2016 年 82 巻 10 号 p. 901-906
We propose an efficient estimation method of three dimensional profiles of through-silicon vias (TSVs) from one-shot projection images captured with a nano-focus X-ray microscope with a resolution of 0.25 μm aiming at an in-line inspection system. Complicated three dimensional profiles after etching are represented with a simplified model that consists of 8 dimensional parameters. X-ray projection images are simulated with varied parameters determined by extraction of generated uniform random datasets. Machine learning by support vector regression algorithm is applied to form a transfer function to get the geometrical parameters from data of the images. This method gives a repeatability of estimation of < 0.2μm for profiling of TSVs with a targeted size of φ8μm × 90μm depth.