New design of SPC control limit setting flow for super large amount of engineering data

摘要

SPC (Statistic Process Control) Control charts are widely used to establish and maintain statistical process control in many domains for process performance shift detection. But the assumption of process data distribution is a potential risk to cause unexpected alarm ratio for general Shewhart Control Chart when we use inadequate control limit setting method. Special in semiconductor manufactory practice, we found many process engineering data do not follow normal distribution. So one by one to analyzes and data transmit to make it standardizing to normal is hard to execute in practice for time and manpower concerns, because of the very large amount of engineering data count. Therefore, in this paper we propose to use different control methods or control chart for the different data types, rather than do data transmit. Basically, 4 types of data distribution are mainly focused: fix valued data (just only one constant value occurs in normality), discrete distribution, normal distribution and continuous non-normal distribution. We design to use data level counts and normal distribution test to split them base on some statistical methods and engineering concerns. Depending on the result of classification, we assign the suitable quality control method for each class. Of course, the false alarm rate is considered to balance all data types into comparable level.