Minimizing equipment downtime through advances in wafer handling
09/01/2005
Production ramps in 300mm fabs have increased the emphasis on equipment availability. Productivity gains from larger wafers result in fab models of record (MOR) requiring fewer tools than needed to produce an equivalent number of devices on 200mm substrates. New fabs are also employing fewer shared and redundant tools as process complexity grows, resulting in more unique fabrication steps. Consequently, advances in automated wafer-handling systems are needed to minimize both scheduled and unscheduled downtime in fabs. This article describes several gains being made in automation that enable quick replacement and accurate calibration of wafer-handling robots to maximize uptime in plants.
Fab equipment downtime can be divided into scheduled and unscheduled states, based on the Semi E10 standard (Fig. 1) [1]. Scheduled downtime has become a “necessary evil” for semiconductor manufacturers wanting to stay ahead of the equipment productivity curve by scheduling preventive maintenance or changing out consumables. During process development and early pilot production, scheduled downtime for preventive maintenance or equipment monitoring generally occurs at predefined time intervals. Scheduled downtime proves more difficult as production ramps, however, and it is certainly not an option for high-volume manufacturing, where the servicing of tools is typically based on the number of wafers processed. On the other hand, unscheduled downtime from failures can be extremely costly, resulting in unexpected lost capacity along with the need to repair tools in a hurry.
 Figure 1. How Semi E10 defines fab equipment downtime. |
Recent advances in wafer-handling systems are helping to reduce both scheduled and unscheduled downtime for greater operational efficiency of equipment as a whole. While unscheduled downtime remains reactive and scheduled downtime is generally proactive, both can be anticipated and better managed through the use of new concepts, such as data logging, system-health monitoring, predictive diagnostics, and fault-tolerance designs. These capabilities offer advanced warning of a part failure and instill confidence in pushing out preventive maintenance. For wafer-handling systems, additional measures are being introduced to ensure that production equipment can be quickly returned to a fully functional state if systems are shut down for maintenance or repair. This article focuses on several of these advanced features - kinematic mounted hardware designs, absolute motor encoders, and “hands-free” station teaching of wafer robots.
Kinematic coupling
The use of kinematic coupling is not new to semiconductor manufacturing, although to many in the industry, the term “kinematics” may be. To find kinematics at work in a 300mm fab, one only needs to observe the underside of a wafer-carrying front-opening unified pod (FOUP) and the corresponding alignment pins on loadports. Indeed, using a three-V-groove kinematic-coupling concept is an economical and dependable way of achieving high placement accuracy and repeatability of wafer FOUPs.
 Figure 2. Application of kinematic coupling for an atmospheric robot. |
Properly designed kinematic couplings are deterministic: They only make contact at a number of points equal to the number of degrees of freedom that are to be restrained [2]. Since only three points are needed to level and establish a plane, this can be achieved by machining a combination of three features such as cones, slots, or flats into a surface (Fig. 2) and using balls or cylinders with hemispherical ends to mate with these features (Fig. 3). Proper selection of these grooves and spherical couplings allows ease of mounting while not allowing the system to be overly constrained.
 Figure 3. a) Ball and b) cone kinematic features help achieve high placement accuracy and repeatability in wafer FOUPs. |
The benefits of kinematic coupling are now being extended beyond the loadport-to-FOUP interface to include mounting of loadports, robots, traversers, and aligners to equipment frontend modules (EFEM). By applying these machine-design techniques to automation modules, unit-to-unit variations are minimized and accuracy is precise enough to eliminate the need for reteaching (i.e., recalibrating) the system when a component is replaced. As a result, downtime is significantly reduced when hardware is replaced.
Historically, when a nonkinematically coupled robot was installed, up to 30 min could be spent verifying and adjusting the normality of the vertical z axis to the wafer transfer plane. Since most applications require access to the front (loadport) and backside of the EFEM, robot perpendicularity is necessary. In addition, setup time is involved, requiring the end effector to be precisely leveled, unless kinematic coupling features have been designed into the components. In advanced material handling systems, sensors also can help to detect the magnitude and direction of adjustment needed to calibrate an end effector into a level position. By enabling rapid interchange of automation systems, a field-replaceable unit (FRU) strategy can be developed at the component level to eliminate the need for troubleshooting on the tool. Downtime can simply be minimized by replacing the parts and then restarting fab tools. A cost comparison of troubleshooting failing units on tools vs. a FRU strategy is covered later in this article.
Absolute encoders
An encoder is an electrical mechanical device that can monitor motion or position. A typical encoder uses optical sensors to provide a series of pulses that can be translated into motion, position, or direction [3]. These devices are typically mounted on motor shafts or linear rails to count the number of revolutions or distance traveled. Early examples include resolvers and incremental encoders that have been employed to monitor the position of wafer-handling robots and other axes of motion, such as wafer positioning stages and aligners. While adequate in the past, these monitoring and positioning devices are running into problems in today’s critical applications in fabs and the nanometer era.
Resolvers have been used in industrial equipment for more than 50 years and operate on the principal of resolving the mechanical angle of the motor shaft relative to a reference. Unfortunately, these devices are susceptible to gap changes and electrical noise immunity, rendering them almost useless for demanding equipment applications. Incremental encoders, on the other hand, offer more robustness and resolution than resolvers because they use two out-of-phase sets of signals that indicate direction and one signal that counts the number of revolutions. A limitation of incremental encoders is that they require an external home flag and homing sequence to reference a known starting point from which the counting sequence begins. Position data, or counts, are lost from buffer storage (or external counter) every time power is turned off. During homing steps, motors work through a preprogrammed range of motion in search of the flag that indicates the home position per the home sensor. This homing sequence adds time to system recovery and also endangers the safety of product wafers and equipment because the range of motion (location or condition) is unknown.
Absolute encoders increasingly demonstrate higher levels of performance for advanced equipment applications. Absolute encoders use a unique set of codes for any position of motion, and they “know” their location at any time, even after power-up. This is necessary for safe stopping and starting after system interlock or EMO circuit failure. It also allows for continued system operation after an unexpected interruption. A home position still must be defined; however, there is no need to re-reference a system with a homing routine when power turns off, as is the case with incremental encoders.
To fully leverage the power of absolute encoders, integrated diagnostics capabilities can be added to monitor system health. When used with external sensors, wafer-handling component issues, like loosening belts or thermal expansion, can be detected and corrected before failures occur. Encoders can also now monitor themselves.
Hands-free teaching
A paradox in semiconductor manufacturing is the ongoing need to “teach” material handling systems precisely where to pick and place wafers while attempting to minimize human intervention. More stringent microcontamination requirements inside wafer handling environments make it difficult to introduce external means for calibrating the system in pick-and-place stations. System teaching is traditionally accomplished manually with a teach pendent. This approach can take up to 15 min/station or up to 90 min for a full atmospheric handling system. These barriers, coupled with the need for more accurate and reproducible results, are making automated teaching, or hands-free teaching, a required feature in fabs.
Teaching techniques that embody all or some of the attributes described in “What makes up hands-free teaching” (see p. 36) use hand location methods, autotouch (learning positions from touching objects in workstations), and noncontact sensors. Early teach techniques for large, industrial robotics were applied to the semiconductor industry with some initial success in the 1980s. Robots were moved by hand to desired locations or manually dragged through a path of motion. Although still valid today for some industries, this technique is losing favor in advanced fabs, where human intrusion in the wafer environment can no longer be tolerated due to disruption of operations and risk of microcontamination in chambers.
Microcontamination is also a concern for automated teach routines that involve robots making physical contact with objects to determine location. In these cases, a robot moves until it runs into an obstacle, which causes it to slow down and stop. In another example, the robot drags or “wipes” the end effector along a feature or wall to determine relative location. Neither of these situations is acceptable in advanced process applications that have stringent contamination requirements. In the future, only clean, noncontact sensing technologies will be able to meet and support process node requirements.
Even as automated teaching becomes more mainstream, it will not completely displace teach pendents. Methods to manually teach the system will still be required on an exception basis when human intervention is required. For example, new process and tool requirements might require additional stations to be added to a workspace; a teach pendent could aid in defining new wafer placement locations.
Cost savings
Figure 4 shows potential cost and time savings from kinematic coupling, absolute encoders, and hands-free teaching in wafer handling systems when servicing components in a fab. The cost of downtime for tools in a leading-edge fab could reach hundreds of thousands of dollars. For this model, we are using an average rate of $100,000/hr. This figure eclipses the costs associated with labor ($100/hr) and spare parts, which are included in this overall estimate.
 Figure 4. Potential time and cost savings from downtime reduction improvements. |
In a FRU strategy, fabs will swap out a robot instead of spending hours troubleshooting and repairing the existing system in the tool. This can significantly reduce downtime. Based on Brooks Automation studies, this strategy will incur a $25,000 cost for a new robot, some of which could be recouped if the original robot were refurbished and returned to the fab as a spare. Setup of the new robot with kinematic coupling leaves only the end effector to be leveled, if required. Hands-free teaching with direct drive and absolute encoders drastically reduces the teach time from 90 min to 10 min. The time it takes to mount a robot onto a frame remains about 10-15 min.
Conclusion
Previously offered as individual features of automation components, autoteaching, absolute encoders, and kinematic coupling are now available at the automation system level. By designing these advances into new systems, their benefits can be fully leveraged by equipment OEMs and semiconductor manufacturers. For fabs experiencing 90 down events/year related to tool automation issues, $27 million can be saved annually by installing systems that employ these advanced technologies, based on current estimates.
Acknowledgments
The author would like to thank Anthony Robson, Ulysses Gilchrist, and Bob Caveney in Brooks Automation’s Technology Automation Division for technical contributions to this paper.
References
- Semi E10-0304 standard, “Specification for Definition and Measurement of Equipment Reliability, Availability, and Maintainability (RAM),” 2004.
- A.H. Slocum, Precision Machine Design, SME, 1992.
- T.E. Kissell, Industrial Electronics, Prentice Hall, 1999.
Kurt Greissinger is a product marketing manager at Brooks Automation Inc., 15 Elizabeth Dr., Chelmsford, MA 01824; ph 978/262-7737, e-mail [email protected].
What makes up hands-free teaching
Many different hands-free teaching techniques have been developed, but they generally embody some or all of the following attributes:
- Sensors provide a necessary means of feedback to the system during the automated teach process. They can take many forms including optical, proximity, touch, and motor current, all in either analog or digital versions. Appropriately positioned sensors on and around the robot and workspace can provide meaningful information about the robot’s location with regard to different stations. When this data is coupled with absolute known counts of a robot axis, it can result in even more meaningful data to the system and facilitate faster, more repeatable teaching.
- Aligners are often used to fine-tune the wafer placement to various stations. By leveraging the eccentricity algorithms used for an aligner’s standard function, offset data can be fed back to the robot to correct any wafer placement inaccuracies and provide a finer resolution of teaching to the system, free of human error.
- Teach fixtures are a carryover from manual teaching. Fixtures were invaluable in ensuring proper handoff between robots, aligner chucks, and inside wafer carriers, and continue to find a role in autoteaching routines. Their use on material handling systems is proving more difficult in advanced fabs, which cannot tolerate particle levels resulting from human operators entering minienvironments to install and remove teaching fixtures.