Study On Prediction Models For Integrated Scheduling In Semiconductor Manufacturing Lines

Lu Guo, Jinghua Hao, Min Liu

Quality prediction of lot operations is significant for integrated scheduling in semiconductor production line. The modeltraining algorithm needs to be fast and incremental to satisfy the online applications where data comes one by one or chunk by chunk. This paper presents a novel prediction model referred to as Incremental Extreme Least Square Support Vector Machine (IELSSVM), which transforms the data into ELM feature space and then minimizes the structural risk like LSSVM. The transformation into ELM feature space can be regarded as a good dimensionality reduction. The incremental formula is proposed for on-line industrial application to avoid retraining when data comes one by one or chunk by chunk. Detailed comparisons of the IELSSVM algorithm with other incremental algorithms are achieved by simulation on benchmark problems and real overlay prediction problem of lithography in semiconductor production line. The results show that IELSSVM has better performance than other incremental algorithms like OS-ELM.