Mark detection device, mark learning device, and substrate processing device

Abstract

Disclosed are a mark detection device, a mark learning device, and a substrate processing device. The mark detection device includes: an imaging unit configured to generate an alignment mark image by imaging an alignment mark on an object; a detection unit configured to detect the alignment mark in the alignment mark image; and an adjustment unit configured to adjust a parameter related to imaging based on a learning model generated by learning using an alignment mark image in which the alignment mark cannot be detected and a first parameter as a parameter used for imaging of an alignment mark image in which the alignment mark can be detected. The adjustment unit acquires a second parameter as a result of an inference process based on the learning model. The imaging unit performs imaging in a state in which the parameter is adjusted to the second parameter.

Description

Technical Field

[0001] This invention relates to a technique for detecting alignment marks from image data obtained by imaging alignment marks. More specifically, this invention relates to a mark detection apparatus, a mark learning apparatus, a substrate processing apparatus, a mark detection method, and a method for manufacturing an article. Background Technology

[0002] For example, the alignment marks mentioned above are detected in an exposure apparatus used to manufacture semiconductor devices or display devices using photolithography. The exposure apparatus projects a pattern of a master image (such as a mask and intermediate mask) onto a substrate (such as a wafer and glass plate) via a projection optics system and transfers the pattern. When transferring multiple layers with different master images onto the same substrate, highly accurate alignment is required to prevent the actual transfer position from deviating from the target transfer position. The alignment marks are detected using image data (alignment mark images) obtained by imaging the alignment marks, and the position is corrected based on the results.

[0003] Japanese Patent Publication No. (“JP”) 2003-338455 discloses a method for detecting the position of alignment marks in an alignment mark image through template matching. This method employs an optimization approach that uses automatic deformation and brightness changes of a template corresponding to the deformation of the alignment mark. This method self-learns to use the optimized template for the next match and improves the detection rate of the alignment mark.

[0004] However, even the method disclosed in JP 2003-338455 requires the operator to first manually adjust the alignment mark position to the imaging position, adjust multiple parameters (such as the wavelength and illuminance of the imaging light), and maintain the alignment mark in a good imageable state. At this point, the operator needs to spend time determining the combination of parameters to use. Because the state of the alignment mark in the alignment mark image changes according to the physical properties of the resist applied to the substrate, the operator needs sufficient knowledge to determine the combination.

[0005] Some exposure devices can automatically adjust parameters, but these types of devices automatically and sequentially change the combination of parameters and determine if the combination is suitable. Therefore, determining the appropriate combination of parameters takes a long time. Summary of the Invention

[0006] The present invention provides an alignment mark detection device, etc., which can automatically and quickly adjust imaging-related parameters to obtain good alignment mark images.

[0007] A mark detection apparatus according to one aspect of the present invention includes: an imaging unit configured to generate an alignment mark image by imaging alignment marks on an object; a detection unit configured to detect alignment marks in the alignment mark image; and an adjustment unit configured to adjust imaging-related parameters based on a learning model generated by learning an alignment mark image in which alignment marks cannot be detected and a first parameter as a parameter for imaging an alignment mark image in which alignment marks can be detected. The adjustment unit obtains a second parameter as a result of inference processing based on the learning model. The imaging unit performs imaging with the parameter adjusted to the second parameter. Mark detection methods corresponding to the above-described mark detection apparatus, as well as substrate processing apparatuses and article manufacturing methods using the above-described mark detection apparatus, also constitute other aspects of the present invention.

[0008] According to another aspect of the invention, a marker learning apparatus includes: an acquisition unit configured to acquire an alignment mark image as an image generated by imaging alignment marks on an object, wherein alignment marks in the image cannot be detected, and a first parameter as an imaging-related parameter in the imaging of the alignment mark image in which alignment marks can be detected; and a learning unit configured to generate a learning model for use with inference processing, the learning model being learned by using the acquired alignment mark image and the first parameter to output a second parameter as a parameter.

[0009] Other features of the invention will become clear from the following description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0010] Figure 1 This is a block diagram illustrating the configuration of the marker detection device according to the first embodiment.

[0011] Figure 2 The diagram illustrates the relationship between the parameters to be adjusted and the adjustment unit according to the first embodiment.

[0012] Figure 3 This is a flowchart illustrating the parameter inference process based on the first embodiment.

[0013] Figure 4 The illustration shows images of alignment marks with different viewing illumination levels.

[0014] Figure 5 The illustration shows alignment mark images with different alignment mark positions in the field of view.

[0015] Figure 6 This is a flowchart illustrating the parameter learning process according to the first embodiment.

[0016] Figure 7The diagram illustrates the network configuration of the learning sharing server according to the second embodiment.

[0017] Figure 8 The illustration shows the configuration of an exposure apparatus including a marker detection device according to the first embodiment. Detailed Implementation

[0018] A description of embodiments of the present invention will now be given with reference to the accompanying drawings.

[0019] First Embodiment

[0020] Figure 1 The illustration shows a configuration (marker detection device) related to alignment mark measurement in an exposure apparatus according to a first embodiment. The exposure apparatus includes an operation terminal 100, a control unit 110, a learning server 120, and a drive unit 130. The sequence controller 111 in the control unit 110 issues measurement and drive commands to the measurement controller 112 and the drive controller 113 respectively based on commands from the operation terminal 100 operated by the operator, and performs sequence control.

[0021] The measurement controller 112 communicates with the drive unit 130 to control the measurement unit 131, the observation light unit 133, and the wavelength correction unit 134 within the drive unit 130. The measurement unit 131, acting as an imaging unit, images alignment marks provided on an object such as a mask or plate (substrate) within the observation field of view (imaging area) and generates an alignment image. The observation light unit 133 includes multiple light sources, such as LEDs, having different wavelengths from each other. The wavelength correction unit 134 controls reflected light from the mask or plate to correct the wavelength of the light to be imaged by the measurement unit 131.

[0022] The drive controller 113 communicates with the drive unit 130 to control the objective lens drive unit 132, mask stage drive unit 135, and plate stage drive unit 136 within the drive unit 130. The objective lens drive unit 132 moves the objective lens to transfer the mask pattern onto the plate through exposure. The mask stage drive unit 135 and the plate stage drive unit 136 move the stages supporting the mask and the plate, respectively.

[0023] The operating terminal (operating unit) 100 includes an operating unit 101 that enables adjustment of multiple parameters related to imaging for acquiring an alignment mark image, and adjusts the parameters according to the operator's operation of the operating unit 101. The multiple imaging-related parameters are values, and when the parameters change, the state of the alignment mark in the alignment mark image (such as brightness, contrast, and position) changes.

[0024] The learning server (alignment mark learning device) 120 includes a learning model 121 for obtaining (inferring) a suitable combination of parameters, and a learning unit 122 for updating the learning model 121.

[0025] In adjustment Figure 2 After setting the parameters and imaging conditions as shown, the measurement unit 131 performs imaging in the exposure apparatus to obtain an image of the alignment mark. More specifically, the position of the alignment mark in the observation field of view of the measurement unit 131 is adjusted by driving the objective lens drive unit 132, the mask stage drive unit 135, and the stage drive unit 136. During imaging, the observation field of view is illuminated by observation light. The wavelength of the observation light is adjusted by switching the current light source to other light sources with different wavelengths in the observation light unit 133, by adjusting the aperture stop (aperture) deployed on the observation light path to cut out light in a specific wavelength band, and by using the wavelength correction unit 134 to suppress specular reflection light. The illuminance of the observation light is adjusted by increasing or decreasing the current value of the light source input to the observation light unit 133 to change the amount of light emitted.

[0026] Figure 3 The flowchart illustrates the inference process using a learning model 121 to infer a suitable combination of parameters. This learning model 121 receives an alignment mark image as input from imaging by the measurement unit 131. A control unit 110, including a computer (comprising a sequence controller 111, a measurement controller 112, and a drive controller 113), executes this process according to a computer program. The control unit 110 corresponds to the detection device and the adjustment unit.

[0027] First, the sequence controller 111 sends commands to the measurement controller 112 and the drive controller 113 to measure the position of the alignment mark. In response, the drive controller 113 acquires a combination of parameters (hereinafter referred to as initial parameters) pre-held by the operation terminal 100 and sets them to the respective units in the drive unit 130. As a result, the alignment mark position moves to a position corresponding to the initial parameters, either within or outside the field of view, and the wavelength and illuminance of the observation light from the observation light unit 133 are set according to the initial parameters.

[0028] In step S300, the measurement controller 112 causes the measurement unit 131 to image the alignment mark. The alignment mark image obtained by imaging is set as the first alignment mark image.

[0029] In step S301, the sequence controller 111 inputs the first alignment mark image into the learning model 121 in the learning server 120. The learning model 121 outputs a combination of appropriate parameters (the second parameter: referred to below as the inference parameter) as the inference result.

[0030] The learning model 121 is generated through machine learning using various alignment mark images of past measurement failures and parameters (first parameters) used for imaging alignment mark images of successful measurements. However, the learning model 121 does not necessarily have to be generated in the exposure apparatus according to this embodiment; it can be a learning model generated by another exposure apparatus or a learning model generated using logs of alignment mark images and parameters obtained from an external source.

[0031] Machine learning can be performed using, for example, convolutional neural networks. A convolutional neural network is an algorithm that has an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer, which takes an image as input and outputs an inference result. This embodiment discusses a learning model 121 using a convolutional neural network, but learning model 121 can be generated using other models or algorithms.

[0032] In step S302, the sequence controller 111 applies the combination of inferred parameters output from the learning model 121 to each unit in the drive unit 130 through the measurement controller 112 and the drive controller 113. Thus, the imaging-related parameters are adjusted to inferred parameters.

[0033] Then, in step S303 after applying the inferred parameters, the sequence controller 111 causes the measurement unit 131 to image the alignment mark again via the measurement controller 112. The alignment image obtained by imaging is set as the second alignment mark image.

[0034] Figure 4 Examples of a first alignment mark image and a second alignment mark image are illustrated. A first alignment mark image 400 with insufficient brightness or low contrast is obtained when the illuminance of the observation light is as low as, for example, 20% with initial parameter settings. A second alignment mark image 401 with high brightness or high contrast is obtained when the illuminance of the observation light is as high as, for example, 70% by applying inference parameters obtained by inputting the first alignment mark image 400 into the learning model 121.

[0035] Figure 5 Another example of a first alignment mark image and a second alignment mark image is illustrated. A first alignment mark image 500, where the alignment mark, set by initial parameters, is at least partially outside the field of view, is obtained. When the first alignment mark image 500 is input to the learning model 121, a stage drive amount is output as an inference parameter. When the stage drive unit 136 drives the stage with the mounted plate at this stage drive amount, the alignment mark is located at the center of the field of view. Thus, a second alignment mark image 501, where the alignment mark is centered, is obtained.

[0036] In step S304, the sequence controller 111 measures the position of the alignment mark in the second alignment mark image. Then, in step S305, the sequence controller 111 determines whether the measurement was successful (in other words, whether the alignment mark was detected). If the alignment mark position has been successfully measured, the inference process terminates. If the measurement of the alignment mark position is unsuccessful (failure), the process proceeds to the learning process described below.

[0037] This embodiment can automatically and quickly adjust imaging-related parameters to obtain a good alignment mark image by using a learning model that accepts alignment mark images as input. Furthermore, this embodiment can use a good alignment mark image to position a mask or plate with high precision.

[0038] Figure 6 The flowchart illustrates the learning process (learning model generation method). In this description, the learning process is performed when the measurement of the alignment mark position in the second alignment mark image obtained by imaging using a combination of inference parameters obtained through the inference process is unsuccessful. The circumstances under which the learning process is performed are not limited to this case.

[0039] First, in step S600, the sequence controller 111 automatically adjusts the combination of parameters that enable measurement according to a predetermined adjustment process. At this time, the sequence controller 111 searches for the combination of parameters that enable measurement while simultaneously driving the objective lens drive unit 132, the observation light unit 133, and the wavelength correction unit 134 according to predetermined rules via the measurement and drive controllers 112 and 113. In this step, the sequence controller 111 can obtain the combination of parameters input by the operator through the operation unit 101.

[0040] Next, in step S601, the sequence controller 111 applies the combination of parameters obtained in step S600 to each unit in the drive unit 130 through the measurement and drive controllers 112 and 113, so that the measurement unit 131 images the alignment mark.

[0041] Next, in step S602, the sequence controller 111 measures the position of the alignment mark in the alignment mark image obtained through imaging. Then, in step S603, the sequence controller 111 determines whether the measurement was successful. If the measurement was successful, the sequence controller 111 causes the learning unit 122 to update the learning model 121 in step S604. The learning unit 122 corresponds to the acquisition unit and the learning unit.

[0042] In the update of learning model 121, sequential controller 111 causes learning unit 122 to learn using a combination of the first alignment mark image that failed to be measured and the parameters that made the measurement successful in the aforementioned step S602 (hereinafter referred to as success parameters) as input. For example, when the learning is machine learning using a convolutional neural network, the error backpropagation method can be applied.

[0043] Machine learning using convolutional neural networks sets the combination of successful parameters as ground truth data and the combination of inferred parameters obtained by learning model 121 as output data. It adjusts the weights between neurons to reduce the error between ground truth data and output data. The error between ground truth data and output data can be expressed as the sum of squares error, as shown in expression (1), or as the cross-entropy error, as shown in expression (2). In each expression, E is the error value, y is the output data value, and t is the correct data value.

[0044]

[0045] E=-∑ k t k log y k (2)

[0046] At this point, the learning model 121 is updated using the first alignment marker image obtained in the inference process.

[0047] If the measurement of the alignment mark is unsuccessful in step S602, the process returns to step S600 to repeat the above process.

[0048] This embodiment takes an alignment mark image as input and can automatically or with minimal burden on the operator optimize the learning model used to infer imaging parameters.

[0049] This embodiment has been applied to an exposure apparatus as a substrate processing apparatus, which exposes a substrate positioned by using alignment marks detected by a mark detection device to form a pattern on the substrate. However, this embodiment can be applied to other substrate processing apparatuses. For example, this embodiment is also applicable to an embossing apparatus that uses a mold to form a pattern made of an embossing material on a substrate, or a drawing apparatus such as one that irradiates a substrate with a beam of charged particles to form a pattern on the substrate. This embodiment is also applicable to substrate processing apparatuses such as a coating machine for coating a photosensitive medium onto the surface of a substrate, and a developer for developing the photosensitive medium on which the pattern has been transferred. This embodiment is also applicable to substrate processing apparatuses such as film forming apparatuses (such as CVD apparatuses), processing apparatuses (such as laser processing apparatuses), inspection apparatuses (such as overlay inspection apparatuses), and measuring apparatuses (such as mark measuring apparatuses).

[0050] Figure 8 The configuration of the exposure apparatus 10 is illustrated. The exposure apparatus 10 projects an image of the pattern of the mask M onto the substrate W via a projection optics system (objective lens) 14, thereby exposing the substrate W. The substrate W and the mask M are provided with alignment marks detected by a mark detection device.

[0051] The Z-axis direction is set to be parallel to the optical axis of the projection optical system 14, and the X-axis and Y-axis directions are set to be two directions orthogonal to each other in a plane perpendicular to the Z-axis direction. θX, θY, and θZ are set to rotations about the X-axis, rotations about the Y-axis, and rotations about the Z-axis, respectively.

[0052] The exposure apparatus 10 includes a light source 11, an illumination optics system 12, a mask stage 13, a projection optics system 14, a stage 15, and a main control unit 16. The exposure apparatus 10 includes a mask stage driving unit 21 for driving the mask stage 13. Figure 1 135) and the objective lens driving unit 22 for driving the lens (optical element) 14a in the projection optical system 14. Figure 1 (132 in the text). The exposure apparatus 10 includes a stage driving unit 51 for driving the stage 15. Figure 1 (136 in the text). The mask stage driving unit 21, objective lens driving unit 22, and stage driving unit 51 are respectively comprised of components included in, for example, 136. Figure 1 The mask stage control unit 31, lens control unit 32, and plate stage control unit 41 in the measurement controller 112 shown are controlled. The components from the light source 11 to the plate stage control unit 41 correspond to the processing unit used to perform processing on the plate W (forming a pattern by exposure).

[0053] The main control unit 16 controls the overall operation of the exposure apparatus 10, and includes... Figure 1 The control unit 110 and learning server 120 are shown. Figure 1 The operating terminal 100 shown is connected to the main control unit 16.

[0054] Light source 11 emits exposure light. Illumination optics system 12 uses the light emitted from light source 11 to illuminate mask M. Mask stage 13 holds mask M and moves it in the XY plane orthogonal to the optical axis of projection optics system 14 via mask stage drive unit 21. Projection optics system 14 projects an image of the pattern of mask M illuminated by illumination optics system 12 onto plate W. Projection optics system 14 includes a lens 14a movable in the X-axis direction via objective lens drive unit 22. Plate stage 15 holds plate W and translates or rotates it in the XY plane via plate stage drive unit 51.

[0055] The main control unit 16 controls the mask stage drive unit 21, the objective lens drive unit 22, and the plate stage drive unit 51 via the mask stage control unit 31, the lens control unit 32, and the plate stage control unit 41, based on the measurement results of the alignment mark position detected by the mark detection device. As a result, the mask M and the plate W are positioned in the XY plane.

[0056] Exposure apparatus 10 is suitable for manufacturing articles, such as microdevices, like semiconductor devices, devices with fine structures, and flat panel displays. Methods for manufacturing articles include processing a substrate using exposure apparatus 10 and manufacturing articles from the substrate processed in the processing steps. Manufacturing methods may include well-known steps such as exposure, oxidation, film deposition, vapor deposition, doping, planarization, etching, resist stripping, dicing, bonding, and encapsulation. Methods for manufacturing articles using exposure apparatus 10 are more advantageous than conventional methods in at least one aspect of article performance, quality, productivity, and production cost.

[0057] Second Embodiment

[0058] Figure 7 The diagram illustrates a network configuration according to the second embodiment. Exposure apparatus 701, along with other exposure apparatuses, is connected to learning and sharing server 700 via a network.

[0059] In the first embodiment, the learning model in a single exposure device (701) is used to obtain inference parameters or update the learning model. On the other hand, this embodiment connects multiple exposure devices, including exposure device 701, to a learning-sharing server 700, and uses the learning model in the learning-sharing server 700 to obtain inference parameters and update the learning model. That is, the learning model is shared by multiple exposure devices.

[0060] By sharing the learning model among multiple exposure units, inference parameters can be obtained even for newly installed exposure units that have not yet been trained. Learning accuracy can be further improved by using alignment marker images and success parameters obtained from multiple exposure units.

[0061] The learning shared server 700 can be installed inside or outside a factory where exposure equipment is installed. The same learning shared server 700 can be used by exposure equipment in multiple factories via a network.

[0062] Other embodiments

[0063] One or more embodiments of the present invention can also be implemented by a computer of a system or apparatus, wherein the computer reads and executes computer-executable instructions (e.g., one or more programs) recorded on a storage medium (which may also be more fully referred to as a "non-transitory computer-readable storage medium") to perform the functions of one or more of the above embodiments, and / or the computer includes one or more circuits (e.g., application-specific integrated circuits (ASICs)) for performing the functions of one or more of the above embodiments, and can also be implemented by a computer of a system or apparatus by, for example, reading and executing computer-executable instructions from a storage medium to perform the functions of one or more of the above embodiments and / or controlling one or more circuits to perform the functions of one or more of the above embodiments. The computer may include one or more processors (e.g., a central processing unit (CPU), a microprocessor unit (MPU)) and may include a network of individual computers or individual processors to read and execute the computer-executable instructions. The computer-executable instructions may be provided to the computer from, for example, a network or a storage medium. The storage medium may include, for example, a hard disk, random access memory (RAM), read-only memory (ROM), storage devices for distributed computing systems, optical discs (such as compact discs (CDs), digital variety discs (DVDs), or Blu-ray discs (BDs)). TM One or more of the following: flash memory devices, memory cards, etc.

[0064] Other embodiments

[0065] The embodiments of the present invention can also be implemented by providing software (programs) that perform the functions of the above embodiments to a system or device via a network or various storage media, and the computer or central processing unit (CPU) or microprocessor unit (MPU) of the system or device reads out and executes the program.

[0066] By using a learning model that accepts alignment mark images as input, the embodiments are able to automatically and quickly adjust imaging-related parameters to obtain good alignment mark images.

[0067] While the invention has been described with reference to exemplary embodiments, it should be understood that the invention is not limited to the disclosed exemplary embodiments. The scope of the following claims should be given the broadest interpretation to cover all such modifications and equivalent structures and functions.

Claims

1. A marker detection device, comprising: The imaging unit is configured to generate an alignment mark image by imaging alignment marks on an object; The detection unit is configured to detect alignment marks in the alignment mark image; as well as The adjustment unit is configured to adjust parameters related to imaging conditions, which include at least one of the following: the position of the alignment mark in the field of view, the wavelength of the imaging light, and the illuminance of the imaging light. The adjustment unit is configured to adjust parameters based on the output of a learning model, the output of which is a parameter estimated by means of an alignment mark image imaged using initial parameters as input. The learning model is generated by learning an alignment mark image in which alignment marks cannot be detected and a first parameter used for imaging an alignment mark image in which alignment marks can be detected. The adjustment unit obtains a second parameter as the result of inference processing based on the learning model, which is output from the learning model. The imaging unit performs imaging when the parameters are adjusted from the initial parameters to the second parameters.

2. The marker detection device according to claim 1, wherein, The learning model outputs the position of the alignment mark in the observation field of view, the wavelength of the imaging light, and the illuminance of the imaging light as a second parameter. The adjustment unit is configured to adjust only the necessary imaging conditions that should be adjusted in the initial parameters.

3. The marker detection device according to claim 1, further comprising: The learning unit is configured to learn in order to generate a learning model.

4. The marker detection device according to claim 3, wherein, The learning unit performs learning while adjusting the parameters to the second parameter, provided that the alignment marks in the alignment mark image generated by imaging cannot be detected.

5. A label detection method, comprising the following steps: An alignment mark image is generated by imaging the alignment marks on the object; Detect alignment marks in an alignment mark image; as well as Adjust parameters related to imaging conditions, which include at least one of the following: the position of the alignment mark in the field of view, the wavelength of the imaging light, and the illuminance of the imaging light. The adjustment step includes adjusting parameters based on the output of a learning model, the output of which is a parameter estimated by taking an alignment mark image image as input using initial parameters. The learning model is generated by learning an alignment mark image in which alignment marks cannot be detected and a first parameter used for imaging an alignment mark image in which alignment marks can be detected. The adjustment step obtains a second parameter as the result of the inference process based on the learning model, which is output from the learning model. Imaging is performed when the parameters are adjusted from the initial parameters to the second parameters.

6. The marker detection method according to claim 5, wherein, The learning model outputs the position of the alignment mark in the field of view, the wavelength of the imaging light, and the illuminance of the imaging light as a second parameter. The adjustment step only adjusts the necessary imaging conditions that should be adjusted in the initial parameters.

7. A computer-readable storage medium storing a computer program that causes a computer to perform the marker detection method according to claim 5 or 6.

8. A substrate processing apparatus, comprising: The mark detection device according to any one of claims 1 to 4; as well as The processing unit is configured to process a substrate positioned by using alignment marks detected by a mark detection device.

9. A method for manufacturing an article, the method comprising the following steps: The substrate is processed using the substrate processing apparatus according to claim 8; and Articles are manufactured from processed substrates.

Images