CN-121997757-A - Semiconductor wafer film pasting parameter optimization processing method and system

CN121997757ACN 121997757 ACN121997757 ACN 121997757ACN-121997757-A

Abstract

The application relates to the technical field of semiconductor wafer film sticking, and discloses a semiconductor wafer film sticking parameter optimization processing method and system. The method comprises the steps of training a film pasting quality prediction model based on historical production data, constructing a quality evaluation function, aiming at maximizing comprehensive quality score, searching an optimal technological parameter combination by using an optimization algorithm under a given target condition, screening key parameters from equipment parameters, carrying out cluster analysis on wafer characteristic parameters to generate virtual wafer characteristic parameters, training a parameter mapping model, establishing a mapping relation between the equipment parameters and the technological parameters, constructing a parameter optimization model, aiming at minimizing a key parameter correction quantity norm and meeting performance guarantee and robustness constraint conditions, generating the optimized equipment parameters based on correction quantity, and outputting an actual technological parameter combination through the parameter mapping model. The application can improve the stability of the film quality.

Inventors

HE JINGJING
Shi hainan
LU AIJUN

Assignees

浙江丽水中欣晶圆半导体科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (8)

1. The semiconductor wafer film pasting parameter optimization processing method is characterized by comprising the following steps of: Step S1, training based on historical production data to generate a film pasting quality prediction model, wherein the film pasting quality model takes a process parameter combination, an environment parameter and a wafer characteristic parameter as input parameters, and outputs film pasting quality indexes; s2, constructing a quality evaluation function, taking a film pasting quality index as input, outputting a comprehensive quality score, taking the maximization of the comprehensive quality score as a target, and searching for an optimal technological parameter combination by using a first optimization algorithm under a given target condition; Step S3, acquiring an equipment parameter analysis data set, and screening key parameters from a plurality of equipment parameters based on the equipment parameter analysis data set, wherein the key parameters are equipment parameters with high contribution to realizing the optimal process parameter combination; s4, carrying out cluster analysis on the wafer characteristic parameters in the historical production data, and extracting vertex coordinates at the generated boundary vertexes of each characteristic cluster to generate a plurality of virtual wafer characteristic parameters; s5, training a parameter mapping model based on historical production data, wherein the parameter mapping model takes equipment parameter combination, environment parameters and wafer characteristic parameters as inputs, and outputs predicted process parameter combination; s6, taking the correction quantity of the key parameter as an optimization variable, constructing a parameter optimization model, minimizing the norm of the correction quantity of the key parameter as an optimization target by the parameter optimization model, simultaneously meeting a first constraint condition and a second constraint condition, and solving the parameter optimization model to obtain an optimal solution for solving the parameter optimization model, wherein the optimal solution represents the correction quantity of the key parameter; And S7, generating optimized equipment parameters based on the correction amounts of the key parameters, inputting the target environment parameters, the target wafer characteristic parameters and the optimized equipment parameters into a parameter mapping model, and combining the output of the parameter mapping model as actual process parameters.
2. The method of claim 1, wherein the first constraint is: And inputting the optimized equipment parameters, the target environment parameters and the historical wafer characteristic parameters into a parameter mapping model to obtain a difference between a predicted process parameter combination and an optimal process parameter combination, wherein the difference is smaller than a first preset threshold value.
3. The method of claim 1, wherein the second constraint is: For each virtual wafer characteristic parameter, inputting the optimized equipment parameter, the target environment parameter and the virtual wafer characteristic parameter into a parameter mapping model to obtain a first prediction process parameter combination, inputting the first prediction process parameter combination, the target environment parameter and the virtual wafer characteristic into a film pasting quality model to obtain a first prediction comprehensive quality index, inputting the first comprehensive quality index into a quality evaluation function to obtain a first comprehensive quality score, wherein the comprehensive quality score is larger than or equal to a second comprehensive quality score, and the second comprehensive quality score is obtained based on the equipment parameter before optimization.
4. The method of claim 1, wherein screening a plurality of key parameters that contribute highly to achieving an optimal process parameter combination comprises: Extracting production data before and after the equipment hardware is changed from historical production data, correlating the production data with equipment parameters to form an equipment parameter analysis data set, taking the equipment parameters as interpretation variables, respectively taking all process parameters of process parameter combinations in the corresponding production data as target variables, constructing a plurality of multiple regression models, fitting the multiple regression models based on the equipment parameter analysis data set, calculating a standardized regression coefficient of each equipment parameter for each target variable, adding absolute values of regression coefficients of each equipment parameter in each multiple regression model to obtain comprehensive contribution degree of the equipment parameters, and taking the first N equipment parameters with the highest comprehensive contribution degree as key parameters.
5. The method of claim 1, wherein generating a plurality of virtual wafer characteristic parameters comprises: Clustering historical wafer characteristic parameters by using a clustering algorithm to form a plurality of characteristic clusters, calculating convex hulls of each characteristic cluster in a corresponding characteristic space, acquiring vertex coordinates of the convex hulls, and taking the vertex coordinates as virtual wafer characteristic parameters.
6. The method of claim 1, wherein finding an optimal combination of process parameters using a first optimization algorithm comprises: Obtaining a process parameter combination in historical production data, regenerating a plurality of new process parameter combinations different from the historical process parameter combination, respectively combining the historical process parameter combination and the generated new process parameter combination with given target conditions to generate a plurality of input parameters of a film pasting quality model, inputting the input parameters into the film pasting quality model to obtain corresponding film pasting quality indexes, inputting the film pasting quality indexes into a quality evaluation function to obtain corresponding comprehensive quality scores, eliminating part of process parameter combinations based on the comprehensive quality scores, and optimizing the rest of process parameter combinations by using a first optimization algorithm to obtain an optimal process parameter combination.
7. The method of claim 1, wherein training to generate a film quality prediction model comprises: Performing first hierarchical clustering on the historical production data based on the technological parameter combination to obtain a plurality of first data sets, defining a first label for each first data set, performing second hierarchical clustering on each first data set to obtain a plurality of second data sets, and defining a second label for each second data set; Predefining lightweight models of a plurality of different structures to generate a model pool, and calculating key features for each second data set; training all models in the model pool in turn for each second data set, evaluating the performance of each model, and marking the model with the best performance as the preferred model corresponding to the second label; The key features are taken as input, and the types of the corresponding preferred models are taken as output to train the classification model.
8. A semiconductor wafer film parameter optimization processing system for implementing a semiconductor wafer film parameter optimization processing method as claimed in any one of claims 1 to 7, wherein the system comprises: The model training unit is used for training and generating a film pasting quality prediction model based on historical production data, wherein the film pasting quality model takes a process parameter combination, an environment parameter and a wafer characteristic parameter as input parameters, and outputs film pasting quality indexes; the target optimizing unit is used for constructing a quality evaluation function, taking a film pasting quality index as input, outputting a comprehensive quality score, taking the maximization of the comprehensive quality score as a target, and searching an optimal technological parameter combination by using a first optimizing algorithm under a given target condition; The parameter screening unit is used for acquiring an equipment parameter analysis data set, screening key parameters from a plurality of equipment parameters based on the equipment parameter analysis data set, wherein the key parameters are equipment parameters with high contribution to realizing the optimal process parameter combination; The sample generation unit is used for carrying out cluster analysis on the wafer characteristic parameters in the historical production data, extracting vertex coordinates at the generated boundary vertexes of each characteristic cluster, and generating a plurality of virtual wafer characteristic parameters; The mapping training unit is used for training a parameter mapping model based on the historical production data, wherein the parameter mapping model takes the equipment parameter combination, the environment parameter and the wafer characteristic parameter as inputs, and outputs a predicted process parameter combination; The parameter optimization unit is used for taking the correction quantity of the key parameter as an optimization variable, constructing a parameter optimization model, minimizing the norm of the correction quantity of the key parameter as an optimization target by the parameter optimization model, simultaneously meeting a first constraint condition and a second constraint condition, and solving the parameter optimization model to obtain an optimal solution for solving the parameter optimization model, wherein the optimal solution represents the correction quantity of the key parameter; And the parameter generation unit is used for generating optimized equipment parameters based on the correction amount of the key parameters, inputting the target environment parameters, the target wafer characteristic parameters and the optimized equipment parameters into a parameter mapping model, and combining the output of the parameter mapping model as actual process parameters.

Description

Semiconductor wafer film pasting parameter optimization processing method and system Technical Field The application relates to the technical field of semiconductor wafer film sticking, in particular to a semiconductor wafer film sticking parameter optimization processing method and system. Background In semiconductor wafer lamination processes, conventional methods typically optimize process parameters based on specific lot wafers and stable environmental conditions. However, in actual production, the wafer materials often change (e.g., from one vendor to another or from lot to lot), the environmental conditions (e.g., humidity, temperature) fluctuate, and long-term drift in the hardware state of the equipment can occur. These factors lead to film yield fluctuation, and quality problems such as bubble increase, uniformity decrease, insufficient adhesive strength and the like appear. Whenever such problems occur, the conventional practice needs to be stopped and parameters are debugged again by experience according to the characteristics of the new wafer and the current environment, and a large number of trial and error are performed, which not only results in low production efficiency, but also brings high quality risks. The prior art lacks of system modeling of dynamic nonlinear relation between equipment hardware parameters and process parameters, and also fails to effectively identify key equipment parameters with the greatest influence on film quality. In addition, the conventional optimization method is only aimed at typical working conditions in historical data, and cannot cope with the extreme wafer characteristics possibly appearing in the future, so that the performance of the optimized parameter scheme is drastically reduced under unknown working conditions. Therefore, a parameter optimization method capable of adapting to wafer variation, environmental fluctuation and equipment state drift and having strong robustness is needed to improve the stability and generalization capability of the film-sticking process. Therefore, the invention provides a semiconductor wafer film sticking parameter optimization processing method and system. Disclosure of Invention The embodiment of the specification provides the following technical scheme: Step S1, training based on historical production data to generate a film pasting quality prediction model, wherein the film pasting quality model takes a process parameter combination, an environment parameter and a wafer characteristic parameter as input parameters, and outputs film pasting quality indexes; s2, constructing a quality evaluation function, taking a film pasting quality index as input, outputting a comprehensive quality score, taking the maximization of the comprehensive quality score as a target, and searching for an optimal technological parameter combination by using a first optimization algorithm under a given target condition; Step S3, acquiring an equipment parameter analysis data set, and screening key parameters from a plurality of equipment parameters based on the equipment parameter analysis data set, wherein the key parameters are equipment parameters with high contribution to realizing the optimal process parameter combination; s4, carrying out cluster analysis on the wafer characteristic parameters in the historical production data, and extracting vertex coordinates at the generated boundary vertexes of each characteristic cluster to generate a plurality of virtual wafer characteristic parameters; s5, training a parameter mapping model based on historical production data, wherein the parameter mapping model takes equipment parameter combination, environment parameters and wafer characteristic parameters as inputs, and outputs predicted process parameter combination; s6, taking the correction quantity of the key parameter as an optimization variable, constructing a parameter optimization model, minimizing the norm of the correction quantity of the key parameter as an optimization target by the parameter optimization model, simultaneously meeting a first constraint condition and a second constraint condition, and solving the parameter optimization model to obtain an optimal solution for solving the parameter optimization model, wherein the optimal solution represents the correction quantity of the key parameter; And S7, generating optimized equipment parameters based on the correction amounts of the key parameters, inputting the target environment parameters, the target wafer characteristic parameters and the optimized equipment parameters into a parameter mapping model, and combining the output of the parameter mapping model as actual process parameters. Compared with the prior art, the invention has the following beneficial effects: According to the technical scheme, the optimal technological parameter combination is systematically found under the given target condition through the film pasting quality prediction model and the quality evalua