ABSTRACT
This chapter discusses how we can treat and analyze multiple variables and outputs in statistical analyses. Important applications are the calculation of sensitivities and correlations from MC results. It focuses on an overview and several specific methods related to variation-aware design. Designers often ask for the sensitivities of the different performances to the different “design” variables, like transistor width of M1 or to sheet resistance or temperature, etc. The simplest multivariate analysis is a linear regression, so the result allows a linear approximation of the circuit behavior. The effort in model creation depends highly on the number of variables and nonlinearities. An immediate application of a multivariate analysis is finding the sensitivities and the relative contributions of a design.
