| Abstrakt: |
With microchip sales exceeding US $500 billion in 2023, improving semiconductor yield is a valuable proposition. Toward this goal, this paper proposes a novel adaptation of Naive Bayes methods using tool visits and process variables, combined with k-means clustering of yield, to quickly identify poorly performing microchip manufacturing equipment. As a result, managers can better target maintenance during production to increase yield. The proposed Dayaratna/McFarlane (DM) method was developed and tested on a simulated dataset of over 300,000 wafers entangled among nearly 8000 process stations. By a factor of two, the DM method outperforms the current practice, tool commonality, and a process variable-free version of the method. Additionally, the DM method also outperforms standard Naive Bayes methods drastically. This article includes a discussion of implications to both managers and policymakers. |
Full text from SpringerLink