An automated robot teaching method improves process control
07/01/2009
Craig C. Ramsey, CyberOptics Semiconductor Inc., Beaverton, OR USA
Legacy robot teaching methods contribute to increased cycle time, increase the risk of particle generation and contamination, and introduce variation into the task because of differences in skill, motivation and training. An automated robot-teaching method was evaluated by a 300mm European fab manager and his team; they were able to reduce the time to conduct preventive maintenance, mean-time-to-repair (MTTR), and downtime.
The thin-film group at a 300mm European fab wanted to reduce the tool and production downtime resulting from a time-consuming metrology process that called for multiple technicians to vent and dismantle Endura/Centura tool sets and employ manual methods to teach wafer-transfer coordinates to robots. The process took ~24 hours.
The legacy robot-teaching method required techs to follow a manual process to set up, maintain and troubleshoot the Endura and Centura tools. The labor-intensive process produced no measurements in micrometers, inaccurate alignments and wafer mishandling in the tools, and ultimately, it reduced the group’s film-deposition uniformity and die yield per wafer due to particle generation, resulting in wafer scrap.
Techs in the thin-film group used legacy teaching methods that employed bubble-levels, calipers and dowel pins to manually set robots’ wafer-transfer coordinates. The manual methods required the group’s manager to dedicate multiple technicians to cool, vent and dismantle tools. Many of the procedures relied on line-of-sight, eye balling, and technician consensus to calibrate robots.
The techs’ ongoing work with the off-line Endura and Centura tools increased the risk of particle generation and contamination and made inline and impromptu handoff checks impossible, thereby increasing maintenance overhead. The legacy process put techs in danger when it required them to reach hard-to-access tools or set blind handoffs, i.e., assign calibration data found in an open chamber to closed chambers.
The techs primarily set wafer-transfer coordinates in the Endura and Centura tools by aligning dowel pins through the end-effector, chuck and custom Plexiglas fixture holes. The techs made the most of their process and essentially expended a great deal of time working with off-line tools to get imprecise data. Accurate teaching is required because these tools can place a wafer with one robot and pick the same wafer with another robot. More than 30 handoff coordinates must all be accurately taught in some tools to maintain wafer handling process control. If one or both robots use inaccurate handoff coordinates, errors build and propagate — quickly leading the wafer handling process to drift out of the desired operating limits.
Each technician followed the group’s manual teaching process differently depending on training, skill and overall experience. The highly variable calibration methods prevented the thin-film group from establishing a repeatable teaching process for the Endura and Centura tool sets. None of the group’s calibrations were performed under vacuum, so the measurements they collected didn’t reflect true production conditions.
The legacy robot-teaching process left the thin-film group without reliable or complete metrology data to improve their maintenance and production processes or establish uniform teaching standards for the Endura and Centura tool sets across the fab. Additionally, once techs completed their manual teaching process, the thin-film group continued to remain off-line to reassemble the Endura and Centura tool sets and restore both vacuum and production processes, which, in many cases, took 4???8hrs.
Testing automated robot teaching
To address the flawed legacy robot-teaching process, the European fab’s thin-film manager wanted to identify an automated method that would allow techs to quickly teach wafer-transfer robots in the Endura and Centura tool sets and ensure repeatability. The manager set up an internal evaluation of a wireless, wafer-like robot-teaching device with an on-board camera. During the evaluation, the manager placed the wireless device in one of the tools to obtain real-time wafer-transfer coordinates via the device’s companion software and set handoff positions. He asked a team of techs to simultaneously calibrate a second tool using the group’s legacy robot-teaching process.
The manager loaded the wireless device on the end-effector blade’s centering hole. The tool transferred the device as if it were a wafer to all process locations for efficient handoff teaching and quality checks. The group’s manager reported that, while working side-by side with engineering staff on two adjacent tools, his team was able to complete troubleshooting and adjusting the wafer handoffs (10 chambers checked and two adjusted) in two hours with the automated teaching system. The other team took six hours on its tool to look at two chambers using the traditional method, and still had issues beyond that, which meant two days to get the tool back on-line.
The automated teaching device obtained live video from inside the tool and reported in real-time its three-axis (X, Y, Z) coordinates in relation to a target via the GUI of its companion software. The device reported X- and Y-axis offsets with an accuracy to 100??m, which allowed the group to obtain precise wafer-transfer data — including go/no-go data — to analyze, improve and set process standards.
 Comparison of the time needed to complete preventive maintenance and the mean-time-to-repair with respect to legacy robot teaching vs. automated wireless teaching. |
With the device, one tech was able to set up, maintain and troubleshoot the Endura and Centura tools while reducing typical maintenance time from 24 hours to 3???4 hours (figure). The process did not require the tech to cool, vent or dismantle the tool. The wireless device worked in environments up to 120??C for up to five minutes and was vacuum compatible.
The device automatically moved through the entire tool set, and its wireless transmission of data from inside the tool eliminated the need for dangerous blind robot teaching by techs. The techs ultimately had less to worry about with the automated method, and focused on other maintenance-related issues and optimizing their tools — re-characterizing their process.
The European fab’s thin-film group used the automated device to replace legacy robot-teaching methods for the Endura and Centura tool sets and, based on the data, saved hundreds of thousands of dollars per year by adopting automated robot teaching. The savings came largely from the thin-film group’s ability to use the device to reduce, per tool, robot-teaching downtime by 20 hours. The device allowed the group to reduce the number of techs required for robot teaching from three to one and save ~3,500 staff hours per year, while increasing per-wafer die yield.
Techs used the precise wafer-transfer coordinates they obtained with the device to set accurate tool alignments and improve overall wafer handling, including on-demand handoff checks. The Endura and Centura tool sets remained under vacuum during teaching and did not require the group to set blind handoffs, i.e., assign calibration data found in an open chamber to closed chambers.
The automated robot-teaching process helped the thin-film group increase film-deposition uniformity and die yield per wafer by reducing particle generation and wafer scrap. Additionally, real-time data obtained by the automated device allowed the thin-film group to establish a repeatable robot-teaching process for the Endura and Centura tool sets and uniform process controls, eliminating tech-to-tech variances and a metrology process founded on manual trial-and-error. All techs were able to automatically set up, maintain and troubleshoot handoff positions for wafer-transfer robots, regardless of their fab experience.
Conclusion
Accurate wafer handling robot handoff coordinates are required to control the location of wafers in process chambers. Even small errors in only one robot handoff coordinate can easily accumulate when multiple robots cooperate during tool operation. Using a wafer-like wireless teaching camera to measure handoff offsets removes technician judgment from the handoff coordinate teaching process and yields “as found” and “as left” wafer handling process data that can be used to implement an effective process control regimen. When the teaching device is additionally vacuum-compatible, further tool and labor savings can result. A high-volume semiconductor device manufacturer saved significant operating expense.
Craig C. Ramsey received his BS in chemistry from Purdue U. and his MS and PhD in bio-physics from The Ohio State U. and is the general manager of CyberOptics Semiconductor, Inc., 13555 SW Millikan Way, Beaverton, OR 97005 USA; ph.: (503) 495-2200; [email protected].