LithoSeg: A Coarse-to-Fine Framework for High-Precision Lithography Segmentation

TL;DR

Lithography segmentation of SEM images demands pixel-accurate contour delineation and robust performance across diverse process windows, yet dense annotations and contour smoothing hinder practical metrology. The authors introduce LithoSeg, a two-stage coarse-to-fine framework that first uses a Human-in-the-Loop Bootstrapping of Segment Anything Model (SAM) to obtain robust coarse masks with minimal supervision, then refines contours via a point-wise 1D regression using a lightweight MLP along groove normals. The coarse stage reduces annotation cost through prompt-based prompts, human curation, and prompt-free finetuning, while the fine stage achieves pixel-level precision by predicting displacements of contour points from 1D features derived from SEM physics. Experiments show LithoSeg outperforms state-of-the-art baselines on segmentation accuracy and metrology metrics (LER/LWR, CD) across easy to extreme conditions, with substantially faster training times and strong robustness to novel patterns and suboptimal process windows, making it practical for industrial lithography metrology.

Abstract

Accurate segmentation and measurement of lithography scanning electron microscope (SEM) images are crucial for ensuring precise process control, optimizing device performance, and advancing semiconductor manufacturing yield. Lithography segmentation requires pixel-level delineation of groove contours and consistent performance across diverse pattern geometries and process window. However, existing methods often lack the necessary precision and robustness, limiting their practical applicability. To overcome this challenge, we propose LithoSeg, a coarse-to-fine network tailored for lithography segmentation. In the coarse stage, we introduce a Human-in-the-Loop Bootstrapping scheme for the Segment Anything Model (SAM) to attain robustness with minimal supervision. In the subsequent fine stage, we recast 2D segmentation as 1D regression problem by sampling groove-normal profiles using the coarse mask and performing point-wise refinement with a lightweight MLP. LithoSeg outperforms previous approaches in both segmentation accuracy and metrology precision while requiring less supervision, offering promising prospects for real-world applications.

Figures (4)

• Figure 1: This figure illustrates the proposed LithoSeg framework: a two-step process for metrology-ready segmentation of SEM images. Stage 1 (Coarse Segmentation via human-in-the-loop Bootstrapping) involves using SAM to generate initial segmentation masks, followed by Human Curation to classify good and bad results, which are then used to finetune SAM from pretrained weight. Stage 2 (Fine-Grained Contour Segmentation) refines the coarse segmentation by processing each contour point, predicting contour fineness, and generating a detailed fine segmentation mask.
• Figure 2: Visualization of LithoSeg's input, output and intermediate outputs. This coarse-to-fine pipeline progressively reduces artifacts and sharpens boundaries, closely matching the ground truth mask and its boundary roughness.
• Figure 3: Comparison of segmentation model performance across different difficulty levels. The columns show five levels of dataset difficulty, from "Easy" to "Extreme". The rows display the original SEM images (input), segmentation output masks overlay on the SEM images, and the ground truth masks. Please zoom in to see the detailed quality differences.
• Figure 4: Performance vs. Noise Rate in Bootstrapping Ablation Study