Reducing RF Parameter Tests for Test Cost Reduction: Device Model, Hypothesis Testing and Experiment Verification

摘要

This paper discusses the approach of reducing RF parametric test by using statistical techniques. Linear regression and logistic regression are used to investigate the relationship between the yield and defect level as well as the correlation among parametric tests. The analysis is then performed on manufacturing data collected from an low noise amplifier (LNA) product built using IBM Silicon Germanium (SiGe) technologies. Separate data sets were used for model training and model validation. During the validation process, the model was checked for its predictive capabilities and its impact on both the yield and defect level. Results show that eliminating all the RF testing will negatively impact the yield, the defect level or both. On the other hand, removing some RF tests and keeping a few selected others can successfully reduce test time without having a negative impact on the defect level. Based on the data sets analyzed, it was concluded that parametric tests should be carefully designed based on device physics in order to reduce the number of parametric tests required.