The idea of using radio frequency,

LotTrack: RFID-Based Process Control in the Semiconductor Industry A real-time identification and localization system uses RFID and ultrasound sensor technologies to improve tracking visibility for inbound logistics in a wafer fabrication cleanroom. Frédéric Thiesse and Elgar Fleisch University of St. Gallen Markus Dierkes Intellion The idea of using radio frequency, infrared, ultrasound, or combinations of these technologies for indoor location sensing, or localization, has been around for quite some time. 1 Although industry and academia have done a lot of research in this area, commercially available systems are still rare. One reason is that many approaches have technical drawbacks, such as insufficient robustness in harsh environments and limitations on the number of items that can be located simultaneously. Furthermore, few economically feasible application scenarios have emerged to justify the substantial infrastructure and systems integration investments that localization would require. 2 Against this background, we describe the design and implementation of a real-time localization solution that Infineon Technologies uses in its wafer fabrication facility in Villach, Austria. The system combines active RFID, passive RFID, and ultrasound sensors to track plastic wafer boxes and wafer cassettes in the company s chip-manufacturing process. It s been in use for almost two years and has proved to be reliable and efficient in localizing and communicating with a large number of moving physical objects. Custom fabrication requirements Infineon Technologies is one of the largest semiconductor manufacturers in the world, with 36,000 employees and annual net sales totaling 7.19 billion in 2004. Its strategic focus is on separate business units for communications, automotive electronics, and memory products. Like other companies in the industry, Infineon seeks to increase automation in its production logistics so that it can reduce stock and improve the efficiency of its transport processes. While several highly automated transport system solutions exist, they mostly address the needs of large fabrication facilities (fabs) with relatively static processes, limited product range, and high-volume production. In contrast, fabs with stronger customer orientation require more flexible, operator-centric automation approaches to handle the large number of varying production processes and frequent rearrangements of the fab s machinery and other technical equipment. Infineon s Villach facility exemplifies these requirements. Headquarters for the company s Automotive and Industrial business group, the Villach unit develops and produces car components primarily for use in engine management, transmission control, comfort and safety management, and infotainment. The fab operates 24/7 with 2,000 employees and manufactures about 800 different products a total volume of 10 billion chips per year. Because of the enormous variety of processes, the operators transport production lots manually from one of its 600 machines to another, rather than using conveyor belts. For this purpose, lots are kept in plastic boxes, each containing a wafer cassette 1536-1268/06/$20.00 2006 IEEE Published by the IEEE CS and IEEE ComSoc PERVASIVEcomputing 47

Figure 1. (a) A lot box and (b) a wafer cassette holder. (photo courtesy of Entegris) (a) holder for up to 25 silicon wafers (see figure 1). These boxes are stored on metal shelves throughout the cleanroom before and after processing. In this system, an average of about 400 production steps occur per lot before the chips are eventually tested. A central manufacturing execution system controls the actual production procedure in each machine. While the MES knows each and every production step in the machine, the intermediate transport process wasn t at all transparent. Analyses showed that improved transparency of inbound logistics could decrease lead times and increase machine utilization ratios. Furthermore, the planned operation sequence was hard for operators to follow if the next lot for processing wasn t in place. Infineon Technologies decided to start the LotTrack project, equipping the fab with a real-time identification and localization system that would make a lot s entire production process visible and thus controllable. The project s main goals were decreased lead times, complete elimination of handling faults, and a reduction of non-value-adding activities. Moreover, the project was also regarded as one important step toward a paperless fab, which would further reduce particles in the air. (b) Requirements analysis In the first project phase, the project team had to select a suitable identification and localization technology. Choosing the best technology requires precise knowledge of the desired solution s requirements. In the case of lot box tracking in a chip fab s cleanroom, the following requirements and constraints had to be taken into account: The total number of wafer boxes that the tracking system must localize in the fab is greater than 1,000. Especially in places with many shelves, the number of wafer boxes can be greater than 100 boxes per 6 square meters. To control the process in real time, the time lag between physical movements in the real world and position updates in the system should be significantly less than one minute. The system s location accuracy should be less than 1 meter to make the search for a specific lot box as efficient as possible. Walls, floors, and ceilings in the fab are made of metal that can complicate RF communications and generate electromagnetic reflections. A direct line of sight between the box and a reader device on a wall or on the ceiling isn t always present the technology should be able to tolerate varying box positions and orientations on the shelf. Energy consumption of the tracking devices that are attached to a lot box should allow for a battery lifespan of at least two years. In case of system failures or other critical errors, a simple shutdown-andsystem-restore procedure for the whole system should take no more than five minutes. Furthermore, during the requirements analysis phase, Infineon decided that the system should not only collect data from the fab for planning purposes but also enable direct communication with operators to ensure more control over cleanroom operations. Thus, the system was to completely replace traditional paperbased working plans with, for example, text messages on an LED screen. For this purpose, it had to be able to transfer information to operators within 1.5 seconds, on average, in parallel with its localization functionality. While localization of lot boxes primarily aims to minimize search times, the need to decrease the risk of processing errors at each operation also requires identification on the level of single wafer cassettes. Therefore, the system needed a way to identify cassettes when an operator places them into a machine. By this means, the system could detect accidentally missing production steps and prevent repeated steps from occurring. Localization concept The limitations in various existing systems (see the sidebar Related Work in Location Systems ) made clear that only a combination of RF and ultrasound technologies could resolve the trade-off between high location accuracy and high communication rate together with a large number of wafer boxes. The design 48 PERVASIVEcomputing www.computer.org/pervasive

Related Work in Location Systems Localization is usually associated with GPS, which can achieve localization accuracy up to about 3 meters. Unfortunately, we can t use this classic location technology inside buildings because the radio signals sent by GPS satellites are too weak to penetrate walls. 1 However, previous work has demonstrated successful indoor localization using infrared, ultrasound, or radio frequency technology: The Active Badge Location System is an example of an IR-based solution. 2 The system was originally developed for tracking people in office buildings. In this scenario, a badge worn by a person sends a unique IR signal every 10 seconds. Sensors placed at known positions pick up the signals and relay them to a central location manager. The line-of-sight requirement and short-range signal transmission are two major limitations of this approach. Cricket is an ultrasound-based positioning system that uses active beacons to send periodic chirps on an RF channel. 3 Each beacon also sends an ultrasonic signal at the same time as the RF message. Passive listeners that are attached to a host device, such as a laptop PC or a mobile handheld, receive both signals. Thus, listeners can estimate their distance to the beacon from the time difference between the arrival of the RF and the ultrasound signal. Active Bat uses an approach similar to Cricket s but instead equips mobile devices with emitters and installs listeners on the ceiling. 4 Active RFID transponders are used by Landmarc, which requires signal strength information from each tag to readers. 5 Readers detect RFID-equipped objects and forward all collected data to a central lateration algorithm. Furthermore, the system uses stationary reference tags to improve its location accuracy. Other RF-based alternatives to RFID are the use of signal strength measurements from an existing wireless LAN infrastructure as implemented in the RADAR system 6 or time-of-flight calculations of ultrawideband signals. 7 Positioning and tracking systems are fundamentally different. As the name suggests, LotTrack was meant to be an object-tracking application that is, an application in which a central system observes objects. This contrasts with a positioning application, in which the physical object determines its own location. 8 In either case, the simplest way to generate the actual location information is by associating the object with the controller device for example, an RFID reader that has detected the object and whose position is known. If an application requires more finely grained information, it must use lateration that works with distance measurements between the physical object and one or more controller devices. 9 Each of the related technologies we ve mentioned has its own strengths and weaknesses regarding location accuracy, flexibility, costs, and so on. In the LotTrack project, the design team rejected a solution based on passive RFID at the beginning because of the metallic environment and the limited read range of passive transponders. Nor does an active RFID infrastructure work properly in metallic environments: While the achievable communication rate for data transfers to a transponder is satisfactory, reflections and overlaying signals result in highly volatile distance measurements. An IR-based solution was rejected because it requires line of sight for object identification. Furthermore, location changes of shelves or production equipment would have led to a costly readjustment of installed IR sensors. Ultrasound is very accurate in distance measurement but, again, would have been unsuitable for tracking numerous objects simultaneously. A system like the Active Bat, for instance, doesn t perform well if many objects are located at the same place and are sending lots of ultrasound signals to the same sensor. Furthermore, ultrasound can cause problems when factory noise in the relevant frequency band reaches a sufficiently high level, especially when the distance between sender and object exceeds 10 meters. REFERENCES 1. G. Borriello et al., Delivering Real-World Ubiquitous Location Systems, Comm. ACM, vol. 48, no. 3, 2005, pp. 36 41. 2. R. Want et al., The Active Badge Location System, ACM Trans. Information Systems, vol. 10. no. 1, 1992, pp. 91 102. 3. A. Smith et al., Tracking Moving Devices with the Cricket Location System, Proc. 2nd Int l Conf. Mobile Systems, Applications, and Services (MobiSys 2004), ACM Press, 2004, pp. 190 202. 4. M. Addlesee et al., Implementing a Sentient Computing System, Computer, vol. 34, no. 8, 2001, pp. 50 56. 5. L.M. Ni et al., Landmarc: Indoor Location Sensing Using Active RFID, Wireless Networks, vol. 10, no. 1, 2004, pp. 701 710. 6. P. Bahl and V.N. Padmanabhan, RADAR: An In-Building RF-Based User Location and Tracking System, Proc. IEEE Infocom 2000, IEEE CS Press, 2002, pp. 775 784. 7. R.J. Fontana and S.J. Gunderson, Ultra-Wideband Precision Asset Location System, Proc. IEEE Conf. Ultra Wideband Systems and Technologies (UWBST 2002), IEEE CS Press, 2002, pp. 147 150. 8. S. Helal et al., Enabling Location-Aware Pervasive Computing Applications for the Elderly, Proc. 1st IEEE Int l Conf. Pervasive Computing and Communications (PerCom03), IEEE CS Press, 2003, pp. 531 538. 9. J. Indulska and P. Sutton, Location Management in Pervasive Systems, Proc. Australasian Information Security Workshop Conf. ACSW Frontiers, vol. 21, Australian Computer Soc., 2003, pp. 143 151. team selected active RFID as the desired solution s foundation mainly because of the battery requirements and communication performance. As the first tests demonstrated, identification in itself worked fine whereas localization based on RF field strength failed to a large degree because of insoluble problems with reflections. JANUARY MARCH 2006 PERVASIVEcomputing 49

Corridor B p 4 Figure 2. Calculating lot box locations using the localization algorithm. Corridor A p 1 u 4 l B ID and a distance value is enough information to find any place in the fab. Besides, it s much easier for fab operators to orient themselves using this solution than using 3D coordinates. The resulting algorithm works as follows: u 1 l A u 2 u 3 For this reason, LotTrack uses ultrasound for fine-grained localization. Ultrasound emitters on the ceiling periodically send a signal that is received by active RFID transponders, which are equipped with ultrasound sensors. The tags calculate the time-of-flight for all received signals and store these values in their read/write memory. RFID controllers read this data from the tags and transfer it to a central server that generates location information for all objects in the system. To make the whole system work, every component must be fully synchronized. For this purpose, controllers provide tags with a clock signal ensuring that they all operate on the same time basis. The localization procedure is based on a 10-second cycle that is divided into three phases: 1. Transponders listen to a predefined pattern of signals from ultrasound emitters in their environment. Since the transponder is in sync with controllers, it knows exactly when the signal was sent and thus calculates its distance from the emitter. 2. The controller device uses the RFID p 2 p 3 protocol to scan its environment for tags and reads their collected distance measurements. 3. In the remaining time slot, tags and controllers communicate as needed. In this solution, RFID serves three different purposes: The RFID protocol provides an efficient way to identify even a large number of tags in the environment. The measured RFID signal strength and the coarse-grained distance information that the system can derive from it serve as an additional plausibility check for the localization algorithm s results. RFID enables reading distance measurements and all other tag access operations. To minimize the amount of ultrasound emitters and to simplify the lateration algorithm, the design team regarded bilateration along a line as sufficient for localization in a chip fab. The fab s layout contains no wide rooms but only interconnected corridors, so a corridor 1. Distance measurements define circles around the coordinates of a given ultrasound emitter u i. The algorithm calculates intersection points p j for all possible combinations of two circles. In the example in figure 2, we get six partially duplicate intersection points from four different distance measurements. 2. The algorithm then performs plausibility checks and eliminates all evidently incorrect positions. In our example, p 4 is eliminated because it s located behind a wall. 3. LotTrack repeatedly calculates the average value of all positions and eliminates the ones whose distance from this mean exceeds a configurable threshold. Eventually, Lot- Track considers the average of all remaining positions as the object s actual position. In our example, it selects only p 1 because the object seems to be at rest and all circles intersect at the same location. 4. LotTrack projects the position onto the centerline of all corridors in which the ultrasound emitters u i are located. Thus, a 2D position is transformed into one or more 1D locations l k. In our example, we get l A on corridor A s centerline and l B on corridor B s centerline. Because l B is not within corridor B s logical borders, which are predefined in the LotTrack database, the algorithm selects l A as the final result. One specific problem that results from this simplified localization technique is the calculation of positions at the corner 50 PERVASIVEcomputing www.computer.org/pervasive

Figure 3. DisTag features. of T-crossings. As our example demonstrates, tags sometimes receive ultrasonic signals from emitters in two corridors. These signals are processed and result in conflicting positions some in corridor A, others in corridor B. Unfortunately, the set of signals that the tag receives isn t always the same within every localization cycle even if the box is at rest. Sometimes the tag receives a few more signals from one or the other corridor, which eventually leads the localization algorithm to conflicting decisions. The effect is an apparent blinking that is, the box seems to jump back and forth from time to time. LotTrack prevents this behavior by comparing the algorithm s sets of p-values from the current and previous localization cycles. Hardware components Infineon s partner in this project was Intellion, a Swiss-based automation solution provider. Intellion was responsible for project management, systems design, development, and implementation of the LotTrack system. The foundation for the LotTrack-specific localization tags was a long-range technology for active RFID transponders that could operate in the UHF band and reach a maximum read range of up to 100 meters. The newly developed localization transponder, called the DisTag, also features an ultrasound sensor, a two-colored LED, a mechanical flipdot, a no-power display, and four keys for user interaction. So the DisTag is not only an active ID tag and a tracking device but also the system s interface to the cleanroom operators (see figure 3). The display replaces the traditional paper checklists and provides information on the lot ID number and the next production step. The ultrasound sensor and display contribute LED No-power display Keys Ultrasound sensor Flipdot most of the DisTag s energy consumption rate increase. Nevertheless, depending on environmental conditions, the Dis- Tag s lifespan is about two years before it requires a battery change. Although its many special features make the embedded device significantly more expensive compared to a standard long-range transponder, the price of one single Dis- Tag is still relatively low compared to the value of a wafer carrier s contents. The second system component is the controller device. The RFID controller is connected to four RF antennas and a variable number of ultrasound emitters. To reduce installation activities in the cleanroom to a minimum, one RF antenna and three ultrasound emitters were integrated into each standardized antenna module with preconfigured cabling that can easily be mounted on the ceiling. The design team chose the distance between ultrasound emitters to ensure that a DisTag always receives at least three signals from different emitters for bilateration. A central server controls the system, managing network connections to every operational LotTrack controller (see figure 4). The server receives all the collected controller data that serves as a basis for the localization algorithm. The server is responsible not only for bilateration calculations but also for plausibility checks and other filtering procedures that guarantee high data quality. The Lot- Track server transfers the results to the messaging middleware that supports a publish/subscribe mechanism. The middleware is also necessary to submit commands from client systems to LotTrack for example, read/write operations on a DisTag s display. LotTrack s clients include the central MES and the Fab- Viewer, a visualization tool that gives a graphic overview of all the fab s wafer boxes. Furthermore, LotTrack continuously records management indicators to monitor the actual status of readers, antennas, and DisTags that can be accessed from an administration console. User functionality Wafer production processes are organized according to dispatch lists that contain all waiting lots that fab operators must process in a specific type of machine ordered by priority. Operators can select the lot they wish to process next directly on screen. The DisTag s flipdot indicates when an operator has reserved a lot box. LotTrack provides JANUARY MARCH 2006 PERVASIVEcomputing 51

Controller 1 TCP connection to every operational RFID reader Data Data collection from controller #1 Database for archiving Controller 2 Controller n Controller selection based on signal strength Data collection from controller 2 Data collection from controller n DisTag communication Message buffer Every 10 seconds Location calculation Periodic database updates In-memory tag repository Position updates XML messaging middleware Publication FabViewer Manufacturing Execution System (MES) Tag access operations Figure 4. The LotTrack system architecture. information on the box s position, such as Corridor A, 2.7 m. The operator can also make the LED blink, making it easy to find the box. Finally, the system detects if a box is currently at rest or moving, which helps an operator decide whether to wait for a box to arrive or actively search for it. The operator then picks up the wafer box and brings it to a machine. The information on the display indicates the machine to which the system has assigned that lot. The operator opens the box, puts the wafer cassette into the machine, and starts the processing. To make sure that the right lot is processed, Infineon chose to equip wafer cassettes with standard passive RFID tags as a complement to the active DisTags on the box. A reader device in the machine identifies these passive tags. In this way, the system can detect duplicate and missed process steps before they occur. When the processing has finished, the operator takes the cassette out of the machine and puts it back into the box. The system automatically moves the lot to another dispatch list and updates the DisTag display. Then the operator puts the box on a shelf where it waits to be transported to another machine. Lessons learned Over the course of almost two years operations, LotTrack has proved its reliability and efficiency. The following results indicate the system s capacities: Performance. The installation in Villach comprises about 100 controller devices and more than 1,000 DisTags. The system is installed on a standard two-cpu server and receives about 3 billion distance measurement messages from DisTags per day. Processing this huge amount of data leads to 270 million positions. Because LotTrack only publishes position changes that exceed a configurable threshold, the net total of positions delivered to its clients is 500,000 per day. Location accuracy. The first system prototype provided location information within an accuracy of 5 cm under laboratory conditions. These results are consistent with the characteristics of other ultrasound location systems. While the production system still meets Infineon s requirements, its accuracy reaches only 30 cm. This is due to the dynamic production environment with several moving people and objects that cause ultrasonic reflections and slightly incorrect distance measurements that our plausibility checks can t recognize. Other factors are background noise (some machines in the fab generate noise in the ultrasound spectrum) and discrepancies between reference coordinates in the LotTrack database and the real positions of ultrasound emitters, but these influences are virtually negligible. Time lag. The time lag between physical movement and the corresponding position message is no more than 30 seconds. Actually, the system performs even better and generates position updates after 10 seconds. After the first tests with operators, we ve nev- 52 PERVASIVEcomputing www.computer.org/pervasive

the AUTHORS ertheless chosen to merge location information from three consecutive localization cycles to make better decisions about whether an object is at rest or moving. This takes more time for new positions to become visible, but the results are also more convenient for the user. Communication operations. Whenever the server must transfer data to a DisTag, it starts a thread that tries to perform a tag access operation via the closest RF antenna. In 80 percent of cases, the operation succeeds at first try. In the other 20 percent, which fail because of moving boxes, the operation always succeeds after two retries. Frédéric Thiesse is associate director of the Auto-ID Lab at the University of St. Gallen and project manager of the M-Lab, a joint research project of the University of St. Gallen and the ETH (Swiss Federal Institute of Technology) Zürich. His research interests include RFID integration with corporate information systems, RFID s role in operations management, and the design of pervasive products and services. Thiesse received his PhD in business administration from the University of St. Gallen. Contact him at the Inst. for Technology Management, Univ. of St. Gallen, Dufourstrasse 40a, 9000 St. Gallen, Switzerland; frederic.thiesse@unisg.ch. Markus Dierkes is chief technology officer of Intellion, a spin-off of the University of St. Gallen that specializes in RFID and location-based automation solutions for manufacturing and logistics. His activities focus on ubiquitous computing s potential for transforming automation concepts and business models. He received his PhD in business administration from the University of St. Gallen. Contact him at Intellion, Schuppistrasse 8, 9016 St. Gallen, Switzerland; markus.dierkes@intellion.com. Elgar Fleisch is a professor of technology management at the University of St. Gallen, a professor of information management at the ETH Zurich, and cochair of the international Auto-ID Lab Network. His research interests include ubiquitous computing s impact on products, services, business processes, business models, and users. He received his PhD in business administration from the Vienna University of Economics and Business Administration. Contact him at the Inst. for Technology Management, Univ. of St. Gallen, Dufourstrasse 40a, 9000 St. Gallen, Switzerland; elgar.fleisch@unisg.ch. All in all, location visibility of fab lots has increased from 65 to nearly 100 percent in covered areas. Formerly manual transport processes, such as moving lots, have become completely transparent. On the basis of these experiences, Infineon extended the Villach installation to a second, larger fab in Regensburg, Germany. Other installations are in the planning phase. The solution s process impact includes fewer handling errors and reduced lead times, which result in lower manufacturing and inventory costs. Furthermore, the fast and rich communication with operators enhanced the flexibility of transport and handling processes. Last but not least, LotTrack is an important part of the paperless fab in that the DisTag displays provide needed information and so replace the traditional paper lot traveler. To integrate the enriched information quality of accurate, reliable lot positions with the existing manufacturing system required significant training and change management. Operators had to learn how to use LotTrack in their working environment. Optimizing logistics processes and adjusting the entire production automation system took a lot of time. Nevertheless, the full potential of Lot- Track and similar systems is still waiting to be leveraged. Over the long term, these systems can lead to a more strategic use of production flexibility as well as enabling customer-driven production systems and mass customization concepts. REFERENCES 1. J. Hightower and G. Borriello, Location Systems for Ubiquitous Computing, Computer, vol. 34, no. 8, 2001, pp. 57 66. 2. M. Hazas, J. Scott, and J. Krumm, Location-Aware Computing Comes of Age, Computer, vol. 37, no. 2, 2004, pp. 95 97. For more information on this or any other computing topic, please visit our Digital Library at www. computer.org/publications/dlib. IEEE Pervasive Computing IEEE Pervasive Computing delivers the latest developments in pervasive, mobile, and ubiquitous computing. With content that s accessible and useful today, it acts as a catalyst for realizing the vision of pervasive (or ubiquitous) computing Mark Weiser described more than a decade ago the creation of environments saturated with computing and wireless communication yet gracefully integrated with human users. SUBSCRIBE NOW! www.computer.org/pervasive/subscribe.htm JANUARY MARCH 2006 PERVASIVEcomputing 53