Plasmonic lithography fast imaging model based on the decomposition machine learning method under arbitrary illumination system.

Fast imaging models are the cornerstone of computational lithography technology. Addressing the limitation that existing fast imaging models for surface plasmonic lithography (SPL) struggle to handle complex illumination conditions, this paper proposes a general fast imaging model based on the decomposition machine learning method, which is applicable to arbitrary illumination systems. First, the model utilizes rigorous electromagnetic field (EMF) simulation to construct a complete training library containing 81 reference point sources. By training the imaging transfer matrix (ITM), it achieves fast mapping of mask features. Furthermore, an approximation method based on inverse spatial distance weighting is proposed to achieve high-precision prediction for arbitrary non-reference point sources and partially coherent illumination systems. Simulation experiments demonstrate that the model exhibits excellent robustness under both TE and TM polarization states. The predicted photoresist images (PRI) are highly consistent with the calculation results of rigorous EMF simulation, with the root mean square error (RMSE) consistently controlled within 0.075. In terms of computational efficiency, for reference point sources within the library and partially coherent illumination based on them, the calculation speed of the model is improved by 30 to 60 times compared to rigorous simulation; for arbitrary non-reference point sources requiring approximation calculation and partially coherent illumination based on them, the calculation speed is improved by 7 to 26 times. This work significantly reduces computational costs while guaranteeing sub-wavelength imaging accuracy, providing a key efficient simulation tool for source-mask optimization (SMO) and optical proximity correction (OPC) in plasmonic lithography.