Estimating heights of defects in a wafer by scaling a 3D model using an artificial neural network
The present disclosure relates to a method for estimating heights of defects in a wafer. The method comprises creating an un-calibrated 3D model of a defect in a wafer, determining one or more attributes associated with the un-calibrated 3D model, transforming the un-calibrated 3D model to a calibrated 3D model, and estimating a height of the defect using the calibrated 3D model. Creating an un-calibrated 3D model corresponds to a defect present in a wafer based on a plurality of Scanning Electron Microscope (SEM) images of the defect. Transforming the un-calibrated 3D model to a calibrated 3D model uses a scaling factor corresponding to the determined one or more attributes associated with the un-calibrated 3D model. A height of the defect is estimated based on the calibrated 3D model of the defect.
1. A method comprising:
providing a plurality of Scanning Electron Microscope (SEM) images of a defect in a wafer;
determining an un-calibrated 3D model corresponding to the defect, based on the plurality of SEM images of the defect;
determining one or more size attributes associated with the un-calibrated 3D model;
transforming the un-calibrated 3D model to a calibrated 3D model using a scaling factor corresponding to the determined one or more size attributes associated with the un-calibrated 3D model;
estimating a height of the defect based on the calibrated 3D model associated with the defect; and
performing a manufacturing process on the wafer based on the estimated height of the defect.
2. The method as claimed in claim 1 , wherein the transforming further comprises:
providing the determined one or more size attributes as an input to a trained model;
determining, by the trained model, the scaling factor based on the one or more size attributes; and
modifying a scale of the un-calibrated 3D model based on the determined scaling factor, wherein the calibrated 3D model is obtained based on the modified scale of the un-calibrated 3D model.
3. The method as claimed in claim 1 , further comprising:
generating a normal map based on the plurality of SEM images, wherein the normal map comprises first shape information of the defect;
generating a candidate normal map based on the calibrated 3D model, wherein the candidate normal map comprises second shape information of the defect;
comparing the first shape information and the second shape information; and
validating the calibrated 3D model based on the comparison.
4. The method as claimed in claim 1 , wherein estimating the height of the defect further comprises:
determining a maximum z-value of the calibrated 3D model; and
selecting the maximum z-value of the calibrated 3D model as the height of the defect.
5. A method comprising:
providing a plurality of Scanning Electron Microscope (SEM) images of a defect in a wafer;
determining an un-calibrated 3D model corresponding to the defect, based on the plurality of SEM images of the defect;
determining one or more size attributes associated with the un-calibrated 3D model;
performing a convolution operation based on the one or more size attributes of the uncalibrated 3D model using a convolutional neural network to obtain a scaling factor;
transforming the un-calibrated 3D model to a calibrated 3D model using the scaling factor corresponding to the determined one or more size attributes associated with the un-calibrated 3D model;
estimating a height of the defect based on the calibrated 3D model associated with the defect: and
performing a manufacturing process on the wafer based on the estimated height of the defect.
6. The method as claimed in claim 5 , wherein the transforming further comprises:
providing the determined one or more size attributes as an input to a trained model;
determining, by the trained model, the scaling factor based on the one or more size attributes; and
modifying a scale of the un-calibrated 3D model based on the determined scaling factor, wherein the calibrated 3D model is obtained based on the modified scale of the un-calibrated 3D model.
7. The method as claimed in claim 5 , further comprising:
generating a normal map based on the plurality of SEM images, wherein the normal map comprises first shape information of the defect;
generating a candidate normal map based on the calibrated 3D model, wherein the candidate normal map comprises second shape information of the defect;
comparing the first shape information and the second shape information; and
validating the calibrated 3D model based on the comparison.
8. The method as claimed in claim 5 , wherein estimating the height of the defect further comprises:
determining a maximum z-value of the calibrated 3D model; and
selecting the maximum z-value of the calibrated 3D model as the height of the defect.