CN113139702A - Method and system for predicting work-in-process quantity - Google Patents
Method and system for predicting work-in-process quantity Download PDFInfo
- Publication number
- CN113139702A CN113139702A CN202110591292.0A CN202110591292A CN113139702A CN 113139702 A CN113139702 A CN 113139702A CN 202110591292 A CN202110591292 A CN 202110591292A CN 113139702 A CN113139702 A CN 113139702A
- Authority
- CN
- China
- Prior art keywords
- data
- work
- correlation coefficient
- quantity data
- input feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 268
- 239000004065 semiconductor Substances 0.000 claims abstract description 42
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 26
- 238000004140 cleaning Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000012216 screening Methods 0.000 claims description 42
- 238000012545 processing Methods 0.000 claims description 36
- 238000007781 pre-processing Methods 0.000 claims description 17
- 238000013507 mapping Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 abstract description 5
- 238000004519 manufacturing process Methods 0.000 description 12
- 238000013528 artificial neural network Methods 0.000 description 9
- 230000008901 benefit Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000004088 simulation Methods 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 3
- 235000012431 wafers Nutrition 0.000 description 3
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- Databases & Information Systems (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Manufacturing & Machinery (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Factory Administration (AREA)
Abstract
The method for predicting the quantity of work-in-process comprises the following steps: acquiring a plurality of work-in-process quantity data and a plurality of corresponding chip yield data in a semiconductor process route; cleaning the data set by taking the data of the quantity of the plurality of products in process and the data of the yield of the plurality of chips as the data set so that the data set meets the preset requirement of a semiconductor process route; obtaining a plurality of input feature groups and a plurality of corresponding output feature groups in a semiconductor process route; extracting a training sample set by using a first preset algorithm by taking a plurality of input feature groups and a plurality of output feature groups as sample sets so as to construct a prediction model; work-in-process quantity data is calculated from an average of a portion of the data in the plurality of input feature sets and based on a predictive model. Compared with the prior art, the method and the device can remove the data of the quantity of the products in process and the corresponding chip yield data with high abnormal rate in the data set by cleaning the data set, thereby improving the accuracy and the stability of the prediction model.
Description
Technical Field
The present invention relates to the field of semiconductor technologies, and in particular, to a method and a system for predicting the number of products in process.
Background
At present, a semiconductor manufacturing factory generally uses a neural network technology to build a prediction model based on the neural network technology according to the relation between the historical production factors such as the yield of chips, the available time rate of machines, the required processing time of each Process station, the number of machines working, and the like, and the WIP (Work In Process, the number of chips In Process). The average data in the last week or a period of time, including the average data of the yield of the chip, the available time rate of the machine, the average data of the processing time required by each process station and the average data of the working number of the machine, are input into the constructed prediction model as predicted values to obtain the maximum value of the accumulation benefits of the process capacity of each device on the number of products.
Because the neural network has a complex nonlinear relation and the data abnormal rate is higher, the constructed prediction model is extremely difficult to select the point of the maximum value of the accumulation benefit of the process capacity of each equipment in the product quantity, a large amount of manual intervention means is needed to ensure the accuracy of the maximum value of the accumulation benefit of the process capacity of each equipment in the product quantity, and the prediction model cannot achieve the purpose of automation; a prediction model constructed based on a neural network technology needs a prediction value with a large base number as an input quantity, so that a step of data cleaning is not carried out, and the deviation of the maximum value of the accumulation benefit of the process capacity of each output device in the product quantity is large due to the occurrence of heterogeneous data in the prediction value. For a plant with a single product type and average production line capacity, the accuracy of the maximum stacking benefit of the number of products in each equipment process capability output by the prediction model constructed based on the neural network technology is high, but the accuracy of the prediction value is relatively depended on, and for a plant with multiple product types and unstable production line, the prediction model constructed based on the neural network technology is obviously not applicable.
Disclosure of Invention
The invention aims to provide a method and a system for predicting the quantity of work-in-process, one aim of the invention is to solve the problem that the accuracy of the conventional prediction model for predicting the quantity of work-in-process is low, and the other aim of the invention is to solve the problem that the conventional prediction model is not suitable for factories with various product types and unstable production lines.
In order to solve the above technical problems, according to an aspect of the present invention, a method for predicting quantity of work in process is provided, which is applied to a semiconductor process route, and includes:
acquiring a plurality of work-in-process quantity data in a semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data one to one;
cleaning the data set by taking the plurality of work-in-process quantity data and the corresponding plurality of chip yield data as the data set so that the data set meets the preset requirements of a semiconductor process route;
acquiring a plurality of input feature groups and a plurality of output feature groups corresponding to the input feature groups one by one in a semiconductor process route; the input feature set comprises processed quantity data, available machine time rate data, processing time data of each process station and available machine quantity data of the cleaned data set, and the output feature set comprises chip yield data of the cleaned data set;
taking the obtained multiple input feature groups and the corresponding multiple output feature groups as a sample set, and extracting and training the sample set by using a first preset algorithm so as to construct a prediction model;
and calculating the quantity data of the work-in-process based on the prediction model according to the average value of the available time rate data of the machine stations in the plurality of input feature groups, the average value of the processing time data of each process station and the average value of the available quantity data of the machine stations.
Optionally, the step of cleaning the data set with a plurality of the work-in-process quantity data and a plurality of the corresponding chip yield data as the data set so that the data set meets the preset requirements of the semiconductor process route includes:
extracting a mapping relation of the training data set by using a second preset algorithm, and fitting in a coordinate system to obtain a screening function; wherein the horizontal axis represents the work-in-process quantity data and the vertical axis represents the chip yield data;
respectively calculating the longitudinal axis deviation between a plurality of coordinate points simulated in the coordinate system by the data set and the screening function; if the absolute value of the deviation of the longitudinal axis is larger than the expected deviation, deleting the quantity data of the products in process and the yield data of the chips corresponding to the coordinate points; and if the absolute value of the deviation of the longitudinal axis is less than or equal to the expected deviation, storing the quantity data of the work-in-process and the chip yield data corresponding to the coordinate points.
Optionally, the second preset algorithm includes a least square method.
Optionally, after the step of fitting the coordinate system to obtain the screening function, the method for predicting the quantity of work-in-process further includes: comparing the absolute value of the correlation coefficient of the screening function with a preset expected correlation coefficient; if the absolute value of the correlation coefficient of the screening function is greater than or equal to the expected correlation coefficient, retaining the screening function; if the absolute value of the correlation coefficient of the screening function is smaller than the expected correlation coefficient, adjusting the data set and extracting the data set after training adjustment until the absolute value of the correlation coefficient of the screening function after fitting is larger than or equal to the expected correlation coefficient.
Optionally, the first preset algorithm includes polynomial regression; the step of calculating the work-in-process quantity data based on the prediction model according to the average value of the machine available time rate data, the average value of the processing time data of each process station and the average value of the machine available quantity data in the plurality of input feature groups comprises the following steps:
converting the prediction model into a prediction regression function of the output feature group about the quantity data of the work-in-process based on an average value of machine available time rate data, an average value of processing time data of each process station and an average value of machine available quantity data in a plurality of input feature groups;
solving a partial derivative of the predictive regression function with respect to the work-in-process quantity data, and calculating the work-in-process quantity data based on an extreme value of the partial derivative.
Optionally, the method for predicting the quantity of work-in-process further includes: and adjusting the weight factor of the work-in-process quantity data, the weight factor of the machine available time rate data, the weight factor of the processing time data of each process station and the weight factor of the machine available quantity data of the input feature group of the prediction model.
Based on another aspect of the present invention, the present invention further provides a work-in-process quantity prediction system applied to a semiconductor process route, comprising:
the system comprises a first data acquisition module, a second data acquisition module and a third data acquisition module, wherein the first data acquisition module is used for acquiring a plurality of work-in-process quantity data in a semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data one to one;
the data cleaning module is used for cleaning the data set by taking a plurality of the work-in-process quantity data and a plurality of corresponding chip yield data as data sets so that the data sets meet the preset requirements of a semiconductor process route;
the second data acquisition module is used for acquiring a plurality of input feature groups and a plurality of output feature groups corresponding to the input feature groups in a semiconductor process route one by one; the input feature set comprises processed quantity data, available machine time rate data, processing time data of each process station and available machine quantity data of the cleaned data set, and the output feature set comprises chip yield data of the cleaned data set;
the model construction module is used for taking the obtained multiple input feature groups and the corresponding multiple output feature groups as a sample set, extracting and training the sample set by utilizing a first preset algorithm, and constructing a prediction model;
and the data prediction module is used for calculating the quantity data of the work-in-process according to the average value of the machine available time rate data, the average value of the processing time data of each process station and the average value of the machine available quantity data in the plurality of input feature groups and based on the prediction model.
Optionally, the data cleansing module includes:
the first preprocessing unit is used for extracting a mapping relation of the training data set by using a second preset algorithm and fitting the mapping relation in a coordinate system to obtain a screening function; wherein the horizontal axis represents the work-in-process quantity data and the vertical axis represents the chip yield data;
a second preprocessing unit, configured to respectively calculate longitudinal axis deviations between a plurality of coordinate points of the data set simulated in the coordinate system and the screening function; if the absolute value of the deviation of the longitudinal axis is larger than the expected deviation, deleting the quantity data of the products in process and the yield data of the chips corresponding to the coordinate points; and if the absolute value of the deviation of the longitudinal axis is less than or equal to the expected deviation, storing the quantity data of the work-in-process and the chip yield data corresponding to the coordinate points.
Optionally, the data cleansing module further includes a third preprocessing unit, configured to compare an absolute value of a correlation coefficient of the screening function with a preset expected correlation coefficient; the third preprocessing unit is configured to retain the filter function if the absolute value of the correlation coefficient of the filter function is greater than or equal to the expected correlation coefficient; if the absolute value of the correlation coefficient of the filter function is smaller than the expected correlation coefficient, the third preprocessing unit is configured to adjust the data set and extract the training-adjusted data set until the absolute value of the correlation coefficient of the filter function after fitting is greater than or equal to the expected correlation coefficient.
Optionally, the first preset algorithm includes polynomial regression; the data prediction module comprises;
a function building unit, configured to convert the prediction model into a prediction regression function of the output feature group with respect to the work-in-process quantity data based on an average value of machine available time rate data, an average value of processing time data of each process station, and an average value of machine available quantity data in the plurality of input feature groups;
a data calculation unit for solving a partial derivative of the predictive regression function with respect to the work-in-process quantity data and calculating the work-in-process quantity data based on an extreme value of the partial derivative.
Optionally, the data prediction module further includes a weight adjustment unit, configured to adjust a weighting factor of the work-in-process quantity data, a weighting factor of the machine-available time rate data, a weighting factor of the processing time data of each process station, and a weighting factor of the machine-available quantity data of the input feature set of the prediction model.
In summary, the method for predicting the quantity of work-in-process applied to the semiconductor process route provided by the invention comprises the following steps: acquiring a plurality of work-in-process quantity data in a semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data one to one; cleaning the data set by taking the plurality of work-in-process quantity data and the corresponding plurality of chip yield data as the data set, so that the data set meets the preset requirements of a semiconductor process route; acquiring a plurality of input feature groups and a plurality of output feature groups corresponding to the input feature groups one to one in a semiconductor process route; inputting the quantity data of the processed products of the cleaned data set, the available time rate data of the machine, the processing time data of each process station and the available quantity data of the machine in the characteristic set, and outputting the chip yield data of the cleaned data set in the characteristic set; extracting a training sample set by using the obtained multiple input feature groups and the corresponding multiple output feature groups as a sample set by using a first preset algorithm so as to construct a prediction model; and calculating the quantity data of the work-in-process based on the prediction model according to the average value of the available time rate data of the machines, the average value of the processing time data of each process station and the average value of the available quantity data of the machines in the plurality of input feature groups. Compared with the prior art, the data set is cleaned to meet the preset requirements of a semiconductor process route, and the data of the quantity of work-in-process products with high abnormal rate and the corresponding chip yield data in the data set can be removed, so that the accuracy and the stability of a prediction model are improved; the prediction model configured by the input feature group and the output feature group has automatic calculation capability, is suitable for factories with various product types and unstable production lines, and increases the factory capacity.
Drawings
It will be appreciated by those skilled in the art that the drawings are provided for a better understanding of the invention and do not constitute any limitation to the scope of the invention. Wherein:
FIG. 1 is a flow chart of a method for predicting work-in-process quantities in accordance with an embodiment of the present invention;
FIG. 2 is a schematic diagram of a work in process quantity prediction system according to an embodiment of the present invention;
FIG. 3 is a simulation diagram of a prior art work-in-process quantity prediction method;
FIG. 4 is a simulation diagram of a method for predicting work-in-process quantities in accordance with an embodiment of the present invention.
Detailed Description
To further clarify the objects, advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. It is to be noted that the drawings are in greatly simplified form and are not to scale, but are merely intended to facilitate and clarify the explanation of the embodiments of the present invention. Further, the structures illustrated in the drawings are often part of actual structures. In particular, the drawings may have different emphasis points and may sometimes be scaled differently.
As used in this application, the singular forms "a", "an" and "the" include plural referents, the term "or" is generally employed in a sense including "and/or," the terms "a" and "an" are generally employed in a sense including "at least one," the terms "at least two" are generally employed in a sense including "two or more," and the terms "first", "second" and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicit to the number of technical features indicated. Thus, features defined as "first", "second" and "third" may explicitly or implicitly include one or at least two of the features, "one end" and "the other end" and "proximal end" and "distal end" generally refer to the corresponding two parts, which include not only the end points, but also the terms "mounted", "connected" and "connected" should be understood broadly, e.g., as a fixed connection, as a detachable connection, or as an integral part; can be mechanically or electrically connected; either directly or indirectly through intervening media, either internally or in any other relationship. Furthermore, as used in the present invention, the disposition of an element with another element generally only means that there is a connection, coupling, fit or driving relationship between the two elements, and the connection, coupling, fit or driving relationship between the two elements may be direct or indirect through intermediate elements, and cannot be understood as indicating or implying any spatial positional relationship between the two elements, i.e., an element may be in any orientation inside, outside, above, below or to one side of another element, unless the content clearly indicates otherwise. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
The invention provides a method and a system for predicting the quantity of work-in-process (WIP), which aims to solve the problem that the conventional prediction model has low accuracy in predicting the quantity of WIP, and aims to solve the problem that the conventional prediction model is not suitable for factories with various product types and unstable production lines.
The method and system for predicting the quantity of work in process according to the present embodiment will be described with reference to the drawings. FIG. 1 is a flowchart of a work-in-process quantity prediction method according to an embodiment of the present invention, and FIG. 2 is a schematic diagram of a work-in-process quantity prediction system according to an embodiment of the present invention.
The present embodiment provides a method for predicting work-in-process quantity applied to a semiconductor process route, including the steps of S1: a plurality of work-in-process quantity data in a semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data in a one-to-one mode are obtained.
Accordingly, the present embodiment provides a work-in-process quantity prediction system applied to a semiconductor process route, which includes a first data acquisition module for acquiring a plurality of work-in-process quantity data in the semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data one to one.
It is understood that Work In Process (WIP) refers to the input of a wafer into a chip product, and a significant number of wafers are registered In a FAB (factory, plant), collectively referred to as Work In Process (npl) In the factory. The yield of chips is related to the quantity of work-in-process, and the quantity of wafers registered in a factory needs to be in an optimal value, so that the quantity of the yield of chips is optimal.
The method for predicting the work-in-process quantity comprises the step two S2: and cleaning the data set by taking the plurality of work-in-process quantity data and the corresponding plurality of chip yield data as the data set so that the data set meets the preset requirements of a semiconductor process route.
Specifically, step two S2 may include:
s21: extracting a mapping relation of the training data set by using a second preset algorithm, and fitting in a coordinate system to obtain a screening function; wherein the horizontal axis represents the work-in-process quantity data and the vertical axis represents the chip yield data;
s22: respectively calculating the longitudinal axis deviation between a plurality of coordinate points simulated in the coordinate system by the data set and the screening function; if the absolute value of the deviation of the longitudinal axis is larger than the expected deviation, deleting the quantity data of the products in process and the yield data of the chips corresponding to the coordinate points; and if the absolute value of the deviation of the longitudinal axis is less than or equal to the expected deviation, storing the quantity data of the work-in-process and the chip yield data corresponding to the coordinate points.
Accordingly, the work-in-process quantity prediction system comprises a data cleaning module, wherein the data cleaning module is used for cleaning a data set by taking a plurality of work-in-process quantity data and a corresponding plurality of chip yield data as the data set, so that the data set meets the preset requirements of a semiconductor process route.
Further, the data cleaning module comprises a first preprocessing unit and a second preprocessing unit, wherein the first preprocessing unit is used for extracting a mapping relation of the training data set by using a second preset algorithm and fitting the mapping relation in a coordinate system to obtain a screening function; wherein the horizontal axis represents the work-in-process quantity data and the vertical axis represents the chip yield data;
the second preprocessing unit is used for respectively calculating the longitudinal axis deviation between a plurality of coordinate points simulated in the coordinate system by the data set and the screening function; if the absolute value of the deviation of the longitudinal axis is larger than the expected deviation, deleting the quantity data of the products in process and the yield data of the chips corresponding to the coordinate points; and if the absolute value of the deviation of the longitudinal axis is less than or equal to the expected deviation, storing the quantity data of the work-in-process and the chip yield data corresponding to the coordinate points.
Understandably, the coordinate simulation of the data set in the coordinate system is recorded as (A, B), the coordinate of the screening function on the vertical axis at the coordinate of the horizontal axis is A is B ', and the absolute value of the deviation of the vertical axis is equal to the absolute value of (B-B'); here, the expected deviation is set to be a positive number, and for a specific value thereof, those skilled in the art may configure the deviation accordingly according to actual situations, which are not illustrated here. Through the data cleaning steps and the data cleaning module configured above, data with high abnormal rate in the data set, namely data with too large deviation with the screening function, can be removed, so that the precision of the data set is ensured, and the accuracy and the stability of the constructed prediction model are higher.
In an exemplary embodiment, the second predetermined algorithm is a least squares method. The data set includes m WIP data and m corresponding chip yield data, and the data set is a characteristic of the chip yield data with respect to the WIP data and can be written as (x)i,yi) (i ═ 1,2, 3 … …, m), the screening function can be derived based on the principle of least squares:
wherein, theta (theta)0,θ1,θ2,θ3……θn) Are parameters.
Conversion to finding a set of theta (theta)0,θ1,θ2,θ3……θn) Minimizing the sum of squares of the residuals of the screening function with respect to the data set, i.e. solving:
For convenience of description, the description is made hereinFor theta (theta)0,θ1,θ2,θ3……θn) The solution is usually made by successively finding g (theta) with respect to theta0,θ1,θ2,θ3……θnPartial derivatives of, i.e.Then, each partial derivative is zero, and the combined solution equation set is used for solving to obtain theta (theta)0,θ1,θ2,θ3……θn) Then, a screening function h (x) is obtainedi)。
Further, the vertical axis deviation is | h ((x)i)-yiLet D be the expected deviation, compare | h (x) in turn1)-y1|,|h((x2)-y2|,|h(x3)-y3|……|h(xm)-ymAnd | and D, so as to determine the data to be eliminated in the data set.
The method for predicting the quantity of work-in-process includes the third step S3: acquiring a plurality of input feature groups and a plurality of output feature groups corresponding to the input feature groups one by one in a semiconductor process route; the input feature set comprises processed quantity data, available machine time rate data, processing time data of each process station and available machine quantity data of the cleaned data set, and the output feature set comprises chip yield data of the cleaned data set;
correspondingly, the predicting system for the quantity in process comprises a second data acquisition module, a second data acquisition module and a second data acquisition module, wherein the second data acquisition module is used for acquiring a plurality of input feature groups in a semiconductor process route and a plurality of output feature groups corresponding to the input feature groups in a one-to-one mode; the input feature set comprises work-in-process quantity data, machine-usable time rate data, processing time data of each process station and machine-usable quantity data of the cleaned data set, and the output feature set comprises chip yield data of the cleaned data set.
Preferably, after the step of fitting the coordinate system to obtain the screening function, the method for predicting the quantity of work-in-process further includes: comparing the absolute value of the correlation coefficient of the screening function with a preset expected correlation coefficient; if the absolute value of the correlation coefficient of the screening function is greater than or equal to the expected correlation coefficient, retaining the screening function; if the absolute value of the correlation coefficient of the screening function is smaller than the expected correlation coefficient, adjusting the data set and extracting the data set after training adjustment until the absolute value of the correlation coefficient of the screening function after fitting is larger than or equal to the expected correlation coefficient.
Correspondingly, the prediction system of the work-in-process quantity comprises a third preprocessing unit for comparing the absolute value of the correlation coefficient of the screening function with a preset expected correlation coefficient; the third preprocessing unit is configured to retain the filter function if the absolute value of the correlation coefficient of the filter function is greater than or equal to the expected correlation coefficient; if the absolute value of the correlation coefficient of the filter function is smaller than the expected correlation coefficient, the third preprocessing unit is configured to adjust the data set and extract the training-adjusted data set until the absolute value of the correlation coefficient of the filter function after fitting is greater than or equal to the expected correlation coefficient.
It can be understood that the correlation coefficient of the screening function is used to check the fitting degree of the screening function to the data set, and as known in the art, the square of the correlation coefficient is a decision coefficient, also called a decision coefficient or goodness of fit, and is used to check the fitting degree of the screening function, and the value range of the fitting degree is between 0 and 1, including 0 and 1, and the greater the goodness of fit is, the better the fitting degree of the screening function to the data set is, otherwise, the worse the fitting degree is. For the calculation method of the goodness of fit, this embodiment will not be described, and those skilled in the art can learn from the prior art. In this embodiment, the comparison between the correlation coefficient of the filtering function and the expected correlation coefficient is configured to determine whether to refit the filtering function, so that the fitted filtering function is the function with the optimal fitting degree with respect to the data set, and the accuracy of the data set is ensured, thereby ensuring that the complex correlation coefficient of the prediction model constructed later is the maximum. The setting of the expected correlation coefficient is not limited in this embodiment, and those skilled in the art can set the expected correlation coefficient according to the actual production line condition, and in a more preferred embodiment, the expected correlation coefficient is set between 0.7 and 0.9, for example, may be 0.8.
The method for predicting the quantity of work-in-process includes the step four S4: taking the obtained multiple input feature groups and the corresponding multiple output feature groups as a sample set, and extracting and training the sample set by using a first preset algorithm so as to construct a prediction model;
correspondingly, the work-in-process quantity prediction system comprises a model construction module, wherein the model construction module is used for taking the obtained plurality of input feature groups and the corresponding plurality of output feature groups as a sample set, and extracting and training the sample set by utilizing a first preset algorithm so as to construct a prediction model.
In the embodiment, the input feature group and the input feature group are used for establishing the prediction model based on the first preset algorithm, and compared with the prior art, a plurality of independent variables, namely, the number data of the products in process, the available time rate data of the machine table, the processing time data of each process station and the available number data of the machine table are considered, so that the established prediction model has higher accuracy and better stability, and is suitable for factories with various product types and unstable production lines.
The method for predicting the quantity of work-in-process comprises the following steps of S5: and calculating the quantity data of the work-in-process based on the prediction model according to the average value of the available time rate data of the machine stations in the plurality of input feature groups, the average value of the processing time data of each process station and the average value of the available quantity data of the machine stations.
Further, the first preset algorithm comprises polynomial regression; step S5 specifically includes:
s51: converting the prediction model into a prediction regression function of the output feature group with respect to the work-in-process quantity data based on an average value of machine available time rate data, an average value of processing time data of each process station, and an average value of machine available quantity data in the plurality of input feature groups
S52: solving a partial derivative of the predictive regression function with respect to the work-in-process quantity data, and calculating the work-in-process quantity data based on an extreme value of the partial derivative.
Accordingly, the work-in-process quantity prediction system comprises a data prediction module, and the data prediction module is used for calculating the work-in-process quantity data according to the average value of the machine available time rate data, the average value of the processing time data of each process station and the average value of the machine available quantity data in a plurality of input feature groups and based on the prediction model.
Further, the first preset algorithm comprises polynomial regression; the data prediction module comprises;
a function building unit, configured to convert the prediction model into a prediction regression function of the output feature group with respect to the work-in-process quantity data based on an average value of machine available time rate data, an average value of processing time data of each process station, and an average value of machine available quantity data in the plurality of input feature groups;
a data calculation unit for solving a partial derivative of the predictive regression function with respect to the work-in-process quantity data and calculating the work-in-process quantity data based on an extreme value of the partial derivative.
It can be understood that polynomial regression is one of linear regression, and a polynomial regression analysis method between a dependent variable and a plurality of independent variables is researched, and the biggest advantage of the polynomial is that a real measuring point is approximated by increasing high-order terms of the independent variables.
According to the prior art, the unary m-degree polynomial regression equation is:
f(x)=b0+b1x+b2x2+b3x3+…+bnxn
the binary quadratic polynomial regression equation is:
the multiple independent variables of the embodiment include work in process quantity data, machine available time rate data, processing time data of each process station and machine available quantity data, and the prediction model established according to polynomial regression is a quaternary m-th polynomial regression equation. In practice, the problem can be solved by a univariate m-degree polynomial regression equation, and the problem is rather complicated by considering a plurality of independent variables, based on which, the embodiment inputs the calculated average value of the machine available time rate data, the average value of the processing time data of each process station and the average value of the machine available quantity data into the independent variables in the corresponding prediction model, so that the prediction model is converted into a univariate m-degree polynomial regression equation of the chip yield data relative to the work-in-process quantity data, namely, the prediction regression function. The partial derivative of the predictive regression function with respect to the work-in-process quantity data is further solved, and the work-in-process quantity data at this time is solved with the partial derivative being 0. For the coefficients (parameters, b (b) in the prediction regression function0,b1,b2,b3……bn) Can be solved according to the value range of the quantity data of the products in process in the input characteristic group by combining the least square method partial derivative solving parameter way, which is not described here; or when a prediction model is constructed, the prediction regression function can be generated by setting independent variable related parameters according to the values of each data in the input feature group and substituting the average value of the available time rate data of the machine, the average value of the processing time data of each process station and the average value of the available quantity data of the machine. For factories with various product types and unstable production lines, the yield of chips can be predicted to be optimal when the quantity of products is the same, so that the capacity of the factories is improved. In addition, compared with the neural network in the prior art, the polynomial regression adopted by the embodiment has better algorithm selection aspect and higher accuracy of the constructed prediction model.
Preferably, the method for predicting the quantity of work-in-process further comprises: and adjusting the weight factor of the work-in-process quantity data, the weight factor of the machine available time rate data, the weight factor of the processing time data of each process station and the weight factor of the machine available quantity data of the input feature group of the prediction model.
Correspondingly, the data prediction module further comprises a weight adjusting unit for adjusting the weighting factor of the work-in-process quantity data, the weighting factor of the machine-available time rate data, the weighting factor of the processing time data of each process station and the weighting factor of the machine-available quantity data of the input feature set of the prediction model.
And each weight factor is adjusted, so that the constructed prediction model is more accurate, the fitting degree of the transformed prediction regression function on the years sample set is better, and the characteristic relation between the chip yield and the quantity of products in process can be reflected more accurately. Specifically, in the embodiment, the work-in-process quantity data having the largest relationship with the chip yield data is adjusted to be larger, and the weighting factor of the machine-usable time rate data, the weighting factor of the processing time data of each process station, and the weighting factor of the machine-usable quantity data are adjusted to be larger, smaller, or kept unchanged according to the actual situation.
Referring to fig. 3 and 4, fig. 3 is a simulation diagram of a work-in-process quantity prediction method (based on neural network technology) in the prior art, and fig. 4 is a simulation diagram of a work-in-process quantity prediction method in an embodiment of the present invention, in which a horizontal axis represents work-in-process quantity data and a vertical axis represents chip yield data; outout represents the curve fitted to the raw data (unprocessed), predictioned Output in FIG. 3 represents the curve fitted based on neural network techniques, and predictioned Output in FIG. 3 represents the fitted curve of the prediction method for work-in-process quantity of the present embodiment. As can be seen from fig. 3 and 4, the fitting degree of the product quantity data and the chip yield data is higher than that of the prior art, and the method is close to the actual production condition of the semiconductor process line, thereby being beneficial to maximizing the productivity benefit of a factory.
In summary, the method for predicting the quantity of work-in-process applied to the semiconductor process route provided by the invention comprises the following steps: acquiring a plurality of work-in-process quantity data in a semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data one to one; cleaning the data set by taking the plurality of work-in-process quantity data and the corresponding plurality of chip yield data as the data set, so that the data set meets the preset requirements of a semiconductor process route; acquiring a plurality of input feature groups and a plurality of output feature groups corresponding to the input feature groups one to one in a semiconductor process route; inputting the quantity data of the processed products of the cleaned data set, the available time rate data of the machine, the processing time data of each process station and the available quantity data of the machine in the characteristic set, and outputting the chip yield data of the cleaned data set in the characteristic set; extracting a training sample set by using the obtained multiple input feature groups and the corresponding multiple output feature groups as a sample set by using a first preset algorithm so as to construct a prediction model; and calculating the quantity data of the work-in-process based on the prediction model according to the average value of the available time rate data of the machines, the average value of the processing time data of each process station and the average value of the available quantity data of the machines in the plurality of input feature groups. Compared with the prior art, the data set is cleaned to meet the preset requirements of a semiconductor process route, and the data of the quantity of work-in-process products with high abnormal rate and the corresponding chip yield data in the data set can be removed, so that the accuracy and the stability of a prediction model are improved; the prediction model configured by the input feature group and the output feature group has automatic calculation capability, is suitable for factories with various product types and unstable production lines, and increases the factory capacity.
The above description is only for the purpose of describing the preferred embodiments of the present invention, and is not intended to limit the scope of the present invention, and any variations and modifications made by those skilled in the art according to the above disclosure are within the scope of the present invention.
Claims (11)
1. A method for predicting work-in-process quantity, which is applied to a semiconductor process route, is characterized by comprising the following steps:
acquiring a plurality of work-in-process quantity data in a semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data one to one;
cleaning the data set by taking the plurality of work-in-process quantity data and the corresponding plurality of chip yield data as the data set so that the data set meets the preset requirements of a semiconductor process route;
acquiring a plurality of input feature groups and a plurality of output feature groups corresponding to the input feature groups one by one in a semiconductor process route; the input feature set comprises processed quantity data, available machine time rate data, processing time data of each process station and available machine quantity data of the cleaned data set, and the output feature set comprises chip yield data of the cleaned data set;
taking the obtained multiple input feature groups and the corresponding multiple output feature groups as a sample set, and extracting and training the sample set by using a first preset algorithm so as to construct a prediction model;
and calculating the quantity data of the work-in-process based on the prediction model according to the average value of the available time rate data of the machine stations in the plurality of input feature groups, the average value of the processing time data of each process station and the average value of the available quantity data of the machine stations.
2. The method of claim 1, wherein the step of cleaning the data set with a plurality of the work-in-process quantity data and a corresponding plurality of the chip yield data as the data set such that the data set meets a predetermined requirement of a semiconductor process route comprises:
extracting a mapping relation of the training data set by using a second preset algorithm, and fitting in a coordinate system to obtain a screening function; wherein the horizontal axis represents the work-in-process quantity data and the vertical axis represents the chip yield data;
respectively calculating the longitudinal axis deviation between a plurality of coordinate points simulated in the coordinate system by the data set and the screening function; if the absolute value of the deviation of the longitudinal axis is larger than the expected deviation, deleting the quantity data of the products in process and the yield data of the chips corresponding to the coordinate points; and if the absolute value of the deviation of the longitudinal axis is less than or equal to the expected deviation, storing the quantity data of the work-in-process and the chip yield data corresponding to the coordinate points.
3. The method of predicting work-in-process quantity as set forth in claim 2, wherein said second predetermined algorithm comprises a least squares method.
4. The method of predicting work-in-process quantity as set forth in claim 2, wherein after the step of fitting a screening function in the coordinate system, the method further comprises: comparing the absolute value of the correlation coefficient of the screening function with a preset expected correlation coefficient; if the absolute value of the correlation coefficient of the screening function is greater than or equal to the expected correlation coefficient, retaining the screening function; if the absolute value of the correlation coefficient of the screening function is smaller than the expected correlation coefficient, adjusting the data set and extracting the data set after training adjustment until the absolute value of the correlation coefficient of the screening function after fitting is larger than or equal to the expected correlation coefficient.
5. The method of predicting work-in-process quantity according to claim 1, wherein the first predetermined algorithm comprises a polynomial regression; the step of calculating the work-in-process quantity data based on the prediction model according to the average value of the machine available time rate data, the average value of the processing time data of each process station and the average value of the machine available quantity data in the plurality of input feature groups comprises the following steps:
converting the prediction model into a prediction regression function of the output feature group about the quantity data of the work-in-process based on an average value of machine available time rate data, an average value of processing time data of each process station and an average value of machine available quantity data in a plurality of input feature groups;
solving a partial derivative of the predictive regression function with respect to the work-in-process quantity data, and calculating the work-in-process quantity data based on an extreme value of the partial derivative.
6. The method of predicting work-in-process quantity according to claim 5, further comprising: and adjusting the weight factor of the work-in-process quantity data, the weight factor of the machine available time rate data, the weight factor of the processing time data of each process station and the weight factor of the machine available quantity data of the input feature group of the prediction model.
7. A work-in-process quantity prediction system applied to a semiconductor process route is characterized by comprising the following components:
the system comprises a first data acquisition module, a second data acquisition module and a third data acquisition module, wherein the first data acquisition module is used for acquiring a plurality of work-in-process quantity data in a semiconductor process route and a plurality of chip yield data corresponding to the work-in-process quantity data one to one;
the data cleaning module is used for cleaning the data set by taking a plurality of the work-in-process quantity data and a plurality of corresponding chip yield data as data sets so that the data sets meet the preset requirements of a semiconductor process route;
the second data acquisition module is used for acquiring a plurality of input feature groups and a plurality of output feature groups corresponding to the input feature groups in a semiconductor process route one by one; the input feature set comprises processed quantity data, available machine time rate data, processing time data of each process station and available machine quantity data of the cleaned data set, and the output feature set comprises chip yield data of the cleaned data set;
the model construction module is used for taking the obtained multiple input feature groups and the corresponding multiple output feature groups as a sample set, extracting and training the sample set by utilizing a first preset algorithm, and constructing a prediction model;
and the data prediction module is used for calculating the quantity data of the work-in-process according to the average value of the machine available time rate data, the average value of the processing time data of each process station and the average value of the machine available quantity data in the plurality of input feature groups and based on the prediction model.
8. The work-in-process quantity prediction system of claim 7, wherein the data cleansing module comprises:
the first preprocessing unit is used for extracting a mapping relation of the training data set by using a second preset algorithm and fitting the mapping relation in a coordinate system to obtain a screening function; wherein the horizontal axis represents the work-in-process quantity data and the vertical axis represents the chip yield data;
a second preprocessing unit, configured to respectively calculate longitudinal axis deviations between a plurality of coordinate points of the data set simulated in the coordinate system and the screening function; if the absolute value of the deviation of the longitudinal axis is larger than the expected deviation, deleting the quantity data of the products in process and the yield data of the chips corresponding to the coordinate points; and if the absolute value of the deviation of the longitudinal axis is less than or equal to the expected deviation, storing the quantity data of the work-in-process and the chip yield data corresponding to the coordinate points.
9. The work-in-process quantity prediction system of claim 8, wherein the data cleansing module further comprises a third preprocessing unit for comparing an absolute value of a correlation coefficient of the screening function with a preset desired correlation coefficient; the third preprocessing unit is configured to retain the filter function if the absolute value of the correlation coefficient of the filter function is greater than or equal to the expected correlation coefficient; if the absolute value of the correlation coefficient of the filter function is smaller than the expected correlation coefficient, the third preprocessing unit is configured to adjust the data set and extract the training-adjusted data set until the absolute value of the correlation coefficient of the filter function after fitting is greater than or equal to the expected correlation coefficient.
10. The work-in-process quantity prediction system of claim 7, wherein the first predetermined algorithm comprises a polynomial regression; the data prediction module comprises:
a function building unit, configured to convert the prediction model into a prediction regression function of the output feature group with respect to the work-in-process quantity data based on an average value of machine available time rate data, an average value of processing time data of each process station, and an average value of machine available quantity data in the plurality of input feature groups;
a data calculation unit for solving a partial derivative of the predictive regression function with respect to the work-in-process quantity data and calculating the work-in-process quantity data based on an extreme value of the partial derivative.
11. The system of claim 10, wherein the data prediction module further comprises a weight adjustment unit configured to adjust a work-in-process quantity data weight factor, a machine-available time rate data weight factor, a process time data weight factor for each process site, and a machine-available quantity data weight factor for the input feature set of the prediction model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110591292.0A CN113139702B (en) | 2021-05-28 | 2021-05-28 | Method and system for predicting number of products in process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110591292.0A CN113139702B (en) | 2021-05-28 | 2021-05-28 | Method and system for predicting number of products in process |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113139702A true CN113139702A (en) | 2021-07-20 |
CN113139702B CN113139702B (en) | 2024-04-30 |
Family
ID=76815821
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110591292.0A Active CN113139702B (en) | 2021-05-28 | 2021-05-28 | Method and system for predicting number of products in process |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113139702B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103310285A (en) * | 2013-06-17 | 2013-09-18 | 同济大学 | Performance prediction method applicable to dynamic scheduling for semiconductor production line |
US20130346024A1 (en) * | 2012-06-22 | 2013-12-26 | National Cheng Kung University | Method for forecasting work-in-process output schedule and computer program product thereof |
CN105988424A (en) * | 2015-01-28 | 2016-10-05 | 中芯国际集成电路制造(上海)有限公司 | Semiconductor technology process information generation method and device |
CN106846805A (en) * | 2017-03-06 | 2017-06-13 | 南京多伦科技股份有限公司 | A kind of dynamic road grid traffic needing forecasting method and its system |
CN109034540A (en) * | 2018-06-29 | 2018-12-18 | 长安大学 | A kind of lathe serial arrangement dynamic prediction method based in article process stream |
CN115730697A (en) * | 2021-08-26 | 2023-03-03 | 力晶积成电子制造股份有限公司 | Method and system for estimating production schedule of semiconductor manufacturing process |
-
2021
- 2021-05-28 CN CN202110591292.0A patent/CN113139702B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130346024A1 (en) * | 2012-06-22 | 2013-12-26 | National Cheng Kung University | Method for forecasting work-in-process output schedule and computer program product thereof |
CN103310285A (en) * | 2013-06-17 | 2013-09-18 | 同济大学 | Performance prediction method applicable to dynamic scheduling for semiconductor production line |
CN105988424A (en) * | 2015-01-28 | 2016-10-05 | 中芯国际集成电路制造(上海)有限公司 | Semiconductor technology process information generation method and device |
CN106846805A (en) * | 2017-03-06 | 2017-06-13 | 南京多伦科技股份有限公司 | A kind of dynamic road grid traffic needing forecasting method and its system |
CN109034540A (en) * | 2018-06-29 | 2018-12-18 | 长安大学 | A kind of lathe serial arrangement dynamic prediction method based in article process stream |
CN115730697A (en) * | 2021-08-26 | 2023-03-03 | 力晶积成电子制造股份有限公司 | Method and system for estimating production schedule of semiconductor manufacturing process |
Also Published As
Publication number | Publication date |
---|---|
CN113139702B (en) | 2024-04-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4584295B2 (en) | Two-stage virtual measurement method | |
CN105789812B (en) | A kind of cavity body filter Automatic adjustment method and system | |
JPH07191737A (en) | Apparatus and method for control of factory | |
CN112711218A (en) | Industrial equipment data acquisition method | |
US20140135968A1 (en) | Method of maintaining a semiconductor production line | |
US20230359183A1 (en) | Method, system and non-transitory computer-readable medium for reducing work-in-process | |
CN115275420A (en) | Decommissioning system and method for power battery | |
CN112433472B (en) | Semiconductor production control method and control system | |
CN113139702A (en) | Method and system for predicting work-in-process quantity | |
CN117289668B (en) | Distributed speed reducer network cooperative control method, device, equipment and storage medium | |
CN115237056B (en) | Multi-tool rapid deviation rectifying method for industrial robot | |
CN109344448B (en) | fuzzy-FQD-based helical bevel gear shape collaborative manufacturing optimization method | |
CN112736380A (en) | Automatic tuning system, automatic tuning method and storage device of dielectric filter | |
CN111745465A (en) | Self-adaptive positioning method and positioning system for workpiece of numerical control machine tool | |
WO2010119939A1 (en) | Process parameter selection apparatus, process parameter selection method, process parameter selection program, and program storage medium | |
CN108943024A (en) | A kind of Six-DOF industrial robot absolute position error estimation method | |
CN111854822B (en) | Semiconductor process data correction method based on correlation entropy and deep neural network | |
CN112132428A (en) | Big data-based predictive control decision method for steam heating network steam distribution station | |
CN117170307B (en) | Multi-axis parallel machine tool error compensation method, device, equipment and storage medium | |
CN117086175B (en) | Deformation compensation control method and system for stamping die | |
CN117970857B (en) | Remote intelligent industrial control method and system | |
CN115946123B (en) | Multi-tool center point and zero point calibration method for industrial robot | |
CN116680833B (en) | Drilling, tapping and milling design image generation system based on big data sharing | |
CN117369560B (en) | Industrial automation intelligent control method and system | |
CN114221599B (en) | Alternating current motor speed regulating system and control method thereof |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |