ABSTRACT
In the semiconductor industry the desired quality and effectiveness in the process of assembling integrated circuits is nowadays at the limit and without safety margin. To achieve important competitive advantages, this process must be continuously optimized and adjusted. Such process is indeed strongly dependent on parameters that are distributed among various control technology assemblies, materials, and the environment. However, the current inspection tools deployed for defect detection through assembly and packaging process are mainly based on rigid and simple rules. The latter are handcrafted by engineers, which can only extract shallow features. Therefore, the accuracy of classification by tools is quite low, which provides incomplete information for root cause investigation and can cause yield-loss costs due to over reject. Hence, automatic inspection tools for visual defect detection, acting as final quality gate before shipping to end customers is very demanding. Therefore, a deviation detection model based on machine learning is developed. On the other side, due to the lack of existing labelled images, an anomaly detection is proposed, in some cases as an assistant tool for collecting defect images with less effort. Results show that artificial intelligent (AI) solutions can achieve a better performance than the classical tools and overcome the human ability in detecting the deviation in the data.
162Hence, AI can be used for decreasing the yield-loss, improving quality of the product and greatly reduce labour intensity.
