A design methodology for analog circuits based on global optimization will be presented. The focus of the talk will be on an efficient strategy for global optimization problems that occur when describing circuits and on the careful modeling of the circuit components used in circuits.

The optimization strategy is based on convex optimization with non-convex extensions, in order to separate the computationally "hard" constraints from the computationally "easy" constraints. Convex optimization is very efficient, but severly limits the type of design constraints that can be expressed. A branch-and-bound optimizer on top of a convex optimizer provides the flexibility needed to handle non-convex constraints. The transistor model that will be described is based on the charge sheet model, but simplified for the different regions of analog operation.