Method, apparatus and storage medium for simulating dry etching of thin film material

Abstract

The application provides a method, device and storage medium for simulating dry etching of a thin film material, and the method comprises the following steps: determining a real etching case and a simulation etching case; deriving original process parameters into a plurality of groups of simulation process parameters according to target text information; inputting first target global image data and the target text information into a preset first generative adversarial network to reconstruct second target global image data; intercepting a plurality of regions in the second target global image data to obtain target local image data; inputting the target local image data into a preset second generative adversarial network to magnify and obtain target microscopic image data; generating a confidence degree according to the simulation process parameters, the second target global image data and the target microscopic image data; and generating a simulation etching result of a target thin film material according to the simulation process parameters, the second target global image data and the target microscopic image data selected according to the confidence degree. The generalization of the simulation dry etching is effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of semiconductor technology, and in particular relates to a method, apparatus and storage medium for simulating dry etching of thin film materials. Background Technology

[0002] Dry etching is a pattern transfer process in semiconductor and display panel manufacturing, and its precision affects device performance and product yield. As display panels develop towards higher generations (G8.5 and above), 4K / 8K high resolution, and flexibility, the etching feature size is reduced to the sub-micron level, placing higher demands on the process.

[0003] To shorten the process development cycle, the industry has successively developed simulation technologies such as physical model-based and pure data-driven simulation. Pure data-driven simulation refers to constructing a neural network based on process parameters and etching effects, and performing dry etching simulation based on the neural network.

[0004] However, dry etching is a complex process involving more than a hundred influencing factors. A small number of process parameters can be precisely measured and controlled, while the majority of the remaining factors are implicit factors that cannot be directly measured. These factors lead to significant differences in etching effects under the same process parameters, resulting in poor generalization ability and low simulation accuracy of neural networks. Summary of the Invention

[0005] In view of this, the present invention provides a method, apparatus and storage medium for simulating dry etching of thin film materials, so as to improve the generalizability of the simulated dry etching and thus improve the simulation accuracy of dry etching.

[0006] A first aspect of the present invention provides a method for simulating dry etching of thin film materials, comprising: Identify real-world etching cases and simulated etching use cases; the real-world etching cases include the first target global image data of the target thin film material after dry etching detected by an automated optical inspection device and the original process parameters for dry etching of the target thin film material; the simulated etching use cases include target text information describing the changes in the morphology of the target thin film material. Based on the target text information, the original process parameters are derived into multiple sets of simulation process parameters; The first target global image data and the target text information are input into a preset first generative adversarial network to reconstruct the second target global image data; Multiple regions are extracted from the second target global image data to obtain target local image data; The target local image data is input into a preset second generative adversarial network for amplification to obtain target microscopic image data; For each set of simulation process parameters, a confidence level is generated based on the simulation process parameters, the second target global image data, and the target microscopic image data; Based on the confidence level, the simulation process parameters, the second target global image data, and the target microscopic image data are selected to generate the simulation etching result of the target thin film material.

[0007] A second aspect of the present invention provides an apparatus for simulating dry etching of thin film materials, comprising: The etching data determination module is used to determine real etching cases and simulated etching use cases. The real etching cases include the first target global image data of the target thin film material after dry etching detected by an automated optical inspection device and the original process parameters for dry etching of the target thin film material. The simulated etching use cases include target text information describing the changes in the morphology of the target thin film material. The simulation process parameter derivation module is used to derive multiple sets of simulation process parameters from the original process parameters based on the target text information. A target global image data reconstruction module is used to input the first target global image data and the target text information into a preset first generative adversarial network to reconstruct the second target global image data. The target local image data cropping module is used to crop multiple regions from the second target global image data to obtain target local image data; The target microscopic image data magnification module is used to input the target local image data into a preset second generative adversarial network for magnification to obtain target microscopic image data; The confidence generation module is used to generate a confidence score for each set of simulation process parameters based on the simulation process parameters, the second target global image data, and the target microscopic image data. The simulation etching result generation module is used to select the simulation process parameters, the second target global image data, and the target microscopic image data based on the confidence level to generate the simulation etching result of the target thin film material.

[0008] A third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method for simulating dry etching of thin film materials as described in the first aspect above.

[0009] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for simulating dry etching of thin film materials as described in the first aspect above.

[0010] A fifth aspect of the present invention provides a computer program product that, when run on a computer, causes the computer to perform a method for simulating dry etching of thin film materials as described in the first aspect above.

[0011] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: In this embodiment, real etching cases and simulated etching use cases are determined. Real etching cases include first target global image data of the target thin film material after dry etching using automated optical inspection equipment and the original process parameters for dry etching of the target thin film material. Simulated etching use cases include target text information describing the changes in the morphology of the target thin film material. Based on the target text information, the original process parameters are derived into multiple sets of simulated process parameters. The first target global image data and the target text information are input into a preset first generative adversarial network to reconstruct second target global image data. Multiple regions are extracted from the second target global image data to obtain target local image data. The target local image data is input into a preset second generative adversarial network for magnification to obtain target microscopic image data. For each set of simulated process parameters, a confidence level is generated based on the simulated process parameters, the second target global image data, and the target microscopic image data. Based on the confidence level, the simulated process parameters, the second target global image data, and the target microscopic image data are selected to generate the simulated etching result of the target thin film material. This embodiment evolves the macroscopic and microscopic morphological changes of thin film materials based on real-world etching cases. It assesses feasibility by relying on the relationship between the macroscopic and microscopic morphological changes of thin film materials and explicit process parameters and implicit factors, effectively improving the generalizability of simulated dry etching, thereby improving the simulation accuracy of dry etching and providing a new approach to solving the complex modeling and optimization problems of dry etching processes. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of a method for simulating dry etching of thin film materials provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a process for simulating dry etching of thin film materials according to an embodiment of the present invention; Figure 3 This is a schematic diagram of another process detection model provided in an embodiment of the present invention; Figure 4This is a schematic diagram of an apparatus for simulating dry etching of thin film materials provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the present invention. However, those skilled in the art will recognize that the present application may be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted to avoid unnecessary detail that could obscure the description of the present application.

[0015] The technical solution of the present invention will be illustrated below through specific embodiments.

[0016] Reference Figure 1 The diagram illustrates a method for simulating dry etching of thin film materials according to an embodiment of the present invention, which may specifically include the following steps: Step 101: Determine real etching cases and simulated etching use cases.

[0017] In a quality management system (QMS), it is possible to target SiO2 (silicon dioxide) and SiN. x A database of dry etching process maintenance cases for PECVD (plasma-enhanced chemical vapor deposition) thin film materials such as silicon nitride, p-Si (polycrystalline silicon), and a-Si (amorphous silicon), as well as metal thin film materials such as Mo (molybdenum), MoAlMo (molybdenum aluminum molybdenum), and TiAl (titanium aluminum), serves as prior knowledge for simulating dry etching.

[0018] The case library contains records of multiple real-world etching cases, such as... Figure 2 As shown, real-world etching examples include image data (denoted as first target global image data 201) of the target thin film material after dry etching using an Automated Optical Inspection (AOI) device, and process parameters for dry etching of the target thin film material (denoted as original process parameters 202), such as gas flow rate, RF power, bias voltage, chamber pressure, etc.

[0019] When developing dry etching processes for a specific thin film material (denoted as the target thin film material), researchers can generate simulated etching test cases for the target thin film material. For example... Figure 2As shown, the simulated etching use case includes target text information 203 describing changes to the morphology of the target thin film material. That is, researchers can use natural language to describe how to modify the morphology of the target thin film material to change its performance, such as improving etching uniformity to within ±2%, adjusting the taper angle to 74°-76°, reducing pits at the interface of the intermediate aluminum layer, etc.

[0020] Step 102: Based on the target text information, derive multiple sets of simulation process parameters from the original process parameters.

[0021] In this embodiment, as Figure 2 As shown, guided by the target text information, the original process parameter 202 can be used as a baseline to derive multiple sets of potential process parameters, which are denoted as simulation process parameters 2021.

[0022] In one derivative approach, entities and constraints are identified in the target text information using methods such as entity dictionaries and large language models (LLM).

[0023] The entities include material entities, entities related to etching effect indicators, action modification entities, process parameter entities, etc. For example, material entities include SiO2 and SiN. x The entities related to etching performance indicators include etching rate, etching uniformity, etching selectivity, tape angle, sidewall roughness, and interface pits. The entities related to action modification include increasing, decreasing, enlarging, reducing, adjusting, suppressing, and optimizing. The entities related to process parameters include RF power, bias voltage, CF4 flow rate, chamber pressure, and etching time.

[0024] Constraints include interval constraints, upper limit constraints, lower limit constraints, and relative change constraints. For example, the fixed interval is the taper angle of 74°-76°, the upper limit constraint is the pressure ≤30mTorr, and the relative change constraint is the etching rate decrease ≤10%, etc.

[0025] Based on real etching cases in the case library, the correlation between each process item and each entity can be mined offline. Then, by traversing each entity, the first target item and the second target item can be distinguished in the same set of original process parameters. The first target item is the process item related to the entity in the original process parameters, and the second target item is the process item unrelated to the entity in the original process parameters.

[0026] The nominal range of the first objective item is truncated according to the constraints, that is, the intersection between the constraints and the nominal range of the first objective item is taken to obtain the objective range.

[0027] Latin hypercube sampling is performed within the target range to obtain simulation terms.

[0028] The simulation item is combined with the second objective item to form a set of simulation process parameters.

[0029] Compared to uniform grid sampling, this method covers a larger parameter space with the same sample size, has no redundant samples, and has lower inference time. Compared to simple random sampling, this method does not have parameter clusters or local blank areas, and will not miss potential optimal process ranges.

[0030] Step 103: Input the first target global image data and target text information into a preset first generative adversarial network to reconstruct the second target global image data.

[0031] In this embodiment, as Figure 2 As shown, a generative adversarial network (GAN, denoted as the first generative adversarial network) can be pre-constructed, which modifies the first target global image data 201 according to the semantics of the target text information 203 to reconstruct the second target global image data 204.

[0032] The first generative adversarial network includes a first generator and a first discriminator.

[0033] The first generator includes a first encoder and a first decoder. The first encoder is used to extract image features from the first target global image data and semantic features from the target text information, respectively. It fuses the image features of the first target global image data and the semantic features of the target text information and further extracts features to obtain multimodal features. The first decoder is used to decode the multimodal features into second target global image data.

[0034] The first discriminator is used to simultaneously determine the authenticity of the image output by the first generator and whether the image output by the first generator matches the description in the text.

[0035] In one embodiment of the present invention, in the process of training the first generative adversarial network, the thin film material in the real etching case is denoted as the sample thin film material, and the real process parameters in the real etching case are denoted as the sample process parameters. The preset training set includes image data (denoted as sample global image data) of the sample thin film material after dry etching using an automated optical inspection (AOI) device and the original process parameters for dry etching of the sample thin film material. The process of training the first generative adversarial network based on the training set is as follows: Step S31: Select two frames of sample global image data with similar appearances as baseline global image data and standard global image data.

[0036] In this embodiment, the global image data of the samples in the training set is traversed, and descriptors such as SIFT (Scale Invariant Feature Transform) are calculated for the global image data of the samples. Two frames of global image data are matched according to the descriptors to obtain sample image pairs. The similarity such as SSIM (Structural Similarity Index) is calculated for the two frames of global image data in the sample image pairs. Sample image pairs with similarity in a preset range (such as 0.8~0.9) are selected to obtain two frames of global image data with similar appearance. In the sample image pair, either frame of global image data is recorded as the baseline global image data, and the other frame of global image data is recorded as the standard global image data.

[0037] Step S32: Generate sample text information.

[0038] In this embodiment, baseline global image data and standard global image data can be input into a VLM (Visual Language Model) finely tuned in the semiconductor field to generate sample text information, wherein the sample text information is used to describe the morphological changes of the sample thin film material from the baseline global image data to the sample thin film material corresponding to the standard global image data.

[0039] Furthermore, a verification task can be generated based on baseline global image data, standard global image data, and sample text information, allowing annotators to verify the sample text information and improve its accuracy.

[0040] Step S33: Input the baseline global image data and sample text information into the preset first generative adversarial network to reconstruct the global image data.

[0041] In this embodiment, the baseline global image data and sample text information are input into a preset first generative adversarial network, and the baseline global image data is modified according to the semantics of the sample text information to reconstruct the recombined global image data.

[0042] Step S34: Generate the first adversarial loss value based on the reconstructed global image data and the standard global image data.

[0043] In this embodiment, the reconstructed global image data and the standard global image data are substituted into the native loss function of the generative adversarial network to obtain the first adversarial loss value.

[0044] Step S35: Calculate the first constraint loss value on the etching rate based on the sample text information, sample process parameters, reconstructed global image data and standard global image data.

[0045] In this embodiment, the first constraint loss value on the etching rate can be calculated based on the sample text information, sample process parameters, reconstructed global image data and standard global image data, so that the etching rate is conserved and the overall etching depth meets the adjustment requirements of the text description.

[0046] For example, based on the Large Language Model (LLM) to understand the semantics of sample text information on the basis of sample process parameters, the sample text information and sample process parameters are used to generate a first adjustment coefficient representing the etching depth, wherein the etching depth is positively correlated with the etching rate, that is, the greater the etching rate, the deeper the etching depth.

[0047] Generally, based on the etching reaction mechanism of different materials, a physical mapping relationship between gas composition and etching rate is established. For example, for PECVD thin film materials etched with fluorine-containing gas, the concentration of fluorine free radicals is constrained to be related to SiO2 and SiN. x The positive correlation and relative proportion of etching rates of p-Si and a-Si; for metal thin film materials etched with chlorine gas, the correspondence between the concentration of constrained chlorine radicals and the etching rates of Mo, Al, and Ti.

[0048] The product of the average gray level of each pixel in the standard global image data and the first adjustment coefficient is calculated as the first image etching factor.

[0049] The absolute value of the difference between the average gray level of pixels in the reconstructed global image data and the first image etching factor is taken to obtain the first constraint loss value, in order to prevent virtual images that violate the plasma reaction law, such as the process parameters remaining unchanged but the overall etching depth deviating significantly, or the etching rates of different thin film materials being reversed.

[0050] In this example, the first constraint loss value can be expressed as: Among them, L rate The first constraint loss value is represented by a smaller value, indicating that the overall etching depth of the generated reconstructed global image data conforms to the process control logic; H and W represent the height and width, respectively; i and j represent the x and y coordinates of the pixel, respectively; I gen (i,j) represents the grayscale value of pixel (i,j) in the reconstructed global image data. The grayscale value linearly maps to the etching depth; a higher grayscale value indicates a greater amount of etching removal. real (i,j) represents the grayscale value of pixel (i,j) in the standard global image data; k is the first adjustment coefficient used to adjust the etching depth. When k>1, it means to increase the etching depth and increase the etching rate. When k<1, it means to decrease the etching depth and decrease the etching rate. When k=1, it means to maintain the original etching amount.

[0051] Step S36: Calculate the second constraint loss value on etching uniformity based on the sample text information, sample process parameters, reconstructed global image data and standard global image data.

[0052] In this embodiment, a second constraint loss value for etching uniformity can be calculated based on sample text information, sample process parameters, reconstructed global image data and standard global image data. Based on the gas flow and plasma distribution patterns in the etching chamber, the etching uniformity distribution pattern of the generated image is constrained. For example, the gas spray head structure affects the etching rate difference from the center to the edge, the edge etching is enhanced or weakened due to the chamber wall effect, the local etching characteristics caused by the exhaust port position, etc., to prevent random, non-physically compliant spot-like or stripe-like non-uniformities.

[0053] For example, based on the Large Language Model (LLM), the semantics of the sample text information are understood based on the sample process parameters, and a second adjustment coefficient representing etching uniformity is generated using the sample text information and sample process parameters.

[0054] The product of the standard deviation of pixel gray levels in the standard global image data and the second adjustment coefficient is calculated as the second image etching factor.

[0055] The absolute value of the difference between the standard deviation of pixel gray levels in the reconstructed global image data and the second image etching factor is taken to obtain the second constraint loss value.

[0056] In this example, the second constraint loss value can be expressed as: L uniform =|std(I gen )-αstd(I real )|, where L uniform The second constraint loss value is represented by a smaller value, indicating that the etching depth distribution of the entire thin film material conforms to the flow field law of the chamber; std is the standard deviation of the gray level of the image pixel; I gen To reconstruct global image data, I real For standard global image data, α is the second adjustment coefficient. If improved uniformity is required, then α < 1; if moderate uniformity degradation is allowed, then α > 1.

[0057] Step S37: Calculate the third constraint loss value on edge continuity based on sample text information, sample process parameters, reconstructed global image data and standard global image data.

[0058] In this embodiment, the third constraint loss value on edge continuity can be calculated based on sample text information, sample process parameters, reconstructed global image data and standard global image data to ensure the continuity of etched pattern edges and avoid unnatural abrupt changes.

[0059] For example, based on the Large Language Model (LLM), the semantics of the sample text information are understood based on the sample process parameters, and the sample text information and sample process parameters are used to generate a third adjustment coefficient representing photoresist erosion.

[0060] In general, an etching selectivity database of different types of photoresists can be introduced to constrain the relationship between the degree of photoresist erosion in the generated image and the process conditions. For example, the higher the bias voltage and the stronger the physical sputtering effect, the faster the photoresist erosion rate; the higher the proportion of passivation gas, the thicker the protective layer formed on the photoresist surface, and the slower the erosion rate, etc., to prevent unreasonable results such as excessive consumption of photoresist, pattern collapse or abrupt changes in sidewall angle.

[0061] Based on the target detection operation, the local area where the photoresist is located is located in the reconstructed global image data and the standard global image data respectively. The product between the gradient value of the local area where the photoresist exists in the standard global image data and the third adjustment coefficient is calculated as the third image etching factor.

[0062] The first norm is taken to obtain the third constraint loss value by taking the difference between the gradient value of the local area containing photoresist in the reconstructed global image data and the second image etching factor.

[0063] In this example, the third constraint loss value can be expressed as: L pr =|grad(I gen_pr )-γgrad(I real_pr )|1, where L pr The third constraint loss value controls the degree of mask consumption in the virtual image to match the process; grad is the image gradient operator, and the gradient magnitude represents the degree of lateral erosion of the photoresist sidewalls; I gen_pr To reconstruct the localized areas of photoresist in the global image data, I real_pr For standard global image data, there is a local range of photoresist. γ is the third adjustment coefficient, which is positively correlated with process parameters such as bias voltage and passivation gas flow rate. That is, if the bias voltage increases, the ion sputtering is enhanced, then γ increases, and the lateral erosion of the photoresist is intensified; if the passivation gas flow rate increases, the sidewall protective layer is thickened, then γ decreases.

[0064] Step S38: Combine the first adversarial loss value, the first constraint loss value, the second constraint loss value, and the third constraint loss value into a first total loss value.

[0065] In this embodiment, the first adversarial loss value, the first constraint loss value, the second constraint loss value, and the third constraint loss value can be fused into a first total loss value using a linear or nonlinear method.

[0066] For example, the first total loss value can be expressed as: L total1 =Lgen1 +w1L rate +w2L uniform +w3L pr , where L rate L is the first constraint loss value. uniform L is the second constraint loss value. pr The third constraint loss value is represented by w1, w2, and w3, which are all weights.

[0067] Step S39: Update the first generative adversarial network based on the first total loss value.

[0068] In this embodiment, the parameters in the first generative adversarial network can be updated using optimization algorithms such as SGD (stochastic gradient descent) and Adam (adaptive momentum) based on the first total loss value.

[0069] Step 104: Extract multiple regions from the second target global image data to obtain the target local image data.

[0070] In this embodiment, as Figure 2 As shown, multiple local regions can be extracted from the second target global image data 204 to obtain target local image data 205.

[0071] In a specific implementation, the global image data of the second target can be input into the Attention layer based on Transformer, and multiple regions can be determined in the global image data of the second target according to the attention mechanism.

[0072] For each region, calculate the difference in grayscale values ​​between pixels within the region and pixels outside the region.

[0073] For each region, query the probability of adverse events occurring in the region's history, and use this as the defect sensitivity.

[0074] Identify one or more process parameters that show the greatest change in simulated process parameters compared to the original process parameters, and designate them as the change parameters.

[0075] For each region, calculate the correlation between that region and the variable.

[0076] For each region, linear (e.g., weighted summation) or nonlinear methods are used to integrate the degree of variation, defect sensitivity and correlation into an attention value, guiding the focus to areas prone to etching problems, such as multilayer film interfaces, substrate edges, densely patterned areas and large-size patterned areas. For multilayer thin film etching, the attention weight of the interface areas of different material layers is increased; for metal thin film etching, attention is paid to the edges of lines and corner areas.

[0077] In the second target global image data, the regions with the highest attention values ​​are extracted as target local image data. At this time, the target local image data is the key region that has a significant impact on the etching effect, which significantly reduces the amount of computation while ensuring accuracy.

[0078] Step 105: Input the local image data of the target into a preset second generative adversarial network for amplification to obtain the microscopic image data of the target.

[0079] In this embodiment, as Figure 2 As shown, two generative adversarial networks (GANs, referred to as the second generative adversarial network) can be pre-constructed to amplify the target local image data 205 and reconstruct it into target microscopic image data 206, revealing details such as microscopic sidewall contours, taper angles (sidewall tilt angles), and line width roughness.

[0080] The second generative adversarial network includes a second generator and a second discriminator.

[0081] The second generator includes a second encoder and a second decoder. The second encoder is used to extract image features from the target local image data, and the second decoder is used to decode the image features of the target local image data into target microscopic image data.

[0082] The second discriminator is used to determine the authenticity of the image output by the second generator.

[0083] In one embodiment of the present invention, in the process of training the second generative adversarial network, the thin film material in the real etching case is denoted as the sample thin film material, and the real process parameters in the real etching case are denoted as the sample process parameters. The preset training set includes image data (denoted as sample global image data) detected by an automated optical inspection (AOI) device on the sample thin film material after dry etching, the sample process parameters for dry etching of the sample thin film material, and the region obtained by cropping a region from the sample global image data (denoted as sample local image data). The sample microscopic image data detected by a scanning electron microscope (SEM) on the same region in the sample global image data is as follows: Step S51: Input the sample local image data into a preset second generative adversarial network for amplification to obtain simulated microscopic image data.

[0084] In this embodiment, the sample local image data is input into a preset second generative adversarial network to amplify the sample local image data and reconstruct it into simulated microscopic image data.

[0085] Step S52: Generate a second adversarial loss value based on the sample microscopic image data and the simulated microscopic image data.

[0086] In this embodiment, the sample microscopic image data and the simulated microscopic image data are substituted into the native loss function of the generative adversarial network to obtain the second adversarial loss value.

[0087] Step S53: Calculate the fourth constraint loss value at the sidewall tilt angle based on the sample local image data, simulation microscopic image data and sample process parameters.

[0088] In this embodiment, the fourth constraint loss value on the sidewall tilt angle is calculated based on the sample local image data, simulated microscopic image data and sample process parameters to ensure that the taper angle of the generated image conforms to the process rules.

[0089] For example, based on the understanding of sample local image data and sample process parameters in VLM (Visual Language Model), the target angle, i.e. the target value of the taper angle, is determined based on the sample local image data and sample process parameters.

[0090] Generally, a physical relationship can be established between factors such as bias voltage and passivation gas ratio and the taper angle of the etching morphology. For example, the higher the bias voltage, the greater the ion bombardment energy, the more severe the sidewall etching, and the smaller the taper angle; the higher the passivation gas ratio, the thicker the passivation layer formed on the sidewall, and the larger the taper angle, etc., driving the generated taper angle to be consistent with the control law of process parameters.

[0091] The sidewall tilt angle is detected using simulated microscopic image data. The absolute value of the difference between the sidewall tilt angle of the simulated microscopic image data and the target angle is used as the fourth constraint loss value.

[0092] In this example, the fourth constraint loss value can be expressed as: L taper =|θ gen -θ target |, where L taper The fourth constraint loss value is θ; the smaller the value, the closer the virtual topography sidewall tilt angle matches the theoretical process target. gen The sidewall tilt angle for simulating microscopic image data; θ target At the target angle, influenced by factors such as bias voltage and passivation gas ratio, the vertical bombardment of ions is enhanced as the bias voltage increases, θ target Decrease; as the passivation gas flow rate increases, the sidewall passivation layer thickens, θ target Increase.

[0093] Step S54: Calculate the fifth constraint loss value on the continuity of the multilayer film interface based on the simulated microscopic image data.

[0094] For the stacked structure of multilayer thin film materials, the fifth constraint loss value on the continuity of the multilayer film interface can be calculated based on the simulated microscopic image data to constrain the morphological continuity at the interface of different material layers.

[0095] For stacked metal thin film materials, the etching depth and sidewall morphology of the intermediate Al layer can be constrained to prevent interface abrupt changes or excessive pitting that do not conform to physical laws; for dielectric stacked PECVD thin film materials, the etching selectivity between different dielectric layers can be constrained to correspond to the interface step morphology.

[0096] For example, the longitudinal pixel coordinates of the multilayer film interface of the sample thin film material can be queried based on the sample process parameters. The first norm is calculated according to the gradient value of the pixel gray level in the simulated microscopic image data based on the longitudinal pixel coordinates, and the fifth constraint loss value is obtained.

[0097] In this example, the loss value of the fifth constraint can be expressed as: L interface =|grad(I gen (y=y0))|1, where L interface This is the fifth constraint loss value, used to constrain the interface between different materials to have no abrupt changes or excessive pitting. gen For simulating microscopic image data, y0 represents the vertical pixel coordinates of the multilayer film interface, grad represents the image gradient operator, and grad(I gen (y=y0)) represents the gradient value of the gray level of the pixel at the interface of the multilayer film in the simulated microscopic image data. The larger the first norm, the more severe the interface bumps and pits, and the higher the loss. It helps to constrain the etching rate mismatch caused by the density difference of the multilayer material, suppress the deep pits formed by the excessive etching of the Al layer, and adapt to the multi-step morphology evolution law of gradient etching.

[0098] Step S55: Calculate the sixth constraint loss value on the sidewall roughness based on the simulated microscopic image data.

[0099] In this embodiment, the sixth constraint loss value on the sidewall roughness is calculated based on the simulated microscopic image data.

[0100] For example, a baseline value can be determined based on sample process parameters.

[0101] Generally, a correspondence between process parameters and sidewall roughness can be established. For example, the wider the ion energy distribution, the worse the plasma stability, the higher the ion energy dispersion, and the higher the sidewall roughness, corresponding to a rough sidewall morphology, etc. Based on this correspondence, the baseline value corresponding to the process parameters of the current sample can be used to ensure that the generated microscopic surface morphology conforms to the basic rules of the etching process.

[0102] The sidewall roughness is detected by the simulated microscopic image data. The absolute value of the difference between the sidewall roughness of the simulated microscopic image data and the reference value is taken as the sixth constraint loss value.

[0103] In this example, the loss value of the sixth constraint can be expressed as: L rough =|R gen -R param |, where L rough R is the loss value for the sixth constraint. gen R represents the sidewall roughness. param This is the baseline value.

[0104] Step S56: Combine the second adversarial loss value, the fourth constraint loss value, the fifth constraint loss value, and the sixth constraint loss value into a second total loss value.

[0105] In this embodiment, the second adversarial loss value, the fourth constraint loss value, the fifth constraint loss value, and the sixth constraint loss value can be fused into a second total loss value using a linear or nonlinear method.

[0106] For example, the second total loss value can be expressed as: L total2 =L gen2 +w4L taper +w5L interface +w6L rough , where L tape L is the fourth constraint loss value. interface Fifth constraint loss value, L rough The third constraint loss value is w4, w5, and w6, which are all weights.

[0107] Step S57: Update the second generative adversarial network based on the second total loss value.

[0108] In this embodiment, the parameters in the second generative adversarial network can be updated using optimization algorithms such as SGD (stochastic gradient descent) and Adam (adaptive momentum) based on the second total loss value.

[0109] Step 106: For each set of simulation process parameters, generate confidence scores based on the simulation process parameters, the global image data of the second target, and the microscopic image data of the target.

[0110] In practical applications, individual process parameters often fail to reflect the latent state of the chamber, such as the thickness of the chamber wall deposits, electrode aging, gas purity fluctuations, and substrate surface defects. Under the same process parameters, the latent state of the chamber has a significant impact on etching rate and taper angle. Furthermore, individual process parameters often fail to capture the nonlinear relationship of micromorphological evolution, such as sidewall ripples, multilayer film interface pits, photoresist side etching, and micro-loading effects. These micro-defects can affect device performance.

[0111] In this embodiment, the second target global image data and the target microscopic image data serve as fingerprints in the dry etching process. They contain the comprehensive results of various macroscopic and microscopic factors, expressing a complete causal chain from process parameters, microscopic morphological features to macroscopic etching effects. They can learn the characteristic signals of chamber state changes in morphology, indirectly monitor the latent state of the chamber, and also learn the morphological evolution law at the interface of different material layers.

[0112] like Figure 2 As shown, for each set of simulation process parameters 2021, under the conditions of simulation process parameters 2021, the latent state of the chamber, and the morphological evolution law at the interface of different material layers, the confidence level of the thin film material characterized by the simultaneous appearance of the second target global image data 204 and the target microscopic image data 206 during dry etching is evaluated.

[0113] In practice, the process detection model can be trained in advance based on the sample process parameters, global image data of the sample, and microscopic image data of the sample.

[0114] like Figure 3 As shown, the process detection model includes an encoding module, an Attention layer, and a Detection head. The encoding module has a three-way structure: one is an LSTM (Long Short-Term Memory) network, another is a CNN (Convolutional Neural Network), and the third is multiple CNNs and AvgPooling. The Detection head includes a fully connected layer (FC), activation functions (such as Sigmoid), and other structures.

[0115] During the inference of the process inspection model, process features are extracted from the simulated process parameters in LSTM, first image features are extracted from the global image data of the second target in CNN, second image features are extracted from the microscopic image data of the target in each frame in each CNN, and average pooling operation is performed on multiple second image features in the average pooling layer to obtain the third image features.

[0116] In AvgPooling, an attention mechanism is used to fuse process features, first image features, and third image features into multimodal features; In the detection head, multimodal features are mapped and activated as confidence scores.

[0117] Step 107: Select simulation process parameters, second target global image data and target microscopic image data based on confidence level to generate simulation etching results of target thin film material.

[0118] In this embodiment, one or more sets of simulation process parameters and their corresponding second target global image data and target micro image data can be selected according to the confidence level to generate the simulation etching results of the target thin film material. The simulation etching results are then displayed to the developers for viewing. Based on the second target global image data and target micro image data, the potential feasibility of applying the simulation process parameters on the basis of real etching cases can be evaluated.

[0119] Generally, you can select multiple sets of simulation process parameters with the highest confidence level and their corresponding second target global image data and target micro image data. Alternatively, you can select simulation process parameters with a confidence level greater than the threshold and their corresponding second target global image data and target micro image data, and so on.

[0120] In this embodiment, real etching cases and simulated etching use cases are determined. Real etching cases include first target global image data of the target thin film material after dry etching using automated optical inspection equipment and the original process parameters for dry etching of the target thin film material. Simulated etching use cases include target text information describing the changes in the morphology of the target thin film material. Based on the target text information, the original process parameters are derived into multiple sets of simulated process parameters. The first target global image data and the target text information are input into a preset first generative adversarial network to reconstruct second target global image data. Multiple regions are extracted from the second target global image data to obtain target local image data. The target local image data is input into a preset second generative adversarial network for magnification to obtain target microscopic image data. For each set of simulated process parameters, a confidence level is generated based on the simulated process parameters, the second target global image data, and the target microscopic image data. Based on the confidence level, the simulated process parameters, the second target global image data, and the target microscopic image data are selected to generate the simulated etching result of the target thin film material. This embodiment evolves the macroscopic and microscopic morphological changes of thin film materials based on real-world etching cases. It assesses feasibility by relying on the relationship between the macroscopic and microscopic morphological changes of thin film materials and explicit process parameters and implicit factors, effectively improving the generalizability of simulated dry etching, thereby improving the simulation accuracy of dry etching and providing a new approach to solving the complex modeling and optimization problems of dry etching processes.

[0121] It should be noted that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0122] Reference Figure 4 The diagram illustrates a device for simulating dry etching of thin film materials according to an embodiment of the present invention, which may specifically include the following modules: Etching data determination module 401 is used to determine real etching cases and simulated etching use cases; the real etching cases include the first target global image data of the target thin film material after dry etching detected by an automated optical inspection device and the original process parameters for dry etching of the target thin film material; the simulated etching use cases include target text information describing the changes in the morphology of the target thin film material. The simulation process parameter derivation module 402 is used to derive multiple sets of simulation process parameters from the original process parameters based on the target text information. The target global image data reconstruction module 403 is used to input the first target global image data and the target text information into a preset first generative adversarial network to reconstruct the second target global image data. The target local image data cropping module 404 is used to crop multiple regions from the second target global image data to obtain target local image data; The target microscopic image data magnification module 405 is used to input the target local image data into a preset second generative adversarial network for magnification to obtain target microscopic image data; The confidence generation module 406 is used to generate a confidence level for each set of simulation process parameters based on the simulation process parameters, the second target global image data and the target microscopic image data. The simulation etching result generation module 407 is used to select the simulation process parameters, the second target global image data and the target microscopic image data according to the confidence level to generate the simulation etching result of the target thin film material.

[0123] In one embodiment of the present invention, the simulation process parameter derivation module 402 is further configured to: Identify entities and constraints in the target text information; In the same set of original process parameters, a first target item and a second target item are distinguished; the first target item is a process item in the original process parameters that is related to the entity, and the second target item is a process item in the original process parameters that is unrelated to the entity. The target range is obtained by truncating the nominal range of the first target item according to the constraints. Latin hypercube sampling is performed within the target range to obtain simulation terms; The simulation item is combined with the second target item to form a set of simulation process parameters.

[0124] In one embodiment of the present invention, the target local image data cropping module 404 is further configured to: Multiple regions are identified in the global image data of the second target based on an attention mechanism; For each of the regions, calculate the degree of difference in grayscale values ​​between pixels within the region and pixels outside the region; For each of the aforementioned regions, the probability of adverse events occurring in the region in the past is queried and used as the defect sensitivity. Identify one or more process items whose simulated process parameters change the most compared to the original process parameters, and designate them as the change items; For each of the regions, calculate the correlation between the region and the change term; For each of the aforementioned regions, the degree of change difference, the degree of defect sensitivity, and the correlation are fused into an attention value; The regions with the highest attention values ​​are extracted from the second target global image data to form the target local image data.

[0125] In one embodiment of the present invention, the confidence generation module 406 is further configured to: Extract process features from the simulated process parameters; Extract the first image features from the global image data of the second target; Second image features are extracted from the target microscopic image data of each frame, and average pooling is performed on multiple second image features to obtain third image features; An attention mechanism is used to fuse the process features, the first image features, and the third image features into a multimodal feature; The multimodal features are mapped and activated as confidence levels.

[0126] In one embodiment of the present invention, the preset training set includes global image data of the sample thin film material after dry etching detected by an automated optical inspection device and sample process parameters for dry etching of the sample thin film material. The apparatus further includes: The global image data filtering module is used to filter two frames of global image data with similar appearances as baseline global image data and standard global image data. The sample text information generation module is used to generate sample text information; the sample text information is used to describe the morphological changes of the sample thin film material from the baseline global image data to the standard global image data. The global image data reconstruction module is used to input the baseline global image data and the sample text information into a preset first generative adversarial network to reconstruct the global image data. The first adversarial loss value generation module is used to generate a first adversarial loss value based on the recombined global image data and the standard global image data; The first constraint loss value calculation module is used to calculate the first constraint loss value on the etching rate based on the sample text information, the sample process parameters, the reconstructed global image data and the standard global image data. The second constraint loss value calculation module is used to calculate the second constraint loss value in terms of etching uniformity based on the sample text information, the sample process parameters, the reconstructed global image data and the standard global image data. The third constraint loss value calculation module is used to calculate the third constraint loss value on edge continuity based on the sample text information, the sample process parameters, the reconstructed global image data and the standard global image data. The first total loss value fusion module is used to fuse the first adversarial loss value, the first constraint loss value, the second constraint loss value and the third constraint loss value into a first total loss value; The first generative adversarial network update module is used to update the first generative adversarial network based on the first total loss value.

[0127] In one embodiment of the present invention, the first constraint loss value calculation module is further configured to: The sample text information and the sample process parameters are used to generate a first adjustment coefficient representing the etching depth; the etching depth is positively correlated with the etching rate. The product of the average gray level of the pixels in the standard global image data and the first adjustment coefficient is calculated as the first image etching factor; The absolute value of the difference between the average gray level of the pixels in the reconstructed global image data and the first image etching factor is taken to obtain the first constraint loss value; The second constraint loss value calculation module is also used for: The sample text information and the sample process parameters are used to generate a second adjustment coefficient representing etching uniformity. The product of the standard deviation of pixel gray levels in the standard global image data and the second adjustment coefficient is calculated and used as the second image etching factor. The absolute value of the difference between the standard deviation of pixel gray levels in the reconstructed global image data and the second image etching factor is taken to obtain the second constraint loss value. The third constraint loss value calculation module is also used for: The sample text information and the sample process parameters are used to generate a third adjustment factor representing photoresist erosion; The product of the gradient value of the local area where photoresist exists in the standard global image data and the third adjustment coefficient is calculated as the third image etching factor; The first norm is taken for the difference between the gradient value of the local area containing photoresist in the reconstructed global image data and the second image etching factor to obtain the third constraint loss value.

[0128] In one embodiment of the present invention, the preset training set includes global image data of a sample thin film material after dry etching detected using an automated optical inspection device, sample process parameters for dry etching of the sample thin film material, and local image data of a sample obtained by cropping a region from the global image data, and microscopic image data of the sample detected in the same region of the global image data using a scanning electron microscope; the apparatus further includes: The simulated microscopic image data amplification module is used to input the sample local image data into a preset second generative adversarial network for amplification to obtain simulated microscopic image data; The second adversarial loss value generation module is used to generate a second adversarial loss value based on the sample local image data, the sample microscopic image data and the simulated microscopic image data; The fourth constraint loss value calculation module is used to calculate the fourth constraint loss value at the sidewall tilt angle based on the simulated microscopic image data and the sample process parameters. The fifth constraint loss value calculation module is used to calculate the fifth constraint loss value on the continuity of the multilayer film interface based on the simulated microscopic image data. The sixth constraint loss value calculation module is used to calculate the sixth constraint loss value on the sidewall roughness based on the simulated microscopic image data. The second total loss value fusion module is used to fuse the second adversarial loss value, the fourth constraint loss value, the fifth constraint loss value and the sixth constraint loss value into a second total loss value; The second generative adversarial network update module is used to update the second generative adversarial network based on the second total loss value.

[0129] In one embodiment of the present invention, the fourth constraint loss value calculation module is further configured to: The target angle is determined based on the local image data of the sample and the process parameters of the sample. The absolute value of the difference between the sidewall tilt angle and the target angle in the simulated microscopic image data is taken as the fourth constraint loss value. The fifth constraint loss value calculation module is also used for: Query the vertical pixel coordinates of the multilayer film interface for the sample thin film material; The first norm is calculated based on the gradient value of the pixel gray level in the simulated microscopic image data according to the vertical pixel coordinates, and the fifth constraint loss value is obtained. The sixth constraint loss value calculation module is also used for: The baseline value is determined based on the sample process parameters; The absolute value of the difference between the sidewall roughness of the simulated microscopic image data and the reference value is taken as the sixth constraint loss value.

[0130] The present invention provides an apparatus for simulating dry etching of thin film materials. By using the apparatus for simulating dry etching of thin film materials, the various steps in the aforementioned method embodiments for simulating dry etching of thin film materials can be implemented.

[0131] It should be noted that the module division in the various devices for simulating dry etching of thin film materials provided in the above embodiments is illustrative and only represents a logical functional division. In actual implementation, other division methods may also be used. Furthermore, the functional modules in the various embodiments of this invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0132] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the technical solution of the embodiments of the present invention can be embodied in the form of a computer program product, which is stored in a computer storage medium and includes several instructions to cause an electronic device or processor to execute all or part of the steps of the methods in the various embodiments of the present invention. The aforementioned computer storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0133] Furthermore, the apparatus for simulating dry etching of thin film materials and the method for simulating dry etching of thin film materials provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0134] Reference Figure 5 The diagram illustrates an electronic device according to an embodiment of the present invention. Figure 5As shown, the electronic device in this embodiment of the invention includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the method embodiment for simulating dry etching of thin film materials described above. Alternatively, when the processor executes the computer program, it implements the functions of each module in the apparatus embodiment for simulating dry etching of thin film materials described above.

[0135] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which can be used to describe the execution process of the computer program in the electronic device.

[0136] The electronic device may be a desktop computer, a cloud server, or other computing device. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that... Figure 5 This is merely one example of an electronic device and does not constitute a limitation on the electronic device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0137] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0138] The memory can be an internal storage unit of the electronic device, such as a hard drive or RAM. Alternatively, it can be an external storage device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory can include both internal and external storage units. The memory is used to store the computer program and other programs and data required by the electronic device. The memory can also be used to temporarily store data that has been output or will be output.

[0139] This invention also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for simulating dry etching of thin film materials as described in the foregoing embodiments.

[0140] This invention also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for simulating dry etching of thin film materials as described in the foregoing embodiments.

[0141] This invention also discloses a computer program product that, when run on a computer, causes the computer to execute the method for simulating dry etching of thin film materials as described in the foregoing embodiments.

[0142] The embodiments described above are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for simulating dry etching of thin film materials, characterized in that, include: Identify real-world etching cases and simulated etching use cases; the real-world etching cases include the first target global image data of the target thin film material after dry etching detected by an automated optical inspection device and the original process parameters for dry etching of the target thin film material; the simulated etching use cases include target text information describing the changes in the morphology of the target thin film material. Based on the target text information, the original process parameters are derived into multiple sets of simulation process parameters; The first target global image data and the target text information are input into a preset first generative adversarial network to reconstruct the second target global image data; Multiple regions are extracted from the second target global image data to obtain target local image data; The target local image data is input into a preset second generative adversarial network for amplification to obtain target microscopic image data; For each set of simulation process parameters, a confidence level is generated based on the simulation process parameters, the second target global image data, and the target microscopic image data; Based on the confidence level, the simulation process parameters, the second target global image data, and the target microscopic image data are selected to generate the simulation etching result of the target thin film material.

2. The method according to claim 1, characterized in that, The process of deriving multiple sets of simulation process parameters from the original process parameters based on the target text information includes: Identify entities and constraints in the target text information; In the same set of original process parameters, a first target item and a second target item are distinguished; the first target item is a process item in the original process parameters that is related to the entity, and the second target item is a process item in the original process parameters that is unrelated to the entity. The target range is obtained by truncating the nominal range of the first target item according to the constraints. Latin hypercube sampling is performed within the target range to obtain simulation terms; The simulation item is combined with the second target item to form a set of simulation process parameters.

3. The method according to claim 1, characterized in that, The step of extracting multiple regions from the second target global image data to obtain target local image data includes: Multiple regions are identified in the global image data of the second target based on an attention mechanism; For each of the regions, calculate the degree of difference in grayscale values ​​between pixels within the region and pixels outside the region; For each of the aforementioned regions, the probability of adverse events occurring in the region in the past is queried and used as the defect sensitivity. Identify one or more process items whose simulated process parameters change the most compared to the original process parameters, and designate them as the change items; For each of the regions, calculate the correlation between the region and the change term; For each of the aforementioned regions, the degree of change difference, the degree of defect sensitivity, and the correlation are fused into an attention value; The regions with the highest attention values ​​are extracted from the second target global image data to form the target local image data.

4. The method according to claim 1, characterized in that, The step of generating confidence scores based on the simulation process parameters, the second target global image data, and the target microscopic image data includes: Extract process features from the simulated process parameters; Extract the first image features from the global image data of the second target; Second image features are extracted from the target microscopic image data of each frame, and average pooling is performed on multiple second image features to obtain third image features; An attention mechanism is used to fuse the process features, the first image features, and the third image features into a multimodal feature; The multimodal features are mapped and activated as confidence levels.

5. The method according to any one of claims 1-4, characterized in that, The pre-set training set includes global image data of the sample thin film material after dry etching using an automated optical inspection device and sample process parameters for dry etching of the sample thin film material. The method further includes: Two frames of the sample global image data with similar appearance are selected as the baseline global image data and the standard global image data; Generate sample text information; the sample text information is used to describe the morphological changes of the sample thin film material from the baseline global image data to the standard global image data. The baseline global image count and the sample text information are input into a preset first generative adversarial network to reconstruct the recombined global image data. A first adversarial loss value is generated based on the reconstructed global image data and the standard global image data; A first constraint loss value on the etching rate is calculated based on the sample text information, the sample process parameters, the reconstructed global image data, and the standard global image data. A second constraint loss value for etching uniformity is calculated based on the sample text information, the sample process parameters, the reconstructed global image data, and the standard global image data. The third constraint loss value on edge continuity is calculated based on the sample text information, the sample process parameters, the reconstructed global image data and the standard global image data. The first adversarial loss value, the first constraint loss value, the second constraint loss value, and the third constraint loss value are merged into a first total loss value; The first generative adversarial network is updated based on the first total loss value.

6. The method according to claim 5, characterized in that, The calculation of the first constraint loss value on the etching rate based on the sample text information, the sample process parameters, the reconstructed global image data, and the standard global image data includes: The sample text information and the sample process parameters are used to generate a first adjustment coefficient representing the etching depth; the etching depth is positively correlated with the etching rate. The product of the average gray level of the pixels in the standard global image data and the first adjustment coefficient is calculated as the first image etching factor; The absolute value of the difference between the average gray level of the pixels in the reconstructed global image data and the first image etching factor is taken to obtain the first constraint loss value; The calculation of the second constraint loss value on etching uniformity based on the sample text information, the sample process parameters, the reconstructed global image data, and the standard global image data includes: The sample text information and the sample process parameters are used to generate a second adjustment coefficient representing etching uniformity. The product of the standard deviation of pixel gray levels in the standard global image data and the second adjustment coefficient is calculated and used as the second image etching factor. The absolute value of the difference between the standard deviation of pixel gray levels in the reconstructed global image data and the second image etching factor is taken to obtain the second constraint loss value. The calculation of the third constraint loss value on edge continuity based on the sample text information, the sample process parameters, the reconstructed global image data, and the standard global image data includes: The sample text information and the sample process parameters are used to generate a third adjustment factor representing photoresist erosion; The product of the gradient value of the local area where photoresist exists in the standard global image data and the third adjustment coefficient is calculated as the third image etching factor; The first norm is taken for the difference between the gradient value of the local area containing photoresist in the reconstructed global image data and the second image etching factor to obtain the third constraint loss value.

7. The method according to any one of claims 1-4, characterized in that, The pre-set training set includes global image data of a sample thin film material after dry etching, obtained by using an automated optical inspection device; sample process parameters for dry etching of the sample thin film material; and local image data of the sample obtained by cropping a region from the global image data, as well as microscopic image data of the sample obtained by using a scanning electron microscope to detect the same region in the global image data. The method further includes: The sample local image data is input into a preset second generative adversarial network for amplification to obtain simulated microscopic image data; A second adversarial loss value is generated based on the sample local image data, the sample microscopic image data, and the simulated microscopic image data; The fourth constraint loss value at the sidewall tilt angle is calculated based on the simulated microscopic image data and the sample process parameters; The fifth constraint loss value for the continuity of the multilayer film interface is calculated based on the simulated microscopic image data. The sixth constraint loss value for sidewall roughness is calculated based on the simulated microscopic image data. The second adversarial loss value, the fourth constraint loss value, the fifth constraint loss value, and the sixth constraint loss value are merged into a second total loss value; The second generative adversarial network is updated based on the second total loss value.

8. The method according to claim 7, characterized in that, The calculation of the fourth constraint loss value at the sidewall tilt angle based on the sample local image data, the simulated microscopic image data, and the sample process parameters includes: The target angle is determined based on the local image data of the sample and the process parameters of the sample. The absolute value of the difference between the sidewall tilt angle and the target angle in the simulated microscopic image data is taken as the fourth constraint loss value. The calculation of the fifth constraint loss value on the continuity of the multilayer film interface based on the simulated microscopic image data includes: Query the vertical pixel coordinates of the multilayer film interface for the sample thin film material; The first norm is calculated based on the gradient value of the pixel gray level in the simulated microscopic image data according to the vertical pixel coordinates, and the fifth constraint loss value is obtained. The calculation of the fifth constraint loss value on the continuity of the multilayer film interface based on the simulated microscopic image data includes: The baseline value is determined based on the sample process parameters; The absolute value of the difference between the sidewall roughness of the simulated microscopic image data and the reference value is taken as the sixth constraint loss value.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for simulating dry etching of thin film materials as described in any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for simulating dry etching of thin film materials as described in any one of claims 1-8.

Images