Point Cloud Segmentation of Integrated Circuits Package Substrates Surface Defects Using Causal Inference: Dataset Construction and Methodology

TL;DR

The paper tackles the challenge of high-precision 3D segmentation for ceramic package substrates (CPS) where public data are scarce and defects are subtle. It introduces CPS3D-Seg, a high-resolution 3D dataset collected with four line laser scanners and annotated for defect segmentation, and CINet, a causal-inference-based segmentation method built on a Structural Causal Model with backdoor adjustment to mitigate confounding.CINet comprises three modules—Quality Assessment (QA) to model confounder distribution $P(C=c)$ via KDE/GMM, Structural Refinement (SR) to reconstruct $S|X,C$ using a Transformer, and Mapping Attention Detection (MAD) to fuse group embeddings with density features—enabling $P(Y|do(X=x))$ to be estimated despite unobserved confounders.The authors benchmark CPS3D-Seg against a wide range of SOTA 3D segmentation methods and demonstrate that CINet achieves superior performance (e.g., approximately a 4.16% gain in mIoU over strong baselines).This work provides a practical pathway for reliable minor-defect detection in IC packaging and delivers a reproducible dataset and causal-inference framework that can be extended to other industrial 3D segmentation tasks.

Abstract

The effective segmentation of 3D data is crucial for a wide range of industrial applications, especially for detecting subtle defects in the field of integrated circuits (IC). Ceramic package substrates (CPS), as an important electronic material, are essential in IC packaging owing to their superior physical and chemical properties. However, the complex structure and minor defects of CPS, along with the absence of a publically available dataset, significantly hinder the development of CPS surface defect detection. In this study, we construct a high-quality point cloud dataset for 3D segmentation of surface defects in CPS, i.e., CPS3D-Seg, which has the best point resolution and precision compared to existing 3D industrial datasets. CPS3D-Seg consists of 1300 point cloud samples under 20 product categories, and each sample provides accurate point-level annotations. Meanwhile, we conduct a comprehensive benchmark based on SOTA point cloud segmentation algorithms to validate the effectiveness of CPS3D-Seg. Additionally, we propose a novel 3D segmentation method based on causal inference (CINet), which quantifies potential confounders in point clouds through Structural Refine (SR) and Quality Assessment (QA) Modules. Extensive experiments demonstrate that CINet significantly outperforms existing algorithms in both mIoU and accuracy.

Figures (9)

• Figure 1: The demonstration of discrepancies in the presentation effect. The red oval box represents the missing stereo data during the scanning process, and the yellow oval box means the aggregation errors in the structuring process.
• Figure 2: (a) The collection equipment to capture point cloud data. (b) The Keyence LJ-X8020 line laser scanner. (c) The biaxial motion mechanism controls the line laser, scanning from one corner of the sample and covering the full area.
• Figure 3: The concrete examples of CPS3D-Seg.
• Figure 4: The defect proportion statistics of different datasets. CPS3D-Seg shows significant advantages in the defect proportion area, which makes our dataset more applicable and accurate in small defect detection.
• Figure 5: (a) The proposed structural causal model for point cloud. (b) The specific meaning of symbols.