Maintenance Strategy of Semiconductor Equipment under Limited Data Conditions: A Survival Analysis Approach

This study addresses the growing need for precision maintenance in increasingly complex semiconductor manufacturing environments, where real-time event and condition monitoring data are often unavailable. To overcome this limitation, we employ a survival analysis-based approach that leverages historical maintenance records to estimate hazard functions and derive survival functions for individual equipment components. This enables the development of replacement criteria based on survival probability thresholds without relying on real-time condition monitoring. Under constraints of limited data availability, we fit parametric distributions to observed replacement intervals, select the best unimodal model via the Kolmogorov-Smirnov test, and, when necessary, approximate bimodal behavior using Gaussian mixture models. From these survival curves, we propose three replacement criteria that balance early and delayed maintenance risks. Scenario analysis demonstrates that applying the proposed criteria can achieve up to a 40 % reduction in maintenance costs compared to conventional policies. The findings validate the efficacy of our method in supporting maintenance decision-making, reducing unnecessary component replacements, and maintaining process stability. By offering a practical and scalable alternative to data-intensive maintenance strategies, our framework enhances operational efficiency and reliability in semiconductor manufacturing—and serves as a model for predictive maintenance under limited data availability in other high-value process industries.