The Anelva 1015 sputtering system from Analysis Electronics Vacuum (Anelva; Tokyo, Japan) is used by companies such as Advanced Micro Devices (AMD; Austin, TX) for automatic wafer processing. However, the 15-year-old system, which sputter-deposits a variety of metals onto wafers, was not designed to read wafer scribes. This limitation posed a potential miss-processing risk for AMD. To address this risk and improve the traceability of the 6-in. wafers processed in the system, AMD retrofitted the 1015 sputtering system with wafer-reading sensor technology.
"We wanted to automatically read the lot numbers on the wafers to ensure they matched the lot number on the run card and on the box in which the wafers were transported to the process equipment," explains Mark Bradford, AMD senior engineer. "While they almost always match up, someone could put the wrong wafers in the box or mix up the run cards. If the wrong metal were deposited on the wafers, they would be unusable. By adding wafer reading to the 1015 system, the tool can be stopped if the wrong wafers appear."
To perform wafer readings, Bradford selected a Cognex Corp. (Natick, MA) In-Sight 1700 sensor. This self-contained sensor combines optical-character-recognition (OCR) software and on-board image-processing hardware in a hand-held housing. "Sometimes there will be edge-bead removal lines that run right through the middle of the scribe. This adds noise to the image and can reduce the reliability of the vision system. A poor-quality scribe and other processes can also change the appearance of a wafer scribe and make it virtually unreadable to the naked eye and unrecognizable to a vision system," says Bradford. To overcome these problems, the sensor incorporates geometric pattern-matching software tools along with semi-checksum methods to maximize its reading capabilities. In this way, the vision system can deal with a range of variables that could degrade or change a wafer scribe's appearance.
Before wafers are transported by conveyor belt into the load lock of the sputtering system, they are positioned into a flat-finding station, where a wafer is mechanically spun and a sensor is used to locate the flat edge of the wafer. Mounted directly above the flat-finding station and 10 mm from the wafer surface, the In-Sight 1700 sensor is triggered immediately after alignment. In less than a second, the sensor illuminates the wafer using its built-in red-light-emitting diodes, captures an image of the scribe, and analyzes the image.
An automatic image-tuning capability applies the correct lighting and image-processing tools for each specific wafer. "With so much image variability present from wafer to wafer, the same illumination settings and image-processing tools cannot be used; different lighting and tools are required for different wafers," says Bradford. Instead of relying upon the technician, the sensor dynamically balances illumination and image processing based on the visual characteristics of the scribe being read, eliminating the need for setting up filters and other image preprocessing tools to handle image variations.
Just prior to the start of a batch of wafers, the expected lot number for this batch is transmitted over an Ethernet connection to the sensor from a custom Visual Basic program running on a PC. Bradford programmed the sensor to compare the expected lot number and the actual lot number read. If these lot numbers do not agree, wafer processing is stopped.
In addition to verifying that the correct lot is being run, the sensor also tracks the order of the wafers as they are processed through the production equipment. This processing order is used to help determine root-cause problems with the sputtering equipment if they occur. The fabrication line typically runs 48-wafer lots, with 24 wafers divided into left and right cassettes. Currently, the wafers are read one cassette at a time on a separate, stand-alone optical-character-recognition system.
The cassettes are read "left cassette first" and then processed through the deposition tool in that same order. In this way, the processing order through the tool can be deduced from the order of the wafers as read by the OCR system. "But there is always some uncertainty about whether the cassettes are indeed read and processed in the same order," says Bradford. With the wafer reader installed on the tool, the order of wafers can be automatically read and the information sent via Ethernet to a data file, thus eliminating the need for the standalone OCR system. Bradford used the Anelva system as a testbed and sees the potential for the In-Sight 1700 sensor to be incorporated across the whole wafer-fabrication line to further reduce the risk of miss-processing and provide improved wafer traceability.