Search

CN-121980389-A - Training method of overlay error prediction model and overlay error prediction method

CN121980389ACN 121980389 ACN121980389 ACN 121980389ACN-121980389-A

Abstract

The disclosure provides a training method of an overlay error prediction model and an overlay error prediction method. The training method comprises the steps of obtaining basic parameter data and sample alignment error data of a plurality of sample wafers, performing polynomial decomposition on the sample alignment error data to obtain a plurality of sample alignment error components with at least two different orders, constructing corresponding sample data sets under different orders based on the basic parameter data and the sample alignment error components, taking the basic parameter data in the sample data sets as input, taking the sample alignment error components as output, and performing model training by adopting the sample data sets corresponding to the different orders to obtain alignment error prediction models corresponding to the different orders. The method and the device can effectively enhance the capability of the prediction model in distinguishing tiny differences among different wafers, improve the prediction robustness, reduce the overfitting risk, adapt to more complex production scenes and improve the adaptability of the model to long-term process drift.

Inventors

  • ZHU NINGQI

Assignees

  • 半脊科技(上海)有限公司

Dates

Publication Date
20260505
Application Date
20260206

Claims (15)

  1. 1. A training method of an overlay error prediction model, the training method comprising: basic parameter data and sample alignment error data of a plurality of sample wafers are obtained; performing polynomial decomposition on the sample alignment error data to obtain a plurality of sample alignment error components with at least two different orders; Constructing corresponding sample data sets under different orders based on the basic parameter data and the sample alignment error component; And taking the basic parameter data in the sample data set as input, taking the sample alignment error component as output, and respectively performing model training by adopting the sample data sets corresponding to different orders to obtain alignment error prediction models corresponding to different orders.
  2. 2. The training method of claim 1, wherein the step of constructing corresponding sample data sets at different ones of the orders based on the base parameter data and the sample overlay error component comprises: extracting sample measurement features of the sample wafer based on the base parameter data; according to the sample measurement characteristics and the sample alignment error components, constructing sample data corresponding to the same sample wafer under different orders; And obtaining the corresponding sample data set under different orders based on the sample data corresponding to the plurality of different sample wafers under the same order.
  3. 3. The training method of claim 2, wherein the overlay error prediction model comprises a first overlay error prediction model and a second overlay error prediction model; The step of taking the basic parameter data in the sample data set as input, taking the sample overlay error component as output, and respectively performing model training by adopting the sample data sets corresponding to different orders to obtain an overlay error prediction model corresponding to different orders comprises the following steps: taking the sample measurement characteristics in the sample data set as input, taking the sample alignment error components as output, and respectively training a first preset model by adopting the sample data sets corresponding to different orders to obtain the first alignment error prediction models corresponding to different orders; And/or the number of the groups of groups, Obtaining the similarity of the measurement characteristics of the sample wafer in the sample data set and the reference measurement characteristics corresponding to the reference wafer in the preset reference data set, and obtaining the similarity characteristics of the sample wafer in the sample data set; Taking the similarity characteristics in the sample data set as input, taking the sample alignment error components as output, and respectively training a second preset model by adopting the sample data sets corresponding to different orders to obtain the second alignment error prediction models corresponding to different orders; The second preset model comprises a reference overlay error component of the reference wafer under the same order.
  4. 4. The training method of claim 1, wherein the training method further comprises: Adding target parameter data and a predicted overlay error of a wafer to be detected as new sample data into a sample data set to form a new sample data set; the prediction overlay error is obtained based on the target parameter data and the overlay error prediction model; And training and updating the overlay error prediction model based on the sample data set.
  5. 5. The training method of any one of claims 1 to 4, wherein the base parameter data includes at least one of flatness, alignment, correction compensation values.
  6. 6. A method of predicting overlay error, the method comprising: acquiring target parameter data of a wafer to be tested; inputting the target parameter data into an overlay error prediction model to obtain different-order prediction overlay error components corresponding to the wafer to be detected; wherein the overlay error prediction model is obtained using the training method of any one of claims 1 to 5; and obtaining the predicted overlay error of the wafer to be detected based on the different-order predicted overlay error components.
  7. 7. The prediction method according to claim 6, wherein the predicted overlay error component comprises a first predicted overlay error component and/or a second predicted overlay error component; The step of inputting the target parameter data into an overlay error prediction model to obtain different-order prediction overlay error components corresponding to the wafer to be detected comprises the following steps of: extracting target measurement characteristics of the wafer to be measured based on the target parameter data; Inputting the target measurement characteristics into the overlay error prediction model to obtain the first prediction overlay error components corresponding to the wafer to be measured under different orders; And/or the number of the groups of groups, Obtaining target similarity between the target measurement feature and the reference measurement feature of each reference wafer, and obtaining target similarity features of the wafers to be measured; And inputting the target similarity characteristics into the overlay error prediction model to obtain the second prediction overlay error components corresponding to the wafer to be detected under different orders.
  8. 8. The method according to claim 7, wherein the step of obtaining the predicted overlay error of the wafer to be measured based on the different-order predicted overlay error components comprises: acquiring the first prediction overlay error component and the second prediction overlay error component under the same order, and determining a target prediction wafer component under the same order by adopting a voting mechanism; Obtaining the predicted overlay error of the wafer to be detected based on the corresponding target predicted overlay error components under different orders; And/or the number of the groups of groups, Determining at least one target order in different orders based on the preset prediction precision of the wafer to be detected; and obtaining the predicted overlay error of the wafer to be detected according to the first predicted overlay error component and/or the second predicted overlay error component corresponding to the target order.
  9. 9. A method for controlling a wafer fabrication process, the method comprising: Completing fabrication of the current wafer in response to the current fabrication process parameters; Predicting a predicted overlay error of the current wafer using the prediction method of any one of claims 6 to 8; And adjusting the current manufacturing process parameters based on the prediction overlay error to obtain the adjusted current manufacturing process parameters, and adopting the current manufacturing process parameters to manufacture new wafers.
  10. 10. The training system of the overlay error prediction model is characterized by comprising an acquisition module, a decomposition module, a construction module and a training module; the acquisition module is used for acquiring basic parameter data and sample alignment error data of a plurality of sample wafers; The decomposition module is used for performing polynomial decomposition on the sample alignment error data to obtain a plurality of sample alignment error components with at least two different orders; the construction module is used for constructing corresponding sample data sets under different orders based on the basic parameter data and the sample alignment error component; The training module is used for taking the basic parameter data in the sample data set as input, taking the sample alignment error component as output, and respectively carrying out model training by adopting the sample data sets corresponding to different orders to obtain alignment error prediction models corresponding to different orders.
  11. 11. The prediction system of the overlay error is characterized by comprising a parameter acquisition module, a prediction module and a determination module; the parameter acquisition module is used for acquiring target parameter data of the wafer to be tested; The prediction module is used for inputting the target parameter data into an overlay error prediction model to obtain different-order prediction overlay error components corresponding to the wafer to be detected; wherein the overlay error prediction model is obtained using the training system of claim 10; The determining module is used for obtaining the predicted overlay error of the wafer to be detected based on the different-order predicted overlay error components.
  12. 12. The control system of the wafer manufacturing process is characterized by comprising a control module; The control module is used for predicting the predicted overlay error of the current wafer by adopting the prediction system according to claim 11 in response to completing the manufacture of the current wafer based on the current manufacture process parameters; The control module is further configured to adjust the current manufacturing process parameter based on the predicted overlay error, obtain the adjusted current manufacturing process parameter, and perform new wafer manufacturing by using the current manufacturing process parameter.
  13. 13. An electronic device comprising a memory, a processor and a computer program stored on the memory for running on the processor, characterized in that the processor implements the method of training the overlay error prediction model according to any one of claims 1 to 5, or the method of predicting an overlay error according to any one of claims 6 to 8, or the method of controlling the wafer fabrication process according to claim 9, when executing the computer program.
  14. 14. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the method of training an overlay error prediction model according to any one of claims 1to 5, or the method of predicting an overlay error according to any one of claims 6 to 8, or the method of controlling a wafer fabrication process according to claim 9.
  15. 15. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements a method for training an overlay error prediction model according to any one of claims 1 to 5, or a method for predicting an overlay error according to any one of claims 6 to 8, or a method for controlling a wafer fabrication process according to claim 9.

Description

Training method of overlay error prediction model and overlay error prediction method Technical Field The present disclosure relates to the field of semiconductor manufacturing, and in particular, to a training method of an overlay error prediction model and an overlay error prediction method. Background In advanced semiconductor processes, overlay accuracy between layers of a wafer is a key indicator in determining yield and device performance. Measurement of conventional Overlay errors depends on physical measurement equipment (e.g., optical measurement tools), but the number of measurement points is limited, all wafer areas cannot be covered, and a long time is required, so that the overall processing throughput is reduced, and in addition, the measurement result is often delayed from the process, so that timely feedback and process control are difficult. In recent Overlay Virtual Metrology (OVM, virtual metrology of Overlay error) studies have attempted to implement Overlay prediction of unmeasured wafers through historical process data, equipment parameters, reference wafer measurement information, and the like. However, due to the large difference between Lot (production Lot), the accuracy of the existing Lot-based statistical model in predicting single Wafer is insufficient, the Wafer is large in difference, the noise of a machine is complex, the modeling of high-dimensional fluctuation cannot be realized by the traditional linear fitting method, in addition, the characteristic utilization of the existing Overlay prediction method is insufficient, the sufficient modeling of a space mode is lacking in a characteristic engineering level, the generalization capability is insufficient, and the accurate Overlay prediction cannot be realized. Disclosure of Invention The technical problem to be solved by the present disclosure is to provide a training method of an Overlay error prediction model and an Overlay error prediction method, in order to overcome the defect that accurate Overlay prediction cannot be realized in the prior art. The technical problems are solved by the following technical scheme: According to a first aspect of the present disclosure, there is provided a training method of an overlay error prediction model, the training method comprising: basic parameter data and sample alignment error data of a plurality of sample wafers are obtained; performing polynomial decomposition on the sample alignment error data to obtain a plurality of sample alignment error components with at least two different orders; Constructing corresponding sample data sets under different orders based on the basic parameter data and the sample alignment error component; And taking the basic parameter data in the sample data set as input, taking the sample alignment error component as output, and respectively performing model training by adopting the sample data sets corresponding to different orders to obtain alignment error prediction models corresponding to different orders. Optionally, the step of constructing a corresponding sample data set at different orders based on the base parameter data and the sample overlay error component includes: extracting sample measurement features of the sample wafer based on the base parameter data; according to the sample measurement characteristics and the sample alignment error components, constructing sample data corresponding to the same sample wafer under different orders; And obtaining the corresponding sample data set under different orders based on the sample data corresponding to the plurality of different sample wafers under the same order. Optionally, the overlay error prediction model includes a first overlay error prediction model and a second overlay error prediction model; The step of taking the basic parameter data in the sample data set as input, taking the sample overlay error component as output, and respectively performing model training by adopting the sample data sets corresponding to different orders to obtain an overlay error prediction model corresponding to different orders comprises the following steps: taking the sample measurement characteristics in the sample data set as input, taking the sample alignment error components as output, and respectively training a first preset model by adopting the sample data sets corresponding to different orders to obtain the first alignment error prediction models corresponding to different orders; And/or the number of the groups of groups, Obtaining the similarity of the measurement characteristics of the sample wafer in the sample data set and the reference measurement characteristics corresponding to the reference wafer in the preset reference data set, and obtaining the similarity characteristics of the sample wafer in the sample data set; Taking the similarity characteristics in the sample data set as input, taking the sample alignment error components as output, and respectively training a second preset model by