Training machine learning models to generate MRC and process aware reticles

Abstract

Methods and systems are described herein for training a prediction model to predict mask images in which mask rule check (MRC) violations or process violations (e.g., edge placement errors, sub-resolution assist feature (SRAF) printing) are minimized or eliminated. The prediction model is trained according to a loss function indicating (a) a difference between the predicted mask image and the reference image and (b) at least one of an MRC evaluation of the predicted mask image or a simulated image evaluation of the predicted mask image.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. application No. 63 / 393,024, filed on July 28, 2022, and the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present invention relates to designing photolithography masks for semiconductor manufacturing and, more specifically, to training a machine learning model to generate mask patterns. Background Art

[0004] For example, photolithographic projection apparatus may be used to manufacture integrated circuits (ICs). In this case, a patterning device (e.g., a mask) may contain or provide a circuit pattern corresponding to a single layer (a "design layout") of the IC, and the circuit pattern may be transferred to a target portion (e.g., including one or more dies) on a substrate (e.g., a silicon wafer) that has been coated with a layer of radiation-sensitive material (a "resist"), such as by irradiating the target portion through the circuit pattern on the patterning device. Typically, a single substrate contains multiple adjacent target portions, and the photolithographic projection apparatus transfers the circuit pattern to these target portions one at a time. In one type of photolithographic projection apparatus, the circuit pattern on the entire patterning device is transferred to one target portion at a time; such apparatus is commonly referred to as a stepper. In an alternative apparatus, commonly referred to as a stepper-scan apparatus, a projection beam scans the patterning device along a given reference direction (the "scanning" direction) while synchronously moving the substrate parallel or antiparallel to the reference direction. Different portions of the circuit pattern on the patterning device are gradually transferred to one target portion. Since a lithographic projection apparatus typically has a magnification factor M (typically <1), the speed F at which the substrate is moved will be a factor of M greater than the speed at which the projection beam scans the patterning device. More information on the lithographic apparatus described herein may be obtained, for example, from US 6,046,792 (incorporated herein by reference).

[0005] Before the circuit pattern is transferred from the pattern forming device to the substrate, the substrate can undergo various procedures such as priming, resist coating and soft baking. After exposure, the substrate can undergo other procedures such as post-exposure baking (PEB), development, hard baking and measurement / inspection of the transferred circuit pattern. This series of procedures is used as the basis for manufacturing a single layer of a device (e.g., IC). The substrate can then undergo various processes such as etching, ion implantation (doping), metallization, oxidation, chemical mechanical polishing, etc., all of which are intended to complete a single layer of the device. If multiple layers are required in the device, the entire procedure or its variants are repeated for each layer. Ultimately, the device will be present in each target portion on the substrate. These devices are then separated from each other by techniques such as cutting or sawing, so that the individual devices can be mounted on a carrier, connected to pins, etc.

[0006] As mentioned above, photolithography is a core step in integrated circuit manufacturing, where patterns formed on a substrate define the functional elements of an integrated circuit, such as microprocessors, memory chips, etc. Similar photolithography techniques are also used in the formation of flat panel displays, microelectromechanical systems (MEMS), and other devices.

[0007] As semiconductor manufacturing processes continue to advance, the size of functional elements continues to decrease, while the number of functional elements (such as transistors) per device has been steadily increasing for decades, a trend often referred to as "Moore's Law." At the current state of the art, device layers are fabricated using photolithography projection equipment that projects a design layout onto a substrate using illumination from a deep ultraviolet illumination source, thereby creating individual functional elements with dimensions well below 100 nm, i.e., less than half the wavelength of the radiation from the illumination source (e.g., a 193 nm illumination source).

[0008] This process of printing feature sizes smaller than the traditional resolution limit of the lithographic projection equipment according to the resolution formula CD = k1 × λ / NA is generally referred to as low-k1 lithography, where λ is the wavelength of the radiation used (currently 248nm or 193nm in most cases), NA is the numerical aperture of the projection optical system in the lithographic projection equipment, CD is the "critical dimension" - usually the minimum feature size printed - and k1 is an empirical resolution factor. In general, the smaller the k1, the more difficult it is to reproduce on the substrate a pattern similar in shape and size to that planned by the circuit designer to achieve specific electrical functions and performance. In order to overcome these difficulties, complex fine-tuning steps need to be applied to the lithographic projection equipment and / or the design layout. These include, but are not limited to, optimizing NA and optical coherence settings, customizing the illumination scheme, using phase-shifting pattern forming devices, optical proximity correction (OPC, sometimes also called "optical and process correction") in the design layout, or other methods generally defined as "resolution enhancement technology" (RET). The term "projection optics" used herein should be broadly interpreted to include various types of optical systems, including, for example, refractive optical systems, reflective optical systems, apertures, and reflective-refractive optical systems. The term "projection optics" may also include components operating according to any of these design types for collectively or individually directing, shaping or controlling a projection beam of radiation. The term "projection optics" may include any optical component in a lithographic projection apparatus, regardless of where the optical component is located in the optical path of the lithographic projection apparatus. A projection optical system may include optical components for shaping, conditioning and / or projecting radiation from a source before the radiation passes through a patterning device, and / or optical components for shaping, conditioning and / or projecting radiation after the radiation passes through a patterning device. A projection optical system typically does not include a source and a patterning device. Summary of the invention

[0009] In some embodiments, a non-transitory computer-readable medium is provided, the medium having instructions, when executed by a computer, the instructions cause the computer to perform a method for training a machine learning model to generate a mask image to be used to print a target layout on a substrate. The method includes: inputting a target image to a neural network, the target image being associated with a target layout to be printed on a substrate, wherein the neural network is configured to receive a reference image, the reference image corresponding to an optical proximity correction (OPC) mask image of the target image; using the neural network to generate a predicted mask image representing a mask pattern to be used to print the target layout on a substrate; calculating an indicative loss function, the loss function indicating: (a) a difference between the predicted mask image and the reference image, and (b) at least one of an MRC evaluation of the predicted mask image or an evaluation of a first simulated image of the predicted mask image; and modifying the neural network based on the loss function.

[0010] In some embodiments, a non-transitory computer-readable medium is provided, the medium having instructions, when executed by a computer, the instructions cause the computer to perform a method for training a machine learning model to generate a mask image to be used to print a target layout on a substrate. The method includes: inputting a target image set and a reference image set as training data to a neural network, wherein the target image in the target image set includes a target layout to be printed on a substrate, and wherein the reference image in the reference image set corresponds to an optical proximity correction (OPC) mask image of the target image; and training the neural network based on the training data to generate a predicted mask image so that a loss function indicating (a) a difference between the predicted mask image and the reference image and (b) an MRC cost associated with an MRC evaluation of the predicted mask image is minimized.

[0011] In some embodiments, a non-transitory computer-readable medium is provided, the medium having instructions, when executed by a computer, the instructions cause the computer to perform a method for training a machine learning model to generate a mask image to be used to print a target layout on a substrate. The method includes: inputting a target image set and a reference image set as training data to a neural network, wherein the target image in the target image set includes a target layout to be printed on a substrate, and wherein the reference image in the reference image set corresponds to an optical proximity correction (OPC) mask image of the target image; and training the neural network based on the training data to generate a predicted mask image, so that a loss function indicating (a) a difference between the predicted mask image and the reference image and (b) a simulation cost associated with a first simulated image of the predicted mask image is minimized.

[0012] In some embodiments, a method for training a machine learning model to generate a mask image to be used to print a target layout on a substrate is provided. The method includes: inputting a target image to a neural network, the target image being associated with the target layout to be printed on the substrate, wherein the neural network is configured to receive a reference image corresponding to an optical proximity correction (OPC) mask image of the target image; using the neural network to generate a predicted mask image representing a mask pattern to be used to print the target layout on the substrate; calculating a loss function indicating at least one of (a) a difference between the predicted mask image and the reference image and (b) an MRC evaluation of the predicted mask image or an evaluation of a first simulated image of the predicted mask image; and modifying the neural network based on the loss function.

[0013] In some embodiments, a device for training a machine learning model to generate a mask image to be used to print a target layout on a substrate is provided. The device includes: a memory storing an instruction set; and a processor configured to execute the instruction set to cause the device to perform the following method: input a target image to a neural network, the target image being associated with a target layout to be printed on a substrate, wherein the neural network is configured to receive a reference image corresponding to an optical proximity correction (OPC) mask image of the target image; use the neural network to generate a predicted mask image representing a mask pattern to be used to print the target layout on a substrate; calculate a loss function indicating (a) a difference between the predicted mask image and the reference image, and (b) at least one of an MRC evaluation of the predicted mask image or an evaluation of a first simulated image of the predicted mask image; and modify the neural network based on the loss function. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a block diagram of various subsystems of a lithography system consistent with various embodiments.

[0015] Figure 2 The flow of a photolithography process or a patterning simulation method consistent with various embodiments is shown.

[0016] Figure 3 is a block diagram of a system for generating a mask pattern from a target layout using a predictive model consistent with various embodiments.

[0017] Figure 4 is a block diagram of a system for training a mask generator as an MRC-aware or process-aware mask generator consistent with various embodiments.

[0018] Figure 5 is a flow chart of a method for training a mask generator as an MRC-aware or process-aware mask generator consistent with various embodiments.

[0019] Figure 6 is a flow chart of a method for determining mask rule compliance (MRC) loss consistent with various embodiments.

[0020] Figure 7 is a flow chart of a method for determining a first simulated loss consistent with various embodiments.

[0021] Figure 8 is a flow chart of a method for determining a second simulated loss consistent with various embodiments.

[0022] Fig. 9 An example neural network 900 for implementing mask generator 350 is shown, consistent with various embodiments.

[0023] Fig.10 is a block diagram of an example computer system according to one embodiment of the present disclosure. DETAILED DESCRIPTION

[0024] In lithography, in order to print a target pattern (also commonly referred to as a "design layout" or "design" or "target layout") on a substrate, a pattern of a pattern forming device (e.g., a "mask pattern" of a mask) is projected onto a resist layer provided on a substrate (e.g., a wafer). The mask pattern can be projected onto one or more tube cores of the substrate. In some embodiments, the design layout or a portion of the design layout is used to design a mask to be used in semiconductor manufacturing. Generating a mask design (also referred to as a "mask pattern") includes determining mask features based on a mask optimization simulation. Some techniques use a predictive model (e.g., a machine learning (ML) model, such as a neural network) to predict a mask pattern based on a target pattern. For example, a target image set with a target pattern and a corresponding mask image set with a mask pattern are used as ground truth images to train an ML model to generate a mask image. The ML model can learn a transfer function from a target pattern to a mask pattern by focusing on a faithful reconstruction of the ground truth image. However, conventional ML models may have some disadvantages. For example, conventional ML models are not guided by optical proximity correction (OPC) applications and therefore may not be aware of various metrics, such as mask rule checking or mask rule compliance (MRC) rules or other process simulation related metrics (e.g., edge placement error (EPE), sub-resolution assist features (SRAF), or other issues). This disconnect may result in less useful and unstable solutions. In addition, in some cases, critical prediction errors may come from slight deviations from the ground truth and are therefore ignored by the ML model. Some ML models can be configured to consider one or more of the above metrics when predicting mask images, but even these ML models have some shortcomings. For example, the ML models may not be trained using a supervised learning paradigm, that is, they may not be guided by the ground truth images to generate mask images. Although they can accept target images as input, they do not generate mask images, but rather mask layouts (which are polygonal) as output. That is, conventional ML models do not operate in the image-to-image domain and therefore may require additional image processing steps to generate various images to determine the metrics and train the ML model, consuming a lot of computing resources.

[0025] Disclosed herein is a mechanism for improving the prediction of a mask pattern (e.g., having curved mask features) from a target layout using an MRC-aware or process-aware prediction model. The prediction model is configured to predict a mask image (e.g., a mask pattern) from a target image (e.g., a target layout). The prediction model can be trained according to a loss function (or cost function) that considers a first loss function and a second loss function, the first loss function indicating an image reconstruction loss (e.g., the difference between a predicted mask image and a ground truth mask image), the second loss function indicating at least one of: (a) an MRC loss or (b) a simulation loss (e.g., determined based on an evaluation of an image simulated from the predicted mask image to determine a process indicator, such as EPE, SRAF printing, etc.). In some embodiments, the MRC loss indicates an MRC violation, which can be determined by performing an MRC evaluation on the predicted mask image to identify an MRC violation and scoring the identified MRC violation. In some embodiments, the simulated loss indicates a process indicator (e.g., EPE) that can be determined based on the difference between a simulated image (e.g., an aerial image, a resist image, or an etch image) of the predicted mask image and a ground truth mask image (e.g., a near target feature in the simulated image). In some embodiments, the simulated loss also indicates another process indicator (e.g., SRAF print) that can be determined by simulating an image (e.g., an aerial image, a resist image, or an etch image) of the predicted mask image, comparing the pixel values ​​of the simulated image with a threshold, and scoring pixels associated with values ​​exceeding the threshold. After determining at least one of the MRC loss or the simulated loss and the image reconstruction loss, the prediction model (e.g., parameters of the prediction model) can be modified to minimize the loss function. The trained prediction model can then be used to predict a mask image in which MRC violations or other process indicators (e.g., EPE, SRAF print) are minimized or eliminated. The prediction model can include one or more of the following: an ML model (e.g., a neural network), a statistical model, an analytical model, a rule-based model, or any other empirical model.

[0026] Although IC manufacturing may be specifically mentioned herein, it should be clearly understood that the description herein has many other possible applications. For example, it can be used to manufacture integrated optical systems, guidance and detection patterns for magnetic domain memories, liquid crystal display panels, thin film magnetic heads, etc. The skilled person will understand that in the context of such alternative applications, any use of the terms "reticle", "wafer" or "die" herein should be considered interchangeable with the more general terms "mask", "substrate" and "target portion", respectively.

[0027] In this document, the terms "radiation" and "beam" are used to cover all types of electromagnetic radiation, including UV radiation (e.g., having a wavelength of 365nm, 248nm, 193nm, 157nm or 126nm) and EUV (extreme ultraviolet radiation, e.g., having a wavelength range of about 5-100nm).

[0028] The terms "optimize" and "optimize" as used herein refer to or mean adjusting the lithography projection equipment, lithography process, etc. so that the results and / or processes of the lithography have more ideal characteristics, such as higher projection accuracy of the design layout on the substrate, a larger process window, etc. Therefore, the terms "optimize" and "optimize" as used herein refer to or mean the process of identifying one or more values ​​of one or more parameters, which provide improvements in at least one related indicator, such as a local optimum, compared to the initial one or more value sets of these one or more parameters. "Optimal" and other related terms should be interpreted accordingly. In one embodiment, the optimization step can be iteratively applied to provide further improvements in one or more indicators.

[0029] Furthermore, the lithographic projection apparatus may be of a type having two or more worktables (e.g., two or more substrate tables, a substrate table and a measurement table, two or more pattern forming device tables, etc.). In such a "multi-stage" apparatus, multiple worktables may be used in parallel, or preparation steps may be performed on one or more worktables while exposure is performed using one or more other worktables. For example, a dual-stage lithographic projection apparatus is described in US5,969,441, which is incorporated herein by reference.

[0030] The above-mentioned pattern forming device includes or can form one or more design layouts. The design layout can be generated using a CAD (computer-aided design) program, which is generally referred to as EDA (electronic design automation). Most CAD programs follow a set of predetermined design rules to create a functional design layout / pattern forming device. These rules are set by processing and design constraints. For example, the design rules define the spatial tolerances between circuit devices (such as gates, capacitors, etc.) or interconnects to ensure that the circuit devices or lines do not interact in a bad way. One or more design rule restrictions can be referred to as "critical dimensions" (CD). The critical dimension of a circuit can be defined as the minimum width of a line or hole or the minimum space between two lines or two holes. Therefore, CD determines the overall size and density of the design circuit. Of course, one of the goals of integrated circuit manufacturing is to faithfully reproduce the original circuit design on the substrate (via a pattern forming device).

[0031] The terms "mask" or "patterning device" as used herein may be broadly interpreted as referring to a generic patterning device that can be used to impart an incident radiation beam with a patterned cross-section corresponding to the pattern to be created in a target portion of the substrate; the term "light valve" may also be used in this context. In addition to classical masks (transmissive or reflective; binary, phase-shifting, hybrid, etc.), examples of other such patterning devices include:

[0032] Programmable mirror arrays. An example of such a device is a matrix-addressable surface having a viscoelastic control layer and a reflective surface. The basic principle behind such a device is that (for example) the addressed areas of the reflective surface reflect the incident radiation as diffracted radiation, while the unaddressed areas reflect the incident radiation as non-diffracted radiation. Using appropriate filters, the non-diffracted radiation can be filtered out from the reflected beam, leaving only the diffracted radiation; in this way, the beam is patterned according to the addressing pattern of the matrix-addressable surface. The required matrix addressing can be performed using suitable electronic devices. More information about such mirror arrays can be gleaned from, for example, U.S. Patent Nos. 5,296,891 and 5,523,193 (incorporated herein by reference).

[0033] Programmable LCD Array An example of such a configuration is given in US Pat. No. 5,229,872, which is hereby incorporated by reference.

[0034] Briefly introduce, Figure 1 An exemplary lithographic projection apparatus 10A is shown. The main components are a radiation source 12A, which may be a deep ultraviolet excimer laser source or other types of sources, including an extreme ultraviolet (EUV) source (as described above, the lithographic projection apparatus itself need not have a radiation source); illumination optics, which define the partial coherence (expressed as sigma) and may include optics 14A, 16Aa, and 16Ab, which shape the radiation from source 12A; a patterning device 14A; and transmission optics 16Ac, which projects an image of the patterning device pattern onto a substrate plane - 22A. An adjustable filter or aperture 20A at a pupil plane of the projection optics may limit the range of beam angles incident on substrate plane 22A, where the maximum possible angle defines the numerical aperture NA of the projection optics = n sin (Θ max ), n is the refractive index of the medium between the last element of the projection optics and the substrate, Θ maxIt is the maximum angle of the light beam emitted from the projection optical device that can still be incident on the substrate plane 22A. The radiation from the radiation source 12A is not necessarily a single wavelength. Instead, the radiation can be a series of different wavelengths. The range of different wavelengths can be characterized by a quantity called "imaging bandwidth", "source bandwidth" or simply "bandwidth", which are used interchangeably in this article. A small bandwidth can reduce chromatic aberration and related focusing errors of downstream components, including optical devices in the light source (such as optical devices 14A, 16Aa and 16Ab), pattern forming devices, and projection optical devices. However, this does not necessarily lead to a rule that the bandwidth should never be expanded.

[0035] In the optimization process of the system, the quality factor of the system can be expressed as a cost function. The optimization process is reduced to the process of finding a set of system parameters (design variables) that optimize (e.g., minimize or maximize) the cost function. The cost function can have any suitable form, depending on the goal of the optimization. For example, the cost function can be the weighted root mean square (RMS) of the deviation of certain characteristics (evaluation points) of the system relative to the expected values ​​(e.g., ideal values) of these characteristics; the cost function can also be the maximum value of these deviations (i.e., the worst deviation). The term "evaluation point" in this article should be broadly interpreted to include any characteristic of the system. Due to the practicality of the implementation of the system, the design variables of the system can be limited to a limited range and / or interdependent. For lithographic projection equipment, constraints are generally related to the physical properties and characteristics of the hardware (e.g., adjustable range) and / or the manufacturability design rules of the pattern forming device, and the evaluation points can include physical points on the resist image on the substrate, as well as non-physical characteristics (such as dose and focus).

[0036] In a lithographic projection apparatus, a light source provides illumination (i.e., radiation) to a pattern forming device, and projection optics guides and shapes the illumination onto a substrate through the pattern forming device. The term "projection optics" is broadly defined herein to include any optical element that can change the wavefront of a radiation beam. For example, projection optics may include at least some of components 14A, 16Aa, 16Ab, and 16Ac. An aerial image (AI) is a distribution of radiation intensity at the substrate level. A resist layer on a substrate is exposed, and an aerial image is transferred into the resist layer as a potential "resist image" (RI) therein. A resist image (RI) may be defined as the spatial distribution of the solubility of the resist in the resist layer. A resist model may be used to calculate a resist image based on an aerial image, an example of which may be found in U.S. Patent No. 8,200,468, the disclosure of which is incorporated herein by reference in its entirety. The resist model is only related to the properties of the resist layer (e.g., the effects of chemical processes occurring during exposure, PEB, and development). The optical properties of a lithographic projection apparatus (e.g., properties of the source, patterning device, and projection optics) determine the aerial image. Since the patterning device used in a lithographic projection apparatus can vary, it is best to separate the optical properties of the patterning device from the optical properties of the rest of the lithographic projection apparatus (including at least the source and projection optics).

[0037] Figure 2 An exemplary flow chart for modeling and / or simulating various parts of a patterning process is shown. It will be appreciated that these models may represent different patterning processes and need not include all of the models described below. The source model 200 represents the optical properties of the illumination of the patterning device (including the radiation intensity distribution, bandwidth, and / or phase distribution). The source model 200 may represent the optical properties of the illumination, including but not limited to the numerical aperture setting, the illumination sigma (σ) setting, and any particular illumination shape (e.g., off-axis radiation shapes such as toroidal, quadrupole, dipole, etc.), where σ (or sigma) is the outer radial extent of the illuminator.

[0038] The projection optics model 210 represents the optical properties of the projection optics (including changes in the radiation intensity distribution and / or phase distribution caused by the projection optics). The projection optics model 210 may represent the optical properties of the projection optics, including aberrations, distortions, one or more refractive indices, one or more physical sizes, one or more physical dimensions, etc.

[0039] The patterning device / design layout model module 220 captures how design features are laid out in a pattern of a patterning device and may include a representation of detailed physical properties of the patterning device, for example, as described in U.S. Pat. No. 7,587,704, which is incorporated herein by reference in its entirety. In one embodiment, the patterning device / design layout model module 220 represents the optical properties (including changes in radiation intensity distribution and / or phase distribution caused by a given design layout) of a design layout (e.g., a device design layout corresponding to features of an integrated circuit, memory, electronic device, etc.), which is a representation of the arrangement of features on a patterning device or the arrangement of features formed by a patterning device. Since the patterning device used in a lithographic projection apparatus can vary, it is desirable to separate the optical properties of the patterning device from the optical properties of the rest of the lithographic projection apparatus (including at least the illumination and projection optics). The purpose of the simulation is typically to accurately predict edge placement and CD, etc., which are then compared to the device design. The device design is typically defined as a pre-OPC patterning device layout and is provided in a standardized digital file format (such as GDSII or OASIS).

[0040] An aerial image 230 can be simulated from the source model 200, the projection optics model 210, and the patterning device / design layout model module 220. The aerial image (AI) is the radiation intensity distribution at the substrate level. The optical characteristics of the lithographic projection apparatus (e.g., the characteristics of the illumination, patterning device, and projection optics) determine the aerial image.

[0041] A resist layer on a substrate is exposed by an aerial image, and the aerial image is transferred to the resist layer as a potential "resist image" (RI) therein. The resist image (RI) can be defined as the spatial distribution of the solubility of the resist in the resist layer. A resist model 240 can be used to simulate a resist image 250 from an aerial image 230. A resist model can be used to calculate a resist image from an aerial image, an example of which can be found in U.S. Patent Application No. 8,200,468, the disclosure of which is hereby incorporated by reference in its entirety. A resist model typically describes the effects of chemical processes occurring during resist exposure, post-exposure baking (PEB), and development in order to predict, for example, the profile of a resist feature formed on a substrate, and therefore it is typically only related to such properties of the resist layer (e.g., the effects of chemical processes occurring during exposure, post-exposure baking, and development). In one embodiment, optical properties of the resist layer (e.g., refractive index, film thickness, propagation, and polarization effects) can be captured as part of a projection optical model 210.

[0042] Thus, in general, the link between the optical model and the resist model is the simulated aerial image intensity within the resist layer, which results from radiation projected onto the substrate, refraction at the resist interface, and multiple reflections in the resist film stack. The radiation intensity distribution (aerial image intensity) becomes a latent "resist image" by absorbing the incident energy and is further modified by diffusion processes and various loading effects. For full-chip applications, efficient simulation methods that are fast enough approximate the true three-dimensional intensity distribution in the resist stack via a two-dimensional spatial (and resist) image.

[0043] In one embodiment, the resist image may be used as an input to a post-pattern transfer process model module 260. The post-pattern transfer process model module 260 defines the performance of one or more post-resist development processes (eg, etching, developing, etc.).

[0044] For example, simulation of the patterning process can predict the profile, CD, edge placement (e.g., edge placement error), etc. in the resist and / or etch image. Thus, the purpose of the simulation is to accurately predict the edge placement, aerial image intensity slope, and / or CD, etc. of the printed pattern. These values ​​can be compared to the intended design, for example, to correct the patterning process, identify locations where defects are expected to occur, etc. The intended design is typically defined as a pre-OPC design layout that can be provided in a standardized digital file format (e.g., GDSII or OASIS or other file format).

[0045] Therefore, the model formula describes most, if not all, of the known physical and chemical properties of the entire process, and each model parameter corresponds at best to a different physical or chemical effect. The model formula therefore places an upper limit on how well the model can simulate the entire manufacturing process.

[0046] Typically, a mask may have thousands or even millions of mask features that can perform MRC. MRC can be performed for each mask feature. The mask feature can be any of a variety of shapes, such as a curved mask feature. In some embodiments, the MRC specification may include the minimum critical dimension (CD) of a manufacturable mask feature, the minimum curvature of a manufacturable mask feature, the minimum area of ​​a mask feature, the minimum space between two features, or other geometric properties related to the mask feature. When the geometric properties of the mask feature do not meet the constraints specified in the MRC, an MRC violation may occur. For example, when the area of ​​the mask feature is less than the minimum area specified in the MRC, an MRC violation may occur. In another example, when the space between two mask features is less than the minimum space specified in the MRC, an MRC violation may occur.

[0047] Similarly, due to various reasons, including incorrect feature geometry of the mask pattern (e.g., size, shape, position in the mask pattern, etc.), a process-related metric such as EPE may occur, which indicates that a feature or a portion thereof in the resist is offset or changed relative to the expected position of the feature in the target layout. Similarly, due to incorrect feature geometry (e.g., when the size of the SRAF is greater than a threshold size), another process-related metric such as SRAF overprint may occur, which causes certain features (e.g., SRAF) in the mask pattern that are not intended to be printed on the substrate to be printed. The following paragraphs describe configuring a prediction model to minimize MRC violations or patterning process-related metrics (e.g., EPE, SRAF overprint, etc.) in order to predict (e.g., generate) a mask image from a target image.

[0048] Figure 3 3 is a block diagram of a system 300 for generating a mask pattern from a target layout using a prediction model consistent with various embodiments. A target image 305 is input to a mask generator 350, which generates a mask image 315 of a mask pattern. The mask pattern can have a plurality of mask features. The mask features can be any of a variety of shapes, such as curved mask features. In some embodiments, the mask generator 350 is configured to generate a mask pattern such that MRC violations and at least one lithography process indicator (e.g., EPE or SRAF print) are minimized or eliminated.

[0049] The target image 305 may include a target layout to be printed on a substrate. The target layout includes many features, is typically defined as a pre-OPC design layout, and may be provided in a standardized digital file format (eg, GDSII or OASIS or other file formats).

[0050] The mask generator 350 can be implemented as a predictive model, such as an ML model (e.g., a neural network), a statistical model, an analytical model, a rule-based model, or any other empirical model. In some embodiments, the mask generator 350 is implemented as a neural network. For example, a neural network can be based on a large number of neural units (or artificial neurons). A neural network can loosely imitate the way a biological brain works (e.g., through a large number of biological neuron clusters connected by axons). Each neural unit of a neural network can be connected to many other neural units of the neural network. This connection can strengthen or inhibit the activation state of the connected neural units. In some embodiments, each individual neural unit can have a summation function that combines the values ​​of all its inputs. In some embodiments, each connection (or neural unit itself) can have a threshold function so that the signal must exceed the threshold before it propagates to other neural units. These neural network systems can be self-learning and training, rather than explicitly programmed, and can perform better in certain problem-solving areas than traditional computer programs. In some embodiments, a neural network can include multiple layers (e.g., the signal path is from the front layer to the back layer). In some embodiments, a neural network can utilize a back-propagation technique, in which a forward stimulus is used to reset the weights of the "front" neural unit. In some embodiments, the excitation and inhibition of neural networks may be more free-flowing, with connections interacting in more chaotic and complex ways.

[0051] The mask generator 350 is configured as an MRC-aware or process-aware mask generator, that is, the mask generator 350 generates a mask pattern such that MRC violations and at least one lithography process indicator (e.g., EPE or SRAF printing) are minimized or eliminated. A mask having the generated mask pattern can be manufactured using the predicted mask image 315 and can be used in a patterning step to print a pattern corresponding to the target image 305 on a substrate by a lithography process. The process of training the mask generator 350 to generate the mask image 315 is at least based on Figure 4-Figure 8 Described below.

[0052] Figure 4 is a block diagram of a system 400 for training the mask generator 350 as an MRC-aware or process-aware mask generator, consistent with various embodiments. Figure 5Flowchart of a method 500 for training a mask generator 350 as an MRC-aware or process-aware mask generator consistent with various embodiments. At process P502, a target image set 405 and a reference image set 410 are input to the mask generator 350. The target image 405a of the image set 405 can be an image of a target layout to be printed on a substrate. The reference image 410a of the reference image set 410 can be an image of an OPC mask corresponding to the target layout of the target image 405a. The reference image set 410 can serve as a ground truth mask image for training the mask generator 350 to predict a mask image based on the target image. The reference image 410a can be generated in a variety of ways, for example, using an SMO or OPC method.

[0053] In process P504, mask generator 350 generates a mask image corresponding to the target image. For example, mask generator 350 generates mask image 415a representing a mask pattern corresponding to the target layout in target image 405a. As described above, mask generator 350 can be implemented as a prediction model, such as a neural network.

[0054] In process P506, the loss function component 450 calculates a first loss function indicating an image reconstruction loss, which is determined as a difference between the predicted mask image and the reference image. For example, the first loss function 420 may include an image reconstruction loss 420, which may be determined as a difference between the predicted mask image 315 and the reference image 410a.

[0055] In process P508, the loss function component 450 calculates a second loss function 505 that indicates at least one of an MRC evaluation of the predicted mask image or an evaluation of a simulated image of the predicted mask image. For example, the second loss function 505 includes an MRC loss 425 or an MRC cost 425 that indicates an MRC violation of a mask feature of the mask pattern, which MRC violation may be determined based on an MRC evaluation of the predicted mask image 415a, such as by reference to at least Figure 6 In another example, the second loss function 505 includes a first simulation cost or a first simulation loss 430, which indicates a process indicator (e.g., EPE) that can be determined based on a difference between a simulated image (e.g., an aerial image, a resist image, or an etch image) of the predicted mask image 415a and the reference image 410a, such as at least the reference image Figure 7In another example, the second loss function 505 includes a second simulation cost or a second simulation loss 435 indicating a process indicator (e.g., SRAF printing), which can be determined based on pixel values ​​of a simulation image (e.g., an aerial image, a resist image, or an etching image) of the predicted mask image 415a and a threshold, such as at least referring to Figure 8 described.

[0056] In process P510, the mask generator 350 may be modified or updated according to the first loss function and the second loss function. For example, the configuration of the mask generator 350 may be updated to reduce the first loss function 420 and the second loss function 505. The first loss function 420 may include an image reconstruction loss 420. The second loss function 505 may include at least one of an MRC loss 425, a first simulation loss 430, and a second simulation loss 435. In an embodiment where the mask generator 350 is a neural network, updating the configuration of the mask generator 350 includes updating the configuration (e.g., weights, biases, or other parameters) of the neural network according to the loss function. For example, the connection weights may be adjusted to reconcile the difference between the prediction of the neural network (e.g., the predicted mask image 415a) and the reference feedback (reference image 410a). In another use case, one or more neurons (or nodes) of the neural network may need to send corresponding errors backward through the neural network to facilitate the update process (e.g., backward propagation of errors). For example, the update to the connection weights may reflect the magnitude of the error (e.g., loss function) propagated backward after the forward pass is completed. For example, in this manner, mask generator 350 may be trained to generate better predictions (eg, mask images).

[0057] In some embodiments, the method 500 of training the mask generator 350 is an iterative process, wherein each iteration includes generating a predicted mask image (e.g., the predicted mask image 415a), calculating a first loss function (e.g., the image reconstruction loss 420) and a second loss function 505 (e.g., the MRC loss 425, the first simulation loss 430, or the second simulation loss 435), determining whether the first and second loss functions are minimized, and updating the configuration of the mask generator 350 to reduce the first loss function 420 and the second loss function 505. The iteration may be performed until a specified condition is met (e.g., a predetermined number of times, until the first and second loss functions are minimized, or other conditions).

[0058] After the training method 500 is completed, the mask generator 350 is considered to be trained as an MRC-aware or process-aware mask generator, which can be used to generate or predict a mask image representing a mask pattern in which MRC violations and process indicators (e.g., EPE, SRAF printing) are minimized or eliminated, such as at least referring to Figure 3 described.

[0059] Figure 6 6 is a flow chart of a method 600 for determining MRC loss consistent with various embodiments. In some embodiments, method 600 may be performed as part of process P506 of method 500. At process P602, an image is input to loss function component 450. For example, predicted mask image 315 is input to loss function component 450.

[0060] In process P604, the loss function component 450 performs MRC on the predicted mask image 415a to identify the MRC violation 605. In some embodiments, performing the MRC evaluation may involve determining whether the geometric characteristics of the mask features in the mask pattern meet the MRC specification. For example, performing MRC may involve determining whether the size of the mask feature is greater than the minimum size specified in the MRC specification. If the size is less than the minimum size, the loss function component can identify the MRC violation. The loss function component 450 can process multiple mask features of the mask pattern and identify the MRC violation 605.

[0061] In process P606, the loss function component 450 can assign a violation score 610 to each MRC violation 605. The violation score 610 can be determined in a variety of ways. In some embodiments, the violation score 610 can be determined based on the MRC violation 605. For example, the greater the magnitude of the MRC violation, the greater the violation score 610 may be.

[0062] In process P608, the loss function component 450 can determine the MRC loss 425 based on the violation score 610. The MRC loss 425 can be determined in a variety of ways. For example, the MRC loss 425 can be determined as a value between 0 and 1, and the greater the violation score 610, the greater the MRC loss 425 may be.

[0063] Figure 7 is a flow chart of a method 700 for determining a first simulated loss consistent with various embodiments. In some embodiments, method 700 may be implemented as part of process P508 of method 500. In some embodiments, first simulated loss 430 indicates a process indicator, such as EPE.

[0064] In process P702, an image is input to the loss function component 450. For example, the predicted mask image 415a is input to the loss function component 450.

[0065] In process P704, the loss function component 450 simulates a first image 705 from the predicted mask image 415a. The first simulated image 705 can be an aerial image, a resist image, or an etch image. In some embodiments, the first simulated image 705 is an aerial image. The first simulated image 705 can be generated in various ways. For example, at least a reference Figure 2 The process or component described above is used to generate the first simulated image 705. In some embodiments, the predicted mask image 415a may be upsampled to increase the resolution of the predicted mask image 415a. Additionally, in some embodiments, the predicted mask image 415a may be a continuous transmission mask (CTM) image and may be binarized to generate a binarized mask image.

[0066] In process P706, the loss function component 450 simulates a second image 710 based on the reference image 410a. The second simulated image 710 can be an aerial image, a resist image, or an etched image. In some embodiments, the second simulated image 710 is an aerial image. In some embodiments, the second simulated image 710 can be simulated according to the same conditions or constraints used when generating the first simulated image 705.

[0067] In process P708, the loss function component 450 determines the first simulated loss 430 based on the difference between the first simulated image 705 and the second simulated image 710. By obtaining the difference between the first simulated image 705 and the second simulated image 710, the loss function component 450 can determine any EPE of the mask feature in the predicted mask image 415a based on the ground truth mask image 410a. In some embodiments, the loss function component 450 can perform a comparison of the simulated images in an area near (e.g., within a specified proximity range) a target feature (e.g., a feature intended to be printed on a substrate). For example, the loss function component 450 can use a fixed filter (e.g., a transmission cross coefficient (TCC) kernel) to identify areas in the simulated images that are close to the target feature. In some embodiments, differences between simulated images that are close to the target feature have a greater weight than differences between simulated images that are far from the target feature.

[0068] Figure 8 is a flow chart of a method 800 for determining a second simulated loss consistent with various embodiments. In some embodiments, method 800 may be implemented as part of process P508 of method 500. In some embodiments, second simulated loss 435 indicates a process indicator, such as SRAF printing.

[0069] In process P802, an image is input to the loss function component 450. For example, the predicted mask image 415a is input to the loss function component 450.

[0070] In process P804, the loss function component 450 simulates a third image 805 from the predicted mask image 415a. The third simulated image 805 can be an aerial image, a resist image, or an etch image. In some embodiments, the third simulated image 805 is an aerial image. The third simulated image 805 can be generated in various ways. For example, at least a reference Figure 2 The process or components described above are used to generate the third simulated image 805. In some embodiments, the predicted mask image 415a may be upsampled to increase the resolution of the predicted mask image 415a. In addition, in some embodiments, the predicted mask image 415a may be a CTM image and may be binarized to generate a binarized mask image.

[0071] In process P806, the loss function component 450 compares the pixel values ​​of the third simulated image 805 with a threshold value and calculates a score 810 as a function of the pixel values ​​exceeding the threshold value. In some embodiments, the loss function component 450 can use a fixed filter (e.g., a TCC kernel) to identify areas of the simulated image that are far away from the target feature (e.g., features intended to be printed on the substrate) (e.g., outside a specified proximity range) and compare in the identified area. In some embodiments, by comparing in areas far away from the target feature, the loss function component 450 can identify those features in the mask pattern that are not intended to be printed on the substrate (e.g., SRAF). In some embodiments, by comparing in areas far away from the target feature and determining pixel values ​​exceeding the threshold value, the loss function component 450 can identify features in the mask pattern that are larger than a predetermined size so that they are printed on the substrate and penalize these features by assigning appropriate scores. In some embodiments, the portion of the third simulated image 805 that is far away from the target feature and exceeds the threshold value has a greater weight than the portion that is close to the target feature and exceeds the threshold value.

[0072] The threshold value can be obtained in a variety of ways. For example, the threshold value can be obtained by user input. In another example, the threshold value can be obtained by simulating an image from the reference image 410a (e.g., by simulating an aerial image from a ground truth mask image and obtaining a threshold score from the simulated image).

[0073] In process P808, the loss function component 450 can determine the second simulated loss 435 according to the score. The second simulated loss 435 can be determined in a variety of ways. For example, the second simulated loss 435 can be determined as a value between 0 and 1, and the larger the score 810, the larger the second simulated loss 435.

[0074] Fig. 9 An example neural network 900 for implementing mask generator 350 consistent with various embodiments is shown. The neural network may include an input layer, an output layer, and one or more intermediate or hidden layers. In some embodiments, one or more neural networks may be and / or include a deep neural network (e.g., a neural network having one or more intermediate or hidden layers between the input layer and the output layer). As an example, one or more neural networks may be based on a large number of neural units (or artificial neurons). As described above with at least reference to Figure 4 and Figure 5 As described above, when training a neural network to generate a predicted mask image (e.g., predicted mask image 415a), the configuration of the neural network can be updated to minimize the first loss function 420 and the second loss function 505. The neural network can be updated in a variety of ways. For example, the configuration of the neural network (e.g., weights, biases, or other parameters) can be updated for the entire neural network based on the two loss functions. In another example, Fig. 9 As shown in the neural network 900, the configuration of the first part 920 of the neural network 900 can be updated to reduce or minimize the first loss function 420, and the configuration of the second part 930 of the neural network 900 can be updated to reduce or minimize the second loss function 505. That is, in some embodiments, the first part 920 of the neural network 900 can be trained (according to the first loss function 420) to generate the intermediate mask image 905a according to the target image 405a that is faithful to the reference image 410a, and the second part 930 of the neural network 900 can be trained (according to the second loss function 505) to generate the predicted mask image 415a according to the intermediate mask image 905a, so that the MRC violation and at least one process indicator (e.g., EPE, SRAF printing, etc.) are minimized or eliminated in the predicted mask image 415a.

[0075] Fig.10 is a block diagram illustrating a computer system 100 that can assist in implementing the optimization methods and processes disclosed herein. The computer system 100 can be used to implement any entity, component, module, or service depicted in the example of the figure (as well as any other entity, component, module, or service described in this specification). The computer system 100 can be programmed to execute computer program instructions to perform the functions, methods, processes, or services described herein (e.g., the functions, methods, processes, or services of any entity, component, or module). The computer system 100 can be programmed to execute computer program instructions through at least one of software, hardware, or firmware.

[0076] The computer system 100 includes a bus 102 or other communication mechanism for communicating information, and a processor 104 (or multiple processors 104 and 105) coupled to the bus 102 for processing information. The computer system 100 also includes a main memory 106, such as a random access memory (RAM) or other dynamic storage device, which is coupled to the bus 102 for storing information and instructions to be executed by the processor 104. The main memory 106 can also be used to store transient variables or other intermediate information during the execution of instructions to be executed by the processor 104. The computer system 100 also includes a read-only memory (ROM) 108 or other static storage device, which is coupled to the bus 102 for storing static information and instructions for the processor 104. A storage device 110, such as a magnetic disk or optical disk, is provided and coupled to the bus 102 for storing information and instructions.

[0077] The computer system 100 may be coupled via bus 102 to a display 112, such as a cathode ray tube (CRT) or a flat panel or touch panel display, for displaying information to a computer user. An input device 114, including alphanumeric and other keys, is coupled to bus 102 for communicating information and command selections to processor 104. Another type of user input device is a cursor control 116, such as a mouse, trackball, or cursor direction keys, for communicating direction information and command selections to processor 104 and for controlling cursor movement on display 112. The input device typically has two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), allowing the device to specify a position in a plane. A touch panel (screen) display may also be used as an input device.

[0078] According to one embodiment, portions of the optimization process may be performed by computer system 100 in response to processor 104 executing one or more sequences of one or more instructions contained in main memory 106. Such instructions may be read into main memory 106 from another computer-readable medium (e.g., storage device 110). Execution of the sequences of instructions contained in main memory 106 causes processor 104 to perform the processing steps described herein. One or more processors in a multi-processing arrangement may also be employed to execute the sequences of instructions contained in main memory 106. In alternative embodiments, hard-wired circuitry may be used instead of or in combination with software instructions. Therefore, the description herein is not limited to any specific combination of hardware circuitry and software.

[0079] The term "computer-readable medium" as used herein refers to any medium that participates in providing instructions to the processor 104 for execution. Such media can take a variety of forms, including but not limited to non-volatile media, volatile media, and transmission media. Non-volatile media include, for example, optical or magnetic disks, such as storage device 110. Volatile media include dynamic memory, such as main memory 106. Transmission media include coaxial cables, copper wires, and optical fibers, including wires that make up bus 102. Transmission media can also take the form of sound waves or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer-readable media include, for example, floppy disks, flexible disks, hard disks, magnetic tapes, any other magnetic media, CD-ROMs, DVDs, any other optical media, punch cards, paper tapes, any other physical media with hole patterns, RAM, PROMs and EPROMs, FLASH-EPROMs, any other memory chips or cartridges, carrier waves as described below, or any other medium that can be read by a computer.

[0080] Various forms of computer readable media may be used to carry one or more sequences of one or more instructions to processor 104 for execution. For example, the instructions may initially be carried on a disk of a remote computer. The remote computer may load the instructions into its dynamic memory and send the instructions over a telephone line using a modem. A modem local to computer system 100 may receive the data on the telephone line and use an infrared transmitter to convert the data to an infrared signal. An infrared detector coupled to bus 102 may receive the data carried in the infrared signal and place the data on bus 102. Bus 102 carries the data to main memory 106, from which processor 104 retrieves and executes the instructions. The instructions received by main memory 106 may optionally be stored on storage device 110 before or after execution by processor 104.

[0081] The computer system 100 may also include a communication interface 118 coupled to the bus 102. The communication interface 118 provides a two-way data communication coupled to a network link 120 connected to a local network 122. For example, the communication interface 118 may be an integrated services digital network (ISDN) card or a modem to provide a data communication connection to a corresponding type of telephone line. As another example, the communication interface 118 may be a local area network (LAN) card to provide a data communication connection to a compatible LAN. A wireless link may also be implemented. In any such implementation, the communication interface 118 sends and receives electrical, electromagnetic, or optical signals that carry digital data streams representing various types of information.

[0082] The network link 120 typically provides data communication through one or more networks to other data devices. For example, the network link 120 may provide a connection through a local network 122 to a host computer 124 or to data equipment operated by an Internet Service Provider (ISP) 126. The ISP 126, in turn, provides data communication services through a global packet data communication network, now commonly referred to as the "Internet" 128. Both the local network 122 and the Internet 128 use electrical, electromagnetic, or optical signals that carry digital data streams. The signals through the various networks and the signals on the network link 120 and communication interface 118, which carry the digital data to and from the computer system 100, are exemplary forms of carrier waves that transport the information.

[0083] The computer system 100 can send messages and receive data (including program code) through the network, network link 120, and communication interface 118. In the Internet example, server 130 can transmit the requested code of the application program through Internet 128, ISP 126, local network 122, and communication interface 118. For example, one such downloaded application program can provide irradiation optimization of the embodiment. The received code can be executed by processor 104 when received, and / or stored in storage device 110 or other non-volatile memory for later execution. In this way, computer system 100 can obtain application program code in the form of a carrier wave.

[0084] Embodiments of the present disclosure may be further described by the following clauses.

[0085] 1. A non-transitory computer readable medium having instructions which, when executed by a computer, cause the computer to perform a method for training a machine learning model to generate a mask pattern to be used to print a target layout on a substrate, the method comprising:

[0086] inputting a target image to the machine learning model, the target image being associated with a target layout to be printed on a substrate, wherein the machine learning model is configured to receive a reference image corresponding to an optical proximity correction (OPC) mask image of the target image;

[0087] generating a predicted mask image using the neural network, the predicted mask image representing a mask pattern to be used to print the target layout on a substrate;

[0088] calculating a loss function indicating at least one of (a) a difference between the predicted mask image and the reference image, and (b) an MRC evaluation of the predicted mask image or an evaluation of a first simulated image of the predicted mask image; and

[0089] The machine learning model is modified based on the loss function.

[0090] 2. The computer-readable medium of clause 1, wherein calculating the loss function indicative of the MRC evaluation of the predicted mask image comprises:

[0091] A mask rule checking MRC cost is calculated by performing the MRC evaluation on the predicted mask image, the MRC cost indicating an MRC violation.

[0092] 3. The computer-readable medium of clause 2, wherein the MRC violation comprises a violation of at least one of: a critical dimension (CD), a width, an area of ​​a feature in the mask pattern.

[0093] 4. The computer-readable medium of clause 2, wherein performing the MRC evaluation comprises:

[0094] assigning a violation score to the portion of the mask pattern where the MRC violation occurs; and

[0095] The MRC cost is determined based on the violation score.

[0096] 5. The computer-readable medium of clause 1, wherein calculating the loss function indicative of the evaluation of the first simulated image of the predicted mask image comprises:

[0097] The first simulated image is generated based on the predicted mask image.

[0098] 6. The computer-readable medium of clause 5, wherein the first simulation image is at least one of: an aerial image, a resist image, or an etch image.

[0099] 7. The computer-readable medium of clause 5, wherein calculating the loss function comprises:

[0100] A simulation cost is calculated that indicates a difference between the first simulated image and a second simulated image of the reference image.

[0101] 8. The computer-readable medium of clause 7, wherein calculating the simulation cost comprises:

[0102] identifying areas in the first simulated image and the second simulated image that are within a specified proximity of a feature to be printed on a substrate; and

[0103] The simulation cost is calculated to indicate a difference between the first simulated image and the second simulated image within the area.

[0104] 9. The computer-readable medium of clause 5, wherein calculating the loss function comprises:

[0105] calculating a score indicating that a pixel value of each pixel of the first simulated image exceeds a threshold; and

[0106] A simulation cost is calculated based on the scores.

[0107] 10. The computer-readable medium of clause 9, wherein the simulation cost is determined based on a score associated with pixels outside a specified neighborhood of a feature in the first simulated image to be printed on the substrate.

[0108] 11. The computer-readable medium of clause 9, wherein the threshold is determined based on pixel values ​​of a second simulated image of the reference image.

[0109] 12. The computer-readable medium of clause 5, wherein the first simulated image is computed using a fixed filter.

[0110] 13. The computer-readable medium of clause 12, wherein the fixed filter is generated based on a transmission cross-coefficient kernel.

[0111] 14. The computer-readable medium of clause 5, wherein generating the first simulated image comprises:

[0112] The resolution of the predicted mask image is increased to generate the first simulated image.

[0113] 15. The computer-readable medium of clause 14, wherein the predicted mask image is a continuous transmission mask (CTM) image, and wherein the method further comprises:

[0114] A binarized image of the CTM image is generated.

[0115] 16. The computer-readable medium of clause 1, wherein modifying the neural network based on the loss function comprises:

[0116] Modifying parameters of the machine learning model until the loss function is minimized.

[0117] 17. The computer-readable medium of clause 1, wherein the machine learning network comprises a neural network, and wherein modifying the machine learning model based on the loss function comprises:

[0118] modifying parameters of the first portion of the neural network until the loss function indicative of the difference between the predicted mask image and the reference image is minimized; and

[0119] Parameters of a second portion of the neural network are modified until a loss function indicative of at least one of the MRC evaluation of the predicted mask image or an evaluation of the first simulated image of the predicted mask image is minimized.

[0120] 18. The computer-readable medium of clause 1 further comprising:

[0121] inputting a first target image having a first target layout to be printed on a first substrate into a neural network; and

[0122] A first predicted mask image is generated using a neural network, the first predicted mask image representing a first mask pattern for printing a first target layout on a first substrate.

[0123] 19. The computer-readable medium of clause 18 further comprising:

[0124] A mask having a first mask pattern is generated using the first predicted mask image.

[0125] 20. The computer-readable medium of clause 19, further comprising:

[0126] A patterning step is performed using a first mask to print a pattern corresponding to a first target layout on the first substrate by a photolithography process.

[0127] 21. A non-transitory computer readable medium having instructions which, when executed by a computer, cause the computer to perform a method for training a machine learning model to generate a mask image for printing a target layout on a substrate, the method comprising:

[0128] inputting a target image set and a reference image set as training data to a neural network, wherein the target images in the target image set include a target layout to be printed on a substrate, and the reference images in the reference image set correspond to optical proximity correction (OPC) mask images of the target images; and

[0129] Based on the training data, the machine learning model is trained to generate a predicted mask image such that a loss function indicating (a) the difference between the predicted mask image and the reference image and (b) an MRC cost associated with an MRC evaluation of the predicted mask image is minimized, wherein the predicted mask image represents a mask pattern for printing a target layout on a substrate.

[0130] 22. The computer-readable medium of clause 21, wherein training the machine learning model comprises:

[0131] performing an MRC evaluation on the predicted mask image to determine an MRC violation; and

[0132] Calculate MRC costs based on MRC violations.

[0133] 23. The computer-readable medium of clause 22, wherein the MRC violation comprises a violation of at least one of: a critical dimension (CD), a width, an area of ​​a feature in the mask pattern.

[0134] 24. The computer-readable medium of clause 22, wherein calculating the MRC cost comprises:

[0135] assigning a violation score to the portion of the mask pattern where the MRC violation occurs; and

[0136] The MRC cost is determined based on the violation score.

[0137] 25. The computer-readable medium of clause 21, wherein the machine learning model comprises a neural network, and wherein training the machine learning model comprises:

[0138] The parameters of the neural network are modified until the loss function is minimized.

[0139] 26. The computer-readable medium of clause 25, wherein modifying the neural network based on the loss function comprises:

[0140] modifying parameters of the first portion of the neural network until a loss function indicative of a difference between the predicted mask image and the reference image is minimized; and

[0141] Parameters of a second portion of the neural network are modified until the loss function indicative of the MRC cost is minimized.

[0142] 27. The computer-readable medium of clause 21 further comprising:

[0143] inputting a first target image having a first target layout to be printed on a first substrate into a neural network; and

[0144] A first predicted mask image is generated using a neural network, the first predicted mask image representing a first mask pattern for printing a first target layout on a first substrate.

[0145] 28. The computer-readable medium of clause 27, further comprising:

[0146] A mask having the first mask pattern is generated using the first predicted mask image.

[0147] 29. The computer-readable medium of clause 28, further comprising:

[0148] A patterning step is performed using the first mask to print a pattern corresponding to a first target layout on the first substrate by a photolithography process.

[0149] 30. A non-transitory computer readable medium having instructions which, when executed by a computer, cause the computer to perform a method for training a machine learning model to generate a mask image for printing a target layout on a substrate, the method comprising:

[0150] inputting a target image set and a reference image set as training data to a neural network, wherein the target images in the target image set include a target layout to be printed on a substrate, and the reference images in the reference image set correspond to optical proximity correction (OPC) mask images of the target images; and

[0151] Based on the training data, the machine learning model is trained to generate a predicted mask image such that a loss function indicating (a) a difference between the predicted mask image and a reference image and (b) a simulation cost associated with a first simulated image of the predicted mask image is minimized.

[0152] 31. The computer-readable medium of clause 30, wherein the first simulation image is at least one of: an aerial image, a resist image, or an etch image.

[0153] 32. The computer-readable medium of clause 30, wherein training the machine learning model comprises:

[0154] generating the first simulated image based on the predicted mask image; and

[0155] A simulation cost associated with the first simulated image is calculated.

[0156] 33. The computer-readable medium of clause 30, wherein calculating the simulation cost comprises:

[0157] The simulation cost is calculated based on a difference between the first simulated image and a second simulated image of the reference image.

[0158] 34. The computer-readable medium of clause 33, wherein calculating the simulation cost comprises:

[0159] identifying an area in the first simulated image and the second simulated image that is within a specified proximity of a feature to be printed on the substrate; and

[0160] A simulation cost representing a difference between the first simulated image and the second simulated image within the region is calculated.

[0161] 35. The computer-readable medium of clause 32, wherein calculating the simulation cost comprises:

[0162] calculating a fraction of the pixel values ​​for each pixel representing the first simulated image that exceed a threshold; and

[0163] The simulation cost is calculated based on the score.

[0164] 36. The computer-readable medium of clause 35, wherein the simulation cost is determined based on a score associated with pixels outside a specified neighborhood of a feature in a first simulated image to be printed on a substrate.

[0165] 37. A computer-readable medium according to clause 35, wherein the threshold value is determined based on pixel values ​​of the second simulated image of the reference image.

[0166] 38. The computer-readable medium of clause 32, wherein the first simulated image is computed using a fixed filter.

[0167] 39. The computer-readable medium of clause 38, wherein the fixed filter is generated based on a transmission cross coefficient kernel.

[0168] 40. The computer-readable medium of clause 32, wherein generating the first simulated image comprises:

[0169] The resolution of the predicted mask image is increased to generate the first simulated image.

[0170] 41. The computer-readable medium of clause 40, wherein the predicted mask image is a continuous transmission mask (CTM) image, and wherein the method further comprises:

[0171] Generate a binarized image of the CTM image.

[0172] 42. A method of training a machine learning model to generate a mask pattern for printing a target layout on a substrate, the method comprising:

[0173] inputting a target image to a neural network, the target image being associated with a target layout to be printed on a substrate, wherein the machine learning model is configured to receive a reference image corresponding to an optical proximity correction (OPC) mask image of the target image;

[0174] generating a predicted mask image using a neural network, the predicted mask image representing a mask pattern for printing a target layout on a substrate;

[0175] calculating a loss function indicative of at least one of (a) a difference between the predicted mask image and the reference image, and (b) an MRC evaluation of the predicted mask image or an evaluation of the first simulated image of the predicted mask image; and

[0176] The machine learning model is modified based on the loss function.

[0177] 43. An apparatus for training a machine learning model to generate a mask pattern for printing a target layout on a substrate, the apparatus comprising:

[0178] a memory storing an instruction set; and

[0179] A processor configured to execute the instruction set to cause the apparatus to perform the following steps:

[0180] inputting a target image to a neural network, the target image being associated with a target layout to be printed on a substrate, wherein the machine learning model is configured to receive a reference image corresponding to an optical proximity correction (OPC) mask image of the target image;

[0181] generating a predicted mask image using a neural network, the predicted mask image representing a mask pattern for printing a target layout on a substrate;

[0182] calculating a loss function indicative of at least one of (a) a difference between the predicted mask image and the reference image, and (b) an MRC evaluation of the predicted mask image or an evaluation of the first simulated image of the predicted mask image; and

[0183] Modify the machine learning model based on the loss function.

[0184] Although the concepts disclosed herein may be used for imaging on substrates such as silicon wafers, it should be understood that the disclosed concepts may be used in any type of lithographic imaging system, for example, a system for imaging on substrates other than silicon wafers.

[0185] The terms "optimization" and "optimization" used herein refer to or mean adjusting patterning equipment (e.g., lithography equipment), patterning processes, etc., so that the results and / or processes have more ideal characteristics, such as higher projection accuracy of the design pattern on the substrate, a larger process window, etc. Therefore, the terms "optimization" and "optimization" used herein refer to or mean the process of identifying one or more values ​​of one or more parameters, which provide improvements in at least one related indicator, such as local optimum, compared to the initial one or more value sets of these one or more parameters. "Optimization" and other related terms should be interpreted accordingly. In an embodiment, the optimization step can be iteratively applied to provide further improvements in one or more indicators.

[0186] Aspects of the present invention may be implemented in any convenient form. For example, embodiments may be implemented by one or more appropriate computer programs, which may be carried on an appropriate carrier medium, which may be a tangible carrier medium (e.g., a disk) or an intangible carrier medium (e.g., a communication signal). Embodiments of the present invention may be implemented using an appropriate device, which may specifically take the form of a programmable computer that runs a computer program arranged to implement the methods described herein. Therefore, embodiments of the present disclosure may be implemented in hardware, firmware, software, or any combination thereof. Embodiments of the present disclosure may also be implemented as instructions stored on a machine-readable medium, which may be read and executed by one or more processors. A machine-readable medium may include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computing device). For example, a machine-readable medium may include a read-only memory (ROM); a random access memory (RAM); a disk storage medium; an optical storage medium; a flash memory device; an electrical, optical, acoustic, or other form of propagation signal (e.g., a carrier wave, an infrared signal, a digital signal, etc.), etc. In addition, firmware, software, routines, instructions may be described herein as performing certain actions. However, it should be recognized that such descriptions are merely for convenience and that the actions are actually produced by a computing device, processor, controller or other device executing firmware, software, routines, instructions or the like.

[0187] In the block diagrams, the components shown are depicted as discrete functional blocks, but the embodiments are not limited to systems in which the functionality described herein is organized in the manner shown. The functionality provided by each component may be provided by software or hardware modules organized differently than currently shown, for example, such software or hardware may be mixed, combined, replicated, decomposed, distributed (e.g., within a data center or geographically), or organized in other different manners. The functionality described herein may be provided by one or more processors of one or more computers that execute code stored on a tangible, non-transitory, machine-readable medium. In some cases, a third-party content delivery network may host some or all of the information transmitted over the network, in which case, to the extent that the information (e.g., content) is provided or otherwise provided, the information may be provided by sending instructions to retrieve the information from the content delivery network.

[0188] Unless otherwise expressly stated, it will be apparent from the discussion that terms such as "process," "compute," "calculate," "determine," etc., used in the discussion throughout this specification refer to actions or processes of a specific device (e.g., a special purpose computer or similar special purpose electronic processing / computing device).

[0189] The reader should understand that this application describes several inventions. Rather than being separated into multiple separate patent applications, these inventions are grouped as a single document because their related subject matter helps save time in the application process. But the unique advantages and aspects of these inventions should not be lumped together. In some cases, embodiments solve all of the deficiencies described herein, but it should be understood that these inventions are independently useful, while some embodiments solve only a subset of these problems or provide other unmentioned benefits that are obvious to a person skilled in the art who reviews this disclosure. Due to cost constraints, some of the inventions disclosed herein may not be claimed at present, but may be claimed in subsequent applications, such as continuation applications or by amendment of the present claims. Similarly, due to length limitations, neither the abstract nor the summary section of this document should be considered to contain a comprehensive listing of all such inventions or all aspects of such inventions.

[0190] It should be understood that the description and drawings are not intended to limit the disclosure to the particular forms disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present invention as defined by the appended claims.

[0191] In view of this specification, modifications and alternative embodiments of various aspects of the present invention will be clear to those skilled in the art. Therefore, this specification and the accompanying drawings should be interpreted as illustrative only, with the purpose of teaching those skilled in the art the general way of implementing the present invention. It should be understood that the forms of the present invention shown and described herein should be taken as examples of embodiments. Elements and materials can replace the elements and materials shown and described herein, parts and processes can be reversed or omitted, certain features can be used independently, and embodiments or features of embodiments can be combined, all of which will be obvious to those skilled in the art after reading this specification. Changes can be made to the elements described herein without departing from the spirit and scope of the present invention as described in the following claims. The headings used herein are for organizational purposes only and are not intended to limit the scope of the specification.

[0192] In this application, the word "may" is used in a permissive sense (i.e., indicating potential), rather than a mandatory sense (i.e., indicating a must). The words "include", "comprises", and "includes" mean including but not limited to. In this application, unless the content clearly indicates otherwise, the singular forms "one", "a", and "the" include plural references. Thus, for example, references to "one" elements or "a" elements include combinations of two or more elements, although other terms and phrases, such as "one or more", are used for one or more elements. As used herein, unless otherwise expressly stated, the term "or" covers all possible combinations unless not feasible. For example, if a component is specified to include A or B, then unless otherwise expressly stated or not feasible, the component may include A, or B, or A and B. As a second example, if a component is specified to include A, B, or C, then unless otherwise expressly stated or not feasible, the component may include A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

[0193] Terms describing conditional relationships, such as "in response to X, Y", "at X, Y", "if X, Y", "when X, Y", etc., cover causal relationships where the antecedent is a necessary causal condition, the antecedent is a sufficient causal condition, or the antecedent is a contributing causal condition to the effect, for example, "state X occurs when condition Y holds" is common to "X occurs only when Y holds" and "X occurs when Y and Z hold". Such conditional relationships are not limited to effects that occur immediately after the antecedent is obtained, as some effects may be delayed, and in conditional statements, the antecedent is related to its effect, for example, the antecedent is related to the likelihood of the effect occurring. Unless otherwise specified, a statement that maps multiple attributes or functions to multiple objects (e.g., one or more processors perform steps A, B, C, and D) includes both mapping all of these attributes or functions to all such objects and mapping subsets of attributes or functions to subsets of attributes or functions (e.g., all processors perform steps AD, and processor 1 performs step A, processor 2 performs step B and part of step C, and processor 3 performs step C and part of step D). Furthermore, unless otherwise stated, a statement that a value or action is "depending on" another condition or value includes both the case where the condition or value is the only factor and the case where the condition or value is one of multiple factors. Unless otherwise stated, a statement that "every" instance of a set has some property should not be read to exclude the case where some of the same or similar members of the larger set do not have that property, i.e., every does not necessarily mean every. References to range selections include the endpoints of the range.

[0194] In the above description, any process, description or block in the flowchart should be understood to represent a module, code segment or portion, which includes one or more executable instructions for implementing specific logical functions or steps in the process, and alternative implementations are included in the scope of the exemplary embodiments of the present invention, in which the functions may be performed in a different order than shown or discussed, including substantially simultaneously or in reverse order, depending on the functions involved, as understood by those skilled in the art.

[0195] If certain U.S. patents, U.S. patent applications, or other materials (e.g., articles) have been cited, the text of such U.S. patents, U.S. patent applications, and other materials is cited only to the extent that there is no conflict between such materials and the statements and drawings set forth herein. If such a conflict exists, any such conflicting text in such cited U.S. patents, U.S. patent applications, and other materials shall not be cited.

[0196] Although certain embodiments have been described, these embodiments are presented by way of example only and are not intended to limit the scope of the present disclosure. In fact, the novel methods, devices, and systems described herein may be embodied in a variety of other forms; furthermore, various omissions, substitutions, and changes in the form of the methods, devices, and systems described herein may be made without departing from the spirit of the present disclosure. The accompanying claims and their equivalents are intended to cover such forms or modifications as fall within the scope and spirit of the present disclosure.

Claims

1. A non-transitory computer readable medium having instructions which, when executed by a computer, cause the computer to perform a method for training a machine learning model to generate a mask pattern to be used to print a target layout on a substrate, the method comprising: inputting a target image to a neural network, the target image being associated with a target layout to be printed on a substrate, wherein the machine learning model is configured to receive a reference image corresponding to an optical proximity correction (OPC) mask image of the target image; generating a predicted mask image using the machine learning model, the predicted mask image representing a mask pattern to be used to print the target layout on a substrate; calculating a loss function indicating at least one of (a) a difference between the predicted mask image and the reference image, and (b) an MRC evaluation of the predicted mask image or an evaluation of a first simulated image of the predicted mask image; and The machine learning model is modified based on the loss function.

2. The computer-readable medium of claim 1 , wherein calculating the loss function indicative of the MRC evaluation of the predicted mask image comprises: A mask rule checking MRC cost is calculated by performing the MRC evaluation on the predicted mask image, the MRC cost indicating an MRC violation.

3. The computer-readable medium of claim 2, wherein the MRC violation comprises a violation of at least one of: a critical dimension (CD), a width, an area of ​​a feature in the mask pattern, wherein performing the MRC evaluation comprises: assigning a violation score to a portion of the mask pattern where the MRC violation occurs; as well as The MRC cost is determined based on the violation score.

4. The computer-readable medium of claim 1 , wherein calculating the loss function indicative of the evaluation of the first simulated image of the predicted mask image comprises: The first simulated image is generated based on the predicted mask image.

5. The computer-readable medium of claim 1, wherein the first simulation image is at least one of: an aerial image, a resist image, or an etch image.

6. The computer-readable medium of claim 4, wherein calculating the loss function comprises: A simulation cost is calculated that indicates a difference between the first simulated image and a second simulated image of the reference image.

7. The computer-readable medium of claim 6, wherein calculating the simulation cost comprises: identifying an area in the first simulated image and the second simulated image that is within a specified proximity of a feature to be printed on a substrate; as well as The simulation cost is calculated to indicate a difference between the first simulated image and the second simulated image within the area.

8. The computer-readable medium of claim 4, wherein calculating the loss function comprises: calculating a score indicating that a pixel value of each pixel of the first simulated image exceeds a threshold; as well as A simulation cost is calculated based on the scores, wherein the simulation cost is determined based on scores associated with pixels outside a specified neighborhood of a feature in the first simulated image to be printed on the substrate.

9. The computer readable medium of claim 5, wherein the first simulated image is computed using a fixed filter.

10. The computer readable medium of claim 9, wherein the fixed filter is generated based on a transmission cross coefficient kernel.

11. The computer-readable medium of claim 5, wherein generating the first simulated image comprises: The resolution of the predicted mask image is increased to generate the first simulated image.

12. The computer readable medium of claim 11, wherein the predicted mask image is a continuous transmission mask (CTM) image, and wherein the method further comprises: A binarized image of the CTM image is generated.

13. The computer-readable medium of claim 1 , wherein the machine learning model comprises a neural network, and wherein modifying the neural network according to the loss function comprises: modifying parameters of a first portion of the neural network based on the loss function indicative of the difference between the predicted mask image and the reference image; as well as Parameters of a second portion of the neural network are modified based on the loss function, the loss function being indicative of at least one of: an MRC evaluation of the predicted mask image or the evaluation of the first simulated image of the predicted mask image.

14. The computer-readable medium of claim 1, further comprising: inputting a first target image having a first target layout to be printed on a first substrate into a neural network; as well as A first predicted mask image is generated using the neural network, the first predicted mask image representing a first mask pattern to be used to print the first target layout on the first substrate.

15. The computer-readable medium of claim 14, further comprising: A mask having the first mask pattern is generated using the first predicted mask image.

Images