Lithography Model Calibration Via Cache-Based Niching Genetic Algorithms

A set of original model candidates are first divided into groups of original model candidates. Child model candidates are generated by performing crossover on each of the groups of original model candidates without mutation. From the original model candidates and the child model candidates, a set of new model candidates are derived, which includes: selecting a group of new model candidates from each group of the original model candidates and the corresponding child model candidates, selecting an additional new model candidate if adding the additional new model candidate increases overall diversity, and performing niche clearing to keep a number of the new model candidates in each of niches from exceeding a maximum number. The dividing, generating and deriving operations are then iterated. Model caching may be performed by restricting the crossover to the model term level or above.