In semiconductor manufacturing, subtle variations between process chambers can impact the accuracy of AI-driven monitoring. Our team has developed a cross-chamber transfer learning approach that allows models to adapt to these differences without requiring large amounts of new data. By adjusting time and amplitude distributions in sensor traces, this method outperforms traditional Dynamic Time Warping, achieving an area-under-the-curve (AUC) of 0.973 in Fault Detection and Classification (FDC). The result is more reliable AI performance across chambers, ensuring consistent product quality and reducing costly misclassifications.
