Image generation system for alignment error determination, alignment error determination system including same, and alignment error determination method using same

Abstract

Provided are an image generation system for alignment error determination, an alignment error determination system including same, and an alignment error determination method using same, wherein the image generation system includes an image generation unit for generating images for determining an alignment state of a head unit of an optical system. On the basis of the result of being trained on images for each alignment error of the head unit, the image generation unit designs images for analyzing alignment errors of the head unit or separating alignment errors for each degree of freedom of the head unit.

Description

Image generation system for determining alignment errors, alignment error determination system including the same, and alignment error determination method using the same

[0001] The present invention relates to an image generation system for judging alignment errors, an alignment error judgment system including the same, and an alignment error judgment method using the same, and more particularly, to an image generation system for judging alignment errors for judging alignment errors, including an alignment state between a substrate and an image or a cause of alignment errors between multiple images in an image projection optical system such as a DMD (digital micro-mirror device), an alignment error judgment system including the same, and an alignment error judgment method using the same.

[0002] In general, image projection optical systems such as DMD (digital micro-mirror device) or LCos (liquid crystal on silicon) are being used in photolithography processes by projecting images onto a substrate, as in Korean Patent Publication No. 10-2023-0087629, or in projection displays of large-area images, as in Korean Patent Registration No. 10-1199496, and their scope of use has been expanding recently.

[0003] However, in the case of the image projection optical system, the size of the image to be projected is limited, so the area of ​​the image that can be generated with a single image projection optical system is limited. Therefore, in order to implement a large-area image, a method is applied in which the images of a single image projection optical system are connected and stitched, or a method is applied in which an image is implemented by aligning a plurality of single image projection optical systems so that multiple images can be projected simultaneously.

[0004] However, when implementing a large-area image as described above, performing stitching between images or connecting images generated from multiple optical systems by overlapping each other, alignment between the optical systems must be performed accurately to ensure natural connection between the images.

[0005] However, due to various reasons, alignment between the optical systems is difficult to accurately perform, and thus, a technology for accurately determining or analyzing the alignment status and alignment error is required. However, currently, the alignment status is determined only by analyzing the overlapping images themselves, which is time-consuming and has low accuracy.

[0006] Related prior art documents include Republic of Korea Publication No. 10-2023-0087629 and Republic of Korea Registration No. 10-1199496.

[0007] Accordingly, the technical problem of the present invention is conceived from this point, and the purpose of the present invention is to provide an image generation system for determining alignment errors that can more accurately determine the alignment state including the cause of the alignment error between a substrate and an image or the alignment error between multiple images in an image projection optical system such as a DMD (digital micro-mirror device), thereby minimizing the error of the optical system.

[0008] In addition, another object of the present invention is to provide an alignment error judgment system including the image generation system.

[0009] In addition, another object of the present invention is to provide a method for determining an alignment error using the above alignment error determination system.

[0010] An image generation system according to one embodiment of the present invention for realizing the above-described object includes an image generation unit that generates an image for determining the alignment state of a head portion of an optical system. The image generation unit analyzes the alignment error of the head portion based on learning results for images for each alignment error of the head portion, or designs an image for separating the alignment error of each degree of freedom of the head portion.

[0011] In one embodiment, the alignment error of the head portion may be an alignment error with respect to a reference optical system of the head portion, an alignment error with respect to an alignment mark of the head portion, or an alignment error between a pair of adjacent head portions.

[0012] In one embodiment, the image generating unit designs at least one image as an image for analyzing an alignment error of the head unit with respect to a reference optical system, an alignment error of the head unit with respect to an alignment mark, or an alignment error between a pair of adjacent head units, and when there are multiple images, the images may be sequentially sequential images.

[0013] In one embodiment, the alignment error of the head portion may be a translational error in at least one of first to third directions that are perpendicular to each other, a rotational error in which the head portion rotates around at least one of the first to third directions, or an error that includes both the translational error and the rotational error.

[0014] In one embodiment, the image generation unit may design an image for analyzing only one alignment error among the alignment errors of the head unit, design an image for simultaneously analyzing at least two alignment errors among the alignment errors of the head unit, or design an image for separating alignment errors for each degree of freedom when analyzing at least two alignment errors.

[0015] In one embodiment, the image generation unit may include a first image collection unit that provides the learning unit with an image for analyzing only the alignment error or an image for separating the alignment error for each degree of freedom, and a first image design unit that designs an image for analyzing only the alignment error or an image for separating the alignment error for each degree of freedom based on a learning result of the learning unit.

[0016] In one embodiment, the image generation unit may include a second image collection unit that provides an image for analyzing at least two alignment errors simultaneously to the learning unit, and a second image design unit that designs an image for analyzing at least two alignment errors simultaneously based on a learning result of the learning unit and an image for separating alignment errors for each degree of freedom designed by the first image design unit.

[0017] In one embodiment, the image generating unit can be designed to vary at least one of the shape of the image, the size of the image, the gradient of the image, the asychrony of the image arrangement, and the periodicity of the image arrangement.

[0018] In one embodiment, the shape of the image may be any one of a polygon, a circle, an ellipse, and a fan.

[0019] In one embodiment, the method may further include a learning unit that learns the relationship between the image for each alignment error of the head portion and the result of a fast Fourier transform for the image for each alignment error.

[0020] In one embodiment, the image generation unit can generate an image for determining the alignment state of the head of the optical system based on the interdependence between the image for each alignment error and the fast Fourier transform result for the image for each alignment error.

[0021] In one embodiment, the image may be an image generated as a result of an inverse fast Fourier transform of an image generated as a result of the fast Fourier transform, an image generated as a result of a fast Fourier transform of an image for analyzing an alignment error of the head portion, or an image generated as a result of a fast Fourier transform of an image for separating an alignment error for each degree of freedom of the head portion.

[0022] In one embodiment, the learning unit may include an image processing unit that performs predetermined processing and transformation on the alignment error-differentiated image, and a learning execution unit that performs machine learning or deep learning on the image processed and transformed by the image processing unit.

[0023] In one embodiment, the learning unit can perform learning using a neural operators method.

[0024] Another embodiment of the alignment error determination system for realizing the above-described other object of the present invention includes the image generation system, the optical system, and the alignment determination unit. The optical system captures an image designed by the image generation system to generate a captured image. The alignment determination unit determines the alignment status of the head of the optical system based on the captured image.

[0025] In one embodiment, the optical system can be applied to a digital micro-mirror device (DMD), a liquid crystal on silicon (LCoS), a light emitting display (LED), a liquid crystal display (LCD), a nano-imprint head, an inkjet printing head, or an image modulation / non-modulation head.

[0026] In one embodiment, the system further includes an error extraction unit that extracts an alignment error of a head portion of the optical system based on the captured image, and the error extraction unit may include an FFT transform unit that performs a fast Fourier transform on the captured image.

[0027] In one embodiment, the learning unit includes an image judgment unit that regenerates an image by applying a learning result to a fast Fourier transform result of the photographed image, and the error extraction unit can extract an alignment error of the head unit based on the regenerated image.

[0028] In a method for determining an alignment error according to an embodiment of another object of the present invention, based on the learning results for images for each alignment error of the head, the alignment error of the head is analyzed or an image is designed for separating the alignment error for each degree of freedom of the head. The designed image is captured by an optical system to generate a captured image. The alignment state of the head of the optical system is determined based on the captured image.

[0029] In one embodiment, prior to the step of designing the image, the method may further include a step of learning a relationship between an image for each alignment error of the head portion and a fast Fourier transform result for the image for each alignment error.

[0030] According to embodiments of the present invention, in an optical system that performs image projection, the alignment status in the overlapping area between the image acquired by the substrate unit and the image acquired by adjacent head units can be determined in real time. Accordingly, not only can the alignment between the substrate unit and the head unit be performed more accurately based on the alignment error of the optical system, but also the error in stitching images through adjacent head units can be minimized and the alignment status can be corrected.

[0031] In particular, the judgment of the alignment state is performed based on the fast Fourier transform result of the acquired image, and conversely, an image that can more precisely and accurately express the alignment state can be designed from the fast Fourier transform (FFT) result, and by judging the alignment state based on the image designed in this way, the precision and accuracy of the judgment of the alignment state can be improved.

[0032] That is, by learning in advance the relationship between the image for each alignment error of the head section and the FFT result for each alignment error image, it is possible to design an image that is optimal for deriving a specific alignment error from the FFT result, and to make a judgment on the alignment status based on the image designed in this way.

[0033] Of course, the learning result of the above learning unit can also be utilized in the process of determining the alignment status by performing FFT transformation on the image obtained for the alignment status of the head unit using the designed image, thereby improving the speed and accuracy of determining the alignment status.

[0034] At this time, in designing the image, by designing an image that can better highlight each of the six degrees of freedom errors, or by designing an image that can most effectively highlight at least two or more of the six degrees of freedom errors or all of the six degrees of freedom errors, the alignment status of the head can be accurately or quickly determined.

[0035] In this case, since an image can be designed to separate the alignment errors of each of the above six degrees of freedom errors, a faster and more accurate judgment can be made when judging the alignment status of the head from an image having multiple degrees of freedom.

[0036] That is, based on the learning result of the above learning unit, the characteristics of the FFT result for each alignment error of the aligned image are learned, and by reverse engineering this, an image that can most easily reveal the error of a specific degree of freedom in the FFT result, an image that can separate the alignment error for each degree of freedom, or an image that can reveal multiple degrees of freedom is designed, thereby enabling a more accurate alignment state to be determined for not only a specific degree of freedom but also a composite degree of freedom. In particular, in order to quickly determine the alignment state in real time, it is necessary to design an image that can most easily reveal the errors of at least two or more or all six degrees of freedom. At this time, by designing and providing an image that can separate and extract each degree of freedom, it is possible to not only reveal the error from the composite degree of freedom as a whole, but also to individually reveal the error of each degree of freedom from the composite degree of freedom, thereby enabling a quick real-time determination of the alignment state.

[0037] At this time, the image being designed can be derived by varying the shape of the image, the size of the image, the deformation of the image, or the aperiodicity or periodicity of the image arrangement, due to the characteristics of the FFT result, so that the complexity or time taken in designing the image can be minimized.

[0038] Furthermore, in extracting errors from actual captured images, the FFT conversion results of the actual captured images are provided to the learning unit, and the image judgment unit performs so-called image regeneration to remove noise and optimize the image suitable for alignment error analysis, and the errors are extracted based on the regenerated image, so that more accurate alignment judgment is possible.

[0039] FIG. 1 is a block diagram illustrating an alignment error determination system according to one embodiment of the present invention.

[0040] Fig. 2a is a schematic diagram illustrating a state of capturing an image using the optical system of Fig. 1, and Fig. 2b is an example of images captured using the optical system of Fig. 3a.

[0041] Figures 3a to 3d are examples of images captured according to the type of alignment error of the optical system of Figure 2a.

[0042] FIG. 4a is an example of performing FFT transformation by overlapping the captured images of FIG. 2b, FIG. 4b is an example of the FFT transformation result according to the characteristics of the captured images of FIG. 1, and FIG. 4c is an example of the type of FFT transformation obtained from the captured images of FIG. 1.

[0043] Figure 5 is a flowchart illustrating the steps of processing an image in the image processing unit of the learning unit of Figure 1.

[0044] Fig. 6 is a schematic diagram showing an example of an image extracted through the image generation system of Fig. 1 and the result of the FFT transformation accordingly.

[0045] Fig. 7a is an example of designing an image by considering the characteristics of the FFT conversion result through the image generation system of Fig. 1, and Figs. 7b and 7c are graphs illustrating a state in which the result of extracting alignment errors is improved.

[0046] Figure 8 is a flowchart illustrating a method for determining an alignment error using the alignment error determination system of Figure 1.

[0047] <Explanation of symbols>

[0048] 10: Alignment error judgment system 200: Image generation system

[0049] 100: Learning Department 110: Image Learning Department

[0050] 120: Image processing unit 130: Learning execution unit

[0051] 140: Image judgment unit 300: Image generation unit

[0052] 310: 1st Image Collection Department 320: 1st Image Design Department

[0053] 330: Second image collection unit 340: Second image generation unit

[0054] 400: Optical system 410: Head

[0055] 500: Error analysis section 510: FFT conversion section

[0056] 520: Error extraction unit 600: Sorting judgment unit

[0057]

[0058] The present invention is susceptible to various modifications and takes various forms, and thus embodiments are described in detail herein. However, this is not intended to limit the present invention to a specific disclosed form, but rather to encompass all modifications, equivalents, and alternatives falling within the spirit and technical scope of the present invention. Similar reference numerals have been used to designate similar components throughout the description of each drawing. While terms such as "first," "second," etc. may be used to describe various components, these components should not be limited by these terms.

[0059] The above terms are used solely to distinguish one component from another. The terms used in this application are used solely to describe specific embodiments and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise.

[0060] In this application, it should be understood that terms such as “comprise” or “consist of” are intended to specify the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, but do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0061] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and will not be interpreted in an idealized or overly formal sense unless explicitly defined herein.

[0062] Hereinafter, with reference to the attached drawings, a preferred embodiment of the present invention will be described in more detail.

[0063] FIG. 1 is a block diagram illustrating an alignment error determination system according to one embodiment of the present invention.

[0064] First, referring to FIG. 1, the alignment error judgment system (10) according to the present embodiment includes an optical system (400), and is a system that judges the alignment state of the optical system (400) based on an image obtained from the optical system (400).

[0065] At this time, the reason why the optical system (400) and the image acquired through the optical system (400) have an alignment error will be first explained with reference to the subsequent drawings.

[0066] Fig. 2a is a schematic diagram illustrating a state of capturing an image using the optical system of Fig. 1, and Fig. 2b is an example of images captured using the optical system of Fig. 3a.

[0067] First, referring to FIGS. 1 and 2a, the optical system (400) includes a head portion (410) and a substrate portion (420), and a photographed image (430) is acquired through the head portion (410).

[0068] The above optical system (400) is a so-called image projection optical system, and can be applied to a digital micro-mirror device (DMD), a liquid crystal on silicon (LCoS), a light emitting display (LED), a liquid crystal display (LCD), etc. Alternatively, the optical system (400) can also be applied to a nano imprint head, an inkjet printing head, or an image modulation / non-modulation head. However, it is not limited to such an optical system, and can include most optical elements capable of image generation or dynamic conversion.

[0069] In the above image projection optical system (400), a predetermined projection image (i.e., a photographed image) (430) is generated for an incident light source, and the projection image (430) generated in this way is projected onto a substrate (420) and used in a photolithography process or can be used in a large-area image projection display such as a beam projector.

[0070] At this time, the projection image (430) generated through the head portion (410) of the optical system (400) is acquired through the image measurement stage (440).

[0071] Meanwhile, in the case of projection images generated through such image projection optical systems, stitching is required between multiple projection images in order to be implemented on a large area, and alignment errors may occur in the overlapping area between the projection images during such stitching. That is, when performing image stitching, the first and second images (431, 432) acquired from each of the adjacent head units (411, 412) must be overlapped with each other to perform stitching, and in this case, alignment errors may occur between the adjacent head units. In this case, the alignment error between two adjacent head units is exemplified through the drawing, but it is not limited thereto, and it is obvious that if stitching of images is required, it will occur between head units that are adjacent to each other in a plurality of head units.

[0072] Accordingly, the alignment error judgment system (10) according to the present embodiment judges the alignment error between adjacent head parts.

[0073] In addition, in addition to the alignment error when overlapping multiple projection images, an alignment error of the head unit (410) with respect to the reference optical system or the reference alignment mark in the optical system (400) may also occur, and the alignment error determination system (10) according to the present embodiment may determine the alignment error between the reference optical system or the reference alignment mark and the head unit. Furthermore, when performing a predetermined process on the substrate unit (420), an alignment error may occur in the overlapping area between the substrate unit and the projection image, i.e., the head unit, and such an alignment error may also be determined.

[0074] However, for the convenience of explanation, the following description will describe an alignment error occurring between a pair of adjacent head parts (411, 412), and it is obvious that this explanation can be extended to apply to an alignment error with the reference optical system or reference alignment mark, and further, an alignment error with respect to the substrate.

[0075] That is, referring to FIG. 2b, the first and second captured images (431, 432) captured from each of the adjacent head portions (411, 412) are not identical to each other and may include alignment errors.

[0076] An example of the alignment error between the first and second captured images (431, 432) is as shown in the drawing below.

[0077] Figures 3a to 3d are examples of images captured according to the type of alignment error of the optical system of Figure 2a.

[0078] That is, referring to FIG. 3A, the first captured image (431) and the second captured image (432) may move along the first axis (X) and the second axis (Y) perpendicular to the first axis (X), which may cause an error (translation error). At this time, although FIG. 3A illustrates that the second captured image (432) moves with respect to the first captured image (431) along both the first axis (X) and the second axis (Y), an alignment error may also occur by moving along only one of the first axis (X) and the second axis (Y).

[0079] In addition, referring to FIG. 3b, the first captured image (431) and the second captured image (432) may rotate about a third axis (Z) that is perpendicular to the first axis (X) and the second axis (Y), thereby causing an error (rotation error). Due to this rotation error about the third axis (Z), an alignment error occurs in the second captured image (432) in the form of rotation on a plane (XY plane) with respect to the first captured image (431).

[0080] In addition, referring to FIG. 3c, the first captured image (431) and the second captured image (432) may rotate about the first axis (X) as the rotation center, resulting in an error (rotation error). Due to this rotation error about the first axis (X), the second captured image (432) may have an alignment error in the form of a length decreasing in the direction of the second axis (Y) with respect to the first captured image (431), resulting in a blurry image. At this time, although not shown in FIG. 3c, the first captured image (431) and the second captured image (432) may rotate about the second axis (Y) as the rotation center, resulting in an error (rotation error).

[0081] Furthermore, referring to FIG. 3D, the first captured image (431) and the second captured image (432) may move along the third axis (Z), resulting in an error (translation error). Due to this translation error with respect to the third axis (Z), the first captured image (431) may be blurred with respect to the second captured image (432) (or vice versa).

[0082] As described above, as exemplified through FIGS. 3A to 3D, the alignment error in the overlapping region may result in an error (translation error) due to movement along at least one of the X-axis, Y-axis, and Z-axis, or an error (rotation error) due to rotation around at least one of the X-axis, Y-axis, and Z-axis.

[0083] In addition, the above-described translational and rotational errors may occur individually, and at least two of the six errors may occur simultaneously. As described above, the captured images (430) acquired from the optical system (400) may include various types of alignment errors.

[0084] Hereinafter, errors occurring in one degree of freedom are described as single-degree-of-freedom errors, and errors occurring in two or more degrees of freedom are described as compound-degree-of-freedom errors.

[0085] Accordingly, the alignment error judgment system (10) according to the present embodiment judges the alignment error that occurs as described above, and more specifically, the alignment error judgment system (10) includes, in addition to the optical system (400) described above, an image generation system (200), an error analysis unit (500), and an alignment judgment unit (600).

[0086] The above image generation system (200) generates an image required to determine the alignment error described above in the alignment error determination system (10), and includes a learning unit (100), an alignment error selection unit (200), and an image generation unit (300).

[0087] The above learning unit (100) learns the relationship between the image for each alignment error of the head unit (410) of the optical system (400) and the fast Fourier transform (FFT) result for the image for each alignment error, and includes an image learning unit (110) including an image processing unit (120) and a learning execution unit (130), and an image judgment unit (140).

[0088] At this time, the conversion state when FFT is performed on the above image is explained by way of an example through the drawing below.

[0089] FIG. 4a is an example of performing FFT transformation by overlapping the captured images of FIG. 2b, FIG. 4b is an example of the FFT transformation result according to the characteristics of the captured images of FIG. 1, and FIG. 4c is an example of the type of FFT transformation obtained from the captured images of FIG. 1.

[0090] First, referring to FIG. 4a, as described above with reference to FIGS. 2a and 2b, projection images (captured images) (431, 432) acquired through the first and second head units (411, 412) of the optical system (400) are acquired with different alignment states due to the alignment error of the first and second head units (411, 412).

[0091] At this time, the first and second head parts (411, 412) capture the same reference image, and different first and second captured images (431, 432) are obtained as in Fig. 4a due to the alignment error of each head part.

[0092] Accordingly, the first and second captured images (431, 432) are obtained by overlapping each other to obtain an overlapping image (433), and then FFT conversion is performed to obtain an FFT converted image (501).

[0093] That is, the superimposed image (433) of the first and second photographed images (431, 432) is obtained as an FFT converted image (501) including a specific point or a specific shape, and thus, the information contained in the FFT converted image (501) includes relatively very little information compared to the information contained in the superimposed image (433). Therefore, by interpreting the FFT converted image (501), the characteristics of the superimposed image (433), that is, the alignment error information between the first and second photographed images (431, 432) included in the superimposed image (433) can be more easily obtained.

[0094] For example, referring to FIG. 4b, if the captured images (435, 436, 437) include patterns having intervals of, for example, 1λ, 2λ, and 3λ, and each of them is FFT-converted to obtain an FFT-converted image, the FFT-converted images (505, 506, 507) are derived as points at different locations on the FFT domain.

[0095] Therefore, it is easier to interpret the information contained in the photographed image by interpreting the information contained in the FFT transformed image than to interpret the information contained in the photographed image based on the photographed image itself.

[0096] Meanwhile, referring to FIG. 4c, methods for generating such an FFT conversion image include a method for generating an FFT conversion image (508) through amplitude conversion for the photographed image (438), a method for generating an FFT conversion image (509) through phase conversion, etc.

[0097] As described above, a predetermined relationship exists between the captured image and the FFT-converted image obtained by FFT-converting the captured image, and the learning unit (100) learns the relationship between the captured image captured through the head unit (410) of the optical system (400) and the FFT conversion result for the captured image.

[0098] In particular, the photographed image captured through the head unit (410) is acquired as a photographed image that includes an alignment error due to an alignment error between adjacent head units, even if the photographed image is captured with respect to the same reference image, and therefore the FFT conversion result also includes alignment error information included in the corresponding photographed image.

[0099] Accordingly, the learning unit (100) learns the relationship between the image for each alignment error and the FFT result of the alignment error-specific image obtained to include one of the alignment errors described with reference to FIGS. 3a to 3d, or at least two or more alignment errors.

[0100] At this time, the image for each alignment error may be an image including an error in any one of the six degrees of freedom in FIGS. 3A to 3D (a single degree of freedom alignment error image), or may be an image including an error in at least two or more of the six degrees of freedom or an image including all six degrees of freedom (a complex degree of freedom alignment error image).

[0101] Accordingly, the image learning unit (110) performs learning on the characteristics of the FFT results according to the alignment error between the images for each alignment error and the FFT results. This learning can be performed, for example, by machine learning or deep learning. In particular, the learning in the image learning unit (110) can utilize a neural operator method, and examples of the neural operator method include a Deep Operator Network (DeepONet), a Fourier Neural Operator (FNO), etc.

[0102] Meanwhile, images that have an alignment error of one of the six degrees of freedom, as well as images that have an alignment error of at least two or more degrees of freedom, or alignment errors of all six degrees of freedom, can be stored in the database of the image learning unit (110).

[0103] The image including the above alignment error may be a photographed image obtained by photographing a reference image through the head unit (410) having an actual alignment error, or may be an image virtually created by artificially assuming an alignment error that the head unit (410) may have.

[0104] In addition, the database also stores the FFT transformation results for the stored images, as exemplified in Fig. 4c. That is, not only the FFT transformation results for images with one misalignment error among the six degrees of freedom (single-degree-of-freedom FFT transformation images), but also the FFT transformation results for images with at least two misalignments, or misalignments in all six degrees of freedom (complex-degree-of-freedom FFT transformation images) are stored simultaneously.

[0105] Thus, the learning execution unit (130) performs learning on the relationship between an image with an alignment error stored in the database and the FFT transformation result of the image.

[0106] Meanwhile, the image processing unit (120) performs a predetermined preprocessing so that the learning execution unit (130) can perform effective learning on the FFT transformation result derived from FIG. 4c.

[0107] Figure 5 is a flowchart illustrating the steps of processing an image in the image processing unit of the learning unit of Figure 1.

[0108] That is, referring to FIG. 5, the image processing unit (120) performs a predetermined image processing on the result of FFT conversion of the photographed image so that the learning execution unit (130) can perform learning more effectively.

[0109] As described above, the acquired photographed image (438) can be obtained as an amplitude (508) or a phase (509) in the spatial frequency domain by performing a fast Fourier transform. However, the present invention is not limited thereto, and can also be obtained as a power spectrum or a Nyquist plot.

[0110] In the case of the FFT transformed image derived from the spatial frequency domain in this way, after a predetermined image processing is performed through the image processing unit (120), it is stored in the database, thereby making analysis of the FFT transformed image easier. Such image processing can apply techniques such as sharpening, histogram equalization, unsharp masking, image pyramids, multi-scale feature extraction, brightness, contrast, and gamma correction, for example.

[0111] To this end, as illustrated in FIG. 5, for example, the image processing unit (120) may crop (121) the FFT converted image, pre-process the image (122), perform binarization (123) or edge detection (124) of the image, and then perform Hough transform (125).

[0112] However, the image processing unit (120) is exemplary and can process the FFT converted image in a form necessary to perform learning through the learning execution unit (130).

[0113] The image generation unit (300) designs an image that can more accurately induce the alignment error. At this time, the image generation unit (300) designs the image based on the learning results of the learning unit (100).

[0114] Specifically, the image generation unit (300) includes a first image collection unit (310), a first image design unit (320), a second image collection unit (330), and a second image design unit (340).

[0115] First, in order to perform learning of the learning unit (100), the first image collection unit (310) collects images that induce a single degree of freedom alignment error of the head unit (410). At this time, the single degree of freedom alignment error means any one alignment error selected from among the six degrees of freedom alignment errors, as described above.

[0116] Since the learning unit (100) above has learned the relationship between an image having a single degree of freedom alignment error and the FFT transformation result for the image having the single degree of freedom alignment error, a single degree of freedom image that inversely obtains the corresponding FFT transformation result can be selected or generated based on the FFT transformation result of the image having the single degree of freedom alignment error.

[0117] Accordingly, through the first image collection unit (310), it is possible to collect various images having a single degree of freedom alignment error and single degree of freedom FFT transformed images as FFT transformation results for the corresponding images. Thus, the images collected as described above can be considered as a candidate group for images ultimately designed through the first image design unit (320). Meanwhile, the first image collection unit (310) can collect the single degree of freedom alignment error image and the single degree of freedom FFT transformed image thereof through the database of the image learning unit (110), etc.

[0118] Accordingly, the first image design unit (320) designs an image that can more accurately express an alignment error of a specific degree of freedom compared to other alignment errors among the images collected through the first image collection unit (310) based on the learning result of the image learning unit (110).

[0119] At this time, the first image design unit (320) can design an image that can most accurately express an alignment error of a specific degree of freedom by using the learning result of the image learning unit (110) among the images collected through the first image collection unit (310). Alternatively, the first image design unit (320) can separately generate an image that can most accurately express an alignment error of a specific degree of freedom by using the learning result of the image learning unit (110) based on the characteristics of the images collected through the first image collection unit (310).

[0120] Thus, the image finally selected or generated in the first image design unit (320) is an image that can most accurately determine the degree of alignment error of a specific degree of freedom, and can be defined as a single degree of freedom image.

[0121] In addition, the single degree of freedom image that is finally selected or generated through the first image design unit (320) may be an image having the characteristic of periodicity of image arrangement, as exemplified in the above-described FIG. 4b. That is, the periodicity of the image arrangement may be designed as an image having different periodicities, such as 1λ, 2λ, 3λ, etc., and if an image arrangement of a specific period is effective in determining an alignment error of a specific degree of freedom, an image having an image arrangement of the corresponding period may be designed.

[0122] In contrast, the single degree of freedom image may be designed with variations in the shape, size, gradient, and non-periodicity of the image. That is, if a specific shape, size, or gradient is effective in determining the alignment error of a specific degree of freedom, a single degree of freedom image having the corresponding characteristics may be designed. In this case, the shape of the image is not particularly limited, and may include various shapes such as polygons such as triangles, squares, and pentagons, as well as circles, ovals, and fan shapes.

[0123] Furthermore, the image finally selected or generated in the first image design unit (320) may be an image generated as a result of an inverse fast Fourier transform of an image generated as a result of a fast Fourier transform. In addition, it may be an image generated as a result of a fast Fourier transform of an image for analyzing an alignment error of the head portion, or an image generated as a result of a fast Fourier transform of an image for separating an alignment error for each degree of freedom of the head portion.

[0124] As described above, in the first image design unit (320), the single degree of freedom image is designed, and the single degree of freedom image designed in this way may be an optimal image for determining the alignment error of the specific degree of freedom.

[0125] In contrast, the single degree of freedom image designed through the first image design unit (320) may be an image capable of decoupling alignment errors for each degree of freedom included in the complex degree of freedom image designed through the second image design unit (340) described below.

[0126] The above complex degree of freedom image may be an image having two or more degrees of freedom, or even six degrees of freedom. However, it may be difficult to extract errors for all the degrees of freedom at once for such a complex degree of freedom image. Therefore, the single degree of freedom image designed through the first image design unit (320) can be used to separate the alignment errors for each degree of freedom included in the complex degree of freedom image. To this end, the designed single degree of freedom image may be an image optimized for such degree of freedom alignment error separation.

[0127] Furthermore, the single degree of freedom image designed through the first image design unit (320) may be a single image at any time, or may be a sequential image that is sequentially continuous over a specific period of time. In this case, in the case of the sequential image that is sequentially continuous, a single degree of freedom image is designed sequentially, and thus a plurality of single degree of freedom images may be designed.

[0128] As described above, when designing an image through the first image design unit (320), since there is a so-called interdependence between the image for each alignment error of the head unit and the fast Fourier transform (FFT) result for the image for each alignment error, the image design for determining the alignment state of the head unit must be performed by taking this into consideration. That is, there is an interdependence between the real image domain (image for each alignment error) and the spatial frequency domain (spatial frequency domain, fast Fourier transform result) in which the FFT is performed.

[0129] Here, the above interdependence means that, for example, the closer the actual image is to a short pulse, such as a Dirac delta function, the wider the spatial frequency band of the image (broad band), and conversely, the wider the image signal, the narrower the spatial frequency band (narrow band).

[0130] Meanwhile, the second image collection unit (330) collects images that induce two or more complex degrees of freedom alignment errors in advance. At this time, the complex degrees of freedom alignment errors refer to alignment errors that include at least two or more degrees of freedom among the six degrees of freedom alignment errors described above.

[0131] Since the learning unit (100) above has learned the relationship between an image having such a complex degree of freedom alignment error and the FFT transformation result for the image having the complex degree of freedom alignment error, it is possible to select or generate a complex degree of freedom image that inversely obtains the corresponding FFT transformation result based on the FFT transformation result of the image having the complex degree of freedom alignment error.

[0132] Accordingly, through the second image collection unit (330), various images having complex degrees of freedom alignment errors and complex degrees of freedom FFT transformed images as FFT transformation results for the corresponding images can be collected. Thus, the images collected as described above can be considered as a candidate group for images ultimately designed through the second image design unit (340). Meanwhile, the second image collection unit (330) can also collect the complex degrees of freedom alignment error images and the complex degrees of freedom FFT transformed images thereof through the database of the image learning unit (110), etc.

[0133] Accordingly, the second image design unit (340) designs an image that can more accurately reveal the alignment error of the complex degree of freedom among the images collected through the second image collection unit (330) based on the learning result of the image learning unit (110). At this time, the second image design unit (340) can design an image that can most accurately reveal the alignment error of the complex degree of freedom among the images collected through the second image collection unit (330) using the learning result of the image learning unit (110). Alternatively, the second image design unit (340) can separately generate an image that can most accurately reveal the alignment error of the complex degree of freedom among the images collected through the second image collection unit (330) using the learning result of the image learning unit (110).

[0134] Thus, the image finally selected or generated by the second image design unit (340) is an image that can most accurately determine the degree of alignment error of the composite degree of freedom, and can be defined as a composite degree of freedom image. Furthermore, the composite degree of freedom image selected or generated by the second image design unit (340) can utilize a single degree of freedom image selected or generated through the first image design unit (320).

[0135] As previously explained, since the first image design unit (320) can design an image capable of separating alignment errors for each degree of freedom as a single degree of freedom image, the composite degree of freedom image can be designed by utilizing the image capable of separating alignment errors. Thus, the composite degree of freedom image designed by the second image design unit (340) is an image for composite degrees of freedom of two or more degrees of freedom, but can be an image designed to be separated for each degree of freedom in practice.

[0136] Accordingly, in performing subsequent error extraction and alignment judgment, error extraction and alignment judgment can be performed for each separate degree of freedom, thereby improving the accuracy and speed of the alignment judgment result.

[0137] Meanwhile, since the characteristics of the image of the composite degree of freedom selected or generated through the second image design unit (340) are the same as those of the single degree of freedom image described above, redundant descriptions are omitted. Furthermore, the composite degree of freedom image designed through the second image design unit (340) may also be a single image at any time, or may be a sequential image sequentially over a specific period of time. In this case, in the case of the sequential image sequentially, a composite degree of freedom image is designed sequentially, and thus a plurality of composite degree of freedom images may be designed.

[0138] Fig. 6 is a schematic diagram showing an example of an image extracted through the image generation system of Fig. 1 and the result of the FFT transformation accordingly.

[0139] Referring to FIG. 6, when a pair of head parts has a predetermined alignment error (e.g., a rotational error of 1° around the Z-axis), when the output images (431, 432) obtained through the head parts are FFT-converted, the FFT conversion result (501) may also show a relatively small error (1° rotation along the extension axis) related to the alignment error. Therefore, it may be relatively difficult to accurately determine the alignment error between the head parts even through the FFT conversion result.

[0140] However, when designing an image that can determine a more accurate alignment error for a specific alignment error (e.g., rotation error based on the Z-axis) through the image generation unit (300) and performing FFT conversion on the corresponding images (301, 302), the FFT conversion result (511) may show a relatively very large error (rotated by (90+1)˚ along the extension axis) in relation to the alignment error.

[0141] Therefore, by designing an image that can more accurately determine a specific alignment error from the FFT conversion result through the image generation unit (300), the alignment error between the head units can be more accurately determined.

[0142] Meanwhile, the alignment error in the present embodiment includes, as described above, an alignment error between a pair of adjacent head portions, an alignment error between one head portion and a reference optical system or reference alignment mark, or an alignment error between a substrate portion and a head portion.

[0143] Accordingly, in all cases of determining the alignment error between the head portions, determining the alignment error between the head portion and the reference optical system, and determining the alignment error between the substrate portion and the head portion, only one image can be designed through the image generation unit (300) to enable determination of the alignment error. In this case, the one image being designed means one single image, one composite image, single images sequentially successive in time, or composite images sequentially successive in time, meaning that an image having one characteristic is designed.

[0144] In particular, in determining the alignment error between a pair of heads, one image is designed as a reference image, and by acquiring the captured images from the designed reference image from each of the pair of heads, the alignment error between the heads can be determined.

[0145] Alternatively, in determining the alignment error between the pair of head units, a pair of different images may be designed as reference images, and each reference image may be applied to each head unit to acquire captured images, after which the alignment error between the head units may be determined. In particular, when designing different reference images in this way, the designed images may be designed to more accurately reveal a specific alignment error.

[0146] As described above, when an image capable of accurate analysis of a specific alignment error is designed through the image generation unit (300), the optical system (400) acquires a photographed image (430) based on the designed image.

[0147] Accordingly, the acquired photographed image (430) is provided to the error analysis unit (500).

[0148] The above error analysis unit (500) extracts alignment errors by performing a fast Fourier transform (FFT) on a photographed image (430) acquired through the optical system (400), and includes an FFT transform unit (510) and an error extraction unit (520).

[0149] The above FFT conversion unit (510) performs a fast Fourier transform on the photographed image (430) acquired through the optical system (400) to obtain an FFT conversion result. At this time, the FFT conversion result is as exemplified above.

[0150] The above error extraction unit (520) extracts the alignment error of the head unit (410) by using the FFT conversion result and the learning result of the learning unit (100).

[0151] Meanwhile, in the present embodiment, the FFT conversion result converted through the FFT conversion unit (510) is provided to the image judgment unit (140) of the learning unit (100).

[0152] In the image judgment unit (140), the learning result of the image learning unit (110) is applied to the FFT conversion result (image) converted through the FFT conversion unit (510), thereby obtaining a regenerated image.

[0153] That is, the image judgment unit (140) removes noise from the converted FFT converted image and optimizes the FFT converted image from which the noise has been removed into an image suitable for alignment error analysis. At this time, when optimizing the image into an image suitable for alignment error analysis, the pre-learning result of the image learning unit (110) is applied. Thus, based on the pre-learning result, the FFT converted image is processed so that alignment error analysis can be performed more easily and accurately. Such optimization through re-learning of the FFT converted image is performed, and the image regenerated through this is provided to the error extraction unit (520).

[0154] Thus, the error extraction unit (520) can extract, based on the regenerated image, an alignment error with respect to the reference optical system or reference alignment mark of the head unit (410) or an alignment error between the head unit (410) and the substrate unit (420). Furthermore, the alignment error between adjacent head units (411, 412) can also be extracted.

[0155] That is, through the error extraction unit (520), as exemplified above, it is possible to extract an alignment error based on the FFT transformation result for a single output image, as well as to extract an alignment error based on the FFT transformation result in the overlapping region of a pair of output images. In addition, it is possible to extract alignment errors for not only a single degree of freedom but also complex degrees of freedom, as well as alignment errors at any time and alignment errors in a time series where specific times are continuous.

[0156] The above alignment judgment unit (600) judges the alignment state of the optical system (400) based on the alignment error extraction result of the error analysis unit (500).

[0157] That is, based on the alignment error extracted through the error extraction unit (520), the alignment determination unit (600) determines the alignment state of the head unit (410) and further the alignment state of the head units (411, 412). That is, the type and degree of alignment error that the head unit (410) has with respect to the reference optical system or the reference alignment mark can be determined, and the type and degree of alignment error that the head unit (410) has with respect to the substrate unit (420) can be determined. Furthermore, the type and degree of alignment error that the head units (411, 412) have in the overlapping area can be determined.

[0158] Of course, in the case of the present embodiment, since the reference image provided to acquire the above-described photographed image corresponds to an image already designed to more effectively determine a specific alignment error (single degree of freedom and compound degree of freedom), the alignment determination unit (600) may focus on determining a specific alignment error. However, since the head unit (410) may additionally include an unexpected alignment error, various types of alignment errors may be determined in addition to the specific alignment error.

[0159] The results of judging alignment between a pair of first and second head parts (411, 412) having an alignment error of a rotational error of a predetermined angle (α) relative to the third direction (Z) using the alignment error judgment system (10) according to the present embodiment described above are described as follows.

[0160] Fig. 7a is an example of designing an image by considering the characteristics of the FFT conversion result through the image generation system of Fig. 1, and Figs. 7b and 7c are graphs illustrating a state in which the result of extracting alignment errors is improved.

[0161] That is, as in Fig. 7a, for an image designed through the image generation unit (300), a first captured image (301) can be obtained through the first head unit (411), and a second captured image (302) can be obtained through the second head unit (412).

[0162] At this time, the first and second photographed images (301, 302) can be acquired to have different angles (+θ, -θ) as shown, because the images designed for the first and second head parts (411, 412) are different from each other.

[0163] In addition, in the case where the first and second head parts (411, 412) have a rotation error of a predetermined angle (α) based on the third axis (Z) as a predetermined alignment error, the first and second captured images (301, 302) are FFT-converted, and the result of judging the alignment error based on the result is as shown in FIG. 7c.

[0164] That is, referring to Fig. 7b, the rotation error derived as a result of determining the alignment error by performing FFT transformation based on an arbitrary reference image without performing image design is in the range of -0.1˚ to 0.13˚.

[0165] On the other hand, as shown in Fig. 7c, when the alignment error is determined through the alignment error determination system (10) according to the present embodiment, that is, when the alignment error is determined by designing an image that can more effectively highlight the rotational error of the reference image, the resulting rotational error is in the range of -0.034˚ to 0.051˚, and it can be confirmed that the alignment error determination result becomes more precise.

[0166] Through this, when determining alignment errors using designed images, as in this embodiment, the alignment status can be determined with higher precision and accuracy for specific alignment errors.

[0167] Figure 8 is a flowchart illustrating a method for determining an alignment error using the alignment error determination system of Figure 1.

[0168] Referring to FIG. 8, in the alignment error determination method using the alignment error determination system (10) of FIG. 1, first, the learning unit (100) learns the relationship between the image for each alignment error of the head and the fast Fourier transform result for the image for each alignment error (step S10).

[0169] At this time, the image for each alignment error is an image for each of the alignment errors described with reference to FIGS. 3A to 3D, and the relationship between the image for each alignment error and its FFT result is learned. Thus, learning is performed in advance regarding the characteristics of the FFT conversion result for each alignment error when FFT converting an image with each alignment error.

[0170] Thereafter, through the image generation unit (300), an optimal image capable of effectively revealing the specific alignment error is designed. At this time, the designed image may be an image with a single degree of freedom (step S20) or an image with multiple degrees of freedom (step S30).

[0171] When designing the image of the single degree of freedom, the alignment error image of the single degree of freedom and the FFT transformed image of the single degree of freedom are provided to the learning unit (100) (step S21), and the image of the single degree of freedom is designed using the learning result of the learning unit (100) (step S22), as described above.

[0172] Likewise, when designing an image of the above complex degree of freedom, the alignment error image of the complex degree of freedom and the FFT transformed image of the complex degree of freedom are provided to the learning unit (100) (step S31), and the image of the above complex degree of freedom is designed using the learning result of the learning unit (100) (step S32), as described above.

[0173] That is, when an image with a specific alignment error is FFT-transformed through the learning unit (100), the relationship between the image and the FFT transformation has been learned in advance. Therefore, by utilizing the learning results of the learning unit (100) in reverse, an optimal image that can more effectively express a specific alignment error in the FFT transformation results can be designed.

[0174] The image designed in this way can be an image that can effectively reveal one alignment error among the six degrees of freedom errors including the translational errors and rotational errors described with reference to FIGS. 3A to 3D, or an image that can effectively reveal two or more alignment errors simultaneously. Accordingly, an image that can effectively reveal all of the six degrees of freedom errors can be designed. Furthermore, the combination of alignment errors constituting the at least two or more alignment errors is not limited, and thus, two or more alignment errors may be selected only from the translational errors, two or more alignment errors may be selected only from the rotational errors, or two or more alignment errors may be arbitrarily selected from all of the translational errors and the rotational errors.

[0175] In addition, as previously explained, an image can be designed to separate the above six degrees of freedom errors and analyze each error.

[0176] Thus, using the image designed as described above, a photographed image (430) is acquired through the optical system (400) (step S40). At this time, the photographed image (430) acquired through the optical system (400) may be, as described above, a single photographed image, a pair of photographed images, or sequentially sequential photographed images, and the types of these images may vary depending on the target of judging the alignment status.

[0177] Thereafter, the acquired photographed image (430) is FFT-converted again (step S50) and provided to the image judgment unit (140) of the learning unit (100) (step S60). Accordingly, the image judgment unit (140) reuses the learning result of the learning unit (100) to regenerate an image that is more optimized for alignment error analysis and provides this to the error extraction unit (520). Thus, the error extraction unit (520) extracts the error of the optical system (400) using the image regenerated from the FFT conversion result (step S70).

[0178] Thereafter, based on the error extraction results from the error analysis unit (500), the alignment determination unit (600) determines the alignment state of the optical system (400) (step S80). Of course, in determining this alignment state, since an image designed to more effectively highlight a specific alignment state is used, the determination is made based on the degree of the specific alignment state, but the determination results for other alignment states may also be included.

[0179] According to the embodiments of the present invention as described above, in an optical system that performs image projection, the alignment status in the overlapping area between the image acquired by the substrate unit and the image acquired by the adjacent head units can be determined in real time. Accordingly, not only can the alignment between the substrate unit and the head unit be performed more accurately based on the alignment error of the optical system, but also the error in stitching images through adjacent head units can be minimized and the alignment status can be corrected.

[0180] In particular, the judgment of the alignment state is performed based on the fast Fourier transform result of the acquired image, and conversely, an image that can more precisely and accurately express the alignment state can be designed from the fast Fourier transform (FFT) result, and by judging the alignment state based on the image designed in this way, the precision and accuracy of the judgment of the alignment state can be improved.

[0181] That is, by learning in advance the relationship between the image for each alignment error of the head section and the FFT result for each alignment error image, it is possible to design an image that is optimal for deriving a specific alignment error from the FFT result, and to make a judgment on the alignment status based on the image designed in this way.

[0182] Of course, the learning result of the above learning unit can also be utilized in the process of determining the alignment status by performing FFT transformation on the image obtained for the alignment status of the head unit using the designed image, thereby improving the speed and accuracy of determining the alignment status.

[0183] At this time, in designing the image, by designing an image that can better highlight each of the six degrees of freedom errors, or by designing an image that can most effectively highlight at least two or more of the six degrees of freedom errors or all of the six degrees of freedom errors, the alignment status of the head can be accurately or quickly determined.

[0184] In this case, since an image can be designed to separate the alignment errors of each of the above six degrees of freedom errors, a faster and more accurate judgment can be made when judging the alignment status of the head from an image having multiple degrees of freedom.

[0185] That is, based on the learning result of the above learning unit, the characteristics of the FFT result for each alignment error of the aligned image are learned, and by reverse engineering this, an image that can most easily reveal the error of a specific degree of freedom in the FFT result, an image that can separate the alignment error for each degree of freedom, or an image that can reveal multiple degrees of freedom is designed, thereby enabling a more accurate alignment state to be determined for not only a specific degree of freedom but also a composite degree of freedom. In particular, in order to quickly determine the alignment state in real time, it is necessary to design an image that can most easily reveal the errors of at least two or more or all six degrees of freedom. At this time, by designing and providing an image that can separate and extract each degree of freedom, it is possible to not only reveal the error from the composite degree of freedom as a whole, but also to individually reveal the error of each degree of freedom from the composite degree of freedom, thereby enabling a quick real-time determination of the alignment state.

[0186] At this time, the image being designed can be derived by varying the shape of the image, the size of the image, the deformation of the image, or the periodicity of the image arrangement, due to the characteristics of the FFT result, so that the complexity or time taken in designing the image can be minimized.

[0187] Furthermore, in extracting errors from actual captured images, the FFT conversion results of the actual captured images are provided to the learning unit, and the image judgment unit performs so-called image regeneration to remove noise and optimize the image suitable for alignment error analysis, and the errors are extracted based on the regenerated image, so that more accurate alignment judgment is possible.

[0188] Although the present invention has been described above with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various modifications and changes may be made to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below.

Claims

1. Includes an image generation unit that generates an image for determining the alignment status of the head of the optical system, An image generation system characterized in that the image generation unit analyzes the alignment error of the head unit based on the learning results for the image for each alignment error of the head unit, or designs an image for separating the alignment error for each degree of freedom of the head unit.

2. In the first paragraph, the alignment error of the head portion is An image generation system characterized by an alignment error with respect to a reference optical system of the head unit, an alignment error with respect to an alignment mark of the head unit, or an alignment error between a pair of adjacent head units.

3. In the second paragraph, the image generating unit, Design at least one image as an image for analyzing the alignment error with respect to the reference optical system of the head part, the alignment error with respect to the alignment mark of the head part, or the alignment error between a pair of adjacent head parts, A generation system characterized in that, when the above images are multiple, they are sequential images.

4. In the second paragraph, the alignment error of the head portion is A translation error in at least one of the first to third mutually perpendicular directions, A rotational error that rotates around at least one of the first to third directions, or An image generation system characterized in that the error includes both the above-mentioned movement error and the above-mentioned rotation error.

5. In the first paragraph, the image generating unit, Design an image to analyze only one of the alignment errors among the alignment errors of the above head part, or An image generation system characterized in that it designs an image for analyzing at least two alignment errors simultaneously among the alignment errors of the above head part, or designs an image for separating alignment errors for each degree of freedom when analyzing at least two alignment errors.

6. In paragraph 5, the image generation unit, A first image collection unit that provides an image for analyzing only the alignment error or an image for separating the alignment error for each degree of freedom to the learning unit; and An image generation system characterized by including a first image design unit that designs an image for analyzing only the alignment error or an image for separating the alignment error for each degree of freedom based on the learning results of the learning unit.

7. In paragraph 6, the image generating unit, A second image collection unit that provides the learning unit with images for simultaneously analyzing at least two alignment errors; and An image generation system characterized by including a second image design unit that designs an image for simultaneously analyzing at least two alignment errors based on the learning results of the learning unit and an image for separating alignment errors for each degree of freedom designed by the first image design unit.

8. In the first paragraph, the image generating unit, An image generation system characterized in that it is designed to vary at least one of the shape of the image, the size of the image, the gradient of the image, the asymmetry of the image arrangement, and the periodicity of the image arrangement.

9. In paragraph 8, the shape of the image is An image generation system characterized by having any one of a polygon, circle, ellipse, and fan shape.

10. In paragraph 1, An image generation system further comprising a learning unit that learns the relationship between the image for each alignment error of the head section and the fast Fourier transform result for the image for each alignment error.

11. In the 10th paragraph, the image generating unit, An image generation system characterized in that it generates an image for determining the alignment state of the head of the optical system based on the interdependence between the image for each alignment error and the fast Fourier transform result for the image for each alignment error.

12. In paragraph 10, the image is An image generated as a result of the inverse fast Fourier transform of the image generated as a result of the above fast Fourier transform, or An image generation system characterized by being an image of a fast Fourier transform result of an image for analyzing an alignment error of the head portion, or an image of a fast Fourier transform result of an image for separating an alignment error for each degree of freedom of the head portion.

13. In paragraph 10, the learning unit, An image processing unit that performs a predetermined processing and conversion on the image based on the above alignment error; and An image generation system characterized by including a learning execution unit that learns the image processed and converted in the image processing unit through machine learning or deep learning.

14. In paragraph 10, the learning unit, An image generation system characterized by performing learning using a neural operator method.

15. Image generation system of paragraph 1; An optical system that generates a photographed image by capturing an image designed in the image generation system; and An alignment error judgment system including an alignment judgment unit that judges the alignment status of the head of the optical system based on the above-mentioned photographed image.

16. In the 15th paragraph, the optical system, An alignment error judgment system characterized by being applied to a DMD (digital micro-mirror device), LCos (liquid crystal on silicon), LED (light emitting display), LCD (liquid crystal display), nano imprint head, inkjet printing head, or image modulation / non-modulation head.

17. In paragraph 15, It further includes an error extraction unit that extracts an alignment error of the head of the optical system based on the above-mentioned photographed image, An alignment error judgment system characterized in that the error extraction unit includes an FFT transform unit that performs a fast Fourier transform on the photographed image.

18. In paragraph 15, The above learning unit includes an image judgment unit that regenerates an image by applying the learning result to the fast Fourier transform result of the photographed image. An alignment error judgment system characterized in that the error extraction unit extracts the alignment error of the head unit based on the regenerated image.

19. A step of analyzing the alignment error of the head part based on the learning results for each image of the alignment error of the head part or designing an image for separating the alignment error of each degree of freedom of the head part; A step of generating a photographed image by photographing the designed image in an optical system; and An alignment error determination method including a step of determining the alignment status of the head of the optical system based on the above-mentioned photographed image.

20. In paragraph 19, before the step of designing the image, A method for judging alignment errors, further comprising a step of learning the relationship between the image for each alignment error of the head portion and the result of a fast Fourier transform for the image for each alignment error.

Images