Ind. Eng. Chem. Res. 2008, 47, 3907–3911
3907
PROCESS DESIGN AND CONTROL Real-Time End-Point Detection Using Modified Principal Component Analysis for Small Open Area SiO2 Plasma Etching Kyounghoon Han and En Sup Yoon School of Chemical & Biological Engineering, Seoul National UniVersity, Seoul 151-744, Korea
Jaewon Lee and Heeyeop Chae* Department of Chemical Engineering, Sungkyunkwan UniVersity, Suwon 440-746, Gyeonggi-do, Korea
Kwang Hoon Han and Kun Joo Park DMS Co., Ltd., Suwon 445-810, Gyeonggi-do, Korea
Principal component analysis (PCA) was modified for real-time applications and applied to the end-point detection of small open area SiO2 plasma etching. Typically, the end point of plasma etching is determined from a few manually selected wavelengths. Determining the end point of the plasma etching using this approach is quite a challenge when the exposed open area is less than several percent. To increase the sensitivity, information was extracted from the entire spectra of 2755 signals in the range of 200-1100 nm, using a PCA algorithm. In this study, the PCA algorithm was modified to allow real-time applications of end-point detection. The loading vector was determined from the model wafer, and the score vector was determined using the real-time data of the target wafer to reduce the processing time. This algorithm was tested for the small open area of SiO2 etching of a 200 ms sampling period, using the entire optical emission spectra, through a comparison with a defined signal-to-noise ratio. The results were compared with the conventional single wavelength signals of SiF (440.2 nm), CO (482.5 nm), and Si (505.6 nm). The end-point detection of 0.4%-0.8% SiO2 open area was achieved using the suggested algorithm, while the single wavelength showed limitations in the open areas above a few percent. The sensitivity was also increased by a factor of 2.15, compared to the signal-to-noise ratio of the single wavelength method. 1. Introduction Plasma etching is an essential step in defining the microscaleand nanoscale patterns in semiconductor wafer processing.1 Reactive radicals and ions are generated in plasma that reacts with a solid surface, resulting in the removal of selective material. When the target material is removed, overetching is inevitable, because of nonuniformity of the target layers. For this reason, it is essential to detect the end point of etching, to avoid excessive overetching. This event is known as end-point detection (EPD). Severe overetching can damage the underlying layer, which can decrease the yield. The method most commonly used for EPD is to monitor the optical emission trace of the reactive species in plasmas using optical emission spectroscopy (OES).2 The radicals and ions that are present in the plasma can be identified by measuring the intensity of the optical emission signal at specific wavelengths. Most EPD methods using OES focus on identifying a single wavelength that corresponds to a chemical species that shows a pronounced transition at the end point.3–7 When the target layer is cleared by the etching process, the concentration of reaction products from the target layer is reduced and the concentration of products from the underlying layer is increased. This single wavelength method shows a detection limitation when the open area becomes smaller or the signal is too weak. * To whom correspondence should be addressed. Tel.: +82-31-2907342, Fax.: +82-31-290-7272, E-mail:
[email protected].
Furthermore, the EPD call with a single wavelength requires significant system-specific experience of the process engineer. The traditional single wavelength method can detect an EPD with
