With the development of chip manufacturing technology, the demand for nanoscale detection has also increased dramatically. However, traditional optical, electronic microscopy and X-ray detection technologies have their limitations such as insufficient resolution, causing sample damage, and the poor detection of internal structure, which have narrowed the applications of traditional technologies in nanoscale detection. We establish a new X-ray nano three-dimensional (3D) imaging method on the X-ray nano 3D imaging line station (BL18B) of Shanghai Synchrotron Radiation Facility (SSRF). It can obtain the depth information that the traditional X-ray two-dimensional (2D) detection technology cannot gain, avoid the limitation of X-ray computer tomography (CT) technology on the sample, and reduce the time spent on reconstruction. The technology is applied to chip detection to acuqire the accurate size and depth of the micron-level defects inside the chip, showing the great potential of this method in chip detection.

The binocular stereo vision technology is based on the parallax principle. Two images of the object under test are captured from different perspectives by the imaging device, and the position deviation between the corresponding points in the images is calculated to extract the 3D geometric information of the object. The two projection images of the same object are obtained by two X-rays with an angle of θ, which is equivalent to using the single X-ray irradiation and the rotation of the sample by θ to obtain another projection image. Based on the two images, the depth information of the sample is restored using the parallax principle in binocular stereo vision to realize 3D reconstruction of the sample which is applied to defect detection. The experimental steps include selecting an appropriate stereo imaging perspective according to the shape of the sample and the characteristics of the region of interest, determining the binocular projection angle difference, and preprocessing the binocular image to reduce the interference of background noise and enhance image clarity. The normalized cross-correlation (NCC) coefficient is utilized to calculate the gray similarity between pixel points for the feature matching of binocular projection, the result of which is used to obtain binocular parallax maps. The depth value of the pixel in the projection image is calculated according to the geometric relationship of binocular projection and the depth information reconstruction formula in this paper, so as to complete 3D reconstruction.

The NCC algorithm is used to simulate a spiral wire with continuous changes in the depth direction. The experimental results show that the depths of the two leftmost endpoints of the spiral wire are 73 and 74, which are consistent with the continuity of the endpoints in the 2D top view. The expected maximum depth difference is 100, which is also consistent with the real simulation situation (Fig. 5), proving the effectiveness of the NCC algorithm in depth recovery. The depth distribution curve obtained by the algorithm is also highly consistent with the standard curve (Fig. 6). The normalized standard deviation between the restored depth value and the standard one is only 1.441×10^-2, with minimal depth recovery error, which verifies that the depth information recovered by the NCC algorithm is similar to the real depth and highly accurate. Subsequent 3D reconstruction verification experiments are carried out on the nano-resolution target. The results show that the disparity image with an included angle of 20° has the highest similarity with the standard image with a normalized standard deviation of 8.11×10^-4. Given the recovery rate, the accuracy of the restored disparity image is the highest (Table 1). Lastly, a silicon chip sample is detected, and a 600-nm channel-like structure is observed inside the structure, accompanied by tiny impurities. The overall 3D view shows a rough surface with an irregular shape [Figs. 10(b) and 10(c)]. It proves that this technology has great potential in chip defect detection, thus providing an effective detection method for chip quality control in the future.

Based on the principle of binocular stereo vision, an X-ray nano-resolution stereo imaging method is established at Shanghai Light Source Nano-3D Imaging Line Station. By combining X-ray stereo imaging technology with nano imaging system, we realize the depth estimation and 3D reconstruction of the sample by collecting images from two angles, reducing the experimental time required for 3D characterization and removing the limitations of traditional chip defect detection methods. The line width, length, and depth of the standard sample resolution target are obtained through this method. Simulation and experimental results show that this method can obtain the accurate depth information of the sample. Moreover, this method can accurately detect the chip sample, which provides a new nano-resolution non-destructive solution for chip detection and has a huge application prospect in chip defect detection and nano-stereo imaging. Future research will focus on improving the depth resolution and optimizing the matching algorithm to improve the quality of the reconstructed image. Also, more advanced image processing technologies such as machine learning and deep learning methods will be adopted to improve feature matching and image fusion for more accurate 3D imaging.