Data Mining Technology In The Analysis Of Measured Data

Data analysis has become the foundation of company management and people are suffering management changes that are from empirical management to scientific decision. Data analysis is competitive strategy in many excellent companies in the word. They increase competitive advantage through data analysis. A lot of Chinese companies are suffering a plenty of difference. They need new methods to decrease cost and improve quality and design new product. Undoubtedly data analysis is the key of success.Semiconductor manufacture has a lot of technologic parameters and is very complicated. How to find the relationship through probe data is very important. Traditional data analysis methods can't fit needs of actual manufacture. Data mining can solve not only no-line difference and many parameters problem but also actual problems in manufacture. We believe that data mining will make great impact in semiconductor industrial.This article introduce business process and module frame of data mining. It also introduced association analysis and cluster analysis and anomaly detection. It mainly introduce classification include common algorithms such as decision tree and artificial neural network. Article has analyzed faults and goodness of each algorithm. At same time, It analyzed over-fitting problem which occur at growing tree process. Article detailedly introduced pre-pruning and post-pruning technology in order to solve over-fitting problem. In the article evaluation the performance of a classifier and methods for comparing classifiers are also introduced.In one section of article data mining is used to solve two problems suffering in actual manufacture. One is that the yield of transistor's high temperature cutting current is low, another is that the control of ts test can't be precise because ts can't be test automatically. It is common for two problems that the parameter can't be test directly in product line and need indirectly control. Through systemic comparing deferent algorithms, we select the fittest algorithms to find the strong correlation between it and other parameters which can be control directly in product line. We introduce data mining method to probe data analysis through two cases.