Modeling CMOS gate characteristics using independent component analysis

Independent component analysis, a technique for blind source separation is applied to E-test parameter data from a CMOS manufacturing process. Models of variance for channel transistors are developed using the assumption that discrete steps in the process introduce statistically independent variance components through particular physical mechanisms. We calculate confidence intervals for the models, eliminating marginally significant components and chart the percentage of parameter variance explained by each significant component. The isolated variance sources are easily interpretable based on device physics.