Intelligent vision system for grasping error compensation of a wafer manipulation robot.

Stanciu, Mihai ; Nicolescu, Adrian ; Enciu, George 等

1. INTRODUCTION

Most of the industrial robots used in silicon wafer manipulation are SCARA type articulated arms with 3 or 4 degrees of freedom. Generally the silicon wafer manipulation process is performed in vacuum or in a clean room environment. In present the semiconductor industry try to optimize the manufacturing process using 200mm and 300mm diameter wafers and the average manipulation speed is 1.2m/s. The silicon wafer weight is in the range 0.5-0.8kg but considering the very high value of the manipulation speed involving too a high acceleration value on the trajectory, usually in 5-8% of the cases a positioning error of the wafer in the grasping mechanism occurs (***,2001). This error is affecting the final positioning of the wafer in the dropping point (that can be a place in a wafer cassette or processing station).

The authors of this paper developed a specific vision system able to detect the positioning error and to provide the correction parameters trough a feedback loop to the robot controller (Nicolescu et al., 2002). The error evaluation process and correction parameters calculations are performed in the final stage of the trajectory trough the dropping point and the robot controller compensate this error by changing the coordinates of programmed endpoint (Stanciu et al., 1997).

2. VISION SYSTEM ARCHITECTURE

The vision system includes (fig.1):

* 1 or 2 video cameras installed on the bottom side of the robot end effector (for the 3 degrees of freedom version of the robot is required only one video camera due to the robot kinematics allowing the error compensation only on one direction--radial). For the 4 degrees of freedom robot 2 cameras are used to detect two components of the error in an horizontal plane;

* computer having installed the image processing software and two usb free ports;

* 1 or 2 usb cables connecting the cameras with the computer;

* RS232 serial cable connecting the computer and the robot controller.

[FIGURE 1 OMITTED]

Video cameras positions on the robot end effector are correlated with the wafer diameter. If the wafer is right positioned his edge will cover half of video cameras visual field in order to allows the detection of a positive / negative error value. This set-up increases the accuracy of the edge detection routine maximizing the image contrast. In the same time to increase the error detection software speed will be used gray scale images having a reduced size and less color information.

3. VISION SYSTEM CALIBRATION

The calibration process of the above presented vision system is performed automatically when the working cycle of the robot is started. In the first step the robot will pick up the reference wafer from a wafer holder. The vision system will detect the wafer edge and will set this edge (Guerra & Vlsi, 2000) as a calibration reference line (line position is expressed by pr). In the second step the robot will pick up the reference wafer from a wafer holder with an offset of 10mm. The vision system will detect the wafer edge (p10) and will set the vision system conversion parameter (cp) as the difference between this edge and the calibration reference line (fig.2). In the last step is calculated the vision system aspect ratio parameter (ar):

ar = 10/[absolute value of cp] [mm/pixels] (1)

cp = p10 - pr [pixels] (2)

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

Concluding, during the calibration stage is detected the reference line of a zero error positioned wafer and is also calculated the vision system aspect ratio parameter that will be used in error evaluation (fig.3). To optimize the error detection process will be used the reference wafer holder as a working station in the application architecture. In the same time periodically after 50-60 robot working cycles the vision system will be calibrated to increase the accuracy of the error evaluation process.

4. ERROR MEASUREMENT

The positioning error evaluation process is performed in two different ways depending on the number of robot DOF. A 3-axis robot can compensate the positioning error only on one direction. In this case will be used one video camera positioned on the longitudinal symmetry axis of the robot effector. The edge detection algorithm will select the 2 points defining the intersection between the wafer edge and the left and respectively right image limits (Spacek, 1986). To determine the error value is required only one point but using a second one will increase the accuracy of the calculation routine (fig.4).

A four axis robot can compensate both components of the positioning error. In this case will be used two video cameras positioned on the sides of the robot effector. The edge detection algorithm will select the two points defining the intersection between the wafer edge and the bottom and respectively right or left image limits for both images. To determine the error value are required only two points but using a second set will result an increased accuracy of the calculation routine (fig.5). Usually the aspect ratio parameter corresponding to a vision system resolution of about 3-3.2 pixels / mm and the overall accuracy of the positioning error routine will be in the range 0.3-0.35 mm. This value is very rough considering an usual 0.1 mm accuracy of this type of robots. To increase this accuracy can be used a more advanced video camera but the vision system resolution will be limited to 5-6 pixels / mm. To solve this problem the authors developed a mathematical method to increase the resolution by "devised pixel" technique.

In fig.6 is presented an enhanced image of a detected point with his 8 neighbors. On the figure are also marked the light intensity values for every pixel. With a standard resolution algorithm the detected point on the edge of the wafer will be the pixel marked on fig.6 with the light intensity value = 155. Assuming too that the vision system resolution is 3 pixels / mm resulting the reference point position as in fig.7.

[FIGURE 4 OMITTED]

[FIGURE 5 OMITTED]

[FIGURE 6 OMITTED]

[FIGURE 7 OMITTED]

[FIGURE 8 OMITTED]

By applying by "devised pixel" technique is assumed that every pixel weight is proportional with the light intensity as defined in fig.6. In this case the position of the reference point will be the position of the center of gravity of the fig.6. The equations of the center of gravity are presented in (3) and (4).

x = [[summation].sup.9.sub.i=1] [x.sub.i] x [I.sub.i]/ [[summation].sup.9.sub.i=1] [I.sub.i] (3)

y = [[summation].sup.9.sub.i=1] [y.sub.i] x [I.sub.i]/ [[summation].sup.9.sub.i=1] [I.sub.i] (4)

The results for the image presented in fig.6 are: x=0.44mm and y=0.59mm. Assuming that the vision system resolution is the same (3 pixels / mm) resulting the reference point position determined with the "devised pixel" technique as in fig.8.

Experimental research revealed that using "devised pixel" technique is possible to achieve an accuracy of 0.05 mm with a standard video camera (resolution 800x600 pixels).

5. CONCLUSION

Based on the requirement to minimize the positioning error resulting in wafer damage, the authors developed the vision system presented in the paper. The error detection model is based on an original "devised pixel" technique that allows increasing significantly the accuracy with standard hardware and low processing resources.

6. REFERENCES

***. SHR3000--Manual, JEL Corporation, 2001.

Guerra, C.; A Vlsi. Algorithm For The Optimal Detection Of A Curve, CAPAIDM, No.86 /2000, pp. 197-202.

Nicolescu, A.; Popescu, D.; Costian, D. Industrial robot assisted design using Catia P3 V5R7 environment, Constructia de masini, 2002 (54), nr. 6/ 2002, ISSN 0573-7419, OID-ICM, Bucuresti,

Spacek, L.A. Edge Detection And Motion Detection, IVC, No.4, 1986, pp. 43-56.

Stanciu M. D.; Nicolescu A.; Ispas C. Identification of Joint's Elastic Behavior Influence on Volumetric Accuracy of Industrial Robots., "IEEE's 6th International Workshop RAAD'97", pag.163 ... pag.168, STUDIO 22 Edizioni, ISBN 88-87054-00-2, 1997, Cassino, Italy.