• Acta Optica Sinica
  • Vol. 38, Issue 12, 1222002 (2018)
Heng Zhang1、2、*, Sikun Li1、2、*, Xiangzhao Wang1、2、*, and Wei Cheng1、2
Author Affiliations
  • 1 Laboratory of Information Optics and Optelectronic Technology, Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Shanghai 201800, China
  • 2 University of Chinese Academy of Sciences, Beijing 100049, China
  • show less
    DOI: 10.3788/AOS201838.1222002 Cite this Article Set citation alerts
    Heng Zhang, Sikun Li, Xiangzhao Wang, Wei Cheng. 3D Rigorous Simulation of Defective Masks used for EUV Lithography via Machine Learning-Based Calibration[J]. Acta Optica Sinica, 2018, 38(12): 1222002 Copy Citation Text show less
    Schematic of EUVL 3D mask. (a) 3D view; (b) side view
    Fig. 1. Schematic of EUVL 3D mask. (a) 3D view; (b) side view
    Schematic of mask simulation model. (a) Absorber model; (b) multilayer model
    Fig. 2. Schematic of mask simulation model. (a) Absorber model; (b) multilayer model
    Flow chart of training and predicting process employed in machine learning methods
    Fig. 3. Flow chart of training and predicting process employed in machine learning methods
    Comparison of simulation results for fast models with and without model parameter calibration using different machine learning methods
    Fig. 4. Comparison of simulation results for fast models with and without model parameter calibration using different machine learning methods
    Comparison of the simulation accuracy of a parameter calibrated fast model and an advanced single-surface approximation model
    Fig. 5. Comparison of the simulation accuracy of a parameter calibrated fast model and an advanced single-surface approximation model
    Comparison of the aerial images of mask simulations. (a) Rigorous simulation of a defect-free mask; (b) rigorous simulation of defective uncompensated mask; (c) defect compensation using a rigorous model; (d) defect compensation using a fast model
    Fig. 6. Comparison of the aerial images of mask simulations. (a) Rigorous simulation of a defect-free mask; (b) rigorous simulation of defective uncompensated mask; (c) defect compensation using a rigorous model; (d) defect compensation using a fast model
    Data indexMask parameterModel parameter
    Mask pitch /nmContact size /nmhtop /nmwtop /nmhbot /nmwbot /nmgbest
    1176912520301.104
    21447061920421.082
    312860101923241.073
    582144672158241.069
    583176894104129.802
    5841769281120441.098
    Table 1. Training data for different masks
    Data indexMask pitch /nmContact size /nmhtop /nmwtop /nmhbot /nmwbot /nmModel parameter
    OriginalKNNBest
    11769361210211.101.191.23
    217689101818251.101.151.15
    317683898371.101.101.10
    48144809711351.101.091.08
    49200956628331.101.121.10
    501608351011231.101.101.10
    Table 2. Testing data for different masks
    Method nameRMS meanRMS medianRMS minRMS maxRMS SD deviation
    DT2.191.680.5410.21.70
    KNN1.821.550.235.481.09
    RF2.101.560.558.231.68
    Original3.322.530.539.772.58
    Best1.070.980.164.450.62
    Table 3. Comparison of simulation accuracy of different parameter calibration methods
    Heng Zhang, Sikun Li, Xiangzhao Wang, Wei Cheng. 3D Rigorous Simulation of Defective Masks used for EUV Lithography via Machine Learning-Based Calibration[J]. Acta Optica Sinica, 2018, 38(12): 1222002
    Download Citation