Abstract
AbstractThe semiconductor manufacturing process involves long and complex activities, with intensive use of resources. Producers compete through the introduction of new technologies for increasing yield and reducing costs. So, yield improvement is becoming increasingly important since advanced production technologies are complex and interrelated. In particular, Wafer Bin Maps (WBMs) presenting specific fault models provide crucial information to keep track of process problems in semiconductor manufacturing. Production control is often based on the “judgement” of expert engineers who, however, carry out the analysis of map templates through simple visual exploration. In this way, existing studies are subjective, time consuming, and are also limited by the capacity of human recognition. This study proposes a network-based data mining approach, which integrates correlation graphs with clustering analysis to quickly extract patterns from WBMs and then bind them to manufacturing defects. An empirical study has been conducted on real production data for validating the proposed clustering algorithm, which showed a perfect correspondence between the malfunction patterns found by the algorithm and those discovered by human experts, so confirming the validity of our approach in its ability of correctly identifying actual defective patterns to help improving production yield.
Highlights
The semiconductor manufacturing process in- Semiconductor production involves lengthy and complex volves long and complex activities, with intensive use of processes, employing a significant amount of resources.resources
Chia-Yu Hsu [22] uses "clustering ensemble", in which Wafer Bin Map pattern extraction is made in order to recognize systematic defect patterns efficiently
Beyond the above clustering methods, we propose a lems of a device is to run, on an adequate number of method which belongs to the class of "graph based" clus- wafers, a bar graph representing the relative frequencies of tering methods
Summary
Abstract: The semiconductor manufacturing process in- Semiconductor production involves lengthy and complex volves long and complex activities, with intensive use of processes, employing a significant amount of resources. Chia-Yu Hsu [22] uses "clustering ensemble", in which Wafer Bin Map pattern extraction is made in order to recognize systematic defect patterns efficiently This approach integrates data transformation with k−means and particle swarm optimization (PSO) clustering algorithms, assessing results through adaptive response theory (ART) neural networks. Beyond the above clustering methods, we propose a lems of a device is to run, on an adequate number of method which belongs to the class of "graph based" clus- wafers, a bar graph representing the relative frequencies of tering methods It performs a non-exclusive (each wafer electrical failures in descending order, where the bin relacan be assigned to more than one cluster) and partial tive to the functioning chip is removed from the graph. This allows a generalization of the analysis, regardless of the individual spatial distribution of defective chips, mitigating the effects of "excursions" through the monitoring and diagnosis of the causes of failure due to extraordinary events such as incorrect operation, malfunction of machinery or contamination
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
