Root Cause Analysis for Semiconductor Wafer Contamination: Identifying Sub-20 nm Organic and Inorganic Defects with PiFM

Key takeaways:

• PiFM fills a capability gap of SEM/EDX and ToF-SIMS for root cause analysis (RCA). Specifically, the combination of molecular identification with sub-5 nm resolution, and monolayer sensitivity.
• PiFM is nondestructive and effective on both organic and inorganic materials without requiring any special sample preparation, making it an excellent choice to find contaminants without modifying the surface like SEM/EDX or using destructive sample preparation like ToF-SIMS.
• PiFM shortens the RCA loop by quickly identifying the compounds that make up residues and defects, allowing engineers to proceed with correlation experiments.
• The Vista 300 ANDR accepts KLARF defect maps and reviews defects of interest at ~1 minute each, enabling full-wafer chemical-ID surveys in roughly one hour (50 defects).

Organic and Sub-20 nm Defects Challenge Root Cause Analysis

Root cause analysis (RCA) of shrinking device geometries and increasingly complex process integration is difficult because the contaminants that now limit yield are more often monolayer organic residues and sub-20 nm particles which sit below the spatial resolution and outside the chemical specificity of typical analytical techniques. Contaminants originate from process-induced, material-related, environmental, or handling-related sources, and often pass through several process steps before producing measurable effects. RCA aims to trace those effects, which are manifested as parametric drift, reliability failures, or bonding defects in advanced packaging, back to the originating source. In current RCA workflows, inspection tools, such as KLA Surfscan, utilize optical methods to characterize the distribution of defects, which allows engineers to make inferences about defect origin. For example, spatial clustering can suggest equipment contamination or random distributions may indicate airborne exposure. While this information is useful, chemical identification of defects provides necessary causality information to failure analysis teams. As device geometries shrink and heterogeneous integration multiplies material interfaces, RCA must resolve sub-20 nm and organic defects outside of the capability of current analytical techniques.

PiFM for Semiconductor RCA

Photo-induced Force Microscopy (PiFM) is uniquely poised for use in semiconductor RCA because it is the only analytical technique that combines sub-5 nm spatial resolution, molecular identification, monolayer sensitivity, and non-destructive measurement (Table 1); the exact combination of capabilities missing from widely used techniques such as SEM/EDS and ToF-SIMS on sub-20 nm and organic defects. PiFM is a non-contact atomic force microscopy and infrared spectroscopy (AFM-IR) technique that identifies molecular species by directly measuring the photo-induced force between the AFM probe and the sample. PiF-IR produces spectra that matches very well to corresponding bulk FTIR spectra and can be searched directly against commercial or user-created IR libraries for identification purposes. Like FTIR, PiFM identifies organic and inorganic materials, however, since PiFM detects the PiF via the mechanical response of the cantilever, the technique can also render useful information on “IR inactive” materials such as pure metals and 2D materials. Due to near-field enhancement of the IR laser by the AFM probe, the spatial resolution of the IR spectroscopy is enhanced to better than 5 nm. By imaging at multiple IR wavelengths corresponding to absorption peaks of different chemical species, PiFM can spatially map nm-scale patterns of the individual chemical components [1]. For RCA specifically, PiFM delivers three capabilities that other analytical techniques cannot:

Molecular identification at the defect length scale. Sub-5 nm lateral resolution with monolayer sensitivity means a sub-5 nm defect with ~ 1 nm thickness is resolved and chemically identified in the same measurement.

Non-destructive, ambient operation. There is no need for vacuum or damaging sample preparation, such as sputtering or electron-beam exposure.

Identification of organics. Strong IR absorption from organic defects makes photoresist residues, BARC residues, adhesion promoters, and bonding inhibitors easy for PiFM to identify. These are the materials most likely to drive modern yield failures, yet they are invisible or ambiguous to EDS.

Where SEM/EDS, ToF-SIMS, and other Analytical Techniques Fall Short

SEM/EDS, ToF-SIMS, and TEM/STEM-EDS each fail a different step of the RCA workflow when the defect is sub-20 nm, primarily organic, or a monolayer residue.

SEM/EDS delivers excellent morphological imaging at the nanometer scale, but because EDS collects elemental, not molecular information, material’s identity cannot be distinguished. Furthermore, SEM-based EDS spatial resolution is limited by the electron-beam interaction volume to roughly 0.5 µm, more than two orders of magnitude larger than the defects that now determine yield. Finally, e-beam exposure could destroy smaller defects and frequently deposits hydrocarbons on the sample, contaminating the signature an analyst is trying to interpret.

ToF-SIMS provides molecular information and high trace sensitivity, but its lateral resolution of 0.2–10 µm cannot resolve a sub-20 nm defect. ToF-SIMS is also destructive due to ion sputtering and is sensitive to matrix effects that complicate identification of organic materials and thin films.

TEM / STEM-EDS offers atomic-scale imaging but requires destructive FIB cross-sectioning and delivers elemental rather than molecular information.

XPS, TXRF, and Auger are useful in specific contexts, but are either too large in analysis spot size (XPS, TXRF) or too limited in chemistry (Auger) for nanoscale, organics RCA (Table 1).

	IR PiFM	ToF-SIMS	XPS	SEM/EDS	TEM	TXRF	Auger
Species Detected	Molecular	Molecular	Molecular	Elemental	Elemental	Elemental	Elemental
Lateral Resolution	< 5 nm ~ 20 nm*	> 0.2 µm	10 µm – 2 mm	1 nm imaging 0.5 µm EDS	0.2 nm imaging 1-20 nm EDS	~ 1 nm	~ 10 nm
Depth Probed	< 20 nm ~ 500 nm*	1 nm	10 nm	1 µm	~ 100 nm	10 nm	10 nm

How PiFM Shortens the RCA Workflow

PiFM on the Vista 300 integrates with standard fab inspection workflows by accepting KLARF defect maps directly from inline inspection tools such as KLA Surfscan, auto-deskewing to wafer fiducials (usually a few larger defects), and visiting every defect-of-interest (DOI) to acquire topography and PiF-IR spectra in sequence (Figure 1).

Typical throughput on the Vista 300 is approximately 1 – 2 minutes per defect depending on the accuracy of the coordinates from the inspection tool, enabling chemical-ID surveys of 30 to 50 defects in roughly an hour.

PiFM is not positioned as a wholesale replacement for SEM/EDS but operates adjacent to those tools. Current SEM/EDS tools enjoy an order of magnitude higher throughput than Vista 300 ANDR. Thus, when many larger inorganic/metallic defects need to be binned by shapes and elemental content, SEM/EDS can continue to be the workhorse. However, the defects that SEM/EDS cannot classify effectively such as sub-20 nm particles, organics, and monolayer residues should be routed to the Vista 300 ANDR for PiFM analysis. This removes the need for speculative follow-up with ToF-SIMS or destructive TEM.

After characterization, the physical findings must be correlated to the contamination source. Using PiFM, nano-defects can be identified quickly, moving from characterization to correlation studies faster, shortening the RCA loop for the defect classes that most often limit yield at advanced nodes.

RCA with PiFM Case Studies: Silica, PFPE, Cu–BTA

PiFM is often used to resolve RCA scenarios. The three examples below are drawn from previously published data in a Molecular Vista application note [2]. These examples show how PiFM replaces inference with direct identification.

Case 1: 3 nm Silica Particle on Silicon

A particle ~3 nm tall on a silicon surface produced a PiF-IR spectrum with a characteristic peak at 1085 cm⁻¹ (blue spectrum) (Figure 2). A library search against the Wiley KnowItAll database [3] returned a direct match to silica (upper set of spectra). No inference was required; a single PiFM measurement converted an unclassified particle-of-interest on an inspection map into a named contaminant that can be traced to a specific tool, wet-chemistry step, or polishing slurry. SEM/EDS would not have been able to analyze a particle of this size, and even if it were able, it would have reported Si and O but could not distinguish it from the silicon substrate with its native oxide. The red spectrum acquired on the substrate next to the defect shows peaks at ~1100, 1460, and 2940 cm⁻¹, which are associated with the native silicon oxide and hydrocarbon contamination layer that is most likely a monolayer on the surface of the substrate, demonstrating the monolayer sensitivity of IR PiFM.

Case 2: ~1 nm PFPE Monolayer on a “Clean” Wafer  

Five PiF-IR spectra acquired at different locations of a wafer removed from a storage container were repeatable and showed a dominant peak at ~1250 cm⁻¹ in addition to the silicon oxide peak at 1130 cm⁻¹. Averaging the spectra and searching the IR library identified the contaminant layer as perfluoropolyether (PFPE), a contact-transfer contaminant from the storage container (Figure 3a) that may have been fabricated out of PFA (Perfluoroalkoxy alkanes). Post plasma-cleaning, the same wafer showed a different fingerprint, mostly hydrocarbon (Figure 3b), confirming both the effectiveness in removing the PFPE monolayer and the generation of hydrocarbon contaminants by the cleaning method. For an RCA team, this is a critical example: a monolayer contaminant that is invisible to optical inspection, indistinguishable in EDS, and outside the lateral reach of ToF-SIMS, was not merely detected but identified, which allowed the wafer carrier to be determined as the potential root cause and enabling quick corrective action at a specific point  rather than a generic “improve handling” recommendation.

Case 3: Cu–BTA Residue After Chemical Mechanical Planarization (CMP) in Hybrid Bonding

Hybrid bonding for advanced packaging requires copper surfaces free of organic inhibitors and carbides that prevent true Cu–Cu interfacial grain growth. Benzotriazole (BTA), used as an inhibitor to protect recessed copper lines, is known to form a thin Cu(II)–BTA polymer on the Cu surface that is difficult to remove [4]. Figure 4 shows four averaged PiF-IR spectra for (a) as-electroplated Cu, (b) Cu after BTA treatment, (c) Cu after a CMP process, and (d) Cu after Ar-ion cleaning. The Cu–BTA signature, peaks at ~795 and ~755 cm⁻¹ from C–H stretches of the benzene ring in BTA, is shown in both figure 4b (before CMP) and 4c (after CMP). This means that the presence of BTA on the Cu surface persists through the CMP process. Providing direct, process specific evidence of a residue that would otherwise require inference from XPS elemental data or speculative chemistry from ToF-SIMS [4]. After Ar-ion cleaning, the BTA signature is essentially eliminated (Figure 5d), confirming that the corrective action closes the RCA loop on this contamination source.

When to Use PiFM for RCA

PiFM on Vista 300 is the right RCA tool when wafer defects are small (less than 20 nm), primarily organic, a residue, or otherwise “invisible” to SEM/EDS or ToF-SIMS. Even for inorganic defects, when the defects are smaller than 20 nm, the low energy x-ray peak for several oxides may overlap, and the defect cannot be identified with confidence. Usually, other higher energy x-ray peaks would help with identification, but with small defects, increased acceleration voltage could destroy the particles. PiFM operates alongside existing inline ADR pipelines and closes the RCA loop in hours rather than days or weeks due to its capability to deliver molecular rather than elemental identification. Split-lot experiments and controlled process modifications then become tests of a specific hypothesis rather than searches for a plausible one.

Contact us to schedule a demonstration on your own wafers.

References

1. D. Nowak, W. Morrison, H. K. Wickramasinghe, J. Jahng, E. Potma, L. Wan, R. Ruiz, T. R. Albrecht, K. Schmidt, J. Frommer, D. P. Sanders, and S. Park, “Nanoscale chemical imaging by photoinduced force microscopy,” Sci. Adv. 2:e1501571 (2016).
2. Molecular Vista Application Note, “Analyzing sub-20 nm defects and ultrathin (~1 nm thick) residues in semiconductor processes,” May 2025.4
3. KnowItAll IR Spectral Database Collection — Wiley Science Solutions.
4. Baohong Gao, Baimei Tan, Yuling Liu, Chenwei Wang, Yangang He, and Yanyan Huang, “A study of FTIR and XPS analysis of alkaline‐based cleaning agent for removing Cu‐BTA residue on Cu wafer,” Surface and Interface Analysis, 51, 566–575 (2019).