly positioned. Improper recognition of a wafer can result in expensive failures due to the potential for damaging devices on substrates.
Establishing and maintaining proper mapping-system parameters require precision tuning. Wafers can vary in thickness and optical properties depending on the process steps being completed, as well as what type of products are being made. Generally, wafers must be mapped at two angles to ensure that they are properly detected in the correct locations and to enable detection of cross-slotted wafers.
 Figure 2. "Keystoning" can cause wafer-mapping errors when scan and fan angles are incorrectly selected. |
A major factor in the incorrect mapping of wafers is an effect known as "keystoning." This occurs when the mapping scan and fan angles are incorrectly selected for the optical properties of the wafer-mapping device (Fig. 2). By using an automated tuning algorithm to optimize mapping parameters, the keystoning effect is significantly reduced. For example, the difference in wafer thickness between two mapping positions on a sample wafer using the automated tuning algorithm for 1000 mapping cycles was 0.015mm — 2.14% difference in thickness of a Semi-standard wafer.
During operation, mapping parameters are monitored and compared to baseline performance. Significant deviations are recognized and the user is alerted that mapping parameters may need to be retuned. The user can quickly diagnose the condition and optimize mapping parameters.
Mechanical system integrity
A third in situ tool monitors mechanical-system integrity. Wear in a robot's drive mechanism can go unnoticed, resulting in critical failures. Wear is a normal occurrence in any mechanical device. Changes in lubrication conditions also can alter the dynamic properties of wafer-handling actuators. Any change in the mechanical dynamics causes changes in the required energy to move robotic arms, which is directly related to the torque of each motor for a given movement.
In Berkeley's approach, real-time data is tracked for each axis. Motor torque, velocity error, and position error are analyzed for minimum, maximum, mean, and standard deviation relative to baseline performance for an optimum mechanical system. Capturing this information while the robot is in situ enables preventive maintenance prior to system failure.
 Figure 3. The data demonstrate that motion performance alone does not sufficiently inform the user about the mechanical integrity of wafer handlers. |
As an example, a test was performed with two identically designed robots utilizing the same controller, software, and commanded move profile. One robot had a significantly worn drive actuator; the other robot had a new one. Both robots met satisfactory criteria, following the move profile; however, the robot with the worn axis required 38% more energy to complete the same move (Fig. 3). The data demonstrate that motion performance alone does not sufficiently inform the user about the mechanical integrity. Using this in situ tool, the user gains knowledge of trends in the motor torque profile and can recognize mechanical deterioration long before a performance failure threshold is met.
Wafer-handling systems utilizing the described in situ tools have demonstrated MTBFp in excess of 60,000 hrs and more than 2.8 million MCBF at 97.5% confidence level per Semi E10-0701 standards. These results reflect data from both production tools and laboratory testing, including the control of several different robot designs and both vacuum and atmospheric wafer-handling environments.
Conclusion
Determining the quality of wafer-handling device calibration and health while the devices are in situ can reduce downtime and productivity loss. A collection of in situ diagnostic tools recognizes potential failures and diagnoses failed wafer-handling components efficiently on the fab floor.
References
- J. Irwin, IC Irwin Consulting, Austin, TX, 2003.
- Berkeley Process Control patents: "Touch Calibration System for Wafer Transfer Robots" No. 6,242,879; and "Automatic Calibration System for Wafer Transfer Robot" No. 6,075,334.
- K.C. Januc, "The Expanding Role of Robots in Process Tool Productivity," Solid State Techology, Jan. 1999.
Contact Lenson Wong at Berkeley Process Control Inc., 4124 Lakeside Drive, Richmond, CA 94806; ph 510/243-3375, fax 510/222-8737, e-mail [email protected].