Accurate characterization of Mo/Si multilayer film thickness is important for process iteration and analysis. As one of the visualization methods, transmission electron microscopy (TEM) can characterize the thickness of nanofilm deposited on a single crystal Si wafer. It can be calibrated internally through the Si substrate lattice parameters, which is very accurate. However, if we do not pay attention to the crystal orientation of the Si substrate during TEM characterization or we use amorphous substrate materials such as fused quartz, it is difficult to ensure that the cross-section of the sample is exactly perpendicular to the electron beam. As a result, the two-dimensional projection imaging of three-dimensional samples produces artifacts, resulting in unknown measurement errors. Therefore, it is of great significance to study the influence of sample tilting angle on the TEM characterization of nanofilms.

Mo/Si multilayer films are deposited by pulsed direct current sputtering. Cross-section samples for TEM characterization are prepared by ion milling. TEM images and high-resolution TEM images of the multilayer films are obtained by TEM. The TEM cross-section samples are tilted in α and β directions by a double tilting holder. Combined with the profile curves of the images, we obtain the thickness of the multilayer film at different tilting angles, the roughness of the interface, and the thickness of the Mo and Si layers in a single period.

As the Mo/Si multilayer film sample tilting in the α direction, the thickness direction of the film is always perpendicular to the electron beam direction (Z axis), so the thickness does not change. The roughness increases, because the thickness Z of the TEM sample which the electron beam passes increases as tilling in the α direction. It implies more projective superposition at the interface layer (Fig. 4). As tilting in the β direction, the sample cross-section is not perpendicular to the electron beam direction (Z axis), resulting in artifacts during projection imaging and a large deviation (Fig. 7). A formula for measuring the thickness of thin films after the sample tilting in the β direction is proposed. For thin films, the measured thickness increases with the increase of the tilting angle β. For thicker films, the measured thickness first increases and then decreases with the increase of tilting angle β. A thinner film thickness t₀ causes a greater relative error of the measured film thickness after tilting in the β direction (Fig. 8).

As the sample tilting in the α direction, the measured thickness of the Mo and Si layers is almost unchanged while the interface roughness increases. This is because the thickness direction of the film is always perpendicular to the electron beam during rotation, and the thickness Z of the TEM sample which the electron beam through increases. The artifacts caused by the sample cross-section are not perpendicular to the electron beam during tilting, which is too severe to distinguish the Mo layer and the Si layer. The measured total thickness of the multilayer film first increases and then decreases with the increasing tilting angle. The formula for calculating the thickness of the film after the sample tilting in the β direction is presented. For thin films, the measured thickness increases with the increasing tilting angle. For thicker films, the measured thickness first increases and then decreases with the increasing tilting angle. As the film thickness t₀ becomes thinner, the relative error is greater after tilting in the β direction. When the TEM sample thickness Z is 10 nm, the relative error of measuring thickness is small after tilting in the β direction. Therefore, when characterizing the structure and thickness of nanofilm by TEM, Si wafers should be cut in a specific direction [11ˉ0] from the beginning of sample preparation. Then samples should be observed from the crystal band axis [110]. Only in this way, it can ensure that the cross sections of Si wafers and films are exactly perpendicular to the electron beam. Photograph and analysis in the thin area of the TEM sample show that the result obtained by this method is more accurate.