Accuracy issues in chemical and dimensional metrology in the SEM and TEM

Abstract

Listen

Translate article

The scanning electron microscope (SEM) and transmission electron microscope (TEM) are very powerful tools for the nanoscale characterization of materials. To make the best use of these capabilities, it is necessary to understand the nature of the uncertainties in the measurement process and to allow for errors during the data capture and data analysis stages of the measurement. By using replicated measurements and statistical analyses, it is possible to address repeatability and reproducibility issues and to assign a precision to the measurements. In general it is much harder to address accuracy issues in the experiment because some of the errors are systematic and non-statistical in nature and in almost all cases the true value of the measurand is unknown. In this work computer simulations are used to understand systematic errors and accuracy concerns in dimensional and chemical metrology in both the SEM and TEM. Multislice high-resolution TEM (HRTEM) simulations are used to produce synthetic cross-section micrographs of SiO2 gate dielectrics on Si substrates. These synthetic images are processed in the same manner as experimental images to extract a film thickness value from the micrographs. The absolute error in the dielectric film thickness measurement is determined by comparing this ‘measured’ value with the known, true thickness intrinsic to the sample's structural model used for simulation. This process allows the accuracy of the measurement process to be assessed for a given set of microscope, sample, environmental, data capture and data processing parameters. Statistical design of experiment methods is used to screen the influence of variables including beam tilt, along-beam thickness, dielectric thickness, defocus, astigmatism and vibration on the accuracy. The most important main effects are beam tilt, defocus and vibration, and the most significant two-term interactions are (beam tilt–defocus) and (defocus–vibration). Three-dimensional Monte Carlo simulations in Java and Jython are used to simulate the electron transport and x-ray generation of energy-dispersive x-ray (XEDS) experiments in the SEM. This code permits the simulation of synthetic hyperspectral XEDS datasets from complex, chemically heterogeneous nanostructures. Analysis of these datasets reveals the effects of self-absorption, thickness and beam broadening due to multiple scattering on the accuracy of chemical measurements in the SEM and TEM. A synthetic XEDS datacube from a four-phase Raney nickel sample is presented. While the effects mentioned above degrade both TEM-based and SEM-based measurements, their magnitude is larger and more easily visualized in the SEM because of lower beam energies and thicker samples.