For some time, various techniques have been available for detecting the endpoint of chemical mechanical polish (CMP) processes for the removal of metal or dielectric layer thickness. Commercially available systems employ techniques based on motor current, infrared (IR), and various interferometer-based methods. Attempts were made to implement several such commercial systems and they failed for a variety of reasons, ranging from unreliable hardware components to inadequate signal processing features. To provide a reliable, high-performance endpoint detection system, a new endpoint system was developed for a SpeedFam Auriga polisher. Two additional requirements were added in that the resulting system should need essentially no operator intervention and should be able to detect endpoints for any of the several metal polishing processes. The system developed for this project included data monitoring and control capabilities with digital signal processing, algorithm selection for endpoint detection, and fault detection. The carrier speed controller output signal (DAC) was processed through a digital filter to provide smooth curves that had a characteristic suitable for endpoint detection and fault detection. Any disturbances and noise not removed by the digital filtering would have to be accounted for in the endpoint algorithm and avoided as a possible source of detection confusion. Characteristics of the filtered DAC curves varied depending on the layers being polished and the consumables used (including slurries). The SpeedFam tools for which the endpoint system was developed were capable of polishing several wafers simultaneously. Thus it was necessary that the new system would detect either a batch endpoint for all wafers being polished, or control the downforce on individual carriers. The new system was designed to use the output of a carrier speed controller or an JR signal to detect endpoints, depending on the specific process. It was also possible to analyze the filtered carrier DAC signals for indications of faults related to carrier spindle condition and slurry quality. Temperature variations of the polishing pad were measured with an JR detector and found to indicate progress of the polishing process. When a single JR detector was used for measurement of a spot on a rotating pad for a tool with multiple carriers, the detected thermal signal indicated the average polishing progress for all wafers. This project illustrates the use of digital filter analysis to isolate key process characteristics in measurement signals while rejecting signal noise and process variations due to consumable and wafer properties.
|