Bayesian inference for multi-label classification for root cause analysis and probe card maintenance decision support and an empirical study

Probe cards have been employed as intermediary tools between the wafers and automatic test equipment to prevent defect dies from entering the packaging stage to reduce the consumer risk and extra losses. To enhance the partnership, probe card manufacturer is responsible for troubleshooting of the sold probe cards in short time as part of after-sales maintenance to enhance the productivity. In practice, maintenance engineers relied on domain knowledge for probe card maintenance. However, owing the continuously migration of advanced technologies for semiconductor manufacturing and exponentially increasing product complexity, probe card maintenance has become challenging and time-consuming. Most of the existing studies have not addressed the root cause analysis problem for identifying corrective actions for abnormal symptoms, nor considered the effectiveness of maintenance. The selection of appropriate maintenance actions is crucial in root cause analysis. To fill the gaps, this study integrates domain knowledge and data-driven approaches to develop a smart maintenance solution based on Bayesian inference for multi-label classification to derive effective suggestions to enhance after-sales service quality for probe cards. An empirical study was conducted in a leading probe card company for validation. The results have shown practical viability of the developed solution to effectively and efficiently generate a number of maintenance recommendations for the engineers to improving troubleshooting efficiency and service quality while reducing maintenance time and machine downtime.