Methods, apparatus, electronic devices, and storage media for predicting the final morphology of directional self-assembly of block copolymers.
By using molecular dynamics simulations and machine learning model training, the final morphology field of block copolymer directional self-assembly is generated, which solves the problem that existing models cannot directly output complete morphology and achieves efficient and reliable morphology prediction and analysis.
Patent Information
- Application Number
- CN202611328887.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-17
AI Technical Summary
Existing block copolymer directional self-assembly models cannot directly output a complete spatial density field/phase field that can be used for subsequent process simulation or structural analysis, which limits their availability in engineering processes.
By establishing a molecular model and interaction potential, molecular dynamics simulations are performed to generate the final state morphology field at multiple parameter points. The target final state morphology field is then directly output by training and optimizing a machine learning model.
It enables efficient and rapid prediction of the final morphology of oriented self-assembly of block copolymers, improves the reliability and direct usability of the model, and can output key information such as defect distribution, orientation order and domain size, while reducing computational costs.
Smart Images

Figure CN123120097A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of soft matter physics and computational materials, and in particular to a method and apparatus for predicting the final morphology of directional self-assembly of block copolymers, as well as electronic devices and storage media. Background Technology
[0002] Block copolymers (BCPs) are composed of two or more polymer segments with different chemical compositions linked by covalent bonds. Because they can form various self-assembly morphologies at the nanoscale, such as regular layers, pillars, spheres, and bicontinuous networks, they are widely used in fields such as nanolithography templates, nanoporous membranes, photonic crystals, and energy storage and conversion. Directed self-assembly (DSA) of block copolymers, through coupling with guiding structures such as trench templates and chemical pattern templates, can achieve ordered nanopatterns with dimensions sub-20 nm or even smaller.
[0003] In recent years, with the rapid development of machine learning in materials science, related technologies have attempted to use methods such as neural networks to automatically identify the phase state of block copolymers and predict morphology and defect density from experimental or simulated images. However, existing models in these technologies mostly focus on phase state identification (such as layered, columnar, and spherical structures) or predicting a few scalar parameters such as domain spacing, making it difficult to directly output a complete spatial density field / phase field that can be used for subsequent process simulation or structural analysis. This limits the direct usability of the models in engineering processes, requiring further reconstruction or simplification. Summary of the Invention
[0004] In view of this, this disclosure proposes a prediction scheme for the final morphology of the directional self-assembly of block copolymers.
[0005] According to one aspect of this disclosure, a method for predicting the final morphology of a block copolymer undergoing directional self-assembly is provided, comprising: establishing a corresponding molecular model and interaction potential based on a preset block copolymer system and a directional self-assembly scenario; selecting multiple values for multiple preset morphology control parameters to form multiple parameter points; performing molecular dynamics simulations on each parameter point with a random initial configuration as the initial state based on the molecular model and interaction potential, until the block copolymer system reaches morphological stability or a preset statistical convergence, to obtain a first final morphology field corresponding to each parameter point; training a preset model based on the parameter points and the first final morphology field; and using the trained model to predict target morphology control parameters to obtain a target final morphology field.
[0006] In one possible implementation, the preset morphology control parameters include at least one of polymer parameters, template geometry parameters, and annealing process parameters.
[0007] In one possible implementation, multiple values are selected for multiple preset morphology control parameters to form multiple parameter points, including: using a preset sampling method to select multiple value nodes within the value range of each preset morphology control parameter, and combining the value nodes of each preset morphology control parameter to form multiple parameter points.
[0008] In one possible implementation, the method further includes: for each parameter point, concatenating the polymer parameters, template geometry parameters, and annealing process parameters into a one-dimensional feature vector in a preset order; and combining multiple one-dimensional feature vectors into an input feature vector.
[0009] In one possible implementation, training the preset model based on the parameter points and the first final state topography field includes: dividing multiple sample data into a training set, a validation set, and a test set, where each sample data includes the input feature vector and the first final state topography field corresponding to the parameter points; using the input feature vector as the input feature and the first final state topography field as the output label, training the preset model based on the training set.
[0010] In one possible implementation, the method further includes: constructing a composite loss function, the composite loss function including a numerical error term and a physical constraint term; and optimizing the composite loss function based on the validation set.
[0011] In one possible implementation, the method further includes: selecting multiple parameter points for supplementary molecular dynamics simulation based on the prediction error or uncertainty estimate of the model at each parameter point in the validation set, adding the simulated data to the training set, and retraining the model.
[0012] According to another aspect of this disclosure, a device for predicting the final state morphology of block copolymer directional self-assembly is provided, comprising: a construction module for establishing a corresponding molecular model and interaction potential based on a preset block copolymer system and a directional self-assembly scenario; a value selection module for selecting multiple values for multiple preset morphology control parameters to form multiple parameter points; a simulation module for performing molecular dynamics simulation on each parameter point with a random initial configuration as the initial state based on the molecular model and interaction potential, until the block copolymer system reaches morphological stability or a preset statistical convergence, to obtain a first final state morphology field corresponding to each parameter point; a training module for training a preset model based on the parameter points and the first final state morphology field; and a prediction module for using the trained model to predict the target morphology control parameters to obtain the target final state morphology field.
[0013] In one possible implementation, the preset morphology control parameters include at least one of polymer parameters, template geometry parameters, and annealing process parameters.
[0014] In one possible implementation, the value acquisition module is used to select multiple value nodes within the value range of each preset morphology control parameter using a preset sampling method, and combine the value nodes of each preset morphology control parameter to form multiple parameter points.
[0015] In one possible implementation, the device further includes: a feature splicing module, used to splice the polymer parameters, template geometric parameters and annealing process parameters into a one-dimensional feature vector in a preset order for each parameter point; and to combine multiple one-dimensional feature vectors into an input feature vector.
[0016] In one possible implementation, the training module is used to divide multiple sample data into a training set, a validation set, and a test set, where each sample data includes the input feature vector and the first final state morphology field corresponding to the parameter point; using the input feature vector as the input feature and the first final state morphology field as the output label, a preset model is trained based on the training set.
[0017] In one possible implementation, the apparatus further includes: an optimization module for constructing a composite loss function, the composite loss function including a numerical error term and a physical constraint term; and optimizing the composite loss function based on the validation set.
[0018] In one possible implementation, the apparatus further includes: a retraining module, configured to select multiple parameter points for supplementary molecular dynamics simulation based on the prediction error or uncertainty estimate of the model at each parameter point in the validation set, add the simulated data to the training set, and retrain the model.
[0019] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.
[0020] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described method.
[0021] In this embodiment, based on a preset block copolymer system and a directional self-assembly scenario, a corresponding molecular model and interaction potential are established. Multiple values are selected for each of the preset morphology control parameters, forming multiple parameter points. Based on the molecular model and interaction potential, molecular dynamics simulations are performed on each parameter point with a random initial configuration as the initial state, until the block copolymer system reaches morphological stability or a preset statistical convergence, obtaining the first final state morphology field corresponding to each parameter point. Based on the parameter points and the first final state morphology field, the preset model is trained. Using the trained model, the target morphology control parameters are predicted to obtain the target final state morphology field. By using the method of this disclosure, the final state morphology of the directional self-assembly of block copolymers can be predicted efficiently and quickly.
[0022] In this embodiment of the disclosure, by using a random initial configuration as the starting state for molecular dynamics simulation at each parameter point, the consistency and reliability of the self-assembled structure obtained from repeated simulations at the same parameter point can be ensured; by inputting each parameter point and its corresponding morphology field as training data into a preset model and training it, rapid prediction of the self-assembled morphology can be achieved; by directly outputting the target final state morphology field, it is convenient to directly analyze the defect distribution, orientation order, domain size, etc. in the future.
[0023] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description
[0024] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.
[0025] Figure 1 A flowchart illustrating a method for predicting the final morphology of directional self-assembly of block copolymers according to embodiments of the present disclosure is shown.
[0026] Figure 2 A block diagram of a device for predicting the final morphology of directional self-assembly of block copolymers according to an embodiment of the present disclosure is shown.
[0027] Figure 3 A block diagram of an apparatus according to an embodiment of the present disclosure is shown. Detailed Implementation
[0028] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.
[0029] As used herein, the terms “comprising,” “including,” “having,” or variations thereof are open-ended and include one or more of the stated features, integrals, elements, steps, components, or functions, but do not exclude the presence or addition of one or more other features, integrals, elements, steps, components, functions, or groups thereof.
[0030] When an element is referred to as “connected,” “coupled,” “responding,” or a variation thereof relative to another element, it may be directly connected, coupled, or responding to another element, or there may be an intermediate element present.
[0031] Although the terms first, second, third, etc., may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another. Therefore, without departing from the teachings of the inventive concept, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments.
[0032] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.
[0033] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.
[0034] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions.
[0035] Figure 1 A flowchart illustrating a method for predicting the final morphology of directional self-assembly of block copolymers according to embodiments of the present disclosure is shown. Figure 1 As shown, the method includes:
[0036] S11. Based on the preset block copolymer system and directional self-assembly scenario, establish the corresponding molecular model and interaction potential;
[0037] Among them, the block copolymer system can be a system composed of at least one block copolymer, where polymer segments with different chemical properties can be connected by covalent bonds, and the system can assemble into a periodic ordered structure at the nanoscale under appropriate conditions.
[0038] Directional self-assembly can be achieved by placing a block copolymer system on a pre-treated substrate and using a pre-set chemical pattern or physical groove on the substrate surface as a template, so that the block copolymers are arranged in the direction and position set by the template during the self-assembly process to form an ordered structure.
[0039] Exemplarily, the block copolymer system in this disclosure can be a polymer composed of A segments and B segments with different chemical properties linked by covalent bonds, i.e., an A–B diblock copolymer. The A–B diblock copolymer can be polystyrene-block-polymethyl methacrylate (PS-b-PMMA), and this disclosure does not limit the scope of the invention. The block copolymer system and the directional self-assembly scenario can be the directional self-assembly of the A–B diblock copolymer in a groove template or a chemical pattern template.
[0040] In this embodiment of the disclosure, a molecular model and interaction potential corresponding to the block copolymer system can be established using molecular dynamics (MD) based on a preset block copolymer system and a directional self-assembly scenario. The molecular dynamics can be coarse-grained MD or all-atom MD.
[0041] S12. For multiple preset morphology control parameters, select multiple values to form multiple parameter points;
[0042] In this embodiment, there can be multiple preset morphology control parameters. For these preset morphology control parameters, a large number of values can be selected to form multiple complete parameter configuration schemes, i.e., multiple parameter points. These parameter points can also jointly constitute a multi-dimensional space covering all condition ranges, i.e., a parameter space.
[0043] In one possible implementation, the preset morphology control parameters include at least one of polymer parameters, template geometry parameters, and annealing process parameters.
[0044] Specifically, polymer parameters may include the volume fraction of each block. Total aggregation degree Flory-Hutchins interaction parameters Polymer parameters can be adjustable, meaning they can be set according to actual needs, and the self-assembly behavior of the system can be altered by adjusting these parameters. Template geometry parameters can include trench width. ,depth ,cycle The affinity of the sidewalls and bottom with different blocks, etc. Annealing process parameters may include annealing temperature. Annealing time Cooling rate And its time-varying programs, etc.
[0045] In this embodiment of the disclosure, by comprehensively considering polymer parameters, template geometry parameters, and annealing process parameters, and encoding them together as an input condition vector, the model can cover a more complex condition space that is closer to the actual process, thereby enhancing the reliability of model predictions.
[0046] In one possible implementation, multiple values are selected for multiple preset morphology control parameters to form multiple parameter points, including: using a preset sampling method to select multiple value nodes within the value range of each preset morphology control parameter, and combining the value nodes of each preset morphology control parameter to form multiple parameter points.
[0047] In this embodiment of the disclosure, the selection of parameter points can be achieved through a reasonable sampling strategy, such as using Latin hypercube sampling to sample in the parameter space to obtain feature combinations that can cover the entire parameter space, thereby forming multiple parameter points.
[0048] S13. Based on the molecular model and interaction potential, perform molecular dynamics simulation for each parameter point with a random initial configuration as the initial state until the block copolymer system reaches morphological stability or the preset statistical quantity converges, and obtain the first final state morphological field corresponding to each parameter point.
[0049] After selecting multiple parameter points, based on the molecular model and interaction potential obtained above, MD simulation can be performed on each parameter point starting from a random initial configuration until the system reaches morphological stability or the preset statistical quantity converges, thus obtaining the final state first final state morphological field corresponding to each parameter point.
[0050] Specifically, based on the molecular model and interaction potential established above, an initial spatial arrangement can be randomly generated for each parameter point, serving as the starting state for this calculation. Molecular dynamics calculations are then performed on this initial state to bring the system to equilibrium under given conditions, until the system reaches morphological stability or a preset statistical measure converges, yielding the final morphology for each parameter point. The preset statistical measure can be energy, structure factor, etc., and this embodiment does not impose any limitations on this.
[0051] After obtaining the final state morphology corresponding to each parameter point, the final state morphology corresponding to each parameter point can be post-processed to obtain the first final state morphology field corresponding to each parameter point. In this embodiment, for quantitative description and analysis, the spatial region involved in the entire simulation calculation can be discretized into a series of uniformly arranged grid cells of the same size, and the first final state morphology field on the regular spatial grid can be calculated. The final state morphology field can be a volume fraction field or an equivalent phase field representation, and this embodiment is not limited thereto. In this embodiment, the A / B block volume fraction field on the regular spatial grid can be calculated. , Here, x, y, and z represent the coordinates of three mutually perpendicular directions in three-dimensional space.
[0052] In one possible implementation, the method further includes: for each parameter point, concatenating the polymer parameters, template geometry parameters, and annealing process parameters into a one-dimensional feature vector in a preset order; and combining multiple one-dimensional feature vectors into an input feature vector.
[0053] Specifically, polymer parameters, template geometry parameters, and annealing process parameters can be concatenated into a one-dimensional input feature vector in a preset order. , , Then, they are combined into a unified input feature vector. The preset order can be set according to the needs of those skilled in the art, and the embodiments disclosed herein do not impose any restrictions on it.
[0054] In this embodiment of the disclosure, each parameter dimension can be normalized, for example by subtracting the mean and dividing by the standard deviation, or by linearly mapping to... To improve the stability of model training, the output final state topography field is normalized and its size is standardized. For example, the density fields of different systems are interpolated or resampled onto a three-dimensional grid of uniform size (such as 64×64×64 or 128×128×128), and the volume fraction is mapped to a fixed numerical range.
[0055] S14. Based on the parameter points and the first final state morphology field, train the preset model;
[0056] The preset model can be a machine learning model, a deep learning model, etc. Specifically, it can be a network structure capable of mapping low-dimensional inputs to high-dimensional outputs. For example, it can be a Conditional U-Net, a Conditional Generative Adversarial Network (cGAN), or an encoder-decoder structure with skip connections, etc., and this disclosure does not limit the scope of the embodiments.
[0057] In one possible implementation, training the preset model based on the parameter points and the first final-state topography field includes: dividing multiple sample data into a training set, a validation set, and a test set, where each sample data includes the input feature vector and the first final-state topography field corresponding to the parameter points; training the preset model based on the training set using the input feature vector as input features and the first final-state topography field as output labels. After obtaining the first final-state topography field, it can be stored as a three-dimensional tensor or a two-dimensional image of several cross-sections, serving as the output labels for training the preset model.
[0058] Specifically, the input feature vector obtained above can be... The corresponding first final state morphology field is used as sample data, and the sample data is divided into training set, validation set and test set. The specific division ratio can be determined according to the actual needs of those skilled in the art, and this disclosure does not limit it.
[0059] The input feature vector can be used as the input feature, and the first final state topography field can be used as the output label. The training set can be input into the network structure described above for training. In this embodiment of the disclosure, the input feature vector can also be... First, the input is fed into the embedding layer or at least one fully connected layer of the model, where features are extracted and dimensionality is transformed to convert it into a high-dimensional conditional feature vector. Then, this high-dimensional conditional feature vector is fused with the intermediate layer feature map of the network encoder to form a fused vector. The fused features can be input into the decoder to ultimately achieve self-assembled morphology prediction.
[0060] In one possible implementation, the method further includes: constructing a composite loss function, the composite loss function including a numerical error term and a physical constraint term; and optimizing the composite loss function based on the validation set.
[0061] In this embodiment of the disclosure, a composite loss function can be constructed to evaluate the model during training. Specifically, a composite loss function containing a numerical error term and a physical constraint term can be constructed. For example, it can include reconstruction error, structural similarity, and physical constraint loss. Among them, the reconstruction error can be the mean squared error (MSE) or L1 loss between the predicted morphology and the MD-calibrated morphology; the structural similarity can be measured by the structural similarity index (SSIM) to measure the overall morphological structural similarity; the physical constraint loss can be constraining global or local volume fraction conservation, penalizing obviously unphysical high-frequency noise or discontinuous interfaces, etc. The comprehensive loss can be expressed as formula (1):
[0062] (1)
[0063] in, Mean square error, For structural similarity error, For physical constraint loss, These are the weighting coefficients.
[0064] In this embodiment, stochastic gradient descent or adaptive optimization algorithms can be used to train the network parameters. These algorithms iteratively update the network parameters based on the loss value, gradually reducing the loss as the training rounds increase. After each training round, the model's prediction accuracy can be evaluated using a validation set, and the weight coefficients in the loss function can be empirically tuned based on the evaluation results. This ensures that the contributions of the numerical error term, structural similarity term, and physical constraint term to the overall loss are on the same order of magnitude, balancing prediction accuracy, morphological fidelity, and physical plausibility. For example, one could adopt... wait.
[0065] In one possible implementation, an adversarial training mechanism (such as a GAN discriminator) or an attention module can be introduced into the network. The adversarial training mechanism trains the discriminator network and the generator network in a game-like manner, making the generated morphology more similar to the real morphology in terms of structure and detail. The attention module is used to identify the importance of different spatial regions in the morphology image, so that the network focuses on key regions during processing, thereby improving the prediction accuracy of key locations.
[0066] S15. Using the trained model, predict the target morphology control parameters to obtain the target final morphology field.
[0067] In this embodiment of the disclosure, by inputting the target morphology control parameters into the trained model, the corresponding target final state morphology field can be quickly output. .
[0068] In this embodiment of the disclosure, the prediction results of the model can be verified. Specifically, the predicted morphology can be automatically post-processed to calculate key indicators such as domain spacing, orientation order parameters, and defect density. The extracted indicators are then compared with the corresponding indicators obtained from independent MD simulations to obtain comparison results, ensuring that the model predictions are statistically consistent with the real system.
[0069] In one possible implementation, the method further includes: selecting multiple parameter points for supplementary molecular dynamics simulation based on the prediction error or uncertainty estimate of the model at each parameter point in the validation set, adding the simulated data to the training set, and retraining the model.
[0070] If the comparison results are outside the preset deviation range, an active learning strategy can be introduced. Specifically, based on the current model's prediction error and uncertainty estimate for each parameter point in the parameter space, the information gain that point can bring can be quantified by calculating the prediction variance or entropy value of each point. A batch of parameter points with the highest uncertainty or the most unstable prediction results are automatically selected for additional MD simulations. The simulated data is added to the training set, the model is retrained, and the current model is re-evaluated. If the evaluation results still do not meet the requirements, the parameter points with the largest information gain are identified, new MD simulation data is added, the training set is updated again, and training is iterated, thereby continuously expanding the reliable prediction region within the parameter space.
[0071] In this embodiment of the disclosure, by introducing an active learning strategy, high model prediction accuracy can be obtained with less MD simulation computation. That is, targeted additional simulation is performed only in parameter regions with high model prediction uncertainty, thereby significantly reducing the total computational cost required to obtain training data while ensuring prediction accuracy.
[0072] In this embodiment of the disclosure, by outputting a density field / phase field image that can be represented on a regular grid, it is convenient to directly analyze defect distribution, orientation sequence, domain size, etc.
[0073] Application scenario examples
[0074] The core technical solution of this disclosure is:
[0075] First, molecular dynamics simulations are used to generate a large amount of calibration data of "input conditions - final morphology" in a representative parameter space. Then, a conditional generative model that integrates physical constraints (such as conditional U-Net, conditional generative adversarial network, etc.) is constructed to learn the mapping from polymer parameters, template geometry, annealing conditions to the final morphology field. In practical use, only the input parameters need to be given, and the trained model can quickly predict the morphology field, thereby replacing a complete MD simulation in a few seconds.
[0076] In a preferred embodiment, the method includes the following steps:
[0077] Step 1: Construction of DSA molecular dynamics calibration data for block copolymers
[0078] Select typical block copolymer systems and DSA scenarios, such as the directional self-assembly of A–B diblock copolymers (e.g., PS-b-PMMA) in groove templates or chemical pattern templates; establish corresponding molecular models and interaction potentials through coarse-grained MD or all-atom MD.
[0079] Adjustable polymer parameters can be set, including the volume fraction of each block. Total aggregation degree Flory-Hutchins interaction parameters wait.
[0080] Set template geometry parameters: including groove width ,depth ,cycle The affinity of the sidewalls and bottom with different blocks, etc., is encoded using a set of geometric and surface interaction parameter vectors.
[0081] Set annealing process parameters: including annealing temperature Annealing time Cooling rate And its time-varying programs, etc.
[0082] By employing a reasonable sampling strategy (such as Latin hypercube sampling), a large number of parameter points are selected in the above parameter space. For each parameter point, MD simulation is performed starting from a random initial configuration until the system reaches morphological stability or the statistics (such as energy and structure factor) converge.
[0083] The final molecular configuration corresponding to each parameter point is post-processed to calculate the A / B block volume fraction field on a regular spatial grid. , The output labels for model training are represented by a phase field, or an equivalent phase field, and stored in the form of a three-dimensional tensor or a two-dimensional image of several cross sections.
[0084] Step 2: Encoding and Normalization of Input and Output Features
[0085] The polymer parameters, template geometry parameters, and annealing process parameters are concatenated into a one-dimensional input feature vector in a fixed order. , , Then, they are combined into a unified input feature vector. .
[0086] Normalize each parameter dimension, for example, by subtracting the mean and dividing by the standard deviation, or by linearly mapping it to the [0,1] interval, to improve the stability of model training.
[0087] The output final state morphology field is normalized and its size is unified, such as interpolating or resampling the density fields of different systems onto a three-dimensional grid of uniform size (e.g., 64×64×64 or 128×128×128), and mapping the volume fraction to a fixed numerical range.
[0088] Step 3: Building and Training a Physically Constrained Machine Learning Model
[0089] Select a suitable network structure that combines "low-dimensional conditional input + high-dimensional image output", such as Conditional U-Net, Conditional Generative Adversarial Network (cGAN), or encoder-decoder structure with skip connections. Transform the input parameter vector X into high-dimensional conditional features through embedding layers or multilayer perceptrons, and fuse it with the features of the intermediate layers of the network.
[0090] The network output generates a final-state topography field prediction of the same size as the calibration data. Or equivalent phase field representation.
[0091] Construct a composite loss function that includes numerical error terms and physical constraint terms. For example: reconstruction error: mean square error (MSE) or L1 loss between the predicted topography and the MD-calibrated topography; structural similarity: the structural similarity index (SSIM) is used to measure the overall topography structural similarity; physical constraint loss: constraining global or local volume fraction conservation, penalizing obviously non-physical high-frequency noise or discontinuous interfaces, etc. The comprehensive loss can be seen in formula (1).
[0092] In a preferred embodiment, multiple sets of experiments on the validation set are performed... Empirical optimization is performed to ensure that the contributions of numerical error, structural similarity, and physical constraints to the overall loss are of the same order of magnitude, thus balancing prediction accuracy, fidelity to morphology and structure, and physical plausibility. For example, one could adopt... Of course, the present invention is not limited to the specific values mentioned above, and the relevant weighting coefficients can be adjusted equivalently under the same technical concept.
[0093] The network parameters are trained using stochastic gradient descent or adaptive optimization algorithms. The calibration dataset is divided into training, validation and test sets, and overfitting is avoided by means of early stopping and regularization.
[0094] Adversarial training (GAN discriminator) or attention mechanisms can be introduced as needed to further improve the reconstruction quality of local details and long-range ordered features.
[0095] Step 4: Model Inference and Prediction Result Verification
[0096] During the inference phase, only the parameter vector X under given process conditions needs to be input, and the corresponding predicted topography field can be quickly output.
[0097] The predicted morphology is automatically post-processed to calculate key indicators such as domain spacing, orientation order parameters, and defect density. These are then compared with a small number of independent MD verification simulations or experimental results to ensure that the model predictions are statistically consistent with the real system.
[0098] For operating conditions with significant differences, supplementary MD simulations can be triggered to add new data to the training set and iteratively update the model, thereby continuously expanding the reliable prediction region within the parameter space.
[0099] Step 5: Optional Active Learning and Iterative Model Optimization
[0100] To further reduce the cost of generating calibration data, this invention optionally introduces an active learning strategy:
[0101] Based on the prediction uncertainty or error estimation of the current model in different regions of the parameter space, a small number of parameter points with "maximum information gain" are automatically selected for additional MD simulation.
[0102] Add the newly obtained calibration samples to the training set, retrain or fine-tune the model, and improve the prediction accuracy in the key parameter region.
[0103] By repeating the above process, a closed-loop optimization of "MD simulation - ML model training - active data supplementation" is achieved, which can obtain high-quality global prediction capabilities with fewer high-cost MD samples.
[0104] Description of Optional and Equivalent Alternatives
[0105] Block copolymer systems can be extended to any diblock or multiblock copolymers, such as PS-b-PMMA, PE-b-PEO, PS-b-PDMS, triblock A–B–C, etc., only requiring corresponding adjustments to the molecular dynamics model and parameter encoding.
[0106] Molecular dynamics types can include all-atom MD, coarse-grained MD, dissipative particle dynamics (DPD), or other equivalent simulation methods that accurately reflect polymer dynamics.
[0107] The machine learning model structure can be replaced with three-dimensional convolutional neural networks (3D-CNN), Transformer, diffusion models, etc., as long as it can achieve conditional generation from low-dimensional parameters to high-dimensional topographic fields.
[0108] The specific form of the physical constraint term can be adjusted according to the research object. For example, strengthening the conservation of volume fraction for incompressible systems and introducing free energy approximations for specific phase behaviors are all equivalent technical solutions of this invention.
[0109] In this embodiment, compared with existing machine learning prediction methods that rely on long-term molecular dynamics simulations or equilibrium results based on self-consistent field theory (SCFT), this method significantly reduces computational costs while maintaining near-molecular dynamics simulation accuracy. It compresses the prediction process of block copolymer oriented self-assembly systems from given polymer and process conditions to the final morphology from hours or even days to seconds, theoretically achieving an acceleration of approximately 10² to 10³ times. This enables large-scale scanning of multidimensional parameter spaces and optimization of process windows. By using the final morphology obtained from complete molecular dynamics evolution as training labels, the model learns the typical final states reached by the system under given kinetic rules and annealing paths. This naturally includes key kinetic characteristics in actual processes such as path dependence, metastable state retention, and defect residue, making the prediction results closer to real process behavior than models based on pure equilibrium states. This invention directly outputs the final state density field or phase field distribution on a regular spatial grid, facilitating subsequent quantitative calculation of domain spacing, orientation order parameters, and defect density. This method provides key physical quantities and can serve as input for multi-scale simulations or nanographic design tools. Compared to traditional methods that only provide phase labels or a few structural parameters, it offers richer information and broader applications. Furthermore, by introducing physical constraints such as volume fraction conservation and interface smoothing into the loss function and network structure, it effectively suppresses non-physical interpretations that may arise from purely data-driven models, such as overall component non-conservation or severe interface fragmentation, significantly improving the physical reliability of predicted morphology in both structural and statistical terms. In addition, the overall framework of this method has good versatility and scalability. By changing the molecular model, adjusting the parameter encoding method and network structure, it can be extended to different block copolymer systems and various directional self-assembly template scenarios. It can also be combined with strategies such as active learning to continuously iterate and optimize the model with the support of limited newly added molecular dynamics calibration data, maintaining high prediction accuracy under a wider range of material systems and process conditions. Therefore, it provides an efficient, reliable, and continuously evolving virtual experimentation and design platform for semiconductor nanographic design, functional polymer material development, and related process optimization.
[0110] In this embodiment of the disclosure, a vector containing polymer parameters, template geometry parameters, and annealing process parameters is used as input, and the final state density field / phase field obtained by molecular dynamics evolution is used as output label. The above mapping is learned through a machine learning model to achieve rapid morphology prediction that can replace large-scale MD simulation.
[0111] In this embodiment, multi-source information such as polymer composition and chain length, template geometry and surface affinity, and annealing path parameters are uniformly encoded into a continuous vector input; the output directly provides the three-dimensional or two-dimensional density field / phase field distribution on a regular grid, rather than only providing phase labels or a few scalars.
[0112] In this embodiment of the disclosure, a composite loss function is introduced during the training process, which simultaneously includes reconstruction error (such as MSE), structural similarity (such as SSIM), and physical constraints such as volume fraction conservation and interface energy regularization, and / or a corresponding regularization mechanism is introduced into the network structure to ensure the consistency between the predicted morphology in terms of both numerical and physical dimensions.
[0113] In this embodiment, a conditional U-Net, a conditional generative adversarial network, or other encoder-decoder structure with skip connections is used to enable low-dimensional input parameters to effectively modulate the high-dimensional output topography field, thereby achieving simultaneous reconstruction of long-range ordered structures and local details.
[0114] In this embodiment of the disclosure, by evaluating the uncertainty or error index of the model prediction, the point with the most information in the parameter space is adaptively selected for a new round of MD simulation calibration, and the model is iteratively updated, thereby maximizing the improvement of global prediction performance under limited computing resources.
[0115] It should be noted that the embodiments disclosed herein include, but are not limited to, a variety of different block copolymer systems (such as PS-b-PMMA, PS-b-PDMS, PEO-b-PS, etc.), a variety of DSA template types (grooves, chemical patterns, nanopore arrays, etc.), a variety of molecular simulation types (all-atom MD, coarse-grained MD, DPD, etc.), and a variety of alternative neural network structures (3D-CNN, Transformer, diffusion model, etc.).
[0116] It should be noted that the block copolymer system in the embodiments of this disclosure can be extended to any diblock or multiblock copolymer, such as PS-b-PMMA, PE-b-PEO, PS-b-PDMS, triblock A–B–C, etc., only requiring corresponding adjustments to the molecular dynamics model and parameter encoding. The molecular dynamics type can be all-atom MD, coarse-grained MD, dissipative particle dynamics (DPD), or other equivalent simulation methods that accurately reflect polymer dynamics. The machine learning model structure can be replaced with a three-dimensional convolutional neural network (3D-CNN), Transformer, diffusion model, etc., as long as it can achieve conditional generation from low-dimensional parameters to high-dimensional morphology fields. The specific form of the physical constraint terms can be adjusted according to the research object, such as strengthening the volume fraction conservation for incompressible systems, introducing free energy approximations for specific phase behaviors, etc., all of which are equivalent technical solutions of this invention.
[0117] In this embodiment, the cost of large-scale MD simulation can be reduced. While maintaining near-MD simulation accuracy and kinetic physical meaning, the prediction time for the final morphology of block copolymer DSA is reduced from the order of hours / days to the order of seconds, thereby supporting high-throughput scanning of the parameter space and process optimization.
[0118] In this embodiment, the problem that equilibrium models cannot reflect kinetic paths can be overcome. By using the final state morphology obtained from MD kinetic evolution as a training label, the machine learning model learns "the typical final state that the system evolves to over a long period of time under given kinetic rules and process paths," rather than simply the thermodynamic equilibrium phase.
[0119] In the embodiments of this disclosure, end-to-end prediction from "input parameters → complete topographic field" can be achieved. It can not only obtain phase class or a small number of structural parameters, but also output density field / phase field images that can be represented on a regular mesh, facilitating direct analysis of defect distribution, orientation sequence, domain size, etc.
[0120] In this embodiment, key physical constraints can be incorporated into the machine learning process. By introducing physical regularization terms such as volume fraction conservation into the loss function and network structure, the physical rationality of the prediction results in both structural and statistical terms is improved.
[0121] Figure 2 A block diagram of a device for predicting the final morphology of directional self-assembly of block copolymers according to an embodiment of the present disclosure is shown. Figure 2 As shown, the device includes:
[0122] Module 21: Used to establish corresponding molecular models and interaction potentials based on preset block copolymer systems and directional self-assembly scenarios;
[0123] Value selection module 22: used to select multiple values for multiple preset morphology control parameters to form multiple parameter points;
[0124] Simulation module 23: Based on the molecular model and interaction potential, it performs molecular dynamics simulation on each parameter point with a random initial configuration as the initial state until the block copolymer system reaches morphological stability or the preset statistical quantity converges, and obtains the first final state morphological field corresponding to each parameter point.
[0125] Training module 24: used to train a preset model based on the parameter points and the first final state topography field;
[0126] Prediction module 25: Used to predict the target morphology control parameters using the trained model to obtain the target final morphology field.
[0127] This disclosure also provides a non-volatile computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.
[0128] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.
[0129] Figure 3 A block diagram of an apparatus according to an embodiment of the present disclosure is shown. The apparatus 1900 may be provided as a server or a terminal device. (Refer to...) Figure 3 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.
[0130] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output interface 1958 (I / O interface). Device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM macOS X TM Unix TM Linux TM FreeBSD TM Or similar.
[0131] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.
[0132] Computer-readable storage media can be tangible devices capable of holding and storing programs / instructions used by instruction execution devices. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.
[0133] The computer program (or computer-readable program instructions) described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage medium in the respective computing / processing device.
[0134] The computer program (or computer program instructions) used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages, such as Python. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing state information of computer-readable program instructions. These electronic circuits can execute computer-readable program instructions to implement various aspects of this disclosure.
[0135] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.
[0136] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.
[0137] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.
[0138] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0139] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.
Claims
1. A method for predicting the final morphology of directional self-assembly of block copolymers, characterized in that, include: Based on the pre-defined block copolymer system and directional self-assembly scenario, the corresponding molecular model and interaction potential are established. For multiple preset morphology control parameters, multiple values are selected to form multiple parameter points; Based on the molecular model and interaction potential, molecular dynamics simulation is performed on each parameter point with a random initial configuration as the initial state until the block copolymer system reaches morphological stability or the preset statistical quantity converges, and the first final state morphological field corresponding to each parameter point is obtained. Based on the parameter points and the first final state topography field, the preset model is trained; The trained model is used to predict the target morphology control parameters, and the final morphology field of the target is obtained.
2. The method according to claim 1, characterized in that, The preset morphology control parameters include at least one of polymer parameters, template geometry parameters, and annealing process parameters.
3. The method according to claim 1, characterized in that, For multiple preset morphology control parameters, multiple values are selected to form multiple parameter points, including: Using a preset sampling method, multiple value nodes are selected within the value range of each preset morphology control parameter, and the value nodes of each preset morphology control parameter are combined to form multiple parameter points.
4. The method according to claim 2, characterized in that, The method further includes: For each parameter point, the polymer parameters, template geometry parameters, and annealing process parameters are concatenated into a one-dimensional feature vector in a preset order. Multiple one-dimensional feature vectors are combined into an input feature vector.
5. The method according to claim 4, characterized in that, The step of training a preset model based on the parameter points and the first final state topography field includes: Multiple sample data are divided into training set, validation set and test set, and each sample data includes the input feature vector and the first final state morphology field corresponding to the parameter point; Using the input feature vector as input features and the first final state morphology field as output label, a preset model is trained based on the training set.
6. The method according to claim 5, characterized in that, The method further includes: Construct a composite loss function, which includes a numerical error term and a physical constraint term; Based on the validation set, the composite loss function is optimized.
7. The method according to claim 5, characterized in that, The method further includes: Based on the prediction error or uncertainty estimate of the model at each parameter point in the validation set, multiple parameter points are selected for supplementary molecular dynamics simulations. The simulated data is then added to the training set, and the model is retrained.
8. A device for predicting the final morphology of block copolymer directional self-assembly, characterized in that, include: Builder module: Used to establish corresponding molecular models and interaction potentials based on preset block copolymer systems and directional self-assembly scenarios; Value selection module: used to select multiple values for multiple preset morphology control parameters to form multiple parameter points; Simulation module: Based on the molecular model and interaction potential, it performs molecular dynamics simulation for each parameter point with a random initial configuration as the initial state until the block copolymer system reaches morphological stability or the preset statistical quantity converges, and obtains the first final state morphological field corresponding to each parameter point. Training module: used to train a preset model based on the parameter points and the first final state topography field; Prediction module: Used to predict the target morphology control parameters using the trained model, and obtain the final morphology field of the target.
9. A non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer program product comprising a computer program, or a non-volatile computer-readable storage medium carrying a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.
