Using singular value decomposition to study near electromagnetic field of the 3D through silicon via array with high aspect ratio

Abstract

Inspecting the structure of the through silicon via (TSV) with high aspect ratio is important because they are used for 3D IC stacking. In reflectometry, simulation of near field data for TSV hole arrays is used to investigate reflection spectrum for TSV with different geometry parameters such as depth and top critical dimension. We investigate simulation results of electromagnetic field data for different TSV array using the finite-difference time-domain (FDTD) method. Near field simulation data are stored as n by n complex matrices, where n represent the number of simulation region grid points. The matrices are large in dimension, and it is necessary to compress a huge data set by looking for the dominant singular value terms to keep the information as much as possible. We find that the singular value terms shrink fast in the first few terms. It is shown that after using singular value decomposition to compress near field data, the far field reflectivity spectrum is still close to the accurate results. We propose to use data after singular value decomposition for data analysis to investigate the TSV parameters mapping to the near field data.