In the semiconductor industry, maintaining high-quality standards in materials is critical for ensuring consistent performance, optimal yield, and avoiding excursions. Predictive quality is essential in this regard to enable dynamic adjustments in chemical production processes to address natural variations in raw materials and environmental conditions. By using machine learning algorithms to forecast key quality parameters before production, material manufacturers can reduce cycle times, decrease scrap rates, and minimizing environmental impact.
To effectively implement predictive quality, the development of reliable and resilient machine learning systems is essential. These systems must go beyond simply predicting outcomes. They need to automatically address feature drift caused by changes in manufacturing processes in order to maintain forecast accuracy. Additionally, the machine learning models employed in these systems must be interpretable to ensure that manufacturing staff can depend on the veracity of their predictions in day-to-day operations. Over time, model performance drift and the lack of prediction intervals mean that outliers in operations are often no longer caught because the prediction error grows beyond the control limit range.
This presentation will outline how AI enhances semiconductor quality and smart, adaptive manufacturing. We will discuss the data science foundation of EMD Electronics’ predictive quality approach, focusing on how we combine machine learning operations (MLOps) with explainable AI (XAI) and modern uncertainty quantification techniques, such as conformal prediction. These applications will drive operational efficiency and sustainability in semiconductor materials manufacturing. Our predictive quality models and intelligent raw material allocation have already been proven to save millions of dollars in scrap and improved our sustainability footprint as a manufacturer. Reliable AI is necessary to scale these savings without comprising our quality commitment to our customers. Please join our presentation to learn more how predictive Machine Learning models can effectively be used to generate value in a semiconductor materials manufacturing context based on novel MLOps methods.
