Mask optimization framework based on diffusion model

A mask optimization method based on the diffusion model to establish a network framework through integration with a traditional ILT solver in order to mitigate the influence of the optical proximity effect (OPE). Meanwhile, the method can greatly accelerate the optimization process when the input mask pattern is huge and complex. The method initially divides the masks into different classes and learns their features separately using a diffusion model optimized with a specific probability transfer matrix. Subsequently, the learned coarsened masks are computed by the ILT solver to obtain the optimized masks. The simulation results indicate that this method can greatly accelerate the mask optimization process when faced with a large and complex input mask pattern.