Abstract
In wafer fabrication, data is collected and analyzed to prevent process deviations that could affect product quality and wafer yield. However, the high-dimensional, sparse, and imbalanced nature of the data poses significant challenges to yield and quality root cause analysis. Deep Topological Data Analysis (DTDA) is an unsupervised machine learning method that clusters and models the data in the form of geometric objects such as graphs and their higher-dimensional versions. This method reduces the multidimensional dataset to two-dimensional networks or graphs, where each node represents a cluster of samples with similar characteristics, and an edge represents the presence of overlapping characteristics between the connecting nodes. DTDA provides insights into the necessary data elements required to conduct accurate analysis and helps engineers identify the features contributing to yield and quality issues, enabling corrective actions. Moreover, the approach prevents the waste of engineering resources and mitigates the impact on final manufacturing cost.
