Method of verifying error of optical proximity correction model

A method of fabricating a semiconductor device comprising:

• generating an optical proximity correction (OPC) model;
• generating an error prediction model for the OPC model;
• selecting a plurality of first test areas from first layout data including design information of semiconductor devices, a plurality of second test areas from second layout data obtained by applying the OPC model to the first layout data, and a plurality of third test areas from simulation data obtained by executing a simulation based on the second layout data;
• obtaining a plurality of first test images, a plurality of second test images, and a plurality of third test images from the plurality of first test areas, the plurality of second test areas, and the plurality of third test areas, respectively;
• obtaining a plurality of test data by blending the plurality of first test images, the plurality of second test images, and the plurality of third test images, each test data being obtained by blending a corresponding one of the plurality of first test images, a corresponding one of the plurality of second test images, and a corresponding one of the plurality of third test images;
• and
• predicting a critical dimension of each of a plurality of test patterns disposed in the plurality of first to third test areas by inputting the plurality of test data to the error prediction model for the OPC model.

A method of fabricating a semiconductor device comprising:

• generating first layout data;
• applying a first optical proximity correction (OPC) model to the first layout data to generate second layout data;
• obtaining a plurality of pattern images from a plurality of selected areas of a substrate on which a semiconductor process is performed using a mask formed from the second layout data;
• extracting a plurality of input images corresponding to the plurality of pattern images from a plurality of sample areas of the second layout data, the plurality of sample areas corresponding to the plurality of selected areas;
• generating a first error prediction model by training a machine learning model using a data set including the plurality of input images and the plurality of pattern images;
• extracting a plurality of test images from a plurality of test areas of the second layout data, different from the plurality of sample areas, and inputting the plurality of test images to the first error prediction model;
• and
• verifying an error of the first OPC model using a predictive value generated by the first error prediction model.