During the deterministic alignment process with the multi-robot collaboration, several significant functions were implemented, such as the robotic tool change process, visual servoing, target detection, and SU-8 photoresist printing.
A. Robotic Tool Change Process
There are two tools on the tool changer station in NeXus employed in the alignment process, HIWIN XEG-32 electrical gripper and Nordson 794 Auger valve. The tools were picked up by the 6-DOF robotic arm and the 4-DOF robotic arm respectively using the ATI QC-11 tool change coupler. The electrical gripper was used for grasping the metal frame and holding it parallel to the 4-inch wafer sample chuck. The Auger valve was used for dispensing MICROCHEM® SU-8-50 photoresist (MicroChem Corporation) as the UV-curable adhesive on the fabric material to stick it to the wafer after the alignment process. When the alignment process was done, both tools were placed back in the tool change station at their constant locations.
B. Visual Servoing and Target Detection
With the visual feedback of the Dino-lite® camera on the 4-DOF robotic arm, targets, like the intersection for alignment and the MEMS clamps alignment results, can be detected and checked. Also, to improve the precision of alignment results, a visual servoing function was adopted to enhance the accuracy of the alignment between the fabric intersection and the alignment mark. In the visual servoing function, image Jacobian was employed to adjust the center of the alignment mark reaching its corresponding intersection on the fabric based on the coordinate and orientation of the intersection in the image. Eq. (2), (3), and (4) [15] expresses how to calculate the motion of stages of the 6-DOF positioner to move the alignment mark to reach its corresponding intersection on the fabric.
$$\left[\begin{array}{c}\varDelta {P}_{x}\\ \varDelta {P}_{y}\\ \varDelta {P}_{z}\end{array}\right]= {J}_{image} \left[\begin{array}{c}\varDelta X\\ \varDelta Y\\ \varDelta \theta \end{array}\right]$$
2
$${J}_{image}= \left[\begin{array}{ccc}{J}_{11}& {J}_{12}& {J}_{13}\\ {J}_{21}& {J}_{22}& {J}_{23}\\ {J}_{31}& {J}_{32}& {J}_{33}\end{array}\right]$$
3
$$\left[\begin{array}{c}{X}_{new}-{X}_{cur}\\ {Y}_{new}-{Y}_{cur}\\ {\theta }_{new}-{\theta }_{cur}\end{array}\right]= \varDelta s {{J}_{image}}^{-1}\left[\begin{array}{c}{P}_{{x}_{d}}-{P}_{{x}_{cur}}\\ {P}_{{y}_{d}}-{P}_{{y}_{cur}}\\ {P}_{{\theta }_{d}}-{P}_{{\theta }_{cur}}\end{array}\right]$$
4
$${J}_{image}=\left[\begin{array}{ccc}-0.0051597& 0.0050213& -0.0337985\\ 0.00514525& 0.00557106& -0.416211\\ 0& 0& -1\end{array}\right]$$
5
where \(\varDelta X\), \(\varDelta Y\), and \(\varDelta \theta\) are the variations of the 6-DOF positioner in a plane; \(\varDelta {P}_{x}\), \(\varDelta {P}_{y}\), and \(\varDelta {P}_{\theta }\) are variations in the pixel of the target’s center in the Dino-lite camera’s field of view. Since the image Jacobian is a 3 x 3 matrix that has 9 entries, the linear least squares estimation method was applied to find the entries’ values of the image Jacobian. \({X}_{cur}\), \({Y}_{cur}\), \({\theta }_{cur}\), \({X}_{new}\), \({Y}_{new}\), and \({\theta }_{new}\) are the current and new configurations of the 6-DOF positioner; \({P}_{{x}_{d}}\), \({P}_{{y}_{d}}\), \({P}_{{\theta }_{d}}\), \({P}_{{x}_{cur}}\), \({P}_{{y}_{cur}}\), and \({P}_{{\theta }_{cur}}\) are values in pixel of the desired and current position and orientation of the center of the alignment mark in the Dino-lite camera’s FOV, \(\varDelta s\) represents a step size of the stages’ movements. Based on the vision feedback values, the alignment mark can move to the corresponding intersection in a fast and precise method by using the visual servoing technique in Fig. 11.
After the alignment mark on the wafer was aligned with the intersection on the fabric, the Dino-lite® camera was controlled to check random MEMS clamps' alignment results. Figure 12 depicts some MEMS clamps aligned with their corresponding intersections. Compared with the layout design of the mask for the MEMS clamps fabrication in Fig. 2, after the alignment process with the multi-robot collaboration in the NeXus, the center of each MEMS clamp was aligned with its corresponding intersection properly, during the visual servoing process, there are some buffers programmed in translational and rotational direction to adjust the target to move to the desired position and orientation. Those buffers resulted in +/- 20µm in translation and +/-0.5° in rotation in Fig. 13. Moreover, considering adding 35µm of the repeatability of the 6-DOF robotic arm, the alignment errors are still under the grasp tolerance of the MEMS clamps, which matched the yield in the layout design. Therefore, the yield of alignment results was more than 95%.
C. Auger Valve Printing
When the alignment process was done, the 4-DOF robotic arm moved to pick up the Auger valve on the tool change station, the DYMAX® UV curable adhesive was set up on the Auger valve, and a gauge-14 nozzle was mounted on the bottom of the Auger valve. The nozzle position was calibrated with respect to the center of the Dino-lite® camera. Based on the coordinate of the alignment mark intersection in the image of the camera, the center coordinate of the fabric can be calculated with respect to the center of the camera. The nozzle can be homogeneously transformed to the center of the fabric (shown in Fig. 14) to start the SU-8 photoresist printing based on the G-code generated by Inkscape® software. There was a 55mm x 55mm square pattern of SU-8 photoresist was printed along the outline of the fabric. Then the Auger valve was removed from the top of the frame, the UV light was turned on for at least 10 minutes to cure the SU-8 photoresist.
D. MEMS Clamps Release Process
When the SU-8 photoresist was cured thoroughly, the wafer was attached to the fabric stably. The fabric with wafer was transferred to the cleanroom to use XeF2 gas (Xactix, Inc.) to isotopically etch silicon to release the clamp structures. Figure 15 (left) depicts the top view of the whole E-textiles sample after the releasing process with XeF2 gas, and Fig. 15 (right) shows the MEMS clamps pop to grasp the textiles and conductive fibers in the micro-scale. Their purpose is to provide mechanical stability and electrical contact for MEMS devices tethered to functional fibers in a textile. Note that SU-8 photoresist was selected to stick fabric to the wafer because it would not react with XeF2 gas during the Si etching process after experimental tests, there was no polymer layer generation to prevent the Si etching process. Also, Nylon fiber was selected to be the non-conductive base fiber to have the same purpose as the SU-8 photoresist.