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

Bingyang Guo, Qiang Zuo, Ruiyun Yu · 2025

Read Point Cloud Segmentation of Integrated Circuits Package Substrates Surface Defects Using Causal Inference: Dataset Construction and Methodology by Bingyang Guo, Qiang Zuo, Ruiyun Yu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-05853.txt.

Introduction

Bingyang Guo, Qiang Zuo, and Ruiyun Yu build the first public 3D data set for finding surface defects on ceramic chip-package substrates. These substrates are key parts in integrated circuit packaging, but their complex shapes and tiny flaws, plus the lack of any public data set, have held back automated inspection.

Their data set, CPS3D-Seg, holds thirteen hundred point-cloud samples across twenty product types, each with point-level labels and, they report, the best resolution among industrial 3D data sets. They benchmark leading segmentation methods on it. They also propose CINet, a method using causal reasoning to discount confounders in the clouds. In tests CINet clearly beat existing algorithms on overlap and accuracy.

Catalog record

CREATORBingyang Guo, Qiang Zuo, Ruiyun Yu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-05853

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Point Cloud Segmentation of Integrated Circuits Package Substrates Surface Defects Using Causal Inference: Dataset Construction and Methodology Bingyang Guo, Qiang Zuo, Ruiyun Yu 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. Code & Datasets — https://github.com/Bingyang0410/CPS3D-Seg Introduction 3D point cloud segmentation is vital in industrial condition monitoring, as it enables a comprehensive understanding of the geometric structure and intricate characteristics of each point. This capability is increasingly critical in the high-end manufacturing sector of integrated circuits, where precise defect detection and quality assurance are essential for maintaining reliability and performance. Ceramic package substrates, known for their exceptional material properties such as high hardness, elevated elastic modulus, extreme wear resistance, excellent electrical insulation, and superior corrosion resistance, play a crucial role in the integrated circuits industry by ensuring the reliability and performance of electronic devices ( 1 ) . However, manufacturing defect-free ceramic package substrates remains challenging, with common defects such as cracks, holes, bubbles, delamination, scratches, and stains ( 65 ) potentially compromising substrate quality and increasing the likelihood of electronic equipment failure. 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. Existing research ( 19 ; 20 ; 10 ) has made initial advances in detecting surface defects in substrates, primarily relying on 2D image-based methods. While these 2D techniques have shown some effectiveness, they inherently need more depth information, making it difficult to accurately represent circuits’ vertical connectivity and the substrate surfaces’ precise topography. This limitation impedes the reliable detection and analysis of defects lying in the depth dimension of the substrate. Recent advancements in 3D acquisition technologies, such as stereo cameras ( 46 ) , structured light ( 45 ) , LiDAR ( 5 ) , and laser scanners ( 47 ) , have provided a promising solution to this problem by enabling the capture of complex 3D data. These technologies offer a comprehensive view of the ceramic package substrates from multiple perspectives, allowing for a more precise and detailed surface quality assessment. The integration of 3D data provides richer information, facilitating better identification and characterization compared to 2D data. Although several 3D point cloud datasets ( 7 ; 3 ; 27 ; 9 ; 2 ; 4 ; 32 ) are currently available for defect detection, these datasets frequently exhibit substantial differences when compared to ceramic package substrates. Typical 3D datasets may include objects with relatively simple geometries and textures. In contrast, ceramic package substrates feature highly intricate surface circuitry and complex three-dimensional structures ( 29 ) . The point cloud density and accuracy required in integrated circuits are much higher, making data acquisition more challenging. More publicly available point cloud datasets tailored for integrated circuits are needed to address this issue in this field. To address the above problem, we present a high-resolution point cloud segmentation dataset of ceramic package substrates, named CPS3D-Seg. Unlike previous 3D datasets collected by RGB-D cameras or double CCD structured light, the point cloud data of CPS3D-Seg are all obtained by high-precision line laser scanners Keyence LJ-X8020. Specifically, we have built a data acquisition platform that integrates four LJ-X8020s, ensuring accuracy and data acquisition efficiency. Thanks to the improved performance of the collection equipment, the point cloud density and accuracy of the CPS3D-Seg dataset have significantly improved. CPS3D-Seg offers 16K point clouds per object with a point precision of 2.5 {\mu}m , which exceeds the current highest precision dataset Real3D-AD ( 32 ) by an order of magnitude. What’s more, we utilize CloudCompare to accurately annotate each defect sample at the point level, visually presenting concrete defect information. Inspired by previous excellent datasets, we also construct a comprehensive and large-scale benchmark to evaluate the effectiveness of CPS3D-Seg. As part of the benchmark construction, we perform qualitative and quantitative analysis of the data and algorithms. Our research results indicate that the application of specific preprocessing or structuring procedures may result in discrepancies in the presentation effect due to the increased density and accuracy of point clouds. As shown in Fig. 1 , this can be summarized into two folds: i) The red oval box represents that the line laser may bring out information missing during the scanning process due to its angle of collection. The direct application of the point cloud completion approach will yield imprecise data. ii) The yellow oval box means that aggregation errors in the structuring process may combine certain defective and non-defective areas. By comparing these with the mesh representation of the 3D design template, the above issues become more apparent. In essence, all the aforementioned concerns can be encapsulated as the intricacy of the point cloud. Inspired by this discovery, we propose an innovative approach utilizing causal intervention into the formulation of point cloud segmentation, which helps in discerning latent relationships of high-resolution 3D data. Technically, we conduct a structural causal model to explain the potential effect process. In response to the difficulty of measuring or observing the hidden confounder in this problem, we propose a causal intervention approach that relies on back-door adjustment. Our approach uses the CPS3D-Seg dataset for practical ’virtual’ interventions, rather than overly expensive ’physical’ interventions such as changing the collection angle or repeating the collection multiple times. The primary contributions of this paper are as follows: 1. We construct a 3D segmentation dataset suitable for ceramic package substrates surface defect detection, named CPS3D-Seg, which fills the gap in the field of integrated circuits and surpasses existing datasets in point cloud resolution and accuracy. 2. We propose a benchmark for 3D segmentation on CPS3D-Seg, encompassing latest SOTA algorithms. This benchmark evaluates the effectiveness of CPS3D-Seg. 3. We introduce a novel 3D segmentation method based on structural causal model, which achieves causal inference by solving probabilities through learnable modules. Comprehensive experimental results illustrate the superior efficacy of our approach. Related Work 3D Industrial Segmentation Datasets. In recent years, the importance of point cloud has significantly increased in numerous fields, such as manufacturing ( 56 ) , autonomous driving ( 13 ; 15

Read the complete text