Low-coherence scanning interferometry is an effective method used for measuring characteristic parameters of microstructures with the advantage of non-destruction. However, a large range of vertical scanning is required for the sampling points at the bottom of the high-aspect-ratio structure,which causes the offset of light intensity and reduces the contrast of coherence signals. Besides, the aliasing coherence signals induced by the complex diffraction effect appear at the step edge and two main envelopes can be extracted. All these phenomena obstruct the locating of the coherence peak and then mislead the height measurement. Algorithmic processing seems to be an effective method to solve the above problems including suppressing the component of offset and demodulating the aliasing coherence signals. Due to the central shielding effect of Mirau-type objectives, the effective signals returned from the sampling points at the bottom of the high-aspect-ratio structure can be further reduced. Combined with the consideration of the magnification in the measurement of the structure with narrow linewidth, Linnik-type objectives are selected in the interference system. In this paper, the complete ensemble empirical mode decomposition with adaptive noise is used to decompose the original signals into a series of intrinsic mode functions. This algorithm can decompose the original signals into different frequency components without prior knowledge. By extracting the intrinsic mode functions with high-frequency and then replacing the original signals, the offset with low-frequency is filtered out and the contrast of coherence signals is further improved. The complex diffraction effect at the edge of the high-aspect-ratio structure results in the coherence signals containing two sets of the envelopes, among which the abnormal envelope is induced by the additional coherence signals. These additional coherence signals correspond to the interference fringes appearing near the upper edges and extending horizontally into the air. However, their contrast is even higher than that of the corresponding interference fringes appearing at the lower surface of the structure. Thus,the abnormal envelope corresponding to the upper surface has a higher amplitude,which obstructs the coherence peak position of the effective envelope. A more complicated condition is that the sampling points far away from the step edge has normal coherence signals containing the single envelope. Therefore, the received coherence signals in the full field of view are mixed up and not applicable to one processing method. What is necessary is that the current position of each sampling point must be determined. The envelope corresponding to the correct surface of the step structure can be used for coherence peak locating and height measurement. For the amplitude of the coherence signals,the shielding effect of the high-aspect-ratio structure can lead to a small result, relatively. This can be considered as the intensity information which helps to distinguish the lower surface of the structure. Besides,there have been obvious discrepancies between the positions calculated by the centroid method for the envelopes extracted from the coherence signals. This is another way to distinguish the different surfaces and can be regarded as contrast information. In this paper, binarization processing is realized by combining the contrast and intensity information extracted from the coherence signals. The purpose of combining these two kinds of information is to further highlight the discrepancy between the upper and lower surfaces. After identifying the upper and lower surfaces of the high-aspect-ratio structure, the envelope corresponding to the current position of the sampling point is selected as the effective envelope for locating the coherence peak and height measurement. A high-aspect-ratio groove with a depth of 101.77 mu m and a line width of 10.97 mu m is selected to be the measurement sample. The structural parameters of this sample have been certified by the China Metrology Institute and a test report has been issued. Using the algorithm proposed in this paper to conduct ten repeatability measurements, the mean of the height measurement results can be calculated as 101.093 mu m and the relative error is 0.67%. These two parameters demonstrate the accuracy of the algorithm. Besides, the standard deviation is 0.316 mu m which illustrates the robustness and stability of the algorithm. Due to the general trend toward miniaturization, this non-destructive metrology offers significant advantages in height measurement of high-aspect-ratio structures, without introducing any physical upgrade of the instrument.
