Inspection methodologies and machine-learning approaches for defectivity data in semiconductor industry for automotive applications: case study for field-failure prevention

Big-data infrastructure and environment enable the ability to connect and store data to develop and deploy machine-learning activities and analysis in various industries. Equipment manufacturers are implementing new artificial-intelligence capabilities on their tool platforms, thus pushing manufacturers themselves to use data provided by these new functionalities and to link them with other data from their processes. Then, the manufacturer who uses this new artificial-intelligence data is seeking to connect them to some other of internal data, such as the electrical test results per die, in a new type of analysis, with the purpose to improve the screening offered by the electrical test, and thus to increase overall quality. This paper reports new capabilities in automated optical inspection, implemented in automotive-semiconductor manufacturing, which predict a probability of failure per die, computed by the inspection equipment from the features of the observed defects, and presents some significant results in the case of a product implemented in automotive RADAR products.