Mask optimization based on block non-uniform grid convolution

A framework of optimizing mask of large size is developed. By partitioning the mask into blocks and processing each block with adaptive grid size, acceleration of 2D convolution and thus of mask optimization is achieved with tolerable accuracy loss. Simulation results show its correctness and flexibility in trade-off between speed and accuracy. The method in this paper can be theoretically proved and incorporated into a variety of inverse lithography frameworks, especially for those gradient-based.