Sandblasting Defect Inspection of Investment Castings Based on Deep Convolutional Neural Network

Investment castings often have surface impurities and pieces of shell molds can remain on the surface after sandblasting. Identification of defects involves time-consuming manual inspections in working environments of high noise and poor air quality. To reduce labor costs and increase the health and safety of employees, we applied automated optical inspection (AOI) combined with a deep learning framework based on convolutional neural networks (CNNs) to the detection of sandblasting defects. We applied the following four classic CNN models for training and predictive classification: AlexNet, VGG-16, GoogLeNet, and ResNet-34. In terms of predictive classification, AlexNet, VGG-16, and GoogLeNet v1 could accurately determine whether there were defects. Among the four models, AlexNet was the most accurate, with prediction accuracy of 99.53% for qualifying products and 100% for defective products. We demonstrate a direct detection technique based on the AOI and CNN structure with a fast and flexible computational interface.

R-CNN-Based Large-Scale Object-Defect Inspection System for Laser Cutting in the Automotive Industry

A car side-outer is an iron mold that is applied in the design and safety of the side of a vehicle, and is subjected to a complicated and detailed molding process. The side-outer has three features that make its quality inspection difficult to automate: (1) it is large; (2) there are many objects to inspect; and (3) it must fulfil high-quality requirements. Given these characteristics, the industrial vision system for the side-outer is nearly impossible to apply, and indeed there is no reference for an automated defect-inspection system for the side-outer. Manual inspection of the side-outer worsens the quality and cost competitiveness of the metal-cutting companies. To address these problems, we propose a large-scale Object-Defect Inspection System based on Regional Convolutional Neural Network (R-CNN; RODIS) using Artificial Intelligence (AI) technology. In this paper, we introduce the framework, including the hardware composition and the inspection method of RODIS. We mainly focus on creating the proper dataset on-site, which should be prepared for data analysis and model development. Additionally, we share the trial-and-error experiences gained from the actual installation of RODIS on-site. We explored and compared various R-CNN backbone networks for object detection using actual data provided by a laser-cutting company. The Mask R-CNN models using Res-net-50-FPN show Average Precision (AP) of 71.63 (Object Detection) and 86.21 (Object Seg-mentation), which indicates a better performance than that of other models.

Enabling Automated Sample Delayering, Imaging, and Probing Prep with an Adaptive Endpointing Workflow

The development of advanced logic processing technologies has hit a critical slowing period over the past 10 years. Long gone are the booming days of exponential growth seen in chip transistor density as described by Moore's Law back in 1965.[1] With modern logic manufacturers now capable of creating transistors in the 5-7 nm node range, having the ability to isolate, inspect, and probe individual metal and via layers is of utmost importance for defect inspection and design validation. In this realm of failure analysis, it is critical that design manufacturers possess the ability to isolate any given single layer of their logic samples. These isolated layers can be inspected for defects via SEM, provide validation of CAD designs, or tested with electrical probing for failure analysis. The work here-in describes a functional workflow that enables manufacturers to perform this kind of sample preparation in an automated fashion using the Thermo Scientific™ Helios™ G5 PFIB platform. This workflow can be utilized by both the Thermo Scientific Full Wafer and Small Dual Beam PFIB platforms to streamline sample analysis and failure testing in both the lab and fabrication environments.

Defect Inspection Wafer Notch Orientation and Defect Detection Dependency

Defect detection and defect control are crucial for yield improvement in semiconductor industry. A discrepancy between detected defects compared to yield data regarding a common defect type was found. Historical data show a different behavior was seen on different product groups. A product design analysis on affected layer shows a striking difference in the Terminal Metal Layer (TML) line orientation between those product groups. A particle deposition system was used to distribute a fixed number of PSL spheres on to wafers prior etch process and defect inspections with different wafer notch orientations were performed at the final step. Those deposited PSL spheres prior etch process resulted in extra pattern defects at the inspection step. Extra pattern defects were mostly detected using a certain wafer notch orientation recipe to the majority of TML line orientation compared to the other one. This case study discusses the influence of a defect inspection wafer notch orientation to the defect capture rate on TML layer. Based on this result, the industry should consider the majority line orientation of respective layer on each inspection step when creating a new defect scanning recipe, especially for TML layer.