edfas.org 23 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 28 NO. 4 Thus, the positional drift is a critical consideration for the achieved spatial resolution, particularly if long dwell times are required or when imaging nanoscale features. The drift-per-dwell is less than 25 nm for the majority of dwell positions (Fig. 2), suitable for imaging the 130 nm node IC measured here. Despite drifts-per-dwell on the order of tens of nanometers or less, the cumulative drift over the course of an entire scan can be large enough to degrade the final image if not corrected for. To mitigate the cumulative drift, the experiment leverages the advantage of operation in an SEM. After each dwell, a secondary electron image is acquired of a fiducial feature on the metal target surface. This post-dwell image is aligned to the fiducial image from the previous dwell, yielding the total drift in the most recent dwell. That drift is then integrated into the next commanded dwell position to ensure that the expected grid in the region-of-interest is measured. This avoids unexpected gaps in the measured scan grid due to cumulative drift or unusually high single-dwell drift and ensures that the recorded positions match the actual positions to within the amount of drift in a given dwell. ANGLE-TO-ANGLE ALIGNMENTS Seven projection angles spanning 45° were collected in 7.5° steps. This 45° coverage is less than what is conventionally required for nanoscale CT, but is an experimental necessity for semiconductor FA due to the planar nature of semiconductor devices. The seven angles were collected sequentially, with the sample rotated to a new angle after completion of the region-of-interest scan at the current projection angle. When rotating, slight misalignment to the fiducial or angular offsets can lead to an offset in the scanned region from the prior angle. This effect is not corrected for while acquiring data and results in offsets in the region-of-interest acquired at different angles. To correct for this artifact, the researchers perform and align single-angle reconstructions. A single-angle recon- struction refers to generating a 3D image from only a single projection angle, made possible by the large area HPCD subtending a solid angle of 35° vertically and 33.5° horizontally. While the single angle results are not as wellresolved as reconstructions containing multiple angles, they are not degraded by angular misalignment and are suitable for aligning the region-of-interest at different projection angles. A workflow was developed in which each slice in a single-angle reconstruction is aligned to the identical slice in the single-angle reconstruction of the subsequent projection angle, resulting in a mean offset value across all slices for each angle collected. Representative slices from 3D single-angle reconstructions of normal incidence and 7.5° data are shown in Fig. 3, along with an example of an alignment. The offset between scan regions at different projection angles can reach tens to hundreds of nanometers, making this a critical correction for achieving sub-100 nm imaging resolution. The alignments from each angle are then applied to the raw data and the full sevenangle dataset is reconstructed to obtain the final 3D image. IMAGING RESULTS Using this approach, wiring and via layers in a Cu-SiO2 Fig. 3 An example of single-angle reconstructions and angle-to-angle alignment. Slices showing a single wiring layer from the 0° projection (top left) and 7.5° projection (top right). The 7.5° and 0° scan regions overlaid after alignment (bottom). The two angles were misaligned by 818.9 nm in X and 260.3 nm in Y. The squares filling the reconstruction slices around the IC wiring and vias are chemical-mechanical polishing fill. (continued on page 26)
RkJQdWJsaXNoZXIy MTYyMzk3NQ==