Semiconductor Manufacturing Companies In The "bottleneck" Of Equipment Failure Prediction

In semiconductor manufacturing, equipment performance will significantly affect total capacity of a factory. Since equipments in semiconductor manufacturing are too much expensive, properly making preventive maintenance plan which will lead to equipment down time reduction and equipment availability improvement is an effective way to increase factory capacity and reduce cost.In this paper, the bottleneck in module level and equipment level which is lower than module was defined by presenting some key indicators which can scientifically evaluate equipment performance and quantifying each module performance in producing line based on real manufacturing environment in a semiconductor assembly and test plant.Then the equipment down time analysis for bottleneck was discussed in detail. A new indicator, machine failure, was proposed to estimate equipment fault, and the historical data of machine failure was studied. The extreme points in series were all replaced by the third quartile in box plot, and a novel 8 orders moving average method was used to get zero-mean stationary series. Time series AR model was developed and prediction was made. Results show that time series AR model is suitable for transformed equipment down time prediction with mean absolute error(MAE) controlled within 2.73%.In the meanwhile, the equipment failure mode was analyzed by statistical method. According to Inductive Reasoning theory, mean and median justification method and cumulative equipment failure frequency sort method were proposed to forecast equipment failure respectively.Finally, two validation experiments were designed and the equipment down time prediction and equipment failure mode forecasting methods were applied in this semiconductor assembly and test plant to check their accuracy and effectiveness. After 7 months design of experiment(DOE), the results show that methods proposed by this paper have the capability to correctly predict equipment down time and its accordant failure mode with 70% accuracy. Further more, in DOE period, these prediction methods did contribute to equipment performance improvement with equipment down time reduced by 14.8 minutes and machine failure reduced by 2.62%.