Mask Optimization Model Based on Improved MobileNetV2
ZHANG Yu
XU Hui
Abstract：To address the problem that mask optimization accuracy and efficiency were difficult to balance under advanced process nodes,an end-to-end inverse lithography mask optimization model based on the improved lightweight mobile neural network version 2(MobileNetV2)was proposed.Firstly,MobileNetV2 was adopted as the backbone network of the model,giving full play to its advantages of small parameter size and efficient feature extraction,which could effectively adapt to the processing requirements of large-scale integrated circuit data.Secondly,coordinate attention(CA)modules were embedded multiple times in the feature extraction stage,through which spatial position information and channel features were fully fused,effectively enhancing the model′s ability to model mask edge details and target structures.Finally,a content-aware reassembly of features(CARAFE)module was integrated into the decoder part,enabling efficient and dynamic feature reassembly,which effectively alleviated the problems of information loss and edge blurring in traditional upsampling.The results showed that compared with the neural inverse lithography technology(Neural-ILT),attention-accelerated inverse lithography technology(A2-ILT),and differentiable optical proximity correction(DiffOPC)models,the improved MobileNetV2 model significantly reduced the turnaround time by 97.3%,94.2%,and 96.1%respectively while decreasing the process variation band index.The model was able to balance accuracy,efficiency,and process adaptability,effectively meeting the practical needs of integrated circuit mask design and manufacturing.
Keywords：mask optimizationinverse lithographyend-to-endMobileNetV2coordinate attentioncontent-aware reassembly of features
Publication Date：2025-12-30
Online Publishing Date：2025-12-10(First online date of this platform, not the publication date of the document)
Pages：5( 541-545 )