Skip to Main Content
Your organization might have access to this article on the publisher's site. To check, click on this link:http://dx.doi.org/+10.1116/1.584551
Precise registration to substrate fiducial marks is crucial to high density, fine‐line device fabrication. Mark detectability is a function of both the mark contrast with respect to its surrounding substrate and also its interior topography. Since the accuracy and speed of detection of any scheme is related to the signal‐to‐noise ratio (SNR) of the mark image presented to the detection filter, one is compelled to develop higher contrast marks in order to enhance the effective SNR and thereby keep pace with the more stringent registration requirements of today’s processes. Unfortunately, some higher contrast marks, being developed by more advanced processes, are characterized by a rather severe interior topography. Standard edge detectors are hampered by the apparent interior ‘‘edges’’ of such marks. Another complication arises from the pulsed type data acquisition scheme used on the AT&T in‐house EBES4 system that demands high bandwidth detection electronics. This prohibits the reduction of the detection noise power spectral density by low‐pass filtering operations. While this acquisition architecture has the potential for accurate mark location with little electron beam induced mark degradation or errors due to charging, the detection filter must cope with a relatively strong noise energy content. To register to both low contrast and moderate contrast ‘‘rough terrain’’ marks under these circumstances, a two‐stage matched filter has been developed. This variant of the standard matched filter was designed to improve its edge localization characteristics. The approach incorporates computationally intensive signal processing algorithms executed on special purpose hardware. Our detection system consists of a custom digitizer that feeds a VMEbus based array processor. The digitizer is equipped with a large, fast memory buffer so that scan data may - - be processed as a new scan is being acquired. This parallelism results in detection times commensurate with high throughput. Detection accuracy from chrome marks has been investigated using three detection techniques. Optimum ‘‘∥mean∥+3σ’’ overlay errors of ≤0.06 μm were achieved by utilizing the two‐stage filter.