Semiconductor lithography process simulation optimization methods, devices, storage media and electronic devices
By establishing a parameterized unified model and variational analysis, the problems of excessive computation and poor reusability of simulation models in traditional lithography simulation methods have been solved, achieving efficient multi-scenario simulation optimization and improving the accuracy and efficiency of lithography processes.
Patent Information
- Application Number
- CN202611619099.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Traditional photolithography simulation methods suffer from exponentially increasing computational load when dealing with non-nominal process settings, poor reusability of simulation models, and serious impact on design cycles and increased manufacturing costs.
By establishing a parameterized unified model, the differences in non-nominal process scenarios are transformed into quantifiable parameter offsets. Variational analysis and parametric variational network models are then used to achieve rapid prediction and optimization of multi-scenario simulations.
While ensuring accuracy, it significantly improves the efficiency of multi-scenario simulation, reduces the consumption of computing resources, shortens the design cycle, and improves the yield of chip manufacturing.
Smart Images

Figure CN123116884A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of photolithography technology, specifically to a semiconductor photolithography process simulation optimization method, apparatus, storage medium, and electronic device. Background Technology
[0002] As semiconductor technology evolves towards more advanced process nodes, chip feature sizes continue to shrink. Photolithography, as the core process for transferring design patterns to wafers, sees its process window narrow dramatically with the shrinking feature size. In the advancement of extreme ultraviolet lithography towards single-nanometer nodes, factors such as optical aberrations, mask 3D effects, shading effects, and photoresist chemical noise increasingly affect pattern fidelity, leading to increased critical dimension errors and edge placement errors, further reducing the process window. Maintaining a sufficient process window to achieve high-resolution patterning under conditions of continuously shrinking metal pitch has become a major challenge in semiconductor manufacturing.
[0003] To address these challenges, model-based optical proximity correction techniques have become a core means of compensating for pattern distortion caused by optical proximity effects. Resolution enhancement techniques such as light source-mask co-optimization and reverse lithography are also being used to expand the process window. As the foundation of these techniques, photolithography simulation establishes physicochemical models to simulate the propagation and imaging process of light in photoresist and photomasks, providing quantitative basis for mask correction and process parameter optimization, and playing a crucial role in advanced process development.
[0004] However, traditional lithography simulation methods employ a "multi-scenario independent modeling and simulation" approach when handling non-nominal process settings. This involves creating independent models and performing separate calculations for different process scenarios such as defocusing, dose fluctuations, mask errors, light source wavelength drift, and numerical aperture fluctuations. As layout complexity increases, accuracy requirements rise, and the number of non-nominal scenarios to be considered grows, this approach leads to an exponential increase in computational load. Furthermore, the simulation models have poor reusability; changes in process parameters necessitate model reconstruction, severely impacting design cycles and increasing manufacturing costs. Therefore, improving the efficiency of multi-scenario simulation while maintaining accuracy has become a key bottleneck in the development of lithography simulation tools. Summary of the Invention
[0005] This application provides a semiconductor lithography process simulation optimization method, apparatus, storage medium, and electronic device, which can improve simulation efficiency in multiple scenarios while ensuring accuracy.
[0006] In a first aspect, embodiments of this application provide a semiconductor photolithography process simulation optimization method, including: Obtain the baseline physical model and its corresponding baseline imaging quality under the nominal process parameters; The difference between the non-nominal process scenario and the nominal process parameters is converted into a quantifiable parameter offset, and the parameter offset is embedded into the benchmark physical model to establish a parameterized unified model. Based on the aforementioned parameterized unified model, variational analysis is performed in the non-nominal parameter space with the nominal process parameters as the center, in order to quantify the sensitivity of the influence of the parameter offset on the imaging quality and establish a quantitative mapping relationship between parameter variation and imaging quality change. Based on the quantization mapping relationship, a parametric variational network model is constructed, and the imaging quality under any non-nominal scene is predicted through the parametric variational network model. Based on the prediction results of the parameter variational network model, a robust region in the target parameter space where the imaging quality meets the preset process requirements is determined, and the process window optimization results are output.
[0007] In the semiconductor lithography process simulation optimization method provided in this application embodiment, the step of embedding the parameter offset into the benchmark physical model to establish a parameterized unified model includes: Each parameter offset is injected into a different operator of the baseline physical model according to the physical action mechanism; All parameter offsets after injection are uniformly encapsulated with the nominal process parameters to form a multi-dimensional process parameter vector. Based on the multidimensional process parameter vector, by adjusting the value of the parameter offset, the baseline physical model covers all process scenarios, including nominal and non-nominal scenarios, to obtain a parameterized unified model.
[0008] In the semiconductor lithography process simulation optimization method provided in this application embodiment, the step of performing variational analysis centered on the nominal process parameters in a non-nominal parameter space based on the parameterized unified model to quantify the sensitivity of the influence of the parameter offset on the imaging quality and establish a quantitative mapping relationship between parameter variation and imaging quality change includes: In the non-nominal parameter space, with the reference point corresponding to the nominal process parameter as the variation center, a preset perturbation amount is applied to the parameter offset in at least one dimension. Based on the aforementioned parameterized unified model, the first-order gradient sensitivity and second-order sensitivity of imaging quality with respect to each process parameter are solved. The first-order gradient sensitivity and the second-order sensitivity are embedded into the variational prediction formula to establish a quantitative mapping relationship between the parameter variation and the image quality change.
[0009] In the semiconductor lithography process simulation optimization method provided in this application embodiment, the step of solving the first-order gradient sensitivity and second-order sensitivity of imaging quality with respect to each process parameter based on the parameterized unified model includes: Based on the parameterized unified model, a forward nominal simulation is run once under the nominal process parameters to obtain the full field of view physical field distribution. Construct an adjoint source based on the full field-of-view physical field distribution and run an inverse adjoint simulation; Based on the forward nominal simulation and the backward adjoint simulation, the first-order gradient vector of all process parameters is analyzed by the convolution integral of the forward field and the adjoint field to obtain the first-order gradient sensitivity. Based on the full-field physical field distribution and automatic differentiation framework, the Hessian-Free method is used to calculate the product of the Hessian matrix and any vector to obtain the second-order sensitivity.
[0010] In the semiconductor lithography process simulation optimization method provided in this application embodiment, the step of constructing a parameter variational network model based on the quantization mapping relationship includes: The spatial image matrix under the nominal process scenario, the conditional feature map corresponding to the parameter offset, and the spatial attention mask corresponding to the first-order gradient sensitivity are used as inputs to the residual network. The nonlinear residual information in the input features is extracted through the residual network, and the variational residual term is output. The output of the quantization mapping relationship is connected to the output of the residual network to form a parametric variational network model.
[0011] In the semiconductor lithography process simulation optimization method provided in this application embodiment, the step of determining a robust region in the target parameter space where the imaging quality meets the preset process requirements based on the prediction results of the parameter variational network model, and outputting the process window optimization results, includes: Using the aforementioned parameter variational network model, grid sampling is performed in the target parameter space to obtain the imaging quality prediction results for each sampling point; Based on the imaging quality prediction results, evaluate whether the key process indicators corresponding to the sampling points meet the preset constraints. Based on the sampling points that satisfy the preset constraints, a robust region in the target parameter space whose imaging quality meets the preset process requirements is determined. Based on the robust region, determine the optimal process setpoint and process drift warning line, and output the process window optimization results.
[0012] In the semiconductor lithography process simulation optimization method provided in this application embodiment, the step of determining a robust region in the target parameter space whose imaging quality meets the preset process requirements based on sampling points that satisfy the preset constraints includes: Based on the sampling points that satisfy the preset constraints, the boundary of the feasible region is delineated in the target parameter space; The region within the feasible region boundary is considered a robust region.
[0013] Secondly, embodiments of this application provide a semiconductor photolithography process simulation and optimization apparatus, comprising: The acquisition unit is used to acquire the baseline physical model under the nominal process parameters and its corresponding baseline imaging quality; The conversion unit is used to convert the difference between the non-nominal process scenario and the nominal process parameters into a quantifiable parameter offset, and to embed the parameter offset into the benchmark physical model to establish a parameterized unified model. The analysis unit is used to perform variational analysis in the non-nominal parameter space based on the parameterized unified model, with the nominal process parameters as the center, to quantify the sensitivity of the influence of the parameter offset on the imaging quality and establish a quantitative mapping relationship between parameter variation and imaging quality change. The prediction unit is used to construct a parametric variational network model based on the quantization mapping relationship, and to predict the imaging quality of any non-nominal scene through the parametric variational network model. The output unit is used to determine the robust region in the target parameter space where the imaging quality meets the preset process requirements based on the prediction results of the parameter variational network model, and output the process window optimization results.
[0014] Thirdly, this application provides a storage medium storing a plurality of instructions that are adapted for loading by a processor to execute the semiconductor photolithography process simulation optimization method described in any of the preceding claims.
[0015] Fourthly, this application 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 executes the computer program to implement the semiconductor lithography process simulation optimization method described in any of the above claims.
[0016] In summary, the semiconductor lithography process simulation optimization method provided in this application includes: obtaining a baseline physical model and its corresponding baseline imaging quality under nominal process parameters; converting the difference between non-nominal process scenarios and the nominal process parameters into quantifiable parameter offsets, embedding the parameter offsets into the baseline physical model to establish a parameterized unified model; performing variational analysis in the non-nominal parameter space centered on the nominal process parameters to quantify the sensitivity of the parameter offsets to the imaging quality, and establishing a quantitative mapping relationship between parameter variation and imaging quality changes; constructing a parameter variational network model according to the quantitative mapping relationship, and predicting the imaging quality under any non-nominal scenario through the parameter variational network model; determining a robust region in the target parameter space where the imaging quality meets the preset process requirements based on the prediction results of the parameter variational network model, and outputting the process window optimization results. This application embodiment transforms the differences between multiple scenes into parameter offsets by establishing a parameterized unified model, uses variational analysis to establish a quantitative mapping relationship between the offsets and changes in imaging quality to achieve rapid coarse calculation, and uses residual networks to compensate for higher-order errors of variational analysis in strongly nonlinear regions, thereby improving the efficiency of multi-scene simulation while ensuring accuracy. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.
[0018] Figure 1 This is a schematic diagram illustrating an application scenario of the semiconductor lithography process simulation optimization method provided in this application embodiment.
[0019] Figure 2 This is a flowchart illustrating the semiconductor lithography process simulation optimization method provided in the embodiments of this application.
[0020] Figure 3 This is a schematic diagram of the semiconductor lithography process simulation and optimization device provided in the embodiments of this application.
[0021] Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0022] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0023] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.
[0024] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.
[0025] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "module," "part," or "unit" may be used interchangeably.
[0026] In the description of this application, it should be noted that the terms "upper," "lower," "left," "right," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. In addition, terms such as "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0027] Traditional lithography simulation methods employ a "multi-scenario independent modeling and simulation" approach when handling non-nominal process settings. This involves creating independent models and performing separate calculations for different process scenarios such as defocusing, dose fluctuations, mask errors, light source wavelength drift, and numerical aperture fluctuations. As layout complexity increases, accuracy requirements rise, and the number of non-nominal scenarios to be considered grows, this approach leads to an exponential increase in computational load. Furthermore, the simulation models have poor reusability; changes in process parameters necessitate model reconstruction, severely impacting design cycles and increasing manufacturing costs. Therefore, improving the efficiency of multi-scenario simulation while maintaining accuracy has become a key bottleneck in the development of lithography simulation tools.
[0028] Based on this, embodiments of this application provide a semiconductor lithography process simulation optimization method, apparatus, storage medium, and electronic device. Specifically, the semiconductor lithography process simulation optimization apparatus can be integrated into an electronic device, which can be a server or a terminal, etc. The terminal can include mobile phones, wearable smart devices, tablet computers, laptops, and personal computers (PCs), etc., and other computer and auxiliary devices. The server can be a single server or a server cluster composed of multiple servers, and can be a physical server or a virtual server.
[0029] For example, such as Figure 1 As shown, the electronic device can acquire the baseline physical model and its corresponding baseline imaging quality under nominal process parameters; the difference between the non-nominal process scenario and the nominal process parameters is transformed into a quantifiable parameter offset, and the parameter offset is embedded into the baseline physical model to establish a parameterized unified model; based on the parameterized unified model, variational analysis is performed in the non-nominal parameter space with the nominal process parameters as the center to quantify the sensitivity of the parameter offset to the imaging quality, and to establish a quantitative mapping relationship between parameter variation and imaging quality changes; according to the quantitative mapping relationship, a parameter variational network model is constructed, and the imaging quality under any non-nominal scenario is predicted through the parameter variational network model; based on the prediction results of the parameter variational network model, a robust region in the target parameter space where the imaging quality meets the preset process requirements is determined, and the process window optimization results are output.
[0030] The technical solutions shown in this application will be described in detail below through specific embodiments. It should be noted that the order of description of the following embodiments is not intended to limit the priority of the embodiments.
[0031] Please see Figure 2 , Figure 2 This is a flowchart illustrating the semiconductor lithography process simulation optimization method provided in this application embodiment. The specific flow of this semiconductor lithography process simulation optimization method can be as follows: 101. Obtain the baseline physical model and its corresponding baseline imaging quality under the nominal process parameters.
[0032] The nominal process parameters may include the optimal focal length, standard exposure dose, and ideal mask size. These nominal process parameters correspond to the standard settings of the semiconductor lithography process under stable operating conditions.
[0033] Among them, the baseline physical model is a photolithography imaging model built on the rigorous vector imaging theory. This model can use rigorous coupled-wave analysis (RCWA) or finite-difference time-domain (FDTD) as the core of electromagnetic field solution to simulate the propagation and imaging process of light in photoresist and mask.
[0034] Among them, the baseline imaging quality refers to the imaging quality index calculated by the baseline physical model under the nominal process parameters, including but not limited to key size values, edge placement error, image distortion rate, and spatial image intensity distribution.
[0035] 102. Transform the differences between non-nominal process scenarios and nominal process parameters into quantifiable parameter offsets, and embed these parameter offsets into the baseline physical model to establish a parameterized unified model.
[0036] In this application embodiment, non-nominal process scenarios refer to various actual process situations that deviate from the nominal process parameters, including but not limited to defocusing, dose fluctuations, mask errors, wafer surface topography undulations, light source wavelength drift, and numerical aperture fluctuations. These differences between the non-nominal process scenarios and the nominal process parameters are converted into quantifiable parameter offsets, such as defocusing amount Δz, dose fluctuation ΔD, and mask error ΔM. These parameter offsets are injected into different operators of the baseline physical model according to their physical mechanisms.
[0037] Specifically, the difference between non-nominal process scenarios and nominal process parameters can be converted into quantifiable parameter offsets in the following ways: The defocusing amount is converted into a phase modulation parameter of the wavefront distortion. The wavefront distortion corresponding to different defocusing amounts is expressed using Zernike polynomial coefficients, which are then incorporated into the phase modulation factor of the pupil function. The Zernike polynomials are a set of orthogonal polynomials defined on the unit circle, where the Zernike coefficients Z4 describe the curvature change of the wavefront, and the defocusing aberration can be described by the Z4 coefficients. In practical implementation, the wavefront phase delay corresponding to the defocusing amount Δz can be expressed as φ(ρ) = 2π / λ·Δz·ρ 2 , where ρ is the normalized pupil radius and λ is the exposure wavelength. By adjusting this phase factor, wavefront distortion under arbitrary defocus can be simulated without reconstructing the optical propagation model.
[0038] The dose fluctuation is converted into a linear scaling factor for the nominal spatial image intensity. In practice, the change in spatial image intensity corresponding to the dose fluctuation ΔD can be expressed as I' = I0·(1+ΔD / D0), where I0 is the spatial image intensity at the nominal dose and D0 is the standard exposure dose. This linear scaling factor can be used to simulate the effect of dose fluctuation on exposure energy.
[0039] The mask error is converted into the edge offset of the near-field distribution of mask diffraction. In practical implementation, the mask error ΔM changes the boundary Fourier coefficients of the near-field of mask diffraction through the edge offset, thereby simulating the effect of mask manufacturing error on the diffracted light field.
[0040] The light source wavelength drift and numerical aperture fluctuation are transformed into perturbation terms in the optical transfer function. The light source wavelength drift is simulated by changing the wavelength parameter in the imaging integral equation, and the numerical aperture fluctuation is simulated by changing the aperture size in the pupil function.
[0041] In this embodiment of the application, the step of "embedding the parameter offset into the baseline physical model to establish a parameterized unified model" may include: 1021. Inject the offsets of each parameter into different operators of the benchmark physical model according to the physical action mechanism.
[0042] Specifically, the defocusing amount is injected into the pupil function operator in the form of phase modulation, the dose fluctuation is injected into the illumination source operator in the form of intensity scaling, and the mask error is injected into the mask diffraction near-field operator in the form of boundary perturbation.
[0043] 1022. All parameter offsets after injection are uniformly encapsulated with the nominal process parameters to form a multi-dimensional process parameter vector.
[0044] The multidimensional process parameter vector is represented as the superposition of the nominal process parameter vector and the offset vector, i.e., θ = θ0 + Δθ, where θ0 is the nominal process parameter vector and Δθ is the offset vector that includes the offset of all parameters such as defocusing amount, dose fluctuation, and mask error.
[0045] 1023. Based on the multi-dimensional process parameter vector, by adjusting the value of the parameter offset, the benchmark physical model covers all process scenarios, including nominal and non-nominal scenarios, to obtain a parameterized unified model.
[0046] Understandably, this parameterized unified model covers all process scenarios with a single model, completely replacing the traditional approach of building independent models for each non-nominal scenario.
[0047] In this embodiment of the application, by converting the differences of all non-nominal scenarios into quantifiable parameter offsets and embedding them into the same benchmark physical model, the repetitive work of building simulation models separately for each process scenario is avoided, and the model reuse rate is significantly improved.
[0048] 103. Based on the parameterized unified model, variational analysis is performed in the non-nominal parameter space with the nominal process parameters as the center. The sensitivity of the influence of parameter offset on imaging quality is quantified, and a quantitative mapping relationship between parameter variation and imaging quality change is established.
[0049] Specifically, step 103 may include the following steps: 1031. In the non-nominal parameter space, with the reference point corresponding to the nominal process parameter as the variation center, apply a preset perturbation to the parameter offset of at least one dimension.
[0050] The preset perturbation refers to the change in parameter offset that is pre-set in variational analysis. This perturbation is used to evaluate the impact of small parameter changes on image quality.
[0051] 1032. Based on the parameterized unified model, solve for the first-order gradient sensitivity and second-order sensitivity of imaging quality with respect to each process parameter.
[0052] Specifically, a forward nominal simulation can be run once under nominal process parameters based on a parametric unified model to obtain the full field-of-view physical field distribution. This full field-of-view physical field distribution contains electromagnetic field information throughout the entire field of view during photolithography imaging and serves as the physical basis for subsequent sensitivity calculations.
[0053] Next, an adjoint source is constructed based on the full-field physical field distribution, and an inverse adjoint simulation is run. The adjoint source is the derivative of the imaging quality function with respect to the final photoresist profile. By constructing the adjoint source and running the inverse adjoint simulation, sensitivity information can be obtained efficiently.
[0054] Then, based on the forward nominal simulation and the backward adjoint simulation, the first-order gradient vectors of all process parameters are analyzed through the convolution integral of the forward field and the adjoint field to obtain the first-order gradient sensitivity. This first-order gradient sensitivity characterizes the degree of linear influence of the offset of each parameter on the imaging quality. Based on the reciprocity of the Green's function, the gradient update information of each process parameter can be obtained through forward simulation and adjoint simulation.
[0055] Finally, based on the full-field physical field distribution and automatic differentiation framework, the Hessian-free method is used to calculate the product of the Hessian matrix and any vector, obtaining the second-order sensitivity. This second-order sensitivity characterizes the degree of nonlinear influence of the coupling effect between the offsets of various parameters on the imaging quality. The Hessian-free method avoids the complete explicit storage and calculation of the Hessian matrix, significantly reducing computational resource consumption.
[0056] 1033. Embed the first-order gradient sensitivity and second-order sensitivity into the variational prediction formula to establish a quantitative mapping relationship between parameter variation and imaging quality change.
[0057] In this embodiment, let the nominal process parameter be θ0 and the imaging quality function be f(θ), then the imaging quality change corresponding to the parameter variation Δθ = θ - θ0 can be expressed as: f(θ0+Δθ) ≈ f(θ0) + f(θ0)·Δθ + Δθ T H(θ0)Δθ + ...; in, f(θ0) represents the first-order gradient sensitivity, and H(θ0) represents the second-order sensitivity. By solving this variational prediction formula, the impact of different parameter offsets on image quality can be quickly quantified.
[0058] It should be noted that this quantization mapping relationship is used to calculate a rough prediction of the change in imaging quality based on any given parameter offset.
[0059] In the specific implementation process, after substituting the first-order gradient vector and the second-order sensitivity into the above variational prediction formula, when the parameter offset of any non-nominal scene is received, only vector dot product and matrix quadratic operation are needed to obtain the rough prediction value of the change in imaging quality. The calculation time is only in milliseconds, which completely avoids the minutes or even hours of time required to rerun the physical simulation in the traditional method.
[0060] In this embodiment, variational analysis further includes an adaptive decision step: calculating the estimated higher-order truncation margin of the variational prediction; when the higher-order truncation margin is less than a preset threshold, directly using the coarsely calculated prediction value as the current non-nominal scene's imaging quality output; when the higher-order truncation margin reaches or exceeds the preset threshold, marking the current sample as a strongly nonlinear sample and transferring it to the parametric variational network model for correction processing.
[0061] In this embodiment of the application, a quantitative mapping relationship between parameter variation and imaging quality change is established by means of the above method. Thus, variational analysis is used to replace the traditional "discrete scene sampling" simulation mode with analytical calculation, realizing continuous coverage and global analysis of the non-nominal parameter space. At the same time, the key parameters that have the most significant impact on imaging quality are accurately identified, and the amount of computation for multi-scene simulation is greatly reduced.
[0062] 104. Based on the quantization mapping relationship, construct a parametric variational network model, and use the parametric variational network model to predict the imaging quality under any non-nominal scene.
[0063] In some embodiments, the step "constructing a parameter variational network model based on the quantization mapping relationship" may include the following steps: 1041. The spatial image matrix under the nominal process scenario, the conditional feature map corresponding to the parameter offset, and the spatial attention mask corresponding to the first-order gradient sensitivity are used as inputs to the residual network.
[0064] The spatial image matrix under the nominal process scenario is the spatial image intensity distribution matrix under the nominal process parameters, with a size of H×W. It serves as the baseline physical image channel for the residual network, providing physical background information for the network. The conditional feature map corresponding to the parameter offset is a feature map (size H×W×C) of the same size as the spatial image matrix, which is stretched after the parameter offset vector Δθ is increased in dimension through a fully connected layer. It serves as the conditional control signal for the network. The spatial attention mask corresponding to the first-order gradient sensitivity is a spatial attention map generated by normalizing the first-order gradient sensitivity obtained in step 103. This spatial attention mask is multiplied element-wise with the conditional feature map and then input into the residual network, forcing the network to focus its main computational resources on physically sensitive areas such as edges, line ends, and corners.
[0065] 1042. Extract nonlinear residual information from the input features using a residual network and output variational residual terms.
[0066] In this embodiment, the residual network can employ a lightweight ResNet architecture as a variational residual module. The output of the residual network is not the final image quality prediction value, but a variational residual term used to compensate for the approximation error of variational analysis when the parameter variation is large or the nonlinearity is strong.
[0067] 1043. Connect the output of the quantization mapping relationship with the output of the residual network to form a parametric variational network model.
[0068] In practice, the final predicted value of the parametric variational network model is synthesized through residual connections. That is, the coarse predicted value of the imaging quality change calculated from the quantization mapping relationship is added to the variational residual term output by the residual network to obtain the final predicted value of the imaging quality. This "residual learning" strategy allows the neural network to learn only high-order nonlinear residuals that are difficult to fit precisely by physical formulas, greatly simplifying the learning task and improving generalization.
[0069] In this embodiment, a dual-objective loss function, including data loss and physical consistency loss, is also used to train the residual network. The data loss employs mean squared error loss to fit the label data generated by a high-precision physics simulator. The physical consistency loss, as a physical consistency regularization term, penalizes situations where the prediction results violate basic physical monotonicity (e.g., increased dosage leads to an abnormal decrease in critical size values), ensuring that the model predictions conform to physical laws.
[0070] In this embodiment, the parametric variational network model further includes a closed-loop correction step: monitoring the deviation between the predicted value of the residual network and the coarse predicted value of the variational analysis in each local interval of the parameter space; when the deviation continuously exceeds a preset threshold in a certain local interval, marking the local interval as a strongly nonlinear complex region; calling a high-precision physical simulator to supplement training label data only for the strongly nonlinear complex region; and using the supplemented label data to incrementally fine-tune the residual network to correct the system bias caused by the truncation of higher-order terms in the variational analysis.
[0071] In this embodiment, by fusing variational analysis with a residual network, the applicability of variational analysis is extended from small variations to large-range variations (such as the entire range of defocus ±0.2 micrometers). This compensates for the high-order nonlinear effects of traditional variational analysis when the parameter variations are large or the nonlinearity is strong, improving the stability and prediction accuracy of the network model. Simultaneously, the sensitivity results of variational analysis are used as attention weights in the residual network, enabling the model to prioritize focusing on the variational influence of key parameters. The network's prediction results can then correct the approximation errors of the variational analysis, forming a closed-loop optimization of "analysis-data".
[0072] 105. Based on the prediction results of the parameter variational network model, determine the robust region in the target parameter space where the imaging quality meets the preset process requirements, and output the process window optimization results.
[0073] Specifically, a parametric variational network model can be used to perform grid sampling in the target parameter space to obtain the imaging quality prediction results for each sampling point.
[0074] The target parameter space refers to the range of process parameters for which process window analysis is required, such as defocusing amount Δz ∈ [-100nm, +100nm] and dose fluctuation ΔD ∈ [-10%, +10%]. In practice, high-density grid sampling is performed within the target parameter space, with the number of sampling points on the order of tens of thousands. The parameter offset of each sampling point is input into the parameter variational network model to obtain the image quality prediction result corresponding to each sampling point. Since the forward inference speed of the parameter variational network model is on the order of milliseconds, the overall scan only takes a few minutes, while traditional point-by-point simulation takes several weeks to complete a scan of the same density.
[0075] Next, based on the imaging quality prediction results, we evaluate whether the key process indicators of the corresponding sampling points meet the preset constraints.
[0076] The key process parameters include at least one of the following: critical dimension error, edge placement error, and depth of focus. Preset constraints refer to the tolerance range of the photolithography process for the above key process parameters, such as a critical dimension error of less than ±2 nanometers, an edge placement error less than the design rule requirements, and a depth of focus greater than the minimum exposure energy requirement of the photoresist.
[0077] Then, based on the sampling points that meet the preset constraints, a robust region in the target parameter space is determined where the imaging quality meets the preset process requirements.
[0078] Specifically, based on sampling points that satisfy preset constraints, the feasible region boundary can be delineated in the target parameter space, and the region within the feasible region boundary is considered the robust region. In practical implementation, contour extraction algorithms (such as the Marching Squares algorithm) can be used to accurately delineate the feasible region boundary in the target parameter space where all constraints are satisfied. The area or volume of this feasible region represents the robustness of the process window; a larger area indicates a higher tolerance of the production line to process disturbances and a more guaranteed yield.
[0079] Finally, the optimal process setpoint and process drift warning line are determined based on the robust region, and the process window optimization results are output.
[0080] In practical applications, by using the geometric center value of the robust region as the optimal process setting point for the production line (i.e., the optimal setting values for exposure dose and focal length), and using the boundary value of the robust region as the process drift warning line (i.e., issuing an alarm when the process parameters drift to the boundary), a quantitative and visual decision-making basis is directly provided for the optimization of the process window of the lithography process.
[0081] In the embodiments of this application, the high-speed inference capability of the parametric variational network model can be utilized to achieve continuous coverage and global analysis in the parameter space, accurately delineate robust regions that meet process requirements, provide quantified optimal process setpoints and early warning lines for the lithography process, effectively guide the adjustment of production line process parameters, and improve the yield level of chip manufacturing.
[0082] In summary, the semiconductor lithography process simulation optimization method provided in this application includes: obtaining a baseline physical model and its corresponding baseline imaging quality under nominal process parameters; converting the difference between non-nominal process scenarios and nominal process parameters into quantifiable parameter offsets, embedding the parameter offsets into the baseline physical model to establish a parameterized unified model; performing variational analysis in the non-nominal parameter space centered on the nominal process parameters to quantify the sensitivity of the parameter offsets to the imaging quality, and establishing a quantitative mapping relationship between parameter variation and imaging quality changes; constructing a parameter variational network model based on the quantitative mapping relationship, and predicting the imaging quality under any non-nominal scenario through the parameter variational network model; determining a robust region in the target parameter space where the imaging quality meets the preset process requirements based on the prediction results of the parameter variational network model, and outputting the process window optimization results. This application embodiment transforms the differences between multiple scenes into parameter offsets by establishing a parameterized unified model, uses variational analysis to establish a quantitative mapping relationship between the offsets and changes in imaging quality to achieve rapid coarse calculation, and uses residual networks to compensate for higher-order errors of variational analysis in strongly nonlinear regions, thereby improving the efficiency of multi-scene simulation while ensuring accuracy.
[0083] To facilitate better implementation of the semiconductor lithography process simulation optimization method provided in this application, this application also provides a semiconductor lithography process simulation optimization apparatus. The meanings of the terms used are the same as in the semiconductor lithography process simulation optimization method described above, and specific implementation details can be found in the descriptions in the method embodiments.
[0084] Please see Figure 3 , Figure 3 This is a schematic diagram of the semiconductor lithography process simulation and optimization apparatus provided in an embodiment of this application. The semiconductor lithography process simulation and optimization apparatus may include an acquisition unit 201, a conversion unit 202, an analysis unit 203, a prediction unit 204, and an output unit 205. The acquisition unit 201 is used to acquire the reference physical model under the nominal process parameters and its corresponding reference imaging quality; The conversion unit 202 is used to convert the difference between the non-nominal process scenario and the nominal process parameters into a quantifiable parameter offset, and to embed the parameter offset into the benchmark physical model to establish a parameterized unified model. Analysis unit 203 is used to perform variational analysis in a non-nominal parameter space based on the parameterized unified model, with the nominal process parameters as the center, to quantify the sensitivity of the influence of the parameter offset on the imaging quality and establish a quantitative mapping relationship between parameter variation and imaging quality change. The prediction unit 204 is used to construct a parametric variational network model based on the quantization mapping relationship, and to predict the imaging quality of any non-nominal scene through the parametric variational network model. The output unit 205 is used to determine the robust region in the target parameter space where the imaging quality meets the preset process requirements based on the prediction results of the parameter variational network model, and output the process window optimization results.
[0085] For specific implementation methods of each of the above units, please refer to the embodiments of the above semiconductor lithography process simulation optimization method, which will not be repeated here.
[0086] In summary, the semiconductor lithography process simulation and optimization apparatus provided in this application can acquire a baseline physical model and its corresponding baseline imaging quality under nominal process parameters through the acquisition unit 201; the conversion unit 202 converts the difference between the non-nominal process scenario and the nominal process parameters into a quantifiable parameter offset, and embeds the parameter offset into the baseline physical model to establish a parameterized unified model; the analysis unit 203 performs variational analysis in the non-nominal parameter space with the nominal process parameters as the center based on the parameterized unified model to quantify the sensitivity of the parameter offset to the imaging quality and establish a quantitative mapping relationship between parameter variation and imaging quality change; the prediction unit 204 constructs a parameter variational network model according to the quantitative mapping relationship, and predicts the imaging quality under any non-nominal scenario through the parameter variational network model; and the output unit 205 determines the robust region in the target parameter space where the imaging quality meets the preset process requirements based on the prediction results of the parameter variational network model, and outputs the process window optimization results. This application embodiment transforms the differences between multiple scenes into parameter offsets by establishing a parameterized unified model, uses variational analysis to establish a quantitative mapping relationship between the offsets and changes in imaging quality to achieve rapid coarse calculation, and uses residual networks to compensate for higher-order errors of variational analysis in strongly nonlinear regions, thereby improving the efficiency of multi-scene simulation while ensuring accuracy.
[0087] This application also provides an electronic device that may integrate the semiconductor photolithography process simulation and optimization device of this application, such as... Figure 4 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically: The electronic device may include components such as a processor 301 with one or more processing cores and a memory 302 with one or more computer-readable storage media. Those skilled in the art will understand that... Figure 4 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 301 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs stored in the memory 302 and / or the methods provided in this application, and by calling data stored in the memory 302, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. Optionally, the processor 301 may include one or more processing cores; preferably, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operation of the storage medium, user interface, and application programs, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 301.
[0088] The memory 302 can be used to store software programs and the methods provided in this application. The processor 301 executes various functional applications and data processing by running the software programs stored in the memory 302 and the methods provided in this application. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store applications required for operating the storage medium and at least one function; the data storage area may store data created based on the use of the electronic device. In addition, the memory 302 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.
[0089] Although not shown, the electronic device may also include a display unit, an input unit, and a power supply, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to the processes of one or more application programs into the memory 302 according to the following instructions, and the processor 301 runs the application programs stored in the memory 302 to realize various functions, as follows: Obtain the baseline physical model and its corresponding baseline imaging quality under the nominal process parameters; The difference between non-nominal process scenarios and nominal process parameters is transformed into quantifiable parameter offsets, and these parameter offsets are embedded into the baseline physical model to establish a parameterized unified model. Based on the parameterized unified model, variational analysis is performed in the non-nominal parameter space with nominal process parameters as the center. The sensitivity of the influence of parameter offset on imaging quality is quantified, and a quantitative mapping relationship between parameter variation and imaging quality change is established. Based on the quantization mapping relationship, a parametric variational network model is constructed, and the imaging quality under any non-nominal scene is predicted by the parametric variational network model. Based on the prediction results of the parametric variational network model, a robust region in the target parameter space where the imaging quality meets the preset process requirements is determined, and the process window optimization results are output.
[0090] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0091] Therefore, embodiments of this application provide a storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the methods provided in embodiments of this application. For example, the instructions can execute the following steps: Obtain the baseline physical model and its corresponding baseline imaging quality under the nominal process parameters; The difference between non-nominal process scenarios and nominal process parameters is transformed into quantifiable parameter offsets, and these parameter offsets are embedded into the baseline physical model to establish a parameterized unified model. Based on the parameterized unified model, variational analysis is performed in the non-nominal parameter space with nominal process parameters as the center. The sensitivity of the influence of parameter offset on imaging quality is quantified, and a quantitative mapping relationship between parameter variation and imaging quality change is established. Based on the quantization mapping relationship, a parametric variational network model is constructed, and the imaging quality under any non-nominal scene is predicted by the parametric variational network model. Based on the prediction results of the parametric variational network model, a robust region in the target parameter space where the imaging quality meets the preset process requirements is determined, and the process window optimization results are output.
[0092] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0093] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0094] Since the instructions stored in the storage medium can execute the steps of any method provided in the embodiments of this application, the beneficial effects that any method provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.
[0095] The semiconductor photolithography process simulation optimization method, apparatus, storage medium, and electronic device provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for simulation and optimization of semiconductor photolithography process, characterized in that, include: Obtain the baseline physical model and its corresponding baseline imaging quality under the nominal process parameters; The difference between the non-nominal process scenario and the nominal process parameters is converted into a quantifiable parameter offset, and the parameter offset is embedded into the benchmark physical model to establish a parameterized unified model. Based on the aforementioned parameterized unified model, variational analysis is performed in the non-nominal parameter space with the nominal process parameters as the center, in order to quantify the sensitivity of the influence of the parameter offset on the imaging quality and establish a quantitative mapping relationship between parameter variation and imaging quality change. Based on the quantization mapping relationship, a parametric variational network model is constructed, and the imaging quality under any non-nominal scene is predicted through the parametric variational network model. Based on the prediction results of the parameter variational network model, a robust region in the target parameter space where the imaging quality meets the preset process requirements is determined, and the process window optimization results are output.
2. The semiconductor lithography process simulation optimization method of claim 1, wherein, The step of embedding the parameter offset into the baseline physical model to establish a parameterized unified model includes: Each parameter offset is injected into a different operator of the baseline physical model according to the physical action mechanism; All parameter offsets after injection are uniformly encapsulated with the nominal process parameters to form a multi-dimensional process parameter vector. Based on the multidimensional process parameter vector, by adjusting the value of the parameter offset, the baseline physical model covers all process scenarios, including nominal and non-nominal scenarios, to obtain a parameterized unified model.
3. The method of claim 1, wherein the method further comprises: Based on the parameterized unified model, variational analysis is performed in the non-nominal parameter space centered on the nominal process parameters to quantify the sensitivity of the parameter offset to the image quality, and to establish a quantitative mapping relationship between parameter variation and image quality changes, including: In the non-nominal parameter space, with the reference point corresponding to the nominal process parameter as the variation center, a preset perturbation amount is applied to the parameter offset in at least one dimension. Based on the aforementioned parameterized unified model, the first-order gradient sensitivity and second-order sensitivity of imaging quality with respect to each process parameter are solved. The first-order gradient sensitivity and the second-order sensitivity are embedded into the variational prediction formula to establish a quantitative mapping relationship between the parameter variation and the image quality change.
4. The method of claim 3, wherein the optimization of the semiconductor lithography process simulation is performed by a genetic algorithm. The process of solving for the first-order gradient sensitivity and second-order sensitivity of imaging quality with respect to each process parameter based on the parameterized unified model includes: Based on the parameterized unified model, a forward nominal simulation is run once under the nominal process parameters to obtain the full field of view physical field distribution. Construct an adjoint source based on the full field-of-view physical field distribution and run an inverse adjoint simulation; Based on the forward nominal simulation and the backward adjoint simulation, the first-order gradient vector of all process parameters is analyzed by the convolution integral of the forward field and the adjoint field to obtain the first-order gradient sensitivity. Based on the full-field physical field distribution and automatic differentiation framework, the Hessian-Free method is used to calculate the product of the Hessian matrix and any vector to obtain the second-order sensitivity.
5. The method of claim 3 or 4, wherein the method further comprises: determining a plurality of process parameters for the semiconductor lithography process; and determining a plurality of process parameters for the semiconductor lithography process based on the plurality of process parameters and the plurality of process parameters. The step of constructing a parameter variational network model based on the quantization mapping relationship includes: The spatial image matrix under the nominal process scenario, the conditional feature map corresponding to the parameter offset, and the spatial attention mask corresponding to the first-order gradient sensitivity are used as inputs to the residual network. The nonlinear residual information in the input features is extracted through the residual network, and the variational residual term is output. The output of the quantization mapping relationship is connected to the output of the residual network to form a parametric variational network model.
6. The method of claim 1, wherein the method further comprises: Based on the prediction results of the parameter variational network model, a robust region in the target parameter space where the imaging quality meets the preset process requirements is determined, and the process window optimization results are output, including: Using the aforementioned parameter variational network model, grid sampling is performed in the target parameter space to obtain the imaging quality prediction results for each sampling point; Based on the imaging quality prediction results, evaluate whether the key process indicators corresponding to the sampling points meet the preset constraints. Based on the sampling points that satisfy the preset constraints, a robust region in the target parameter space whose imaging quality meets the preset process requirements is determined. Based on the robust region, determine the optimal process setpoint and process drift warning line, and output the process window optimization results.
7. The semiconductor photolithography process simulation optimization method as described in claim 6, characterized in that, The step of determining a robust region in the target parameter space whose imaging quality meets preset process requirements based on sampling points that satisfy the preset constraints includes: Based on the sampling points that satisfy the preset constraints, the boundary of the feasible region is delineated in the target parameter space; The region within the feasible region boundary is considered a robust region.
8. A semiconductor photolithography process simulation and optimization device, characterized in that, include: The acquisition unit is used to acquire the baseline physical model under the nominal process parameters and its corresponding baseline imaging quality; The conversion unit is used to convert the difference between the non-nominal process scenario and the nominal process parameters into a quantifiable parameter offset, and to embed the parameter offset into the benchmark physical model to establish a parameterized unified model. The analysis unit is used to perform variational analysis in the non-nominal parameter space based on the parameterized unified model, with the nominal process parameters as the center, to quantify the sensitivity of the influence of the parameter offset on the imaging quality and establish a quantitative mapping relationship between parameter variation and imaging quality change. The prediction unit is used to construct a parametric variational network model based on the quantization mapping relationship, and to predict the imaging quality of any non-nominal scene through the parametric variational network model. The output unit is used to determine the robust region in the target parameter space where the imaging quality meets the preset process requirements based on the prediction results of the parameter variational network model, and output the process window optimization results.
9. A storage medium, characterized in that, The storage medium stores multiple instructions, which are applicable to the processor for loading to execute the semiconductor photolithography process simulation optimization method according to any one of claims 1-7.
10. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the semiconductor lithography process simulation optimization method as described in any one of claims 1-7.
