In the field of semiconductor advanced packaging, the detection of rare defects with few samples has always been an industry challenge. WeLinkirt provides an effective solution to this problem with its advanced technologies and products.
User Scenario: A leading semiconductor manufacturer's advanced packaging production line, with high - performance chips as the main products. In the packaging process, the detection objects are the tiny structures after chip packaging, such as solder joints and pins. The dimensions of these structures are usually in the micron level, and there are some rare defects, such as tiny foreign objects and cracks.
Pain Points: Traditional detection methods face many difficulties when dealing with rare defects with few samples. On one hand, due to the extremely small number of rare defect samples, the model training is insufficient, resulting in a missed - detection rate as high as 5%. A large number of defective products flow into subsequent processes, increasing production costs and quality risks. On the other hand, the false - alarm rate remains high, reaching 15%. This causes a large number of good products to be misjudged as defective, requiring manual re - inspection, which not only increases labor costs but also reduces production efficiency. In addition, when changing the production line model, traditional methods require a lot of time for reprogramming and debugging, and the model - changing time is as long as 2 hours, seriously affecting the production flexibility.
Technical Principle
WeLinkirt uses an advanced visual foundation model and APDT positive - sample/few - sample learning algorithm to solve this problem. The visual foundation model has a powerful feature - recognition ability and can learn the feature information of defects from a small number of samples. The APDT positive - sample/few - sample learning algorithm can quickly and accurately identify defect patterns through learning from a small number of positive samples. In terms of imaging, the self - developed 3D camera combined with 3D morphology reconstruction technology can obtain the 3D information of the detection object, clearly presenting the morphological characteristics of tiny structures, which helps to find hidden defects. This multi - dimensional information acquisition and analysis method enables the algorithm to more accurately judge the existence of defects, greatly improving the detection accuracy.
- Visual foundation model: Conduct in - depth mining and learning of image features to improve the recognition ability of different types of defects.
- APDT positive - sample/few - sample learning algorithm: Use a small number of positive samples to quickly train the model to meet the detection needs of rare defects.
- Self - developed 3D camera and 3D morphology reconstruction: Obtain 3D information to detect hidden defects and micron - level morphological changes.
WeLinkirt's Solution and Product Introduction
WeLinkirt provides the DaoAI AI AOI software system and the DaoAI 2D / 3D AI AOI equipment. The DaoAI AI AOI software system has a powerful positive - sample/few - sample learning ability. With only 1 - 20 good samples, it can complete zero - code automatic programming in 5 minutes. At the same time, the system also has a semantic false - alarm filtering function, which can effectively reduce the false - alarm rate. The DaoAI 2D / 3D AI AOI equipment uses the self - developed 3D camera and 3D morphology reconstruction technology, which can detect hidden solder joints, coplanarity, and micron - level morphological changes. In the implementation process, first, a small number of good samples are collected and analyzed, and the software system is used for rapid programming and model training. Then, the trained model is deployed in the AOI equipment to perform real - time detection on the production - line products.
WeLinkirt's solution effectively solves the problem of rare defect detection with few samples in semiconductor advanced packaging through advanced technologies and products.
Quantitative Results: After adopting WeLinkirt's solution, the detection rate has increased from the original 95% to 99.2%, and the missed - detection rate has been reduced to <0.8%, greatly reducing the risk of defective products flowing into subsequent processes. The false - alarm rate has decreased from 15% to 3%, a reduction of - 80%, reducing the workload of manual re - inspection. The production - line model - changing time has been shortened from 2 hours to 5 minutes, improving production flexibility and efficiency.