Preparation method and system for optimizing stability of photoresist and photoresist

Abstract

The invention discloses a preparation method and system for optimizing photoresist stability and photoresist, and relates to the technical field of photoresist, the method comprises the following steps: taking identity characteristics of a target production line as constraints, collecting a photoresist preparation log, the log comprising a raw material ratio, a reaction condition and a stability index; constructing a stability evaluation model based on the log, wherein the model is composed of cascaded weak learners corresponding to a plurality of preparation links; in combination with the evaluation model and a global optimization algorithm, iteratively searching preparation parameters, taking a model prediction result as an objective function output, and taking the preparation parameters as an input; and outputting a parameter meeting the stability threshold as a standard preparation parameter, and performing photoresist preparation according to the standard preparation parameter. Therefore, the technical effects of improving the optimization efficiency, reducing the optimization cost and improving the optimization precision are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of photoresists, and particularly to a preparation method, a system and a photoresist for optimizing the stability of a photoresist. Background Art

[0002] With the development of the integrated circuit manufacturing process towards smaller line widths and higher integration levels, higher requirements are put forward for the performance of photoresists, especially their stability, which is directly related to the pattern transfer accuracy, process window width and production yield. The stability of photoresists is affected by various factors, including preparation parameters such as raw material ratio, reaction temperature, stirring speed, reaction time, etc., and there are complex non-linear coupling relationships between the parameters.

[0003] Currently, the preparation process of photoresists mostly relies on experience accumulation and manual parameter adjustment for optimization. In actual production, engineers usually adjust the preparation parameters through methods such as trial-and-error experiments and single-factor analysis in order to obtain better stability performance. There are still deficiencies in the following aspects: traditional methods mostly adopt local adjustment, and it is difficult to achieve global optimal search in the multi-parameter space; some models fail to fully combine the characteristics of the production line, resulting in poor migration between different production lines and low prediction accuracy; some methods fail to clearly convert the model prediction results into executable preparation parameters, making it difficult to guide actual production. Summary of the Invention

[0004] The present invention provides a preparation method, a system and a photoresist for optimizing the stability of a photoresist, so as to solve the technical problems of low optimization efficiency, high cost and insufficient optimization accuracy in the prior art, and achieve the technical effects of improving the optimization efficiency, reducing the optimization cost and improving the optimization accuracy.

[0005] In a first aspect, the present invention provides a preparation method for optimizing the stability of a photoresist, wherein the method includes: Taking the production line identity characteristics of the target production line as a constraint, collecting the preparation logs of the photoresist, wherein the preparation logs include the raw material ratio, reaction condition parameters and corresponding stability index data.

[0006] Based on the preparation logs, constructing a stability evaluation model, wherein the stability evaluation model includes multiple sets of cascaded weak learners, and the multiple sets of weak learners correspond to multiple preparation links of the target production line.

[0007] Combining the stability evaluation model with a global optimization algorithm, performing iterative search on the preparation parameters, wherein the prediction result of the stability evaluation model is used as the output of the objective function, and the input of the objective function is the preparation parameters.

[0008] Output the preparation parameters whose prediction results meet the preset stability threshold as standard preparation parameters, and prepare the photoresist according to the standard preparation parameters.

[0009] In a feasible implementation, constrained by the production line identity characteristics of the target production line, collect the preparation logs of the photoresist, including: Obtain the production line identification information of the target production line, and determine its corresponding process configuration characteristics and production line label characteristics.

[0010] Based on the production line label characteristics, extract the historical records during the photoresist preparation process from the target production line, and obtain a first data set including raw material ratios, reaction condition parameters, and stability index data.

[0011] When the data volume of the first data set is lower than the preset sample threshold, identify homologous production lines with associated labels according to the production line label characteristics, and extract the corresponding preparation logs from the homologous production lines to construct a second data set.

[0012] Merge the first data set and the second data set as the data basis for modeling, and construct a preparation log that meets the training requirements.

[0013] In a feasible implementation, based on the preparation log, construct a stability evaluation model, where the stability evaluation model includes multiple cascaded weak learners, and multiple groups of weak learners correspond to multiple preparation links of the target production line, including: Perform missing value filling, outlier removal, numerical normalization, and categorical variable encoding on the collected preparation log data to obtain a modeling data set for training.

[0014] According to the key link information of the photoresist preparation process flow, segment the modeling data set by process stages to obtain multiple sub-data sets.

[0015] Based on multiple sub-data sets, train multiple groups of weak learners respectively, and combine the ensemble learning method to cascade each group of weak learners in the order of the preparation process to form the stability evaluation model.

[0016] In a feasible implementation, before training multiple groups of weak learners based on multiple sub-data sets respectively, include: Evaluate the co-variation coefficient between multiple process stages and the stability index, and define a first weight coefficient set according to the co-variation coefficient.

[0017] Based on the sequence of multiple process stages in the photoresist preparation process flow, define a second weight coefficient set, where the second weight coefficient decreases along the direction of the photoresist preparation process flow.

[0018] Modify the first weight coefficient set according to the second weight coefficient set, and normalize the modified first weight coefficient set to obtain a magnitude weight coefficient set.

[0019] In a feasible implementation, multiple weak learners are trained based on multiple sub-datasets respectively. Combining with the ensemble learning method, the weak learners in each group are cascaded in the order of the preparation process to form the stability evaluation model, including: Obtain the evaluation computing power index of the target scenario, allocate the evaluation computing power index according to the magnitude weight coefficient set, and define the group capacity constraints of multiple groups of weak learners.

[0020] Using the group capacity constraint as the filling upper limit, add weak learners to multiple groups of weak learners.

[0021] According to the feature dimensions and data sample sizes included in each sub-dataset, use an adaptive sampling strategy to enhance or crop the data, and use cross-validation and early stopping mechanisms to train each group of weak learners until the stability evaluation index convergence requirement is met on the validation set.

[0022] Integrate the trained multiple groups of weak learners, and dynamically fine-tune the weights of the weak learners according to the training error distribution to form the stability evaluation model.

[0023] In a feasible implementation, combine the stability evaluation model with a global optimization algorithm to iteratively search for the preparation parameters. Among them, the prediction result of the stability evaluation model is used as the output of the objective function, and the input of the objective function is the preparation parameter, including: Obtain the preparation parameters in real time and input the preparation parameters into the stability evaluation model.

[0024] The first group of weak learners of the stability evaluation model takes the preparation parameter as the input, evaluates and outputs the first stability index.

[0025] Input the first stability index and the preparation parameter into the second group of weak learners to obtain the second stability index.

[0026] Input the first stability index, the second stability index and the preparation parameter into the third group of weak learners to obtain the third stability index, and so on, until the N groups of weak learners of the stability evaluation model are traversed, and the Nth stability index is obtained as the prediction result.

[0027] In a feasible implementation, before preparing the photoresist according to the standard preparation parameters, it further includes: Conduct a photoresist preparation experiment according to the standard preparation parameters.

[0028] If the lithography resist preparation test results of consecutive first preset test batches meet the stability requirements, lithography resist preparation is carried out based on the standard preparation parameters.

[0029] If the lithography resist preparation test results of consecutive second preset test batches do not meet the stability requirements, the stability evaluation model and the optimization algorithm are dynamically corrected based on the experimental results.

[0030] In a second aspect, the present invention also provides a preparation system for optimizing the stability of a lithography resist. Among them, the system includes: A log collection module, which is used to collect the preparation logs of the lithography resist with the production line identity characteristics of the target production line as a constraint. Among them, the preparation logs include raw material ratios, reaction condition parameters, and corresponding stability index data.

[0031] An evaluation model construction module, which is used to construct a stability evaluation model based on the preparation logs. Among them, the stability evaluation model includes multiple cascaded weak learners, and multiple groups of weak learners correspond to multiple preparation links of the target production line.

[0032] An optimization search module, which is used to iteratively search for preparation parameters by combining the stability evaluation model with a global optimization algorithm. Among them, the prediction result of the stability evaluation model is used as the output of the objective function, and the input of the objective function is the preparation parameters.

[0033] A parameter output module, which is used to output the preparation parameters whose prediction results meet the preset stability threshold as standard preparation parameters, and carry out lithography resist preparation according to the standard preparation parameters.

[0034] In a third aspect, the present invention also provides a lithography resist, which is obtained by any one of the implementation methods of the above-mentioned preparation method for optimizing the stability of a lithography resist.

[0035] The present invention discloses a preparation method, system and lithography resist for optimizing the stability of a lithography resist, including: collecting the preparation logs of the lithography resist with the production line identity characteristics of the target production line as a constraint, and the preparation logs cover raw material ratios, reaction condition parameters, and corresponding stability index data; constructing a stability evaluation model based on the preparation logs, and the model consists of multiple cascaded weak learners, and each weak learner corresponds to multiple preparation links of the target production line; iteratively searching for preparation parameters by combining the stability evaluation model with a global optimization algorithm, using the prediction result of the stability evaluation model as the output of the objective function and the preparation parameters as the input; finally, determining the preparation parameters whose prediction results meet the preset stability threshold as standard preparation parameters, and carrying out lithography resist preparation accordingly. The present invention solves the technical problems of low optimization efficiency, high cost and insufficient optimization accuracy, and realizes the technical effects of improving optimization efficiency, reducing optimization cost and improving optimization accuracy. Brief Description of the Drawings

[0036] Figure 1 It is a schematic flow chart of a preparation method for optimizing the stability of photoresist according to the present invention.

[0037] Figure 2 It is a schematic structural diagram of a preparation system for optimizing the stability of photoresist according to the present invention.

[0038] Description of reference numerals: Log collection module 11, evaluation model construction module 12, optimization search module 13, parameter output module 14. Detailed Embodiments

[0039] The above technical solutions will be described in detail below in conjunction with the drawings of the specification and specific embodiments to better understand the above technical solutions. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments for explaining the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention. In addition, it should be noted that only the parts related to the present invention rather than all are shown in the drawings for the convenience of description.

[0040] Example 1, as Figure 1 It is a schematic flow chart of a preparation method for optimizing the stability of photoresist according to the present invention, wherein the method includes: Collect the preparation logs of the photoresist with the production line identity characteristics of the target production line as a constraint, wherein the preparation logs include raw material ratios, reaction condition parameters, and corresponding stability index data.

[0041] Specifically, the preparation log refers to a historical data set recording all relevant information in the photoresist preparation process, which at least includes the raw material ratio (the dosage ratio of each component), the reaction condition parameters (such as temperature, pressure, stirring intensity, etc.), and the corresponding stability index data (such as the storage stability of the photoresist).

[0042] Specifically, by collecting the preparation logs with the production line identity characteristics of the target production line as a constraint, the accuracy and consistency of the data can be ensured, and the model deviation caused by inconsistent data sources can be avoided.

[0043] In some embodiments, collecting the preparation logs of the photoresist with the production line identity characteristics of the target production line as a constraint includes: Obtain the production line identification information of the target production line, and determine its corresponding process configuration characteristics and production line label characteristics; based on the production line label characteristics, extract the historical records in the photoresist preparation process from the target production line, and obtain a first data set including raw material ratios, reaction condition parameters, and stability index data; when the data volume of the first data set is lower than the preset sample threshold, identify homologous production lines with associated labels according to the production line label characteristics, and extract the corresponding preparation logs from the homologous production lines to construct a second data set; merge the first data set and the second data set as the data basis for modeling, and construct a preparation log that meets the training requirements.

[0044] Specifically, the production line identity characteristics are the unique identification information of the target production line, including process configuration characteristics (such as equipment type, process parameter range, etc.) and production line label characteristics (such as the process type to which the production line belongs, equipment model, etc.). The sample threshold is the preset minimum number of samples to ensure that the data used for modeling has statistical representativeness and avoid insufficient samples. Homologous production lines refer to other production lines with similar process configurations or label characteristics to the target production line, which are used to supplement data when the data of the target production line itself is insufficient.

[0045] Specifically, first, obtain the unique identifier (such as ID number, area number) of the target production line through database query or data interface, and determine its corresponding process configuration characteristics (such as the equipment type is "lithography machine A type") and production line label characteristics; then, based on the production line label characteristics, extract relevant data in the photoresist preparation process from the historical database of the target production line to form a first data set. For example, extract all photoresist preparation records in the past year, including raw material ratios: such as solvent A: reactant B: additive C = 45:40:15; reaction condition parameters: such as temperature 85°C, stirring 300 rpm, reaction time 6 h; stability indicators: such as sedimentation rate, particle size change percentage, photosensitive decay rate, etc. after 30 days.

[0046] Furthermore, if the number of the first data set is less than the set threshold (for example, less than 50 records), the homologous production line identification mechanism is activated. Through the label matching algorithm, find other production lines with a process label matching degree higher than the set value (such as 90%), extract the corresponding photoresist preparation logs from these production lines to construct a second data set, and clean and standardize the first data set and the second data set and then merge them to finally form a preparation log that meets the modeling requirements.

[0047] The above method steps can improve the integrity and diversity of the data by supplementing the data of homologous production lines, thereby enhancing the generalization ability and prediction accuracy of the model. In addition, this data collection method can effectively combine production line characteristics, provide high-quality data support for subsequent model construction and parameter optimization, and ultimately improve the stability and efficiency of photoresist preparation.

[0048] Based on the preparation log, a stability evaluation model is constructed, wherein the stability evaluation model includes multiple cascaded weak learners, and the multiple weak learners correspond to multiple preparation links of the target production line.

[0049] Specifically, the stability evaluation model is an analysis model based on machine learning, which is used to predict and evaluate the stability performance of photoresist under different preparation parameters. The stability evaluation model includes multiple cascaded weak learners, and each group of weak learners focuses on evaluating a specific link in the photoresist preparation process; by cascading multiple groups of weak learners (such as decision trees, neural networks, etc.) in sequence, the influencing factors of the upstream link on the downstream link can be considered, thus helping to improve the evaluation accuracy of the model.

[0050] In some embodiments, based on the preparation log, a stability evaluation model is constructed, wherein the stability evaluation model includes multiple cascaded weak learners, and the multiple weak learners correspond to multiple preparation links of the target production line, including: Perform missing value filling, outlier removal, numerical normalization and categorical variable encoding on the collected preparation log data to obtain a modeling data set for training; according to the key link information of the photoresist preparation process flow, segment the modeling data set by process stage to obtain multiple sub-data sets; based on the multiple sub-data sets, train multiple groups of weak learners respectively, and combine the ensemble learning method to cascade each group of weak learners in the order of the preparation process to form the stability evaluation model.

[0051] Specifically, first, perform missing value filling, outlier removal, numerical normalization and categorical variable encoding on the collected preparation log data to ensure data quality. For example, convert temperature data from Celsius to Kelvin values, encode equipment types as digital labels, etc.; then, based on the typical preparation process of photoresist (for example: raw material premixing → temperature-controlled polymerization → post-treatment → detection), split the modeling data set into several sub-data sets, and each sub-data set corresponds to a process stage for segmentation. For example, divide the data into raw material mixing stage, reaction stage, stirring stage, etc.

[0052] Specifically, use each sub-data set as a class of training data to train multiple groups of weak learners respectively, wherein each group of weak learners focuses on evaluating the stability of a specific preparation link. For example, the first group of weak learners evaluates the stability of the raw material mixing stage, and the second group evaluates the stability of the reaction stage. Preferably, the multiple weak learners included in each group of weak learners are based on multiple machine learning models with different structures and parameters.

[0053] Furthermore, cascade the weak learners of each group based on the process flow sequence, that is, the output of the previous group serves as part of the input of the next group's model, thereby realizing information conduction and forming a complete stability evaluation model. Exemplarily, connect the weak learners of each group corresponding to links such as raw material mixing, reaction, and stirring in sequence, and finally output the overall stability evaluation result of the photoresist. Among them, each group of weak learners outputs the final evaluation result of this group through an ensemble learning method.

[0054] The above method steps construct a cascade of multiple groups of weak learners, fully considering the sequence and degree of the influence of different preparation links on the stability of the photoresist, ensuring the priority of the evaluation of high-influence / front-end links, thereby improving the prediction accuracy of the model. At the same time, by decomposing complex problems into multiple sub-problems, the complexity of a single learner is reduced, and the training efficiency and generalization ability of the model are improved.

[0055] In some implementation manners, before training multiple groups of weak learners based on the multiple sub-datasets, it includes: Evaluate the co-variation coefficient between multiple process stages and stability indicators, and define a first weight coefficient set according to the co-variation coefficient; define a second weight coefficient set based on the sequence of multiple process stages in the photoresist preparation process flow, where the second weight coefficient decreases along the direction of the photoresist preparation process flow; correct the first weight coefficient set according to the second weight coefficient set, and normalize the corrected first weight coefficient set to obtain a magnitude weight coefficient set.

[0056] Specifically, the co-variation coefficient is a measure representing the degree of co-variation between two variables, used to measure the statistical correlation degree between the process stage characteristics and the finished product stability indicators. The larger the co-variation coefficient, the greater the influence degree of the corresponding process stage on the stability of the photoresist can be considered.

[0057] Specifically, the first weight coefficient set is calculated from the co-variation coefficient between each process stage and the target stability indicator, and is used to quantitatively reflect the direct influence weight of different stages on the stability indicator; the second weight coefficient set is a decaying weight set defined according to the sequence relationship of each stage in the photoresist preparation process flow, indicating the structural importance of the process stage in the overall process. Among them, the process stage upstream in the process flow has a higher weight.

[0058] Specifically, the magnitude weight coefficient set is the final weight set obtained by fusing the above two weight factors and through normalization processing, and is used to adjust the contribution ratio of each weak learner in the integrated model.

[0059] Specifically, to evaluate the process-stability correlation and form the first set of weight coefficients, first, perform a statistical analysis on the key process parameters (such as temperature, rotation speed, time, concentration) of each process stage in the modeling dataset and the target stability indicators (such as sedimentation rate, particle size change, light transmittance); for example, use methods such as Pearson covariance or Spearman to calculate the co-variation coefficient between each parameter and the target indicator; then, perform a weighted average of the co-variation coefficients of each parameter and the target indicator in each stage to obtain the first set of weight coefficients representing the contribution degree of each stage.

[0060] Specifically, in the order of the photoresist preparation process flow, assign a second set of weight coefficients to each stage. The second set of weight coefficients is a decreasing weight. For example, it linearly decreases from 1.0 at the front end of the process to 0.4 at the end, so that the process stages in the front receive higher attention; then, use the second set of weight coefficients to correct the first set of weight coefficients to comprehensively consider the process order and direct influence; then, perform a normalization process on the corrected weight coefficients to ensure that the sum of the weights is 1, and obtain the final set of magnitude weight coefficients for adjusting the contribution of the weak learners. The corrected weights can more emphasize the influence of the earlier process stages while retaining the sensitivity to the stability indicators.

[0061] The above method steps can more accurately reflect the true impact of each process stage on the photoresist stability. Among them, the first set of weight coefficients is used to ensure the sensitivity of the model to the key process parameters, and the second set of weight coefficients introduces the sequentiality of the process flow, enabling the model to dynamically adjust the attention degree to different stages. The training of the weak learners guided by the obtained set of magnitude weight coefficients helps to ensure that the model is more scientific and efficient in evaluating the photoresist stability, thereby improving the accuracy and efficiency of the preparation parameter optimization.

[0062] In some implementation manners, based on multiple said sub-datasets, train multiple groups of weak learners respectively, and combine the ensemble learning method to cascade each group of weak learners in the order of the preparation process to form the stability evaluation model, including: Obtain the evaluation computing power index of the target scenario, allocate the evaluation computing power index according to the set of magnitude weight coefficients, and define the group capacity constraints of multiple groups of weak learners; use the group capacity constraints as the filling upper limit to add weak learners to multiple groups of weak learners; according to the feature dimensions and data sample sizes included in each sub-dataset, adopt an adaptive sampling strategy to enhance or crop the data, and use the cross-validation and early stopping mechanisms to train each group of weak learners until the stability evaluation index convergence requirement is met on the validation set; integrate the trained multiple groups of weak learners and dynamically fine-tune the weights of the weak learners according to the training error distribution to form the stability evaluation model.

[0063] Specifically, the evaluation computing power metric is a computing resource capability metric determined by comprehensively considering factors such as the number of CPU cores, GPU video memory, model execution frame rate, or response time, etc., and is assigned to the training and inference of the stability evaluation model. This evaluation computing power metric can be defined as a scalar. According to the evaluation computing power metric and the magnitude weight coefficient set, the maximum number limit (i.e., group capacity constraint) of weak learners in each group of weak learners can be defined to ensure the resource feasibility of the training process in the target scenario.

[0064] Specifically, the adaptive sampling strategy is a process of data augmentation (such as repetition, adding noise, interpolation) or cropping (such as undersampling, clustering compression) of training samples according to the sample quantity and feature dimension of each sub-dataset, which is used to improve the generalization ability of the model; cross-validation is used to evaluate the generalization error of the model, and the early stopping mechanism is used to abort the training when the accuracy of the validation set reaches the convergence condition to prevent overfitting.

[0065] Specifically, first, read or calculate the computing power evaluation metric in the target production line deployment scenario, such as the computing power under FP16, etc., and allocate the overall computing power metric to each group of weak learners proportionally according to the previously generated magnitude weight coefficient set; then, in each group of weak learners, gradually add weak learners with the group capacity constraint as the upper limit until the group capacity upper limit is reached; then, for each sub-dataset, dynamically adjust the sample distribution according to its sample quantity and feature dimension; at the same time, when the sample quantity is too small, use enhancement strategies such as SMOTE synthesis and local Gaussian noise perturbation, and when the sample quantity is too large, perform K-means clustering representative sampling. Then, use cross-validation and the early stopping mechanism to train each group of weak learners to ensure that the model meets the convergence requirements of the stability evaluation metric on the validation set.

[0066] Furthermore, integrate the trained multiple groups of weak learners, and dynamically fine-tune the weights of the weak learners according to the training error distribution to further optimize the model performance and obtain a complete stability evaluation model. For example, if a certain weak learner has a large error on a specific dataset, its weight is appropriately reduced to reduce its impact on the overall model.

[0067] Through the above process, the limited computing power resources can be efficiently utilized to ensure that the weak learners in the key process stages receive sufficient training support. Among them, the group capacity constraint and the adaptive sampling strategy improve the data utilization efficiency, and the cross-validation and the early stopping mechanism prevent overfitting. The dynamic fine-tuning further optimizes the model performance, enabling the stability evaluation model to more accurately predict the stability of the photoresist, providing reliable evaluation support for subsequent parameter optimization, and helping to improve the stability and efficiency of photoresist preparation.

[0068] Combined with the stability evaluation model and the global optimization algorithm, iterative search is performed on the preparation parameters. Among them, the prediction result of the stability evaluation model is used as the output of the objective function, and the input of the objective function is the preparation parameter.

[0069] Specifically, taking the stability evaluation model as the objective function or evaluation function of the global optimization algorithm, iterative global optimization of the preparation parameters is carried out. Among them, the input of the objective function is the preparation parameter, and the output is the prediction result of the stability evaluation model.

[0070] Exemplarily, first, according to the parameter range of the photoresist preparation process and combining with the existing preparation parameters obtained in real time, a set of preparation parameters is initialized as the initial population; then, each parameter combination in the initial population is input into the stability evaluation model to obtain the predicted stability index. For example, the storage stability of the photoresist under a certain parameter combination is 92% (scalar); then, according to the prediction result (the predicted stability index), the parameter combinations with better performance are selected as the new population, and new parameter combinations are generated through crossover operations. For example, the parameter combinations with high stability are crossed with the parameter combinations with medium stability to generate new parameter combinations and perform mutation operations to introduce randomness to explore more possibilities; further, the processes of evaluation, selection, crossover, and mutation are repeated to gradually optimize the parameter combinations until the preset number of iterations or the stability index converges.

[0071] The above steps, combined with the stability evaluation model and the global optimization algorithm, can efficiently search for the optimal preparation parameters in the multi-parameter space. Among them, the stability evaluation model is used to provide accurate prediction support, guiding the optimization algorithm to quickly converge to the parameter region with high stability, which not only improves the optimization efficiency but also helps to reduce the trial-and-error cost.

[0072] In some embodiments, combined with the stability evaluation model and the global optimization algorithm, iterative search is performed on the preparation parameters. Among them, the prediction result of the stability evaluation model is used as the output of the objective function, and the input of the objective function is the preparation parameter, including: Obtain the preparation parameters in real time and input the preparation parameters into the stability evaluation model; the first group of weak learners of the stability evaluation model takes the preparation parameters as the input, evaluates and outputs the first stability index; input the first stability index and the preparation parameters into the second group of weak learners to obtain the second stability index; input the first stability index, the second stability index, and the preparation parameters into the third group of weak learners to obtain the third stability index, and so on, until all N groups of weak learners of the stability evaluation model are traversed, and the Nth stability index is obtained as the prediction result.

[0073] Optionally, the global optimization algorithm includes Particle Swarm Optimization (PSO), Genetic Algorithm (GA), or Bayesian optimization, etc., which is used to search for the combination of preparation parameters that optimizes the objective function in a complex multi-variable space; the stability index output by the stability evaluation model is used as the performance evaluation function of the optimization algorithm to drive the parameter search direction.

[0074] Specifically, first, obtain the preparation parameter X to be evaluated in the current round, and input the parameter X into the first group of weak learners of the stability evaluation model to obtain the first stability index y. 1 , for example, the first group of weak learners focuses on the raw material mixing stage, and the output stability index is the mixing uniformity score; then, take {X, y 1} as the new input data and input it into the second group of weak learners to output the second stability index y. 2 , for example, the second group of weak learners combines the mixing uniformity score and the reaction temperature to output the reaction stability score; then, take {X, y 1 , y 2} and input it into the third group of weak learners to obtain the third stability index y. 3 And so on until all N groups of weak learners are traversed, and finally obtain the Nth stability index as the prediction result of the model to evaluate the quality of the current preparation parameter combination.

[0075] The above method of gradually evaluating stability through cascaded multiple groups of weak learners can make full use of the information in each process stage. Each group of weak learners is specifically oriented to a preparation link, which helps to reduce the model complexity of a single preparation link and improve the training efficiency. The cascaded structure enables the model to dynamically integrate the stability information of different stages, and the finally output prediction result is more reliable.

[0076] Output the preparation parameters whose prediction results meet the preset stability threshold as the standard preparation parameters, and perform photoresist preparation according to the standard preparation parameters.

[0077] Specifically, the standard preparation parameters are a set of preparation process parameters corresponding to the stability prediction results that meet the preset threshold requirements jointly determined by the stability evaluation model and the global optimization algorithm, and are used for photoresist preparation; the preset stability threshold refers to the minimum acceptable stability index set for the target performance indicators (such as solubility, stability, particle size distribution, etc.) during the photoresist preparation process, and is used to screen qualified parameter combinations.

[0078] Specifically, when the final stability index predicted by the model is not lower than the stability threshold, it indicates that the current parameter combination can meet the requirements of actual production for the quality of photoresist. Then, this set of parameters is used as the standard preparation parameters, and the above standard preparation parameters are sent to the photoresist production system to automatically control the execution devices such as reactors, feeding devices, and temperature control systems for actual photoresist preparation.

[0079] Preferably, from multiple parameter combinations that meet the stability requirements, the parameter group with the lowest model prediction error or the best performance of secondary indicators (such as the minimum energy consumption, the shortest reaction time, etc.) is selected as the final standard preparation parameters.

[0080] In some embodiments, before preparing the photoresist according to the standard preparation parameters, it further includes: Conducting a photoresist preparation experiment according to the standard preparation parameters; if the results of the photoresist preparation experiments in a continuous first preset test batch meet the stability requirements, then preparing the photoresist based on the standard preparation parameters; if the results of the photoresist preparation experiments in a continuous second preset test batch do not meet the stability requirements, then dynamically correcting the stability evaluation model and the optimization algorithm based on the experimental results.

[0081] Specifically, the first preset test batch is the threshold of the continuous batch number set for evaluating the recommendation effect of the model. If the stability indicators in a continuous first preset test batch all meet the requirements, it can be considered that the standard preparation parameters are executable; the second preset test batch is the threshold of the continuous failure batch number set for judging the failure of model prediction. If it is continuously not met, it can be considered that the performance of the standard preparation parameters in actual production does not match the results predicted by the model, and the model needs to be corrected and learned.

[0082] Specifically, according to the standard preparation parameters output by the model, a small-scale preparation experiment (i.e., a photoresist preparation experiment) is carried out in a controlled environment, and the corresponding performance indicators are collected, such as thermal stability, dispersibility, shelf life, etc., to verify the feasibility and stability of the recommended parameter combination under actual process conditions. If the results of a continuous first preset batch (such as 3 batches) of experiments all meet the set stability requirement threshold (such as the deviation ≤ 5%), it is considered that the standard preparation parameters have practical execution value and the batch preparation process is started.

[0083] Specifically, if the results of a continuous second preset batch (such as 2 batches) of experiments do not meet the stability requirements, it indicates that the current model prediction error is large or there is a systematic deviation. The experimental data of the failed batches is fed back to the model training module to update the weights, structures of the weak learners or readjust the objective function constraints based on the real performance to improve the accuracy of the next round of prediction and optimization.

[0084] Through the above experimental verification steps, the effectiveness and reliability of the standard preparation parameters can be confirmed before formal production, reducing the risks of mass production, that is, ensuring that only fully verified parameters will be used in actual production, thereby improving the stability and consistency of the photoresist. In addition, the dynamic correction mechanism enables the model and algorithm to be continuously optimized according to the actual test results, enhancing the adaptability and robustness of the overall optimization.

[0085] In summary, the preparation method for optimizing the stability of the photoresist provided by the present invention has the following technical effects: By taking the production line identity characteristics of the target production line as a constraint, collecting the preparation logs of the photoresist, where the preparation logs cover raw material ratios, reaction condition parameters, and corresponding stability index data; based on the preparation logs, constructing a stability evaluation model, the model consists of multiple cascaded weak learners, and each weak learner corresponds to multiple preparation links of the target production line; by combining the stability evaluation model with the global optimization algorithm, performing iterative search on the preparation parameters, taking the prediction result of the stability evaluation model as the output of the objective function and the preparation parameters as the input; finally, determining the preparation parameters whose prediction results meet the preset stability threshold as the standard preparation parameters, and carrying out photoresist preparation accordingly, so as to achieve the technical effects of improving the optimization efficiency, reducing the optimization cost, and improving the optimization accuracy.

[0086] Example 2, as Figure 2 is a schematic structural diagram of a preparation system for optimizing the stability of a photoresist according to the present invention. For example, Figure 1 In the flow schematic diagram of the preparation method for optimizing the stability of a photoresist according to the present invention can be implemented through a structure such as Figure 2 shown.

[0087] Based on the same concept as the preparation method for optimizing the stability of a photoresist in the above embodiment, the present invention further provides a preparation system for optimizing the stability of a photoresist, including: A log collection module 11, configured to collect the preparation logs of the photoresist by taking the production line identity characteristics of the target production line as a constraint, where the preparation logs include raw material ratios, reaction condition parameters, and corresponding stability index data.

[0088] An evaluation model construction module 12, configured to construct a stability evaluation model based on the preparation logs, where the stability evaluation model includes multiple cascaded weak learners, and the multiple weak learners correspond to multiple preparation links of the target production line.

[0089] An optimization search module 13, configured to perform iterative search on the preparation parameters by combining the stability evaluation model with the global optimization algorithm, where the prediction result of the stability evaluation model is taken as the output of the objective function, and the input of the objective function is the preparation parameters.

[0090] A parameter output module 14 is configured to output the preparation parameters whose prediction results meet a preset stability threshold as standard preparation parameters, and perform photoresist preparation according to the standard preparation parameters.

[0091] In some embodiments, the log collection module 11 includes: A production line identification information acquisition unit is configured to acquire the production line identification information of a target production line and determine its corresponding process configuration features and production line label features.

[0092] A first data set acquisition unit is configured to extract historical records during photoresist preparation from the target production line based on the production line label features, and acquire a first data set including raw material ratio, reaction condition parameters, and stability index data.

[0093] A second data set construction unit is configured to, when the data volume of the first data set is lower than a preset sample threshold, identify homologous production lines with associated labels according to the production line label features, and extract corresponding preparation logs from the homologous production lines to construct a second data set.

[0094] A preparation log integration unit is configured to merge the first data set and the second data set as the data basis for modeling, and construct a preparation log that meets the training requirements.

[0095] In some embodiments, the evaluation model construction module 12 includes: A data preprocessing unit is configured to perform missing value filling, outlier removal, numerical normalization, and categorical variable encoding on the collected preparation log data to obtain a modeling data set for training.

[0096] A sub-data set segmentation unit is configured to segment the modeling data set according to the key link information of the photoresist preparation process flow by process stages to obtain multiple sub-data sets.

[0097] Multiple weak learner training units are configured to train multiple groups of weak learners based on the multiple sub-data sets respectively, and combine the ensemble learning method to cascade each group of weak learners in the order of the preparation process to form the stability evaluation model.

[0098] In some implementation manners, the multiple weak learner training units in the evaluation model construction module 12 include: A co-variation coefficient evaluation unit is configured to evaluate the co-variation coefficient between multiple process stages and the stability index, and define a first weight coefficient set according to the co-variation coefficient.

[0099] A second weight coefficient set defining unit, configured to define a second weight coefficient set based on the sequence of multiple process stages in the photoresist preparation process flow, where the second weight coefficient decreases along the direction of the photoresist preparation process flow.

[0100] A magnitude weight coefficient set obtaining unit, configured to correct the first weight coefficient set according to the second weight coefficient set and normalize the corrected first weight coefficient set to obtain a magnitude weight coefficient set.

[0101] In some implementation manners, the multiple weak learner training units in the evaluation model construction module 12 further include: An evaluation computing power index allocation unit, configured to obtain an evaluation computing power index of a target scenario, allocate the evaluation computing power index according to the magnitude weight coefficient set, and define a group capacity constraint for multiple groups of weak learners.

[0102] A weak learner filling unit, configured to add weak learners to multiple groups of weak learners with the group capacity constraint as the filling upper limit.

[0103] A weak learner training unit, configured to enhance or clip data by using an adaptive sampling strategy according to the feature dimensions and data sample sizes included in each sub-dataset, and train each group of weak learners by using a cross-validation and early stopping mechanism until the convergence requirement of the stability evaluation index is met on the validation set.

[0104] A stability evaluation model integration unit, configured to integrate multiple groups of trained weak learners and dynamically fine-tune the weights of the weak learners according to the training error distribution to form the stability evaluation model.

[0105] In some embodiments, the optimization search module 13 includes: A preparation parameter real-time obtaining unit, configured to obtain preparation parameters in real time and input the preparation parameters into the stability evaluation model.

[0106] A stability index evaluation unit, configured to use the first group of weak learners of the stability evaluation model to take the preparation parameters as inputs, evaluate and output a first stability index.

[0107] A multi-level stability index transmission unit, configured to input the first stability index and the preparation parameters into a second group of weak learners to obtain a second stability index. Input the first stability index, the second stability index, and the preparation parameters into a third group of weak learners to obtain a third stability index, and so on, until traversing the N groups of weak learners of the stability evaluation model to obtain the Nth stability index as the prediction result.

[0108] In some embodiments, the parameter output module 14 includes: A photoresist preparation test unit for conducting photoresist preparation tests according to the standard preparation parameters.

[0109] A stability requirement satisfaction determination unit for preparing photoresist based on the standard preparation parameters if the photoresist preparation test results of a continuous first preset test batch meet the stability requirements.

[0110] A model dynamic correction unit for dynamically correcting the stability evaluation model and optimization algorithm based on the experimental results if the photoresist preparation test results of a continuous second preset test batch do not meet the stability requirements.

[0111] Example 3: Based on the same concept as the preparation method for optimizing the stability of a photoresist in the above example, the present application also provides a photoresist, which can be prepared by the method described in any of the above examples.

[0112] It should be understood that the key points of the embodiments mentioned in this specification lie in their differences from other embodiments. The specific embodiments in the above-mentioned Example 1 are equally applicable to the preparation system for optimizing the stability of a photoresist described in Example 2. For the sake of brevity of the specification, no further elaboration will be made here.

[0113] It should be understood that the disclosed embodiments of the present invention and the above descriptions enable those skilled in the art to implement the present invention using the present invention. At the same time, the present invention is not limited to this part of the embodiments mentioned above. It should be understood that those of ordinary skill in the art can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A preparation method for optimizing the stability of a photoresist, characterized in that: The method comprises: Taking the production line identity characteristics of the target production line as a constraint, collecting the photoresist preparation log, wherein the preparation log includes raw material ratio, reaction condition parameters and corresponding stability index data; Based on the preparation log, construct a stability evaluation model, wherein the stability evaluation model includes multiple groups of cascaded weak learners, and the multiple groups of weak learners correspond to multiple preparation links of the target production line; Combining the stability evaluation model with the global optimization algorithm, an iterative search is performed on the preparation parameters, wherein the prediction result of the stability evaluation model is used as the output of the objective function, and the input of the objective function is the preparation parameters; The preparation parameters whose prediction results satisfy a preset stability threshold are output as standard preparation parameters, and photoresist preparation is performed according to the standard preparation parameters.

2. A method for preparing a photoresist for optimizing stability according to claim 1, characterized in that: Based on the production line identity characteristics of the target production line, the photoresist preparation log is collected, including: Obtain the production line identification information of the target production line and determine its corresponding process configuration characteristics and production line label characteristics; Based on the production line label feature, extracting historical records of the photoresist preparation process from the target production line to obtain a first data set including raw material ratio, reaction condition parameters and stability index data; When the data volume of the first data set is lower than a preset sample threshold, identifying homologous production lines with associated labels according to the production line label features, and extracting corresponding preparation logs from the homologous production lines to construct a second data set; The first data set and the second data set are combined as a data basis for modeling, and a preparation log that meets training requirements is constructed.

3. A method for preparing a photoresist for optimizing stability as claimed in claim 2, characterized in that: Based on the preparation log, a stability evaluation model is constructed, wherein the stability evaluation model includes multiple groups of cascaded weak learners, and the multiple groups of weak learners correspond to multiple preparation links of the target production line, including: Filling missing values, removing outliers, normalizing values, and encoding categorical variables on the collected prepared log data to obtain a modeling data set for training; According to the key link information of the photoresist preparation process, the modeling data set is segmented according to the process stage to obtain multiple sub-data sets; Based on the plurality of sub-data sets, a plurality of groups of weak learners are trained respectively, and the groups of weak learners are cascaded in the order of the preparation process in combination with an ensemble learning method to form the stability evaluation model.

4. A method for preparing a photoresist for optimizing stability as claimed in claim 3, characterized in that: Based on the plurality of sub-datasets, a plurality of groups of weak learners are trained respectively, before, including: evaluating a co-variance coefficient between a plurality of process stages and a stability indicator, and defining a first set of weight coefficients according to the co-variance coefficient; Based on the sequence of the multiple process stages in the photoresist preparation process flow, a second weight coefficient set is defined, wherein the second weight coefficient decreases in a direction along the photoresist preparation process flow; The first weight coefficient set is corrected according to the second weight coefficient set, and the corrected first weight coefficient set is normalized to obtain a magnitude weight coefficient set.

5. A method for preparing a photoresist for optimizing stability as claimed in claim 4, characterized in that: Based on the plurality of sub-data sets, multiple groups of weak learners are trained respectively, and the groups of weak learners are cascaded according to the preparation process sequence in combination with an ensemble learning method to form the stability evaluation model, including: Acquire an evaluation computing power indicator of a target scenario, allocate the evaluation computing power indicator according to the magnitude weight coefficient set, and define group capacity constraints for multiple groups of weak learners; adding weak learners to the plurality of groups of weak learners with the group capacity constraint as a filling upper limit; According to the feature dimensions and data sample size contained in each sub-dataset, an adaptive sampling strategy is used to enhance or trim the data, and each group of weak learners is trained using cross-validation and early stopping mechanisms until the stability evaluation index converges on the validation set. The multiple groups of trained weak learners are integrated, and the weights of the weak learners are dynamically fine-tuned according to the training error distribution to form the stability evaluation model.

6. A method for preparing a photoresist for optimizing stability according to claim 1, characterized in that: Combining the stability evaluation model with the global optimization algorithm, an iterative search is performed on the preparation parameters, wherein the prediction result of the stability evaluation model is used as the output of the objective function, and the input of the objective function is the preparation parameters, including: Acquiring preparation parameters in real time, and inputting the preparation parameters into the stability evaluation model; The first group of weak learners of the stability evaluation model takes the preparation parameters as input, evaluates and outputs a first stability index; Inputting the first stability index and the preparation parameters into a second group of weak learners to obtain a second stability index; The first stability index, the second stability index and the preparation parameters are input into a third group of weak learners to obtain a third stability index, and so on, until N groups of weak learners of the stability evaluation model are traversed to obtain an Nth stability index as the prediction result.

7. A method for preparing a photoresist for optimizing stability according to claim 1, characterized in that: The photoresist is prepared according to the standard preparation parameters, and before that, it also includes: Performing a photoresist preparation test according to the standard preparation parameters; If the test results of the photoresist preparation of the first preset test batches meet the stability requirement, the photoresist preparation is performed based on the standard preparation parameters; If the test results of the photoresist preparation of the second consecutive preset test batch do not meet the stability requirement, the stability evaluation model and the optimization algorithm are dynamically modified based on the experimental results.

8. A preparation system for optimizing the stability of a photoresist, characterized in that: A preparation method for optimizing the stability of a photoresist according to any one of claims 1 to 7, comprising: A log collection module, used to collect the preparation log of the photoresist based on the production line identity characteristics of the target production line as a constraint, wherein the preparation log includes raw material ratio, reaction condition parameters and corresponding stability index data; An evaluation model construction module, used to construct a stability evaluation model based on the preparation log, wherein the stability evaluation model includes a plurality of cascaded groups of weak learners, and the plurality of groups of weak learners correspond to a plurality of preparation links of the target production line; An optimization search module, used to combine the stability evaluation model with a global optimization algorithm to iteratively search for preparation parameters, wherein the prediction result of the stability evaluation model is used as the output of an objective function, and the input of the objective function is the preparation parameters; The parameter output module is used to output the preparation parameters whose prediction results meet the preset stability threshold as standard preparation parameters, and prepare the photoresist according to the standard preparation parameters.

9. A photoresist, characterized in that: The photoresist is obtained by a preparation method for optimizing the stability of the photoresist as described in any one of claims 1-7.

Images