On-the-fly circular substrate centering for robotized vacuum systems
09/01/2008
A new method for on-the-fly detection and correction of substrate-centering inaccuracies of a circular substrate is presented. The method employs a set of photoelectric sensors externally mounted on the transport chamber along the substrate transfer path. The robot continuously captures both the position of the end effector and the sensor’s signal when the moving substrate partially blocks them. The collected substrate’s edge profile is to be used by the robot to compensate for the shift in the substrate.
Processing of substrates, a fundamental part of producing integrated circuits, usually consists of a number of pre-determined sequential steps. At certain process steps, a silicon wafer (one form of a circular substrate) is transferred within a fully automated vacuum system. The system typically consists of a vacuum transport chamber (TC) with load locks (LL) and process modules (PM) connected to the TC [1,2]. The TC is serviced by a robot whose end effector holds a wafer on it by means of frictional force.
Operations performed by the robot include elementary and straight line moves that can also be combined into complicated, 3D trajectories to comply with complex workspace geometries and enhanced throughput of wafers through the system [3]. During vacuum processing, the substrate is typically exposed to processes such as chemical and physical vapor depositions, dry etching, and ion implantation.
A wafer may experience limited, but not negligible, motion due to gas flow, pressure changes, temperature variations, and biased electro-static chucks. Consequently, the robot may pick a wafer that exhibits unwanted inaccuracy, introducing a difference between the desired location of the wafer’s center and its present one. This difference needs to be minimized, or else process issues can develop and, under extreme circumstances, wafer breakage may occur.
Substrate centering is an exceptionally important procedure that is used to help in locate the center of the substrate prior to placiment in the subsequent PM. There are a number of types of center-finding systems adopted currently in semiconductor manufacturing applications. Those that determine the location of the substrate center while the substrate is moving are called on-the-fly center-finding systems [4]. They typically consist of sets of through-beam or reflective sensors, arranged on the TC through which substrates pass. The information on the substrate’s location is obtained by recording the encoder position every time a sensor triggers and, depending on the configuration used, the shift of the substrate is calculated based on the data from up to eight sensors.
The accuracy and repeatability of existing methods strongly depend on the sensor selection, accuracy of the sensor, accuracy of sensor placement and their orientation, the need to move the substrate along a specific path, and even the use of calibrated (i.e., must have very precise geometry) substrates [5]. It explains why design simplicity can be considered as a positive factor for achieving high accuracy and repeatability, while increasing the complexity of the method used usually creates new sources of errors.
Method overview
Here we report on the progress of an unconventional method for on-the-fly detection and correction of substrate-centering inaccuracies of a circular substrate. Advanced semiconductor maintenance and processing schemes require a substrate center-finding technique that permits flexible operation without strict requirements to the number of sensors and their positions, the size of the substrate, or a calibration fixture.
 Figure 1. Calculations of ∆X and ∆Y are based on the found positions of the substrate’s center for two different edge profiles. |
The technique that was developed is based on the analysis of a substrate’s edge profile while it is transferred under a light curtain (one-dimensional). The key idea of the patented algorithm is to use this edge profile for calculations of both the substrate’s radius and shifts along the X and Y directions. For the geometry shown in Fig. 1, a center of Cartesian coordinates corresponds to the center of the robot. Extension of the end effector occurs along the Y axes. The shifts ∆X and ∆Y can be found by comparing calculated positions of the substrate’s center for two subsequent operations, for example, place (dotted line) and pick (solid line).
 Figure 2. Signal from the partially blocked receiver represents the moving substrate’s edge profile. |
Figure 2 represents a schematic view of the partially blocked light curtain created by the circular substrate. The signal from the optical amplifier, along with the end-effector’s position, form a unique set of data related to the given substrate and its desired trajectory. More detailed discussion on the algorithm follows.
null
Shift detection algorithm
When the light from the emitter is partially blocked by the moving substrate, the corresponding analog signal, along with the captured substrate’s position, is used to collect the substrate’s edge profile. The number of points on this curve is crucial to obtain good resolution and is determined by the bandwidth of an optical amplifier and the robot capture capabilities. A typical edge profile collected with the use of Keyence FU-12 sensors and FS-V20 amplifier is shown in Fig. 3 [6]. This data allows the calculation of the wafer’s radius by applying:
 |
where c is the calculated distance between points A and B (c=|B-A|), and h is the segment’s height, which can be found from (2):
 |
where I is the signal intensity, and k is the proportionality coefficient that allows the conversion of sensor voltage readings to micrometers. This coefficient was determined in a separate experiment devoted to the calibration of the photoelectric amplifier. Such calibration enables the capture of the Y position at the wafer’s edge relative to the center line of the robot end effector in μm.
Points A and B of curve a on the last figure represent leading and trailing edges of the substrate. Overall precision of the method strongly depends on how accurately these points can be determined. In this work, an averaging procedure among 100 points on the curve was used to get the background level before point A and after point B, and first derivative of the data to locate these points precisely. Yet another possible way to increase the amount of points on the curve is to interpolate an original curve to fix the step on the level of only a few micrometers. An exact position of minima on the curve was found by analyzing the first derivative of the data.
Once the radius of the wafer is determined, as well as positions of leading and trailing edges, they create a unique “fingerprint” of the wafer that can be stored in the robot memory specific to the given station (PM, LL, etc.). On its movement back from the station, the wafer passes through the same light curtain and yet another edge profile can be collected by the robot. They should be identical (ideally) if no shifts were introduced at the station and no significant temperature change was experienced by the wafer. Otherwise, there is a detectable difference between the two edge profiles.
It is worth mentioning that a silicon wafer (D0=300mm) may experience an extension in its diameter of up to 500??m due when under a temperature at ~600??C per the equation:
 |
where α(T) is temperature expansion coefficient [7]. Both curves on Fig. 3 were collected at room temperature.
As an example, curve b on Fig. 3 was acquired after shifting the wafer by 3000μm along the X and Y directions at the station, which was controlled by an independent vision system (Keyence CV2100), capable of measuring displacement as accurate as a few micrometers. By using the above mentioned calibration procedure, the shifts along the X and Y directions were found as 2950μm and 3010μm, respectively. Discussions of repeatability and statistical analysis of the test data follow.
Detection of alignment fiducial and substrate breakage
An alignment fiducial is used to set up proper orientation of a substrate within a cluster tool. In the case of a silicon wafer, the size of the fiducial is specified by SEMI and is close to 2mm. Nevertheless, this small deviation from the circular shape needs to be taken into account, specifically when the notch is oriented towards the sensor the substrate passes through. This case is illustrated in Fig. 4, which shows an edge profile (open dots) of a standard 300mm silicon wafer aligned perfectly on the robot end effector.
The presence of the notch is indicated by a small kink on the otherwise smooth curve, preventing determination of the minima position. To overcome this problem, the first derivative of the data points was taken and the result is presented on the insert. A point where a linear fit of the derivative is equal to zero gives the position of the minima. The small discrepancy between “expected” and calculated positions of minima in the given example was <15μm. All other possible orientations of the alignment fiducial will not affect center-finding calculations.
The method is also suitable for the detection of sudden deviations from the wafer’s circular shape such as large defects or breakage. If, based on the data collected, the algorithm cannot calculate the radius or the value exceeds a threshold, it signals that a significant deviation has been detected. When this happens, possibly due to substrate breakage or an unexpectedly large shift, the robot will report an error condition which can, in turn, be used to abort operation.
The performance of the method was evaluated by using a test system consisting of a Brooks Automation MAG7 vacuum robot with Dual Arm and ceramic end effectors. Photoelectric through-beam sensors (FU12) and visual system (CV2100), both by Keyence, were used to collect data and verify position of the SEMI-compliant 300mm silicon wafer, correspondingly. The sensor was set up at the radial distance of 350mm from the center of the robot, two thirds of the way to the test station. The maximum velocity of the robot during the test was 15000μm/sec due to the data transfer rate between the robot and host PC (for the algorithm verification purposes only, the program was executed on the separate PC, not on the robot).
To evaluate the algorithm, the robot was commanded to perform “delta place???pick with correction???place to verify” cycles to/from the test station. The absolute values of the shifts were ??1000μm along the X and Y axes to test all quadrants of the robot coordinate system. After executing “place to verify” operations, the calculated inaccuracy was collected along with the data from the vision system.
 |
Three cycles of measurements were taken for each direction adding up to approximately 48 data points in total. The Table summarizes the test results. It shows that calculated inaccuracy of the developed algorithm is satisfactorily close to that measured by the vision system. The absolute values of placement errors do not exceed 150μm and meet the performance requirements for the targeted semiconductor manufacturing applications (250μm). Figure 5 shows the distribution of the collected data. The analysis finds that the three-sigma placement error of approximately 180μm was achieved.
As next steps for further development, the authors would like to perform detailed research into the performance of the proposed method while using embedded robot software. The software development effort to continue this study would require adding the following capabilities to the embedded robot software:
- Logic to gather analog data for creating substrate edge profile;
- Logic to analyze the substrate edge profile, compare this profile to a stored calibrated (centered) substrate edge profile and determine the offsets from the ideal center (as described in this paper); and
- Logic to compensate for the center offsets on the fly while a substrate is being placed into the destination position.
This development would allow the robot to transfer centered substrates at the maximum allowable speed.
Conclusion
A method for on-the-fly detection and correction of substrate-centering inaccuracies of a circular substrate transferred within robotized vacuum systems has been developed. The method employs a pair of photoelectric sensors externally mounted on the transport chamber. The sensors are located along the substrate transfer path, allowing the robot to continuously capture both the position of the end effector and sensor’s signal when the moving substrate partially blocks the sensors. The collected substrate’s edge profile can be compared with the one stored previously by the robot to correct the substrate’s shift. This results in the substrate being moved within the vacuum system on the desired trajectory regardless of the amount and direction of the initial shift.
The method can also be used for defect detection purposes. The performance of this method was tested at lower velocity due to the limited data transfer rate between the robot and host PC. Test results for this method meet the requirements for targeted processes in the semiconductor industry and look promising for further development.
Acknowledgments
The authors would like to thank K. Majczak, system architect at Brooks Automation Inc., for valuable help in the concept development and suggestions during preparation of the paper and J. Xu, MIT, for support in data collection and analysis. The method described here was developed by Brooks Automation Inc.
References
- D. Beaulieu et al., “Substrate processing apparatus having a substrate transport with a front end extension and an internal substrate buffer,” US patent 5,882,413, 1999.
- J.C. Davis et al., “Substrate transport apparatus with dual substrate holders,” US patent 5,647,724, 1997.
- M. Hosek et al., “On-The-Fly Substrate Eccentricity Recognition for Robotized Manufacturing Systems,” Jour. of Manufacturing Science and Engineering, V.127, p. 208, 2005.
- S. Sundar et al., “On-the-fly center-finding substrate handling in a processing system,” US patent 6,198,976, 2001.
- J.M. White et al., “Apparatus and method for robotic alignment of substrates,” US patent 6,557,923, 2003.
- Keyence catalog, 2006.
- See for example: J. Fabian et al., “Parameters and Thermal Expansion of Amorphous Silicon: A Realistic Model Calculation,” Phys. Rev. Lett. 79, 1885, 1997.
- A.G. Krupyshev et al., “Substrate alignment apparatus,” provisional US patent 390-012038-US(PAR), 2004.
Alexander Krupyshev is senior director, systems technology development, at Brooks Automation Inc., 15 Elizabeth Dr., Chelmsford, MA 01824 USA; ph 978/262-7784, fax 978/262-2513, e-mail [email protected].
Kurt Greissinger is product marketing manager at Bosch Rexroth Corp., 816 East Third Street, Buchanan, MI 49107 USA; ph 269/697-5211, fax 269/695-5363, e-mail [email protected].
Haniel Olivera is embedded SW manager, tool automation division at Brooks Automation.
Sergei Syssoev is chief scientist, critical components group, at Brooks Automation; ph 978/262-7728, e-mail [email protected].