Where Subtleties Matter: PiFM in Semiconductor Process Characterization

Advanced semiconductor devices rely on complex three-dimensional architectures composed of many layers of different materials, each contributing to overall chip performance. The most common techniques used to characterize these structures are transmission electron microscopy (TEM) and scanning transmission electron microscopy (STEM), both valued for their exceptional spatial resolution [1, 2]. Electron energy loss spectroscopy (EELS) can also provide chemical and bonding information at the nanometer scale by analyzing inelastically scattered electrons [3].

However, achieving the highest spatial resolution possible using these techniques requires samples to be cut into cross-sections thinner than 100 nm. This takes significant time and requires additional expensive and specialized equipment, limiting throughput and accessibility. Even with such time-consuming sample preparation, EELS is further limited by potential beam damage in organic materials, lower sensitivity with high-Z elements, and the complexity of data interpretation [3, 4].

IR PiFM, which offers exceptional sensitivity to local chemical bonding information and sub-5 nm spatial resolution in chemical mapping, could be an excellent technique to characterize the complex processes utilized in fabricating advanced semiconductor devices. As a proof of concept, we created a cross-section (not a lamella) of a power management region of a 3 nm commercially available die/system on chip from a consumer wearable device and characterized it with IR PiFM.

Figure 1 shows the optical image of the cross-section, AFM topography, and PiFM image with the locations of 50 PiF-IR spectra were acquired indicated, which begin in the silicon substrate (spectra 1 – 5) and span four silicon oxide (spectra 6 – 10, 14 – 21, 26 – 34, and 39 – 50) and three metal regions (spectra 11 – 13, 22 – 25, and 35 – 38).

The silicon and metal regions reveal major peaks for siloxane, at 803, 1021, 1100, and 1262 cm⁻¹, and a C-H bending peak at 1462 cm⁻¹ from a hydrocarbon contaminant. The peak at 1462 cm⁻¹ is stronger on the metal compared to the silicon due to larger field-enhancement of the tip-enhanced IR light, making 1462 cm⁻¹ a useful wavenumber to highlight the metallic region.

From spectrum to spectrum in the oxide regions, the broad silicon oxide peak near 1100 cm⁻¹ varies subtly in magnitude, peak position, and shape across the different oxide regions. These differences likely reflect variation in oxide chemistry at different process levels [5]. Because oxide deposition conditions and available precursors can differ from layer to layer, the local chemistry of silicon oxide can also vary, and PiFM is capable of detecting those differences.

In the six PiFM images acquired at the same location using different wavenumbers in Figure 2, the top row images (at wavenumbers corresponding to Si–O bonding) look similar, but have subtle differences in relative intensity and spatial variations. Again, this is likely due to different growth conditions of the different oxide layers. The three dark regions in the image at 1103 cm⁻¹, shown inside the red rectangle, seem to bridge the metallic layers highlighted in the image at 1460 cm⁻¹. Additionally, these regions appear brighter than the surrounding oxide region, yet still darker than the metal regions, suggesting a buried conductive path. This causes the oxide layer above it to be thinner, and that is confirmed by the lowered signal intensity at the same locations in the oxide-highlighting images.

The images at 970 cm⁻¹ and 1200 cm⁻¹ reveal additional finer features. The image at 970 cm⁻¹ highlights a clear sharp edge around the metallic features, including clearly showing a notch measuring ~ 150 nm × 400 nm that was much less clear in topography. The image at 1200 cm⁻¹ also highlights the suggested conducting material clearly, even though it is not obvious what material may be associated with 1200 cm⁻¹. 

Cross-sectional signal-intensity line profiles from the PiFM images in Figure 2 are shown in Figure 3. Moving from the silicon substrate on the left side of the plot rightwards toward subsequent process layers, these profiles support the observations above and reveal several additional features:

1. The shapes of the intensity profiles through each oxide layer are meaningfully different at each wavenumber measured. At 1103 cm⁻¹, the profile is mostly symmetrically parabolic, but the profiles of 1073 and 1115 cm⁻¹ have opposite, more linear slopes travelling through the layers, again suggesting different processing parameters for oxide deposition.
2. The peak at the interface of silicon and the first oxide layer (seen at the profile for 1460 cm⁻¹) indicates that there is a conducting layer at the interface.
3. The profile at 970 cm⁻¹ shows increased signal near the edges of the oxide layers, which may be related to the precursor chemistry used during SiO₂ deposition. A peak at 970 cm⁻¹ can be associated with Si–O–C or Si–CH=CH– bonds.
4. The line profile at 1200 cm⁻¹ generally matches the overall shape of the oxide profiles, suggesting that it is additionally highlighting the oxide. It is most like the 1073 cm⁻¹ profile, as the slope is positive moving from left to right over the oxide layers.

Finally, we examined the fine features observed in the images at 970 and 1200 cm⁻¹. In Figure 4, we zoomed into the 800 nm × 800 nm region that contained the observed notch, which showed two lobes that are highlighted at 970 cm⁻¹ and the buried conducting material at 1200 cm⁻¹, which connects the conducting interface between the silicon and the first silicon oxide layer to the first metal layer. Even though the purpose of the notch is unknown, these images display the potential for interrogating the chemical make-up of a device via cross-sectioning.

In collaboration with the Celano Lab at Arizona State University, scanning-spreading resistance microscopy (SSRM) was performed on the same cross-section. While PiFM primarily revealed information about the oxide trench, SSRM revealed information about the metallic components: in particular, it revealed an electrical junction not visible in PiFM imaging, as well as doped metal right between the junction and the trench, visible by the increased conductivity in comparison to the rest of the metal (Figure 5). In this case, PiFM became a useful complimentary technique to SSRM in analysis of this chip cross-section.

In conclusion, although we do not have access to proprietary chemistry and processes used to create this chip, IR PiFM still reveals significant information about this chip and can agree well with other analysis methods. These results show that PiFM can distinguish different types of formed oxides, locate and identify both particulate and widespread contaminants, and sometimes even locate buried features, all with very high spatial resolution.

A special thank you to Filippo Ozino Caligaris and Dr. Umberto Celano (ASU) for providing this sample, the SSRM data, and for creating Figure 5!

References

1. Advanced Analytical Electron Tomography for Materials Development and Failure Analysis in Semiconductor Devices. NIST 2023.
2. de la Mata, M.; Molina, S. I. STEM Tools for Semiconductor Characterization: Beyond High-Resolution Imaging. Nanomaterials (Basel) 2022, 12 (3), 337. https://doi.org/10.3390/nano12030337.
3. Heller, N. J.; Washington, A. J.; Cushing, S. K. Electron Energy Loss Spectroscopy; ACS In Focus; American Chemical Society: Washington, DC, USA, 2025. https://doi.org/10.1021/acsinfocus.7e9010.
4. Limitations of Electron Energy-Loss Spectroscopy | Microscopy and Microanalysis | Oxford Academic. (accessed 2026-03-10).
5. 7.11: Oxidation of Silicon. Chemistry LibreTexts. (accessed 2026-03-10).