Parameter variation tolerant method for circuit design optimization

Abstract

A method for optimizing the design of a chip or system by decreasing the cost function that encompasses a plurality of constraints in the presence of variations in the design parameters is described. The method makes use of numerical optimization, simulated annealing, or any other objective-driven optimization means, and accounts for uncertainties in the modeling of the design variables and functions. A significant reduction in the number of design constraints which are violated at the end of an optimization process is achieved, even when all the design constraints cannot be satisfied. The optimization also reduces the cycle time at which the design operates and limits the increase in the minimum operational cycle time of a particular implementation in the presence of variations that cannot be modeled or unpredictable variations in delay introduced by elements of the design. The method for optimizing the design includes the steps of: defining an objective function computed from variables and functions of the design of the chip or system; deriving a merit function from the objective function by adding to it a plurality of separation terms; and minimizing the merit function which reduces the expected value of the objective function when confronted with significant variations of the design variables and functions.