Comparative Assessment of Yield Learning Tools Using Information Theory

A model based on information theory, which allows yield managers to choose the optimal strategies for yield management in microelectronic manufacturing, is presented. The data reduction rate per experimentation cycle and data reduction rate per unit time serve as benchmarking metrics for yield learning. These newly defined metrics enable managers to make objective comparisons of apparently unrelated technologies. Four yield analysis tools -electrical testing, automatic defect classification, spatial signature analysis and wafer position analysis -are examined in detail to suggest an optimal yield management strategy for both the R&D and volume production environments.