Data analytics and machine learning: root-cause problem-solving approach to prevent yield loss and quality issues in semiconductor industry for automotive applications

Quality requirements in the semiconductor industry for automotive products are increasing rapidly with the movement to autonomous vehicles and higher levels of safety. It is no longer possible to express maximum failure requirements in parts per million (ppm). Individual failing parts observed in the field and reported by customers can trigger a significant quality response. ‘Zero-defect’ (ZD) is no longer considered a utopian ideal, but a required potentially reachable goal for semiconductor manufacturers. Projects and studies that include artificial intelligence and big data, are seen as key drivers to reach a ZD level of quality. Competing objectives targeted in any industrial project, such as quality improvement and gross margin, must also be considered. Initial projects in machine learning (ML), focusing on yield-loss issues, are being deployed within the manufacturing sites. These projects interconnect typical internal data collected from the manufacturing and assembly lines with engineering, qualification and reliability data. For a specific case study of unexpected abnormally high variability on some parameters, this paper presents a problem-solving approach in a big-data environment. Models implemented and results obtained towards root-cause problem solving for this issue, are discussed. This overall approach may be replicated in other ML projects.