Metrology — Quality Control Methodology 

Applied Technologies

1. Elliptic Fourier Descriptor (EFD)-based Quantitative Characterization

A closed contour extracted from an SEM image (e.g., a nano-hole edge) is decomposed into its x and y coordinate signals, each represented as a Fourier series. This yields harmonic coefficients that quantitatively describe how the contour deviates from an ideal shape, in both x and y directions independently. Unlike conventional polar-coordinate roughness analysis, which cannot separate directional frequency components, EFD enables a genuinely two-dimensional characterization of contact edge roughness (CER) — capturing power spectral density, correlation length, and shape-variation fingerprints that are directly relevant to device performance, even for non-circular or curvilinear contact patterns.

2. Deep-Learning Denoising for Unbiased PSD Extraction (Future Direction)

CD-SEM images inherently contain measurement noise, since only a limited number of frames can be averaged without damaging the resist. This noise manifests as a flat "noise floor" in the high-frequency region of the PSD, causing roughness values to be systematically overestimated — a problem that becomes especially severe for ultrathin EUV resists with low signal-to-noise ratio. Our planned approach applies an unsupervised deep-learning denoiser (a U-Net trained via a Noise2Void-style scheme, requiring no ground-truth clean images) to suppress this noise at the image level before contour extraction. By comparing PSD curves before and after denoising — and cross-validating against AFM measurements — this pipeline aims to recover an unbiased CER PSD, extending the well-established LER/LWR denoising methodology into the largely unexplored CER/CWR domain.