Research on Digital Filters for Si Wafer Surface Profile Measurement - Design of Filters by Total Variation
The underlying data form of a wafer is a matrix of length (or height) measurements. In the presence of noise, evaluation parameters are normally biased. The expectation value such as peak-to-valley and GBIR (global backside ideal range) is systematically larger than the “true” value. Correction and compensation need a large population of measurements to analytically estimate both bias and the uncertainty. In this study, approach to obtain the true value is to extract a “true” profile by filtering noise from the measured data. In previous paper, the digital filter with wavelet transformation (WT) is proposed and efficiency to remove the noise, however, the method is introduced the pseudo-Gibbs effect. Then, we propose the digital filter with new algorithm of total variation (TV). In this paper, the new algorithm of TV is proposed and the digital filter by new TV indicate that data is filtered without the pseudo-Gibbs effect. The digital filters by WT and new TV are applied on the sample data of actual measurement system to investigate their performance of noise reduction.