Abstract
Semiconductor industry develops a series of technology roadmap, for example, International Technology Roadmap for Semiconductors(ITRS) to address rapid changes driven by Moore’s Law. Yield is a factor of keeping competition for the capital-intensive semiconductor companies. In particular, Probe card is a key component to test functionality of dies, that is for the reliebility &. For the drilling process of probe cards manufacturing, tolerances should be tightened to meet higher quality requirement to solve smaller critical dimension and bigger sizes in wafers than products of previous generations. Therefore, this thesis aims to develop a big data mining analytics and quality prediction framework for drilling manufacturing in probe card industry. The framework integrates stepwise regression, Self-Organizing Map, and Multivariate Adaptive Regression Splines method and construct a quality prediction model to explore impact factors in drilling process and provide suggestion to improve drilling quality and yield. Under the framework, probe card technology and concept can be more suitable for ITRS. An empirical study to validate the model, which cooperates with a Taiwanese probe card manufacturing factory, is provided in this thesis. The framework extracts critical parameters among multiple quality indices from a great amount of equipment data. The model provides quality prediction according to product situation and set the validation index. The results indicate that some drilling parameters will affect quality, and demonstrate the adjust process for operators. In addition, the enterprise can accelerate the judgment time for yield by importing the prediction model. Virtual metrology, real time control, and industry 4.0 are the future research directors by combining several sensor data and entire data collection with the model.