BBDM-ILT: Brownian bridge diffusion model for inverse lithography technology

BackgroundWith the shrinking of the technology node of integrated circuits, lithography and optical proximity correction (OPC) requirements continue to become more challenging in terms of compute requirements, fidelity, and process window. Conventional model–based OPC methods face difficulties in performance degradation and large process variation, whereas conventional inverse lithography technology (ILT) methods suffer from computational cost.AimTo balance time consumption with better mask fidelity and a wider process window, we proposed a framework integrating ILT methods into deep learning methods.Approachwe proposed an end-to-end deep learning model called the Brownian bridge diffusion model (BBDM) combined with the ILT methodology to compensate for the nonideal performance of OPC and excessive time consumption of ILT. We train our model with a loss function that can help suppress redundant sub-resolution assist features to help improve the mask generated by our model.ResultsQuantitative results show that our BBDM-ILT achieves better L2 and process variation band metrics in only 0.082 s turnaround time.ConclusionsThis model concentrates on lower turnaround time, better mask fidelity, and wider process window, which can help the industry apply the ILT model to manufacture.