CN-122021355-A - Semiconductor device process parameter optimization method and system based on data driving

CN122021355ACN 122021355 ACN122021355 ACN 122021355ACN-122021355-A

Abstract

The invention provides a method and a system for optimizing process parameters of a semiconductor device based on data driving. The method for optimizing the technological parameters of the semiconductor device comprises the steps of obtaining technological parameters, a data set of corresponding intermediate physical variables and device performance indexes, training to obtain a physical prediction model based on a mapping relation between the technological parameters and the intermediate physical variables, training to obtain a performance prediction model based on the intermediate physical variables and/or the mapping relation between the technological parameters and the device performance indexes, carrying out reasoning calculation on the input technological parameters based on the physical prediction model and the performance prediction model to obtain the corresponding intermediate physical variables and the device performance indexes, and determining a technological parameter range meeting target performance requirements under preset technological constraint conditions based on a prediction result of the device performance indexes.

Inventors

YANG ZHENHAI
GAO QIANHONG
ZHAN YAOHUI
CAO GUOYANG
QIN LINLING
LI XIAOFENG

Assignees

苏州大学

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The method for optimizing the technological parameters of the semiconductor device based on data driving is characterized by comprising the following steps of: acquiring a data set of process parameters, corresponding intermediate physical variables and device performance indexes; Training to obtain a physical prediction model based on the mapping relation between the technological parameters and the intermediate physical variables; Training to obtain a performance prediction model based on the intermediate physical variable and/or the mapping relation between the technological parameter and the device performance index; based on the physical prediction model and the performance prediction model, carrying out reasoning calculation on the input process parameters to obtain corresponding intermediate physical variables and device performance indexes; And determining a process parameter range meeting the target performance requirement under a preset process constraint condition based on the predicted result of the device performance index.
2. The method for optimizing process parameters of a semiconductor device according to claim 1, wherein the acquiring the process parameters and the corresponding data sets of intermediate physical variables and device performance indicators specifically comprises: acquiring multi-source data related to a target device manufacturing process, wherein the multi-source data comprises process parameter data, intermediate physical variable data and performance test data; Performing association matching processing on the multi-source data to establish a corresponding relation among the process parameter data, the intermediate physical variable data and the performance test data; carrying out consistency processing on the correlated data to form training data samples with uniform structures; And constructing a cross-layer associated data set based on the processed data samples, so that each data sample simultaneously contains process parameters, intermediate physical variables and corresponding performance indexes for subsequent model training.
3. The method for optimizing process parameters of a semiconductor device according to claim 2, wherein training to obtain a physical prediction model based on a mapping relationship between the process parameters and the intermediate physical variables specifically comprises: Determining an intermediate physical variable set to be predicted, and constructing a corresponding mapping model structure aiming at each intermediate physical variable or a plurality of intermediate physical variable combinations; Based on the cross-layer association data set, using process parameter data as input characteristics, using corresponding intermediate physical variable data as a supervision tag, and training the mapping model to learn a nonlinear mapping relation between the process parameter and the intermediate physical variable; In the model training process, carrying out normalization processing on input features, dividing training data to form a training subset and a verification subset, and carrying out parameter updating and convergence control on the model by iteratively optimizing model parameters on the training subset and evaluating prediction errors on the verification subset; after the model converges, a physical prediction model for predicting intermediate physical variables according to the process parameters is obtained for the subsequent performance prediction model to call.
4. A method according to claim 3, wherein said training to obtain a performance prediction model based on the mapping between said intermediate physical variables and/or said process parameters and said device performance metrics comprises: Determining a performance index set to be predicted, and constructing a performance mapping model taking the intermediate physical variable as an input characteristic; Based on the cross-layer association data set, taking the intermediate physical variable data as input, taking the corresponding performance test data as a supervision tag, and training the performance mapping model to learn the mapping relation between the intermediate physical variable and the performance index; in the model training process, input features are subjected to standardized processing, and the model is subjected to iterative optimization by dividing training subsets and verification subsets, so that the prediction error of the model on the verification subsets meets the preset precision requirement; after model training is completed, a performance prediction model for predicting performance indicators from the intermediate physical variables is obtained.
5. The method of claim 4, wherein the physical prediction model and the performance prediction model are invoked in a cascaded manner at an application stage, comprising: inputting the technological parameters to be evaluated into the physical prediction model to obtain a corresponding intermediate physical variable prediction result; Transmitting the intermediate physical variable prediction result as input to the performance prediction model to obtain a corresponding performance index prediction result; Screening process parameter combinations meeting conditions by combining preset performance target constraints based on the performance index prediction results; in the screening process, introducing equipment constraint conditions and non-manufacturable parameter combination constraint into a screening rule, and eliminating process parameter combinations which do not meet the constraint conditions; clustering or section merging processing is carried out on the remaining technological parameter combinations meeting constraint conditions, a plurality of candidate parameter sections are generated, and the candidate parameter sections are ordered based on multi-target evaluation indexes, so that a preferable parameter section set is obtained.
6. The method of claim 5, wherein determining a range of process parameters meeting a target performance requirement under a preset process constraint based on the predicted result of the device performance index specifically comprises: Determining selectable value sets of all the process parameters based on the value ranges of the process parameters and the equipment constraint conditions, wherein the selectable value sets comprise continuous interval ranges and/or discrete gear sets; performing combined sampling in the selectable value set to generate a plurality of process parameter combined samples; calling the physical prediction model and the performance prediction model for each process parameter combination sample to obtain a corresponding intermediate physical variable prediction result and a corresponding performance index prediction result; Based on the performance index prediction result and a preset constraint condition, carrying out feasibility screening on the process parameter combination samples, and eliminating samples which do not meet equipment constraint, process constraint or intermediate physical variable constraint; clustering or interval merging is carried out on the screened process parameter combination samples according to parameter space distribution, and adjacent or similar process parameter combinations are merged into parameter intervals, so that a candidate parameter interval set is formed.
7. The method of claim 6, wherein the ranking candidate parameter intervals based on the multi-objective evaluation index comprises: determining a target index set and an evaluation rule of each target index; Selecting a plurality of representative parameter points in each candidate parameter interval, respectively calculating a performance index prediction result corresponding to the representative parameter points, and carrying out statistical aggregation on the performance index prediction result to obtain an interval-level performance evaluation result; Constructing a comprehensive scoring function based on the interval-level performance evaluation result, scoring each candidate parameter interval, or determining a non-inferior solution interval set by adopting a pareto front screening method based on the interval-level performance evaluation result, and secondarily sequencing the non-inferior solution interval set according to a preset priority rule; And selecting the first K candidate parameter intervals as recommended parameter intervals according to the scoring result or the ranking result, and outputting corresponding ranking information and evaluation result.
8. The method of claim 7, wherein the process of performing feasibility screening on the process parameter combination samples further comprises performing constraint screening on candidate parameter intervals based on intermediate physical variable constraints, comprising: determining constraint ranges corresponding to different intermediate physical variables; For each process parameter combination sample, obtaining a corresponding intermediate physical variable prediction result according to the physical prediction model; judging whether the intermediate physical variable prediction result meets the constraint range or not, and eliminating the process parameter combination samples which do not meet the constraint range; Counting the distribution conditions of the corresponding intermediate physical variables of a plurality of parameter points in the candidate parameter interval, and eliminating or adjusting the candidate parameter interval when the parameter points exceeding the preset proportion in the interval do not meet the constraint range; After screening is completed, reserving candidate parameter intervals which simultaneously meet equipment constraint and intermediate physical variable constraint so as to improve manufacturability and stability of the candidate parameter intervals.
9. The method according to any one of claims 1 to 8, further comprising updating and managing model training and candidate parameter intervals based on an external system interface, comprising: Acquiring newly-added process parameter data, equipment operation log data and corresponding performance test data generated in the production line operation process through an external system interface; the new data is merged into a sample data set, the sample data set is updated in an increment mode, and the physical prediction model and the performance prediction model are retrained or updated online; storing the historically generated candidate parameter intervals and the corresponding evaluation results to form a parameter interval library, and recording the applicable conditions and version information of each parameter interval; When detecting that the process condition changes or the equipment state changes, selecting an adapted historical parameter interval from the parameter interval library as a candidate initial interval, or performing expansion generation based on the historical parameter interval; Based on the updated model and data, the candidate parameter intervals are reevaluated and ordered to realize the dynamic optimization of the process window.
10. A data-driven semiconductor device process parameter optimization system, comprising: The data acquisition unit is used for acquiring a process parameter set, equipment constraint data, a target index set and a sample data set; The physical prediction unit is used for establishing a mapping relation between the technological parameters and the intermediate physical variables and predicting the intermediate physical variables according to the technological parameters; The performance prediction unit is used for establishing a mapping relation between the intermediate physical variable and/or the process parameter and the performance index and predicting the performance index; The candidate interval generating unit is used for generating candidate technological parameter intervals on the premise of meeting the equipment constraint conditions; the evaluation sequencing unit is used for performing multi-objective evaluation on the candidate process parameter interval and determining a preferred process parameter interval; and the output unit is used for outputting the preferable process parameter interval and the sequencing result thereof.

Description

Semiconductor device process parameter optimization method and system based on data driving Technical Field The invention relates to the technical field of parameter optimization, in particular to a method and a system for optimizing process parameters of a semiconductor device based on data driving. Background In the diffusion process of the solar cell, various process parameters such as temperature, time, gas flow, pressure, formula parameters and the like are involved, and the parameters are mutually influenced to jointly determine the doping distribution and the electrical performance of the device. In actual production, there are obvious differences in performance corresponding to different parameter combinations, and process engineers generally need to comprehensively trade-off among multiple indexes such as efficiency, open circuit voltage, filling factor, yield and production takt. Meanwhile, due to the limitations of equipment capacity and process conditions, each parameter often has the limitations of a value range, discrete gears and specific combinations, so that the selection of the process parameters has higher complexity. In the prior art, the determination of the technological parameters depends on an empirical formula or a single factor adjustment mode, and the parameters are adjusted and verified through a step-by-step test. The mode has lower efficiency under a multi-parameter coupling scene, and is difficult to comprehensively optimize multiple performance indexes. In addition, although some parameter combinations have better performance in theory, the parameter combinations may be difficult to implement in actual production due to the influence of equipment constraint or stability factors, so that the determination and application of process windows have certain limitations. Disclosure of Invention The invention aims to provide a method and a system for optimizing process parameters of a semiconductor device based on data driving, which are used for effectively modeling the relation between an intermediate physical process and the performance of the device under the condition of multi-process parameter coupling, and determining the process parameter range meeting the requirements of multi-performance indexes on the premise of meeting equipment constraint and process constraint, so that the efficiency of optimizing the process parameters and the feasibility of the result are improved. In a first aspect, the present invention provides a method for optimizing process parameters of a semiconductor device based on data driving, including: acquiring a data set of process parameters, corresponding intermediate physical variables and device performance indexes; Training to obtain a physical prediction model based on the mapping relation between the technological parameters and the intermediate physical variables; Training to obtain a performance prediction model based on the intermediate physical variable and/or the mapping relation between the technological parameter and the device performance index; based on the physical prediction model and the performance prediction model, carrying out reasoning calculation on the input process parameters to obtain corresponding intermediate physical variables and device performance indexes; And determining a process parameter range meeting the target performance requirement under a preset process constraint condition based on the predicted result of the device performance index. Optionally, the constructing of the dataset includes: acquiring multi-source data related to a target device manufacturing process, wherein the multi-source data comprises process parameter data, intermediate physical variable data and performance test data; Performing association matching processing on the multi-source data to establish a corresponding relation among the process parameter data, the intermediate physical variable data and the performance test data; carrying out consistency processing on the correlated data to form training data samples with uniform structures; And constructing a cross-layer associated data set based on the processed data samples, so that each data sample simultaneously contains process parameters, intermediate physical variables and corresponding performance indexes for subsequent model training. Optionally, the establishing of the physical prediction model for predicting the intermediate physical variable based on the process parameter includes: Determining an intermediate physical variable set to be predicted, and constructing a corresponding mapping model structure aiming at each intermediate physical variable or a plurality of intermediate physical variable combinations; Based on the cross-layer association data set, using process parameter data as input characteristics, using corresponding intermediate physical variable data as a supervision tag, and training the mapping model to learn a nonlinear mapping relation between the process parameter