In the process of semiconductor chip production, the detection of packaging cracks is crucial. WeLinkirt (DaoAI) provides an efficient solution for the semiconductor industry with advanced AI vision technology.
User Scenario: A leading semiconductor chip manufacturer, in the packaging process of its chip packaging production line, mainly produces various high - performance semiconductor chips. The detection objects are cracks in the chip package after encapsulation. These cracks may appear on the surface, edges, etc. of the package. The existence of cracks can affect the electrical performance and reliability of the chip, and may even lead to chip failure.
Pain Points: Traditional detection methods face many difficulties in the detection of chip packaging cracks. On the one hand, the miss - detection rate is relatively high, about 3%. This means that some chips with cracks may enter the market, bringing potential after - sales risks and reputation losses to the enterprise. On the other hand, the false - alarm problem is serious, with a false - alarm rate as high as 40%. A large number of false alarms disperse the limited re - inspection resources, resulting in low re - inspection efficiency and a significant increase in labor costs. Moreover, traditional detection methods require a large amount of reprogramming and debugging when changing models, and the model - changing time is as long as 30 minutes, which seriously affects the flexibility and production efficiency of the production line. In addition, traditional methods have difficulty accurately identifying some tiny cracks and low - contrast defects, which cannot meet the enterprise's strict requirements for product quality.
Technical Principle
WeLinkirt (DaoAI) uses advanced AI vision algorithms and imaging technologies to solve the problem of chip packaging crack detection. In terms of algorithms, deep learning algorithms are used for feature extraction and pattern recognition. Deep learning algorithms can automatically learn the features and patterns of cracks from a large amount of sample data. Through the training of multi - layer neural networks, different types and locations of cracks can be accurately identified. Compared with traditional rule - based algorithms, deep learning algorithms have stronger adaptability and robustness, and can process complex and changeable image data.
In terms of imaging, the DaoAI 2D / 3D AI AOI equipment uses a self - developed 3D camera for image acquisition and combines 3D topography reconstruction technology to obtain the 3D information of the chip package surface. The 3D topography information can more comprehensively reflect the characteristics of cracks, including the depth, width and shape of cracks, thus improving the accuracy of crack detection. For some cracks hidden inside the package or with low contrast, the 3D imaging technology can effectively enhance their features and make them easier to detect.
- The multi - layer neural network structure of the deep learning algorithm can extract high - level features of the image layer by layer, from the bottom - layer edge and texture features to the high - level semantic features, making the recognition of cracks more accurate and detailed.
- The high - resolution and high - precision imaging ability of the 3D camera ensures that the collected images contain enough detailed information, providing a reliable data basis for subsequent analysis and detection.
- The 3D topography reconstruction technology can accurately reconstruct the 3D shape of the chip package surface by processing and analyzing images from multiple perspectives, thus more clearly showing the morphology and characteristics of the cracks.
- The semantic false - alarm filtering function uses the deep - learning model to understand the semantics of the image, which can accurately distinguish real cracks from interfering factors similar to cracks, thus effectively reducing the false - alarm rate.
WeLinkirt Solution and Product Introduction
WeLinkirt provides a complete solution, mainly involving the DaoAI AI AOI software system and the DaoAI 2D / 3D AI AOI equipment. The DaoAI AI AOI software system has a strong feature - recognition ability. Based on the visual basic model, it can quickly and accurately identify chip packaging cracks. The system uses APDT positive - sample/few - sample learning technology. With only 1/10 good products, it can complete 0 - code automatic programming within 5 minutes, greatly shortening the model - changing time. At the same time, the system also has a semantic false - alarm filtering function, which can effectively reduce false alarms and focus the limited re - inspection resources on real problems.
The DaoAI 2D / 3D AI AOI equipment is the hardware core. Its self - developed 3D camera combined with 3D topography reconstruction technology can detect hidden solder joints, coplanarity and micron - level topography. In the detection of chip packaging cracks, the equipment can clearly capture the 3D information of the cracks, providing accurate data support for the analysis and judgment of the software system. The equipment and the software system cooperate closely to form an efficient detection system.
WeLinkirt's solution combines advanced software algorithms with high - precision hardware equipment to provide reliable guarantee for semiconductor chip packaging crack detection.
Quantitative Results: By adopting WeLinkirt's solution, the leading semiconductor chip manufacturer has achieved remarkable results. First, the crack detection rate has increased from the original 97% to 99.2%, and the miss - detection rate has been reduced to <0.8%, greatly reducing the risk of chips with cracks entering the market. Second, the false - alarm rate has been reduced by - 70%, from the original 40% to 12%, making the re - inspection resources more effectively utilized and the re - inspection efficiency significantly improved. In addition, the model - changing time has been shortened from the original 30 minutes to 5 minutes, and the flexibility and production efficiency of the production line have been significantly improved.