Abstract:

The equipment failure down time (or the failure rate) in a certain period of time (such as 12 hours), which is greatly influenced by the production planning and scheduling in semiconductor manufacturing factory, can not reflect the true equipment performance. Moreover, the data series of down time is not suitable for being directly used for auto-regressive moving average (ARMA) modeling because it has very strong randomness and undulatory property. An innovative metrics for equipment failure evaluation, named equipment unavailability (EU), is proposed according to this problem. When building an ARMA model, the equipment failure down time is firstly transformed to EU. Then the data is converted into stationary random sequence by outliers’ replacing, and thirdly the trend term of data is removed by using an improved moving average algorithm. So the zero mean stationary random sequence is available to build the ARMA model. The forecasting result is transformed into equipment failure probability in a certain period of time at last. The process of data pretreatment, modeling, forecasting and result transforming is realized to a software application system by using VS.NET. The application in a chipset assembly and test factory shows that the method can predict machine status with the accuracy of 70%, the equipment downtime is average reduced by 14.8 minutes and the machine unavailability is average reduced by 2.62% in one shift (12 hours).