CN-121995875-A - Chip epitaxial wafer picking production control method and system based on deep learning

CN121995875ACN 121995875 ACN121995875 ACN 121995875ACN-121995875-A

Abstract

The invention discloses a chip epitaxial wafer picking and production control method and system based on deep learning, comprising the steps of obtaining a formulated epitaxial wafer picking and production strategy; the epitaxial wafer picking and production strategy comprises a production plan, resource constraint and performance parameter range constraint corresponding to a chip product, epitaxial wafer IDs meeting the resource constraint in an epitaxial semi-finished product library are obtained, corresponding physical position information and/or production equipment information are obtained, the corresponding physical position information and/or the production equipment information are input into a trained output distribution prediction model to predict corresponding performance parameters, the epitaxial wafer IDs meeting the range constraint in the epitaxial wafer picking and production strategy are selected, epitaxial wafer IDs to be produced are screened out based on the production plan, and production is carried out on epitaxial wafers corresponding to the screened epitaxial wafer IDs based on resource constraint control. According to the invention, the epitaxial wafer picking production strategy is optimized by effectively integrating the accurate output distribution data captured by the subsequent process, and the accurate control of the chip output is realized.

Inventors

CAI WENBI
XU JIANXI
GAO YUCONG
HUANG YANG
LIU ZE
YAN ZIYI
GAO YUE
DING GUIGUANG

Assignees

泉州市三安集成电路有限公司
清华大学

Dates

Publication Date: 20260508
Application Date: 20251229

Claims (10)

1. The chip epitaxial wafer picking production control method based on deep learning is characterized by comprising the following steps of: Acquiring a formulated epitaxial wafer picking and production strategy, wherein the epitaxial wafer picking and production strategy comprises a production plan, resource constraint and performance parameter range constraint corresponding to a chip product; Obtaining an epitaxial wafer ID meeting resource constraint in an epitaxial semi-finished product library, and obtaining corresponding physical position information and/or production equipment information; inputting corresponding physical position information and/or production equipment information into a trained yield distribution prediction model to predict corresponding performance parameters; Selecting an epitaxial wafer ID with performance parameters meeting range constraint in an epitaxial wafer production strategy, and screening the epitaxial wafer ID to be produced based on a production plan; and based on resource constraint control, putting the epitaxial wafer corresponding to the screened epitaxial wafer ID into production.
2. The deep learning-based chip epitaxial wafer production control method of claim 1, wherein the training process of the yield distribution prediction model comprises the following steps: Acquiring collected historical data, wherein the historical data comprises production equipment information of an epitaxial wafer, physical position information of a chip on the epitaxial wafer and performance parameters of a chip product produced based on the physical position information of the chip on the epitaxial wafer; Performing cluster analysis on the epitaxial wafer based on performance parameters in the historical data, and classifying the analyzed data of the same chip product; Based on the classified data, physical position information and/or production equipment information are used as input, corresponding performance parameters are used as output, and the constructed yield distribution prediction model is trained to obtain a trained yield distribution prediction model.
3. The deep learning-based chip epitaxial wafer production control method of claim 2, wherein the epitaxial wafer is subjected to cluster analysis based on performance parameters, and data of the same chip product are classified, and the method specifically comprises the following steps: Acquiring absolute deviation values and relative deviation values of each performance parameter and each chip product target value; And if the absolute deviation value and the relative deviation value meet the set deviation threshold, judging that the performance parameters belong to the corresponding chip products, and classifying.
4. The deep learning-based chip epitaxial wafer production control method is characterized in that performance parameters stored in historical data are obtained through subsequent process testing, physical position information stored in the historical data or an epitaxial semi-finished product library is obtained through machine vision algorithm and/or image segmentation algorithm on acquired images, and production equipment information stored in the historical data or the epitaxial semi-finished product library is obtained through sensor acquisition.
5. The deep learning-based chip epitaxial wafer production control method of claim 1, wherein the resource constraints comprise an epitaxial structure and a fast measurement size.
6. The deep learning-based chip epitaxial wafer production control method is characterized in that the performance parameters comprise at least one of full-measurement light output power, full-measurement peak wavelength and full-measurement forward voltage, the physical position information comprises full-measurement coordinates and turns, and the production equipment information comprises an epitaxial machine number and an epitaxial wafer position number.
7. The deep learning-based chip epitaxial wafer production control method of claim 1, wherein the yield distribution prediction model comprises a deep learning model or a machine learning model.
8. The deep learning-based chip epitaxial wafer production control method of claim 1, further comprising: acquiring actual performance parameters acquired in the production process; And on-line adjustment is carried out on the yield distribution prediction model based on the actual performance parameters.
9. Chip epitaxy choosing piece production control device based on deep learning, which is characterized by comprising: The epitaxial wafer picking and production strategy comprises a production plan, a resource constraint and a performance parameter range constraint corresponding to a chip product; the epitaxial semi-finished product library information extraction module is used for acquiring epitaxial wafer IDs meeting resource constraint in the epitaxial semi-finished product library and acquiring physical position information of chips on corresponding epitaxial wafers and/or production equipment information of the epitaxial wafers; The performance parameter prediction module is used for inputting corresponding physical position information and/or production equipment information into the trained output distribution prediction model to predict corresponding performance parameters; The epitaxial wafer selecting module is used for selecting epitaxial wafer IDs with performance parameters meeting range constraint in an epitaxial wafer picking and production strategy, and screening epitaxial wafer IDs to be subjected to production on the basis of a production plan; and the production control module is used for controlling the production of the epitaxial wafers corresponding to the screened epitaxial wafer IDs based on resource constraint.
10. Chip epitaxy choosing piece production control system based on deep learning, which is characterized by comprising: The data acquisition equipment is used for acquiring data and sending the data to the cluster server, wherein the data acquisition equipment comprises production equipment information of the epitaxial wafer, physical position information of the chips on the epitaxial wafer and performance parameters of chip products produced based on the physical position information of the chips on the epitaxial wafer; The data processing and analyzing equipment is used for receiving the data sent by the data acquisition equipment, carrying out data preprocessing and cluster analysis, and then training the constructed output distribution prediction model to obtain a trained output distribution prediction model; The epitaxial wafer decision device is used for executing the method according to any one of claims 1-8 based on the output distribution prediction model trained by the data processing and analyzing device; The production execution and monitoring equipment acquires a production instruction sent by the epitaxial wafer choosing decision equipment, processes the screened epitaxial wafer into a chip finished product, and feeds data in the production process back to the data acquisition equipment.

Description

Chip epitaxial wafer picking production control method and system based on deep learning Technical Field The invention relates to the field of semiconductor chip manufacturing, in particular to a chip epitaxial wafer picking production control method and system based on deep learning. Background Currently, the semiconductor chip manufacturing industry is faced with multiple pressures for improving the production efficiency, reducing the cost and improving the quality of chip products, and the production decision accuracy of an epitaxial growth link is insufficient, so that the semiconductor chip manufacturing industry becomes one of key pain points for limiting the industry to break through bottlenecks. In the traditional production mode, the production decision of epitaxial growth depends on manual experience or a small amount of early data judgment, the inherent correlation between the epitaxial wafer characteristics and the final output distribution cannot be deeply mined, namely, the ' experience-driven ' mode is difficult to match with the process requirements of high precision and high complexity ' of semiconductor manufacturing, and the unbalance problem of ' excessive investment ' or ' insufficient investment ' of epitaxial production is directly caused. On one hand, part of the areas may be excessively put into epitaxial wafers, but the subsequent detection of subsequent processes (such as chip test and finished product quality inspection) shows that the chip yield of the areas is extremely low, so that not only is the waste of epitaxial materials and resources in equipment working hours caused, but also effective energy production is occupied, and on the other hand, some areas with high quality output potential (such as areas with strong process stability in the center of the epitaxial wafer) are insufficient in input due to early production pre-judgment errors, the capacity is not fully released, and economic benefit loss is indirectly caused. The subsequent process is used as a final detection gateway for chip manufacture, and can accurately capture multi-dimensional output distribution information (such as yield, performance parameters, defect types and the like of chips in different areas), and the data are essentially reverse calibration basis for epitaxial production decision-making. Therefore, how to effectively convert the accurate output distribution information acquired in the subsequent process into a 'picking and putting into production guiding scheme' of the epitaxial growth link, namely, through data association, definitely 'which areas need to be added with epitaxial wafers and which areas need to be reduced with putting in', thereby realizing the accurate control of epitaxial putting into production, finally achieving the aims of improving the quality of chip products and optimizing resource allocation, and becoming a core problem to be broken through urgently for semiconductor manufacturing enterprises. Disclosure of Invention The application aims to provide a chip epitaxial wafer picking and production control method and system based on deep learning aiming at the technical problems, and the epitaxial wafer picking and production strategy is optimized by effectively integrating accurate output distribution data captured in the subsequent working procedure, so that the accurate control of chip output is realized, the production efficiency is improved, the cost is reduced and the quality of chip products is improved. On the one hand, the chip epitaxy picking chip production control method based on deep learning comprises the following steps: Acquiring a formulated epitaxial wafer picking and production strategy, wherein the epitaxial wafer picking and production strategy comprises a production plan, resource constraint and performance parameter range constraint corresponding to a chip product; acquiring an epitaxial wafer ID meeting resource constraint in an epitaxial semi-finished product library, and acquiring physical position information of a chip on a corresponding epitaxial wafer and/or production equipment information of the epitaxial wafer; inputting corresponding physical position information and/or production equipment information into a trained yield distribution prediction model to predict corresponding performance parameters; Selecting an epitaxial wafer ID with performance parameters meeting range constraint in an epitaxial wafer production strategy, and screening the epitaxial wafer ID to be produced based on a production plan; and based on resource constraint control, putting the epitaxial wafer corresponding to the screened epitaxial wafer ID into production. Preferably, the training process of the yield distribution prediction model includes: Acquiring collected historical data, wherein the historical data comprises production equipment information of an epitaxial wafer, physical position information of a chip on the epitaxial wafer and performance parameters of a