Photomask correction method and apparatus and training method of layout machine learning model

Abstract

The present application discloses a double pattern exposure photomask correction method, a photomask correction device and a layout machine learning model training method. The double pattern exposure photomask correction method includes the following steps. A photomask layout target image is obtained. The photomask layout target image is decomposed into two photomask layout sub-images. The photomask layout sub-images overlap in a connection area. A layout machine learning model analyzes the size of the connection area according to the photomask layout target image. The layout machine learning model is established according to a three-dimensional information after etching. An optical proximity correction program is performed on the photomask layout sub-images.

Description

TECHNICAL FIELD

[0001] The present application relates to a double patterning photomask correction method, a double patterning photomask correction device and a training method of a layout machine learning model, and particularly relates to a double patterning photomask correction method, a double patterning photomask correction device and a training method of a layout machine learning model. BACKGROUND

[0002] Photolithography is a very important step in semiconductor manufacturing process. In photolithography, a photo mask layout is defined on a photo mask to define exposed regions and unexposed regions, so as to develop a predetermined photoresist layout on a photoresist layer. Then in etching manufacturing process, the photoresist layout can be used to etch a circuit. However, the etched circuit often has defects such as broken lines, short circuits, bridging, etc., and the photo mask layout needs to be corrected by optical proximity correction (OPC). The goal of optical proximity correction is to make the etched circuit as consistent as possible with the predetermined circuit pattern and without any defects. SUMMARY

[0003] The present application relates to a double patterning photomask correction method, a double patterning photomask correction device and a training method of a layout machine learning model, and particularly relates to a double patterning photomask correction method, a double patterning photomask correction device and a training method of a layout machine learning model.

[0004] According to a first aspect of the present application, a double patterning photomask correction method is provided. The double patterning photomask correction method comprises the following steps. A photo mask layout target pattern is obtained. The photo mask layout target pattern is decomposed into two photo mask layout sub-patterns. The photo mask layout sub-patterns overlap at a junction region. A layout machine learning model analyzes the size of the junction region according to the photo mask layout target pattern. The layout machine learning model is established according to an etched three-dimensional information. An optical proximity correction (OPC) procedure is performed on the photo mask layout sub-patterns.

[0005] According to a second aspect of the present disclosure, a double patterning photomask correction device is provided. The double patterning photomask correction device comprises an input unit, a decomposition unit, a layout machine learning model, and an optical proximity correction unit. The input unit is configured to obtain a photomask layout target pattern. The decomposition unit is configured to decompose the photomask layout target pattern into two photomask layout sub-patterns. The photomask layout sub-patterns overlap at a junction region. The layout machine learning model is configured to analyze a size of the junction region based on the photomask layout target pattern. The layout machine learning model is trained based on a post-etch 3D information. The optical proximity correction unit is configured to perform an optical proximity correction procedure on the photomask layout sub-patterns.

[0006] According to a third aspect of the present disclosure, a method for training a layout machine learning model is provided. The method for training a layout machine learning model comprises the following steps. A photomask layout target pattern is obtained. The photomask layout target pattern is decomposed into two photomask layout sub-patterns. The photomask layout sub-patterns overlap at a junction region. Two sub-photomasks are obtained based on the photomask layout sub-patterns. An exposure development procedure and an etching procedure are performed on a training sample using the sub-photomasks. A post-etch 3D information of the training sample is obtained. The layout machine learning model is trained based on the photomask layout target pattern, a size of the junction region, and the post-etch 3D information.

[0007] For better understanding of the above objects and other aspects of the present disclosure, embodiments will be described in detail with reference to the following figures. There are, however, merely shown preferred embodiments of the present disclosure, and the present disclosure is not limited thereto. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 A schematic diagram of a double patterning operation;

[0009] Figure 2 A schematic diagram of an expanded junction region;

[0010] Figure 3 A schematic diagram of incorporating post-etch 3D information into double patterning photomask correction;

[0011] Figure 4 A schematic diagram of a double patterning photomask correction device according to an embodiment;

[0012] Figure 5 A flowchart of a double patterning photomask correction method according to an embodiment;

[0013] Figure 6 A schematic diagram of each step of Figure 5

[0014] Figure 7 ​A flowchart of a training method of a layout machine learning model according to an embodiment;

[0015] Figure 8 For illustration Figure 7 of the steps. DETAILED DESCRIPTION

[0016] As the wiring density gradually increases, the difficulty of exposure and development is also increasing. Therefore, researchers have proposed a double patterning technology. In the double patterning technology, a first exposure / development / etching is performed by a photomask, and then a second exposure / development / etching is performed by another photomask.

[0017] Please refer to Figure 1 for an explanation of the operation of double patterning. The photomask layout target graph MK10 is a predetermined wiring pattern. First, the photomask layout target graph MK10 is divided into a photomask layout subgraph MK11 and a photomask layout subgraph MK12. The photomask layout subgraph MK11 and the photomask layout subgraph MK12 overlap in a stitch region SR10. The stitch region SR10, for example, has a predetermined size, or the stitch region SR10, for example, is set according to a predetermined rule.

[0018] As shown in Figure 1 , the photomask layout target graph MK10 has a minimum line distance D10, the photomask layout subgraph MK11 has a minimum line distance D11, and the photomask layout subgraph MK12 has a minimum line distance D12. Compared with the photomask layout target graph MK10, the wiring density of the photomask layout subgraph MK11 is reduced by half, so the minimum line distance D11 is significantly enlarged. Compared with the photomask layout target graph MK10, the wiring density of the photomask layout subgraph MK12 is reduced by half, so the minimum line distance D12 is significantly enlarged. In this way, the exposure / development / etching operation can be easily completed even if the wiring density is increased.

[0019] The photomask layout subgraph MK11 can be corrected by an optical proximity correction (OPC) program and simulate a two-dimensional profile OPC11 after etching. The photomask layout subgraph MK12 can be corrected by an optical proximity correction program and simulate a two-dimensional profile OPC12 after etching.

[0020] As shown in Figure 1 , the two-dimensional profile OPC11 and the two-dimensional profile OPC12 have no intersection, so it can be inferred that the wiring pattern after the two exposure / development / etching operations will not be able to be successfully connected at the dashed line R1. Therefore, it is necessary to return to expand the stitch region SR10.

[0021] Referring to Figure 2 , an example of the expanded stitching region SR10' is shown. According to the new stitching region SR10', the mask layout target map MK10 is re-segmented into two mask layout sub-maps MK11' and MK12'. The mask layout sub-map MK11' and the mask layout sub-map MK12' overlap at the stitching region SR10'.

[0022] The mask layout sub-map MK11' is corrected using an optical proximity correction procedure and simulated to obtain a post-etch two-dimensional profile OPC11'. The mask layout sub-map MK12' is corrected using an optical proximity correction procedure and simulated to obtain a post-etch two-dimensional profile OPC12'.

[0023] As shown in Figure 2 , the post-etch two-dimensional profile OPC11' and the post-etch two-dimensional profile OPC12' intersect at the dashed line R1', so it can be preliminarily inferred that the line pattern after the two exposures / developments / etchings will be able to successfully connect at the dashed line R1'.

[0024] However, referring to Figure 2 , the post-etch three-dimensional information SV1' corresponding to the cross-sectional line 2-2' is shown. The post-etch three-dimensional information SV1' is, for example, information including cross-sectional information obtained according to a scanning electron microscope (SEM). After the exposure / development / etching operation is actually completed, it can be found that a bump B1 is generated corresponding to the dashed line R1'. When the subsequent metal material is filled into the groove C1, problems such as the metal material being too thin to affect the conductive properties will occur.

[0025] Referring to Figure 3 , a schematic diagram of incorporating the post-etch three-dimensional information SV1' into the mask correction of double patterning exposure is shown. According to the above description, it is not sufficient to consider only whether the post-etch two-dimensional profile OPC11' and the post-etch two-dimensional profile OPC12' intersect to confirm the quality of the etching result. The present embodiment trains a layout machine learning model 430 using the mask layout target map MK10, the size of the stitching region SR10', and the post-etch three-dimensional information SV1', so that the stitching region in the future can be set using the layout machine learning model 430, and the quality of the etching result of double patterning exposure can be ensured.

[0026] Referring to Figure 4Fig. 4 shows a schematic diagram of a double patterning exposure photomask correction device 400 according to an embodiment. The double patterning exposure photomask correction device 400 comprises an input unit 410, a decomposition unit 420, the layout machine learning model 430 described above, and an optical proximity correction unit 440. The functions of the elements are summarized as follows. The input unit 410 is configured to provide a photomask layout target map MK40, such as a database, a storage device, a transmission line, or a wireless network module. The decomposition unit 420 is configured to decompose the photomask layout target map MK10. The layout machine learning model 430 is configured to analyze the junction region SR40 of the photomask layout submap MK41 and the photomask layout submap MK42. The optical proximity correction unit 440 is configured to perform an optical proximity correction procedure. The decomposition unit 420, the layout machine learning model 430, and / or the optical proximity correction unit 440 are, for example, a circuit, a chip, a circuit board, a program code, or a storage device storing a program code. The double patterning exposure photomask correction device 400 of the present embodiment takes into account the three-dimensional information after etching to ensure the quality of the etching result of the double patterning exposure. The operation of the elements described above is explained in more detail below in conjunction with a flowchart.

[0027] Please refer to Figures 5-6 , Figure 5 Fig. 5 shows a flowchart of a double patterning exposure photomask correction method according to an embodiment, Figure 6 illustrating Figure 5 each step. In step S510, the input unit 410 obtains a photomask layout target map MK40. The photomask layout target map MK40 has a minimum line distance D40. The minimum line distance D40 can have already exceeded the limit of the exposure development technology, so it needs to be realized by the double patterning exposure technology.

[0028] Next, in step S520, the decomposition unit 420 decomposes the photomask layout target map MK40 into a photomask layout submap MK41 and a photomask layout submap MK42. The photomask layout submap MK41 and the photomask layout submap MK42 overlap in the junction region SR40. The photomask layout submap MK41 is part of the content of the photomask layout target map MK40, so the line density of the photomask layout submap MK41 is lower, and the photomask layout submap MK41 has a minimum line distance D41. Compared with the photomask layout target map MK40, the minimum line distance D41 of the photomask layout submap MK41 is much larger than the minimum line distance D40 of the photomask layout target map MK40, so it is more conducive to the exposure development manufacturing process.

[0029] Similarly, the photomask layout sub-diagram MK42 is a part of the photomask layout target diagram MK40, and thus the line density of the photomask layout sub-diagram MK42 is lower, and the photomask layout sub-diagram MK42 has a minimum line distance D42. Compared with the photomask layout target diagram MK40, the minimum line distance D42 of the photomask layout sub-diagram MK42 is much larger than the minimum line distance D40 of the photomask layout target diagram MK40, and thus the photomask layout sub-diagram MK42 is more conducive to the exposure and development manufacturing process.

[0030] In step S520, the layout machine learning model 430 analyzes the size of the connection region SR40 according to the photomask layout target diagram MK40. As shown in FIG. 5B, the layout machine learning model 430 can analyze the size of the connection region SR40 according to the photomask layout target diagram MK40. Figure 3 As described above, the embodiment uses the layout machine learning model 430 to accurately analyze the connection region SR40. After accurately analyzing the connection region SR40, the photomask layout sub-diagram MK41 and the photomask layout sub-diagram MK42 can be accurately decomposed from the photomask layout target diagram MK40.

[0031] Then, in step S530, the optical proximity correction unit (OPC unit) 440 performs an optical proximity correction process on the photomask layout sub-diagram MK41 and the photomask layout sub-diagram MK42, and simulates the two-dimensional profile OPC41 and the two-dimensional profile OPC42 after etching. In this step, the optical proximity correction unit 440 can sequentially perform the optical proximity correction process on the photomask layout sub-diagram MK41 and the photomask layout sub-diagram MK42. Alternatively, the optical proximity correction unit 440 can simultaneously perform the optical proximity correction process on the photomask layout sub-diagram MK41 and the photomask layout sub-diagram MK42.

[0032] As shown in FIG. 5C, the two-dimensional profile OPC41 and the two-dimensional profile OPC42 intersect at the dashed line R4, and thus it can be preliminarily inferred that the line pattern after twice exposure / development / etching can be successfully connected at the dashed line R4. Figure 6

[0033] Meanwhile, please refer to FIG. 6A. Figure 6 The three-dimensional information SV4 after etching corresponding to the section line 6-6'. After the actual exposure / development / etching is completed, it can be found that the dashed line R4 has a flat bottom B4. When the subsequent metal material is filled into the groove C4, the thickness of the metal material can be maintained, and the good conductivity can be maintained.

[0034] According to the above embodiment, when the photomask layout target diagram MK40 is decomposed into the photomask layout sub-diagram MK41 and the photomask layout sub-diagram MK42, the layout machine learning model 430 refers to the three-dimensional information after etching, and thus the quality of the etching result of the double pattern exposure can be ensured.

[0035] ​The following further illustrates how the layout machine learning model 430 is trained. Please refer to Figures 7-8 , Figure 7 a flowchart of a training method of a layout machine learning model according to an embodiment, Figure 8 illustrates Figure 7 the steps. In step S710, as shown in Figure 8 , a mask layout target pattern MK80 is obtained.

[0036] Next, in step S720, as shown in Figure 8 , the mask layout target pattern MK80 is decomposed into a mask layout sub-pattern MK81 and a mask layout sub-pattern MK82. The mask layout sub-pattern MK81 and the mask layout sub-pattern MK82 overlap at a joint region SR80. The joint region SR80 is obtained, for example, according to a predetermined rule, by manual setting, or by the layout machine learning model 430.

[0037] Then, in step S730, as shown in Figure 8 , a sub-mask M81 and a sub-mask M82 are obtained according to the mask layout sub-pattern MK81 and the mask layout sub-pattern MK82.

[0038] Next, in step S740, an exposure development process and an etching process are performed on a training sample with the sub-mask M81 and the sub-mask M82. This step is, for example, a 2P2E fabrication process, that is, a first development process and a first etching process are performed with the sub-mask M81, and then a second development process and a second etching process are performed with the sub-mask M82. As shown in Figure 8 , after all the exposure development processes and etching processes are performed, the top-down two-dimensional information TV8 of the training sample can be obtained.

[0039] Then, in step S750, an etching-after three-dimensional information SV8 of the training sample is obtained. The etching-after three-dimensional information SV8 is obtained, for example, according to a scanning electron microscope.

[0040] Next, in step S760, the layout machine learning model 430 is trained according to the mask layout target pattern MK80, the size of the joint region SR80, and the etching-after three-dimensional information SV8.

[0041] According to the above embodiment, the layout machine learning model 430 is established according to the etching-after three-dimensional information SV8. The etching-after three-dimensional information SV8 not only covers the content of the top-down two-dimensional information TV8, but also covers the information of the cross section, which can more accurately set the joint region.

[0042] In conclusion, although the present application is disclosed in combination with the above embodiments, it is not intended to limit the present application. Those skilled in the art to which the present application pertains, without departing from the spirit and scope of the present application, can make various modifications and decorations. Therefore, the protection scope of the present application should be defined by the appended claims.

Claims

1. A method for double patterning photomask correction, comprising: obtaining a photomask layout target pattern; decomposing the photomask layout target pattern into two photomask layout sub-patterns, the two photomask layout sub-patterns overlapping at a junction region, wherein a layout machine learning model analyzes a size of the junction region based on the photomask layout target pattern, the layout machine learning model being trained based on post-etch 3D information, the post-etch 3D information including top-down 2D information and cross-sectional information; and performing an optical proximity correction procedure on the two photomask layout sub-patterns.

2. The method for double patterning photomask correction of claim 1, wherein the post-etch 3D information is obtained based on a scanning electron microscope.

3. The method for double patterning photomask correction of claim 1, wherein in the step of performing the optical proximity correction procedure on the two photomask layout sub-patterns, the two photomask layout sub-patterns are sequentially performed the optical proximity correction procedure.

4. The method for double patterning photomask correction of claim 1, wherein in the step of performing the optical proximity correction procedure on the two photomask layout sub-patterns, the two photomask layout sub-patterns are simultaneously performed the optical proximity correction procedure.

5. An apparatus for double patterning photomask correction, comprising: an input unit configured to obtain a photomask layout target pattern; a decomposition unit configured to decompose the photomask layout target pattern into two photomask layout sub-patterns, the two photomask layout sub-patterns overlapping at a junction region; a layout machine learning model configured to analyze a size of the junction region based on the photomask layout target pattern, the layout machine learning model being trained based on post-etch 3D information, the post-etch 3D information including top-down 2D information and cross-sectional information; and an optical proximity correction unit configured to perform an optical proximity correction procedure on the two photomask layout sub-patterns.

6. The apparatus for double patterning photomask correction of claim 5, wherein the post-etch 3D information is obtained based on a scanning electron microscope.

7. The apparatus for double patterning photomask correction of claim 6, wherein the optical proximity correction unit sequentially performs the optical proximity correction procedure on the two photomask layout sub-patterns.

8. The apparatus for double patterning photomask correction of claim 6, wherein the optical proximity correction unit simultaneously performs the optical proximity correction procedure on the two photomask layout sub-patterns.

9. A method for training a layout machine learning model, comprising: obtaining a photomask layout target pattern; decomposing the photomask layout target pattern into two photomask layout sub-patterns, the two photomask layout sub-patterns overlapping at a junction region; obtaining two sub-photomasks based on the two photomask layout sub-patterns; performing an exposure development procedure and an etching procedure on a training sample based on the two sub-photomasks; obtaining post-etch 3D information of the training sample, the post-etch 3D information including top-down 2D information and cross-sectional information; and training the layout machine learning model based on the photomask layout target pattern, the size of the junction region, and the post-etch 3D information.

10. The method for training a layout machine learning model of claim 9, wherein the post-etch 3D information is obtained based on a scanning electron microscope. ​

Images