Contour-based metrology for layer-specific variability analysis in advanced DRAM patterning

As patterning complexity increases in advanced DRAM manufacturing, yield‑relevant variability is increasingly governed by distribution tails, directional effects, and shape deformation rather than by average critical dimension (CD) alone. Conventional CD‑centric metrology compresses rich scanning electron microscope (SEM) image information into a small set of scalar metrics, limiting observability of these subtle but yield‑critical phenomena. In this work, we present a contour‑based metrology framework for layer‑specific variability analysis that treats full edge contours as statistical objects. By extracting and aggregating dense contour data from CD‑SEM images, statistical contour envelopes and angle‑resolved descriptors are constructed to capture global and local variability beyond conventional critical dimension uniformity (CDU) and line edge roughness/line width roughness (LER/LWR) metrics. Using dense multi‑gauge contour sampling, we demonstrate a substantial enhancement in tail observability, enabling stable detection of ±3σ outliers and higher‑order statistical behavior. Multi‑dimensional analyses incorporating contour envelopes, quntile-quntile (QQ)‑based kurtosis, and angle‑resolved population maps reveal pronounced layer‑dependent and directional variability that remains invisible in average CD analysis. Tracking these descriptors across consecutive process steps further uncovers inheritance and transformation of variability signatures, highlighting process memory effects in multistep dynamic random access memory (DRAM) patterning flows. The proposed framework is also applied to process validation, where nominally equivalent processes—indistinguishable by mean CD—are clearly separated by contour‑based metrics. Compatible with high‑volume manufacturing through automated extraction and statistical aggregation, this contour‑based approach provides a practical and physically interpretable pathway for early detection, root‑cause attribution, and control of yield‑relevant variability in advanced DRAM patterning.