In semiconductor chip production, the detection of laser - marked characters is crucial. WeLinkirt brings new changes to the industry with advanced technologies and solutions.
User Scenario: On the chip production line of a leading semiconductor chip manufacturer, after the chip packaging process is completed, the laser - marked characters on the chip surface need to be detected. These characters contain key information such as the chip model and batch number. The detection objects are the clarity, integrity, and accuracy of these laser - marked characters.
Pain Points: Traditional detection methods based on CNN have many quantitative dilemmas. The miss - detection rate is relatively high, about 5%. This means that about 5 out of every 100 chips with marking problems may flow into the market, affecting product quality and brand image. The false - alarm rate also reaches 12%, which leads to a large number of qualified products needing re - inspection, increasing labor costs and detection time. Moreover, when the chip production is changed to a new type, the traditional method takes a lot of time to reprogram and debug, and the change - over time is as long as 2 hours, seriously affecting production efficiency. In addition, due to the lack of a unified base model, it is difficult to achieve cross - scene generalization, and the adaptability to the marking character detection of different - specification chips is poor.
Technical Principle
WeLinkirt uses an advanced visual foundation model. This model has powerful semantic understanding ability and can accurately identify the features of laser - marked characters. The design of its unified base can achieve cross - scene generalization. By continuously learning from the production line feedback, the detection ability of the model is continuously optimized. In terms of imaging, high - precision imaging equipment is used to clearly capture the details of the marked characters on the chip surface. From the algorithm principle, the model uses a deep neural network for feature extraction and classification, and can accurately judge the clarity, integrity, and accuracy of characters. This method is effective because it gets rid of the dependence of traditional CNN on a large amount of labeled data. Through positive - sample/few - sample learning, the model can be trained with only 1 - 20 good samples, greatly improving the adaptability and accuracy of the model.
- Semantic understanding: It can understand the meaning and features of characters and improve the accuracy of detection.
- Cross - scene generalization: The unified base can adapt to the marking character detection of different - specification chips.
- Continuous learning: Continuously optimize the model from the production line feedback and become more accurate over time.
- Few - sample learning: Reduce the demand for a large amount of labeled data and improve training efficiency.
WeLinkirt's Solution and Product Introduction
WeLinkirt provides the DaoAI AI AOI software system and the DaoAI World model. The DaoAI AI AOI software system has the feature recognition ability of the visual foundation model. It can complete zero - code automatic programming in 5 minutes with one good sample. Through APDT positive - sample/few - sample learning, the model can be trained with only 1 - 20 good samples. At the same time, it also has the function of semantic false - alarm filtering, which effectively reduces the false - alarm rate. The DaoAI World model, as a unified base, has the abilities of semantic understanding, cross - scene generalization, and continuous learning from the production line feedback. Its deployment method is flexible, supporting SDK / API / Docker, and it can achieve 100% local private deployment to ensure that the data does not leave the factory. In the implementation process, first analyze the detection requirements of chip marking characters, then use the DaoAI AI AOI software system for rapid programming and model training, and finally use the DaoAI World model to achieve cross - scene detection and continuous optimization.
WeLinkirt's solution brings an efficient and accurate new choice for semiconductor chip laser marking character detection.
Quantitative Results: After adopting WeLinkirt's solution, the detection rate of chip laser - marked characters has increased to 98%, and the miss - detection rate has decreased to <2%. The false - alarm rate has decreased by - 75%, greatly reducing the re - inspection workload. At the same time, the change - over time has been shortened from the original 2 hours to 5 minutes, significantly improving production efficiency.