In semiconductor manufacturing, the push for greater efficiency and higher yield of silicon semiconductor material is never-ending. As circuit features shrink in size and global price competition intensifies, wafer processes push the physical and operational limits of equipment manufacturers. One result is increasingly narrow tolerances for incoming wafer physical and electrical parameters in delicate process steps, such as mask and etch.
Measuring the bow, warp, and TTV of a wafer requires performing a full dimensional measurement scan of the wafer top and bottom surfaces. This is not only technically challenging, but represents a significant increase in process time compared to simple, single-point measurements that were previously sufficient. For these measurements to be useful, they must match industry-standard benchtop instruments, which have the luxury of taking a great deal of time to ensure measurements are precise. Gigamat’s challenge was to automatically sort wafers from cassettes using full-scan measurements at high throughput rates, with industry-standard accuracy and repeatability.
The measurement process is comprised of two steps: wafer alignment and wafer measurement. Wafer alignment identifies the location and orientation of the wafer relative to a vacuum chuck on which it is held, and repositions the wafer exactly on the chuck center and with its primary fiducial precisely oriented. The second functional step in the measurement process is the performance of the full wafer scan. This step involves acquiring top and bottom distance measurements from many points across the surface of the wafer and performing analysis on them to derive results.
Wafer alignment was performed using three axes of motion and a linescan camera. A wafer was aligned by rotating it in the field of view of the camera. By synchronizing camera scans with chuck rotation, a 6-megapixel image of the wafer edge was composed in a single revolution, which took about one second. Because camera scans were synchronized with chuck position, they were independent of chuck velocity and could be acquired during chuck acceleration and deceleration ramps to save time. The wafer center, flat, and other features were identified from image data using LabVIEW vision, math, and advanced analysis tools. The wafer was then rotated and indexed in two short moves to bring it into perfect alignment for the measurement station.
A full measurement scan was performed by gripping a wafer from beneath with a rotational chuck and spinning it between top and bottom probes, which measured the distance to the wafer surface with a resolution of <0.0001 mm. Accurate and repeatable wafer measurements require high measurement density, and throughput requires high data acquisition rates. Measurements must be associated with the position on the wafer from which they were acquired. The solution was to synchronize the NI PCI-6115 four-channel simultaneous acquisition board with the measurement chuck position to acquire many sets of position-related measurements per chuck revolution. Controlling the position of a servo translational carriage on which the chuck was mounted as data was acquired allowed for full surface scanning. They used NI motion-based contoured moves to create seamless combinations of circular and spiral trajectories employing sometimes one and at other times two axes of motion for optimal throughput and measurement integrity. They then used LabVIEW advanced analysis and math tools extensively to calculate measurement results from the more than many thousands of measurement sets acquired from each wafer.
NI LabVIEW running on a PC was the key to integrating all of the high-performance technologies required to make this project a success. Combining the hardware synchronization of PCI boards controlling eight NI motion axes with two NI DAQ boards and one vision board, the inherent multi-tasking and re-entrant execution capabilities and DAQmx task, timing, and triggering programming simplicity in LabVIEW gave engineers an ideal platform to rapidly implement, test, and validate multiple iterations of process code.
This article was written by Nipun Mathur, Motion Control Product Marketing Manager at National Instruments in Austin, TX. For more information, visit http://info.hotims.com/10974-327.