Corner rounding method of OPC pattern based on deep learning, and OPC method and mask manufacturing method including the corner rounding method

1 . A method of training a transform model to perform a corner rounding method of a deep learning-based optical proximity correction (OPC) pattern, the method comprising:

obtaining a contour of at least one of a photoresist (PR) pattern or an etching pattern on a wafer;

obtaining a square layout of the at least one PR pattern or etching pattern, corresponding to the contour; and

generating a transform model through deep learning with the square layout and the contour such that the transform model is configured to output a rounded layout target with respect to a square layout target.

2 . The method of claim 1 , wherein the deep learning uses a deep convolutional generative adversarial network (DCGAN),

wherein an input of the DCGAN is the square layout, and an output of the DCGAN is the contour, and

wherein the generating the transform model includes training the transform model until the output contour is same as a reference image corresponding to the contour.

3 . The method of claim 2 , wherein the contour is obtained by using a mean value of a plurality of scanning electron microscope (SEM) images of the at least one PR pattern or etching pattern.

4 . The method of claim 2 , wherein the square layout includes square edges and is obtained by performing Manhattanization on the contour.

5 . The method of claim 1 , wherein the PR pattern corresponds to an after develop inspection (ADI) pattern, and

wherein the etching pattern corresponds to an after clean inspection (ACI) pattern.

6 . An optical proximity correction (OPC) method comprising:

generating a retarget layout including straight edges;

performing OPC on the retarget layout such that an OPCed layout is generated;

performing an optical rule check (ORC) on the OPCed layout;

determining whether there is a defect in the OPCed layout based on a result of the ORC;

determining the OPCed layout to be a final OPCed layout based on the determination that there is no defect; and

etching a semiconductor element based on the final OPCed layout,

wherein at least one of the retarget layout or the OPCed layout is generated based on a corner rounding of a deep learning-based OPC pattern,

wherein the corner rounding of the deep learning-based OPC pattern comprises:

using a first transform model generated through deep learning with a first square layout of an etching pattern and a first contour of the etching pattern, wherein the first square layout corresponds to the first contour, or

using a second transform model generated through the deep learning with a second square layout of a photoresist (PR) pattern and a second contour of the PR pattern,

wherein the second square layout corresponds to the second contour.

7 . The OPC method of claim 6 , wherein the generating the retarget layout includes the corner rounding of the deep learning-based OPC pattern and comprises:

obtaining a first rounded layout target corresponding to a first square layout target by using the first transform model, and

wherein the first transform model is generated by

obtaining the first contour,

obtaining the first square layout, and

generating the first transform model such that the first transform model is configured to output the first rounded layout target with respect to the first square layout target.

8 . The OPC method of claim 7 , wherein the deep learning uses a deep convolutional generative adversarial network (DCGAN),

wherein an input of the DCGAN is the first square layout, and an output of the DCGAN is the first contour.

9 . The OPC method of claim 8 , wherein the first contour is obtained using a mean value of scanning electron microscope (SEM) images of the etching pattern on a wafer, and

wherein the first square layout includes square edges and is obtained by performing Manhattanization on the first contour.

10 . The OPC method of claim 7 , wherein the etching pattern corresponds to an after clean inspection (ACI) pattern, and

wherein the generating the retarget layout comprises

performing an inverse correction on the first rounded layout target such that a curvilinear after develop inspection (ADI) layout target is generated, and

performing Manhattanization on the curvilinear ADI layout target such that the retarget layout is generated.

11 . The OPC method of claim 6 , wherein the generating the OPCed layout includes the corner rounding of the deep learning-based OPC pattern and comprises:

obtaining a second rounded layout target corresponding to a second square layout target by using the second transform model, and

wherein the second transform model is generated by

obtaining the second contour,

obtaining the second square layout, and

generating the second transform model such that the second transform model is configured to output the second rounded layout target with respect the second square layout target.

12 . The OPC method of claim 11 , wherein the second contour is obtained using a mean value of scanning electron microscope (SEM) images of the PR pattern on a wafer, and

wherein the second square layout is obtained by performing Manhattanization on the second contour and includes square edges.

13 . The OPC method of claim 11 , wherein the PR pattern corresponds to an after develop inspection (ADI) pattern, and

wherein the generating the OPCed layout comprises performing an inverse correction on the second rounded layout target.

14 . A mask manufacturing method comprising:

generating a retarget layout including straight edges;

performing an optical proximity correction (OPC) on the retarget layout such that an OPCed layout is generated;

performing an optical rule check (ORC) on the OPCed layout;

determining whether there is a defect in the OPCed layout based on a result of the ORC; and

determining the OPCed layout to be a final OPCed layout based on the determination that there is no defect;

transmitting an image of the final OPCed layout as mask tape-out (MTO) design data;

preparing mask data based on the MTO design data; and

exposing a substrate to light, based on the mask data,

wherein at least one of the retarget layout or the OPCed layout is generated based on a corner rounding of a deep learning-based OPC pattern,

wherein the corner rounding of the deep learning-based OPC pattern comprises:

using a first transform model generated through deep learning with a first square layout of an etching pattern and a first contour of the etching pattern, wherein the first square layout corresponds to the first contour, or

using a second transform model generated through the deep learning with a second square layout of a photoresist (PR) pattern and a second contour of the PR pattern,

wherein the second square layout corresponds to the second contour.

15 . The mask manufacturing method of claim 14 , wherein the generating the retarget layout includes the corner rounding of a deep learning-based OPC pattern and comprises:

obtaining a first rounded layout target corresponding to a first square layout target by using a first transform model, and

wherein the first transform model is generated by

obtaining the first contour,

obtaining the first square layout, and

generating the first transform model such that the first transform model is configured to output the first rounded layout target with respect to the first square layout target.

16 . The mask manufacturing method of claim 15 , wherein the deep learning uses a deep convolutional generative adversarial network (DCGAN),

wherein an input of the DCGAN is the first square layout, and an output of the DCGAN is the first contour,

wherein the first contour is obtained using a mean value of scanning electron microscope (SEM) images of the etching pattern on a wafer, and

wherein the first square layout includes square edges and is obtained performing Manhattanization on the first contour.

17 . The mask manufacturing method of claim 15 , wherein the etching pattern corresponds to an after clean inspection (ACI) pattern, and

wherein the generating the retarget layout comprises

performing an inverse correction on the first rounded layout target such that a curvilinear after develop inspection (ADI) layout target is generated, and

performing Manhattanization on the curvilinear ADI layout target such that the retarget layout is generated.

18 . The mask manufacturing method of claim 14 , wherein the generating the OPCed layout includes the corner rounding of the deep learning-based OPC pattern and comprises:

obtaining a second rounded layout target corresponding to a second square layout target by using a second transform model, and

wherein the second transform model is generated by

obtaining the second contour,

obtaining the second square layout, and

generating the second transform model such that the second transform model is configured to output the second rounded layout target with respect the second square layout target.

19 . The mask manufacturing method of claim 18 , wherein the second contour is obtained using a mean value of scanning electron microscope (SEM) images of the PR pattern on the wafer, and

wherein the second square layout is obtained by performing Manhattanization on the second contour and includes square edges.

20 . The mask manufacturing method of claim 18 , wherein the PR pattern corresponds to an after develop inspection (ADI) pattern, and

wherein the generating the OPCed layout comprises performing an inverse correction on the second rounded layout target.