ABSTRACT
In an ever more connected world, semiconductor devices represent the core of every technically sophisticated system. The desired quality and effectiveness of such a system through assembly and packaging processes is high demanding. In order to achieve an expected quality, the output of each process must be inspected either manually or rule-based. The latter would lead to high over-reject rates which require a lot of additional manual effort. Moreover, such an inspection is sort of handcrafted by engineers, who can only extract shallow features. As a result, either more yield-losses due to an increase in the rejection rate or more products with low quality will be shipped. Therefore, the demand for advanced image inspection techniques is constantly increasing. Recently, machine learning and deep learning algorithms are playing an increasingly critical role to fulfil this demand and therefore have been introduced in multiple applications. In this paper, an overview of the potential use of advanced machine learning techniques is explored by showcasing of image and wirebonding inspection in semiconductor manufacturing. The results are very promising and show that AI models can find failures accurately in a complex environment.
