CN103310285A - Performance prediction method applicable to dynamic scheduling for semiconductor production line - Google Patents
Performance prediction method applicable to dynamic scheduling for semiconductor production line Download PDFInfo
- Publication number
- CN103310285A CN103310285A CN2013102395010A CN201310239501A CN103310285A CN 103310285 A CN103310285 A CN 103310285A CN 2013102395010 A CN2013102395010 A CN 2013102395010A CN 201310239501 A CN201310239501 A CN 201310239501A CN 103310285 A CN103310285 A CN 103310285A
- Authority
- CN
- China
- Prior art keywords
- centerdot
- output
- prediction
- scheduling
- input
- 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.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- General Factory Administration (AREA)
Abstract
The invention discloses a performance prediction method applicable to dynamic scheduling for a semiconductor production line. An extreme learning machine (ELM) is applied to prediction and modeling in the performance prediction method. Feeding control and scheduling rules are considered in a unified manner in the method, short-term scheduling key performance indexes such as an equipment utilization rate and a movement step number are predicted on the basis of a real-time state of a system, and a foundation is provided for dynamic real-time scheduling. A novel feed-forward neural network of the ELM is introduced into the semiconductor manufacturing system, and a prediction model is built by the aid of available data of the production line. As shown by test results, ideal prediction results can be quickly acquired by the method implemented by the aid of the ELM, the method has obvious advantages and an obvious application prospect in the aspects of parameter selection and learning speed as compared with the traditional neural network modeling method, and a new idea is provided for online optimal control.
Description
Technical field
The invention belongs to field of semiconductor manufacture, especially a kind of Semiconductor Wafer Fabrication Scheduling system based on extreme learning machine.
Background technology
Semiconductor wafer production line is one of the most complicated production line of generally acknowledging, it has, and equipment is many, product category is many, can reentry, semi-manufacture are scrapped and heavy characteristics such as processing, machine failure, therefore brought great complicacy to scheduling and the control of production line.
A complete Semiconductor Wafer Fabrication Scheduling comprises feed intake control, Job Scheduling.Feed intake control for determining that material enters the speed of production system, and Job Scheduling then refers to competition is used the workpiece of certain equipment, when the free time appears in equipment, what decision-making therefrom to select the next workpiece that will process by.Because forecast model has the following dynamic behavior of display systems and ability, according to system's real-time status, at random provides following control strategy, the exporting change of object of observation performance index under Different Strategies.Performance prediction provides foundation and reference for control evaluation on the one hand, can represent simply from the angle of production management the quality of scheduling result, and then judge whether resulting scheduling scheme is feasible.On the other hand according to some uncertain factors, change such as the variation of equipment situation, customer demand, the uncertain factors such as variation of processing bottleneck, can greatly affect the state of production line, the degrade performance index, need to adjust according to uncertain factor the control strategy of production line this moment, guarantees the carrying out of scheduling scheme, reaches the requirement of real-time control.Therefore for the production system of a given resource structures and scheduling strategy, calculate its performance and be necessary.Instruct to produce significant and practical value by setting up forecast model, for efficient task scheduling provides support, guarantee the high-performance operation of service, thereby improve service quality, so that more scientific and carry out economically activity in production.
The prediction modeling method mainly contains pure computing method, emulation modelling method two class methods at present, yet for the semiconductor fabrication of complexity all multiple constraints and defective is arranged.At first, use traditional modeling method to set up and instruct the mathematical models of production run scheduling to become more and more difficult, and because hypothesis has reduced the actual complex degree, be difficult to again guarantee the precision of institute's established model; Secondly, emulation modelling method needs the working time of a large amount of time and fund and model long, and can't satisfy the needs of Dynamic Uncertain factor in the production scene being made quick effecting reaction.Make and to have produced in a large number from online data in the production run, wherein implied the mass efficient information of reflection actual schedule environmental quality and scheduling knowledge.Artificial neural network has self-organization, self-adaptation, parallel processing and the characteristic such as nonlinear, a kind of as data mining technology, useful information in the extraction mass data is described the intelligent behavior of cognition, Policy making and controlling, is widely used in the manufacturing industry control field.Yet the pace of learning of traditional neural network can not satisfy the requirement of high real-time, can't satisfy the needs that Dynamic Uncertain factor in the production scene is made quick effecting reaction and cause the important bottleneck that becomes its application, especially predicts for real-time online.
Extreme learning machine (ELM) is a kind of Novel learning algorithm that is simple and easy to usefulness, effective single hidden layer feedforward neural network.This algorithm is different from the proposition of traditional learning algorithm, this learning method is when guaranteeing that network has good Generalization Capability, improved dramatically the pace of learning of feedforward neural network, and avoided the many problems based on the Gradient Descent learning method, as local minimum, iterations too much, parameter selects responsive etc.Therefore the ELM algorithm progressively receives publicity, but is in developing stage now, has obtained application at aspects such as spectral analysis, lithology identification, power circuit construction, fingerprint recognitions of face.
Summary of the invention
Adverse effect during for above-mentioned technological deficiency and application the purpose of this invention is to provide a kind of limits of application learning machine, solves the performance prediction problem of Dynamic Schedule of Semiconductor Fabrication Line, guarantees feasibility and the validity of short term scheduling scheme.
Technical scheme of the present invention is as follows:
A kind of performance prediction method that is used for Dynamic Schedule of Semiconductor Fabrication Line based on extreme learning machine comprises:
(1) gathers the semiconductor production line historical data, set up training sample set and test sample book collection.
(2) will encode to feeding mode and the scheduling rule letter symbol of input, network can be accepted.For making the input data that identical dimension be arranged, input quantity is carried out normalized.For making the input data that identical dimension be arranged, input quantity is carried out normalized, in [0,1] interval, the normalization formula is [x-min (x with data limit
i)]/[max (x
i)-min (x
i)], wherein, x refers to input variable, i is sample number.
(3) adopt the extreme learning machine method to make up forecast model.Only need the hidden node number of the neural network determined for ELM, do not need to adjust the input weights of network and biasing and other parameters of hidden unit, adopt method of trial and error to choose suitable hidden node number.So suppose the training set for given N different samples
Wherein x is n dimension input variable, i.e. x
i∈ R
n, t is m dimension output variable, i.e. t
i∈ R
m, i is sample number; Activation function is g (x), and the hidden layer node number is
, then
A) obtain at random the initial input weight w
iWith biasing b
i, i=1 ...,
B) calculate hidden layer output matrix H.
C) calculate output weights β:
T=[t
1..., t
N]
T, wherein
It is the Moore-Penrose generalized inverse matrix of matrix H.
(4) network performance of performance test test sample forecast model, the renormalization that will predict the outcome obtains output valve o after processing
iWherein, i=1 ..., m, m refer to the dimension of output valve o, and m output is namely arranged; With the contrast of test sample book output valve, judge whether to satisfy accuracy requirement.It is benchmark that accuracy requirement can be adopted average relative error, for m output variable,
Wherein n is the test sample book number, and j is sample number, and i is the output variable numbering, and t is the output variable of expection.The average relative error result of test is compared with the relative error of expection, guarantee that maximum average relative error is less than or equal to prediction relative average error, i.e. precision of prediction in m the variable
The wherein accuracy requirement of δ ' for setting.
(5) if the precision of prediction of test result can meet the demands, set up the forecast model process and finish, obtain required forecast model; As not satisfying, then forward step (4) to, the hidden node number of the neural network that reselects is training pattern again.
Described historical data comprises input quantity and output quantity, and wherein input quantity is: feeding mode, scheduling rule and the real-time status buffer zone team leader of system, goods in process inventory WIP, production line produce mobile step number etc. in last dispatching cycle; Output quantity is system's short term scheduling performance index, comprises the amount of movement that produces in day throughput rate, plant factor, the team leader of bottleneck device queue and dispatching cycle.
The invention has the beneficial effects as follows:
The present invention is based on the extreme learning machine algorithm neural network prediction model has been set up in the performance under short-term load prediction of semiconductor production line, can in time react to the real-time change in the system, reduced the needs of reschedule.The present invention regards system as black box, and the modeling method that provides realizes the real-time online optimal control utilizing the obtainable historical data of production line and from online data, excavating wherein useful knowledge.The present invention considers that according to the semiconductor production line feature short-term factor comprises that buffer zone wait workpiece quantity, production line goods in process inventory, production line have produced mobile step number etc., and the performance under short-term load index comprises throughput rate, plant factor, queue length, mobile step number etc., reach the prediction Modeling Research of formation system is considered in the control that feeds intake of semi-conductive scheduling with Job Scheduling is unified.
Performance prediction modeling method efficient provided by the invention is high, real-time good, realization is convenient, and very practical dynamic Real-Time Scheduling is for on-line optimization control provides effective approach.
Description of drawings
Fig. 1 is semiconductor production line simplified model schematic diagram;
Fig. 2 is the process flow diagram of the realization ELM prediction modeling of embodiment of the invention Production Scheduling System;
Fig. 3 is the topology example figure that the embodiment of the invention is set up neural network.
Embodiment
For setting forth better the control method of dispatching system of the present invention, see also Fig. 1.
Fig. 1 is semiconductor production line simplified model schematic diagram, with the production technology classification, three device clusters is arranged, and totally five equipment are respectively diffusion facilities group, ion implantation device group and lithographic equipment group.The diffusion facilities group comprises the first diffusion facilities Ma, the second diffusion facilities Mb; The ion implantation device group comprises the first ion implantation device Mc, the second ion implantation device Md; The lithographic equipment group comprises lithographic equipment Me.Before each device cluster, be respectively equipped with the first buffer zone B_Mab, the second buffer zone B_Mcd and the 3rd buffer zone B_Me and be used for storing the workpiece information to be processed such as needs.By above-mentioned device cluster, can realize six procedures, comprising: diffusion, Implantation, photoetching, Implantation, diffusion, photoetching.Those skilled in the art as can be known, the equipment in the same device cluster can be chosen simultaneously or singlely be chosen work.
Particularly, Minifab simplifies the simple semiconductor production line model of coming according to the actual production line, as shown in the figure.Minifab produces the workpiece of three types, and its machining path is the same.The technological process of workpiece comprises six operations: operation one and operation five can be processed at Ma or Mb for diffusion; Operation two and operation four are Implantation, can process at equipment Mc or Md; Operation three and operation six are photoetching, and Me processes at equipment.Three device clusters all have reenterability.Equipment Ma and Mb are the parallel process equipments of many cards, and maximum manufacturing batch is 3 cards, only have the identical workpiece of operation just can organize batch parallel together processing.Ma and Mb are two equipment of complete interchangeable.Equipment Mc and Md are that single deck tape-recorder is criticized process equipment, and every equipment once can only be processed a card workpiece.Mc and Md also are two equipment of complete interchangeable.Equipment Me also criticizes process equipment for single deck tape-recorder.
Fig. 2 is the process flow diagram of the realization ELM prediction modeling of embodiment of the invention Production Scheduling System.
At first, in data sample, determine input variable (S101) and output variable (S102).Among the step S101, input variable mainly comprises feeding mode P, scheduling rule and system's real-time status.Scheduling rule comprises diffusion facilities group scheduling rule R
AbWith lithographic equipment group scheduling rule R
eReal-time status system of system real-time status is the current generation manufacturing system production status objectively to be described, and comprises production line goods in process inventory WIP, the first buffer zone B_M
AbBuffer zone team leader Q
Ab, the second buffer zone B_Mcd buffer zone team leader Q
Cd, the 3rd buffer zone B_M
eBuffer zone team leader Q
e, the total team leader Q ' of buffer zone, and the mobile step number M that within last dispatching cycle, produces of production line, wherein, M ' refers to the mobile total step number within last dispatching cycle of all silicon chips on the production line.
Wherein, feeding mode P has determined the kind of product and has arrived time and the speed of production line, state and performance on production line have direct impact, optionally, the feeding mode of the present embodiment is: Fixed Time Interval feeds intake (Conrin), Poisson feeds intake (Possion), be fixed on goods feed intake (Conwip).The workpieces to be processed such as scheduling rule decision distribute different processing priority.In the present embodiment, diffusion facilities group scheduling rule R
AbWith lithographic equipment group scheduling rule R
eOne or more in first-in first-out (FIFO), the earliest preferential (EDD), critical value (CR), the shortest residue process time (SRPT) at delivery date, the manufacturing cycle variance fluctuation level and smooth (FSVCT).The buffer zone team leader such as then represents at the Number of Jobs to be processed,
Among the step S102, output variable comprises throughput rate
The queue length of plant factor, bottleneck equipment and total amount of movement.
Throughput rate
Be lot(batch of machining in the unit interval) quantity, its value is higher, and the value that expression is created is higher, therefore it can directly reflect the quality of the day plan of feeding intake.Throughput rate
Wherein P ' is the total production in dispatching cycle, T
cBe time dispatching cycle.The time that the plant factor indication equipment is in machining state accounts for the ratio of its on time, is designated as U
i, i ∈ (a, b, c, d, e) wherein, the identifier of indication equipment, for example U
aThe plant factor of indication equipment Ma.Be described as
T wherein
OpiBe the on time of Mi equipment, T
UiService time during for Mi equipment converted products.In the present embodiment, plant factor refers to Ma plant factor U
a, Mb plant factor U
b, Mc plant factor U
c, Md plant factor U
d, Me plant factor U
eThe queue length of bottleneck equipment such as represents before the bottleneck process equipment at the Number of Jobs to be processed, is generally the workpiece quantity in the buffer zone before the equipment, is designated as Q.If its long meeting causes the increase of average process-cycle, product percent of pass reduction etc.In the present embodiment, as bottleneck equipment, then the queue length of bottleneck equipment refers to the queue length Q of Me equipment with lithographic equipment Me.Total amount of movement refers to the total step number that all silicon chips move within the unit interval, be denoted as M.Every generation one moved further, then respective batch is finished a step job sequence.Total amount of movement is higher, and the processing tasks that production line is finished is higher.
After determining input, output variable, data sample be can collect, sample output variable (S103) and sample output variable (S104) produced accordingly.
After production system is in normal operating conditions, based on the combination of feeding mode and scheduling rule, observe the production line performance index of exporting afterwards a dispatching cycle (the present embodiment is 10 days).Optionally, the feeding mode of this moment is the Conrin mode, and scheduling rule is the FIFO strategy.Suppose, first, second ion implantation device Mc, Md have satisfied process requirements under the FIFO strategy, do not consider rule and choose, and the priority rule of feeding mode and Ma, Mb oxide-diffused device cluster and Me photoetching bottleneck equipment is chosen.Those skilled in the art as can be known, number of samples can change according to different production lines, but not only limits to above-mentioned number, in like manner, also can choose according to actual conditions dispatching cycle.
Then, then in order to allow sample data can unify to process, then need feeding mode and the scheduling rule letter symbol of input are encoded, namely data are carried out pre-service (S105), mainly comprise coding, normalization etc.
In the present embodiment, coded system is as follows:
Feeding mode: 1.Conrin, 2.Possion, 3.Conwip; Ma, Mb scheduling rule: 1.FIFO, 2.EDD, 3.SRPT; Me scheduling rule: 1.EDD, 2.SRPT, 3.CR, 4.FSVCT.
The normalization formula is [x-min (x
i)]/[max (x
i)-min (x
i)], here, i is sample number, x is sample value.So, can set up neural network structure as shown in Figure 3, this structure comprises 9 input variables, 8 output variables.
Data after pretreatment, the sample that can produce in step S103 and step S104 input, output variable are selected training sample (S107) and test sample book (S106).In the present embodiment, control strategy has formed 36 kinds of combinations according to feeding mode and scheduling rule different, to every kind of combining random grasping system state, gathers 50 training samples, forms 1800 groups of inputoutput datas, and other gathers 200 groups as test sample book.
After determining training sample, determine hidden layer neuron number (S109).Because different hidden layer node numbers can cause the difference of estimated performance.
Then, making up ELM neural network (S111) based on the hidden layer neuron number, and determine final ELM neural network model (S113), namely is that the ELM neural network is to be got through the ELM Algorithm for Training by the neural network of determining hidden node.
Choose respectively input data (S108) and output data (S110) in the test sample book, the input data here are input variable data above, and the output data are the output variable data; Then, will input data input value ELM neural network model (S113), as calculated after, can produce accordingly predict the outcome (S115), then this is predicted the outcome with test sample book in output Data Comparison (S112); Then, predict the outcome with test sample book in the error of output data whether satisfy the accuracy requirement (S114) of appointment, if do not satisfy accuracy requirement, then be back to step (S109) and re-start definite hidden layer neuron number, in the present embodiment, take the relative error grand mean of each index of 200 of test sample books as standard, when the hidden layer node number was 135, test error was minimum.If satisfy accuracy requirement, then export this ELM neural network model (S116); In actual production process, semiconductor product line real-time status can be come in and gone out to this model (S116), after this model calculates, get final product the performance index (S117) of the prediction of output.
According to the match performance of test sample book test network, predict the outcome behind the renormalization and the sample value contrast, analyze the relative average error of each desired value.Table 1 is that the average relative error of 200 groups of sample data tests compares, and can find out from predicting the outcome, and the precision of prediction of ELM all is lower than 10%, arrives desirable precision of prediction.By finding with BP neural network and RBF neural net model establishing Contrast on effect, precision of prediction is very approaching, and maximum only differs 0.0638.Being 330.57 times of BP yet the training of ELM should be speed, is 60.45 times of RBF, and test speed is 7.40 times of BP, 9.38 times of RBF, and study and the test speed of visible ELM obviously are better than additive method, show the greater advantage of on-line prediction.
The average relative error of table 1200 group sample data test relatively
Simultaneously for the clearer and more definite prediction effect of seeing, choose four test sample books from test result, the actual value of the system performance index that predicted value and the emulation of ELM network is obtained is compared, as a result shown in the table 2.Therefrom can find out, predict the outcome very approaching, satisfactory for result with actual value.In addition as can be seen from the results, even different real-time system states adopts the same control strategy, the performance that realizes manufacturing system differs larger, perhaps take different control strategies to obtain different effects under the identical system state, this has verified the importance of adjusting control strategy in the dynamic Real-Time Scheduling according to the real-time system state.
The actual value of the system performance index that the predicted value of table 2ELM network and emulation obtain is compared
The above-mentioned description to embodiment is can understand and apply the invention for ease of those skilled in the art.The person skilled in the art obviously can easily make various modifications to these embodiment, and needn't pass through performing creative labour being applied in the General Principle of this explanation among other embodiment.Therefore, the invention is not restricted to the embodiment here, those skilled in the art should be within protection scope of the present invention for improvement and modification that the present invention makes according to announcement of the present invention.
Claims (6)
1. performance prediction method that can be used for Dynamic Schedule of Semiconductor Fabrication Line comprises:
(1) gathers the semiconductor production line historical data, set up training sample set and test sample book collection;
(2) will encode to feeding mode and the scheduling rule letter symbol of input, network can be accepted; For making the input data that identical dimension be arranged, input quantity is carried out normalized;
(3) adopt the extreme learning machine method to make up forecast model; Only need the hidden node number of the neural network determined for ELM, do not need to adjust the input weights of network and biasing and other parameters of hidden unit, adopt method of trial and error to choose suitable hidden node number;
(4) network performance of performance test test sample forecast model obtains output valve and the contrast of test sample book output valve after the renormalization that will predict the outcome is processed, judge whether to satisfy accuracy requirement;
(5) if the precision of prediction of test result can meet the demands, set up the forecast model process and finish, obtain required forecast model; As not satisfying, then forward step (3) to, the hidden node number of the neural network that reselects is training pattern again.
2. the method for claim 1, it is characterized in that: historical data comprises input quantity and output quantity described in the step (1), and wherein input quantity comprises: feeding mode, scheduling rule and the real-time status buffer zone team leader of system, goods in process inventory WIP, production line produce mobile step number in last dispatching cycle; Output quantity is system's short term scheduling performance index, comprises the amount of movement that produces in day throughput rate, plant factor, the team leader of bottleneck device queue and dispatching cycle.
3. the method for claim 1, it is characterized in that: step is carried out normalized to input quantity described in (2), and in [0,1] interval, the normalization formula is [x-min (x with data limit
i)]/[max (x
i)-min (x
i)], wherein, x refers to input variable, i is sample number.
4. the method for claim 1 is characterized in that: suppose the training set for given N different samples in the step (3)
Wherein x is n dimension input variable, i.e. x
i∈ R
n, t is m dimension output variable, i.e. t
i∈ R
m, i is sample number; Activation function is g (x), and the hidden layer node number is
, then
A) obtain at random the initial input weight w
iWith biasing b
i, i=1 ...,
B) calculate hidden layer output matrix H;
C) calculate output weights β:
T=[t
1..., t
N]
T, wherein
It is the Moore-Penrose generalized inverse matrix of matrix H.
5. the method for claim 1, it is characterized in that: output valve is with o described in the step (4)
iExpression, i=1 wherein ..., m, m refer to the dimension of output valve o, and m output is namely arranged.
6. the method for claim 1, it is characterized in that: it is benchmark that accuracy requirement described in the step (4) can be adopted average relative error, for m output variable, average relative error
, wherein n is the test sample book number, and j is sample number, and i is the output variable numbering, and t is the output variable of expection; The average relative error result of test is compared with the relative error of expection, guarantee that maximum average relative error is less than or equal to prediction relative average error, i.e. precision of prediction in m the output variable
The wherein accuracy requirement of δ ' for setting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013102395010A CN103310285A (en) | 2013-06-17 | 2013-06-17 | Performance prediction method applicable to dynamic scheduling for semiconductor production line |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013102395010A CN103310285A (en) | 2013-06-17 | 2013-06-17 | Performance prediction method applicable to dynamic scheduling for semiconductor production line |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103310285A true CN103310285A (en) | 2013-09-18 |
Family
ID=49135472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013102395010A Pending CN103310285A (en) | 2013-06-17 | 2013-06-17 | Performance prediction method applicable to dynamic scheduling for semiconductor production line |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103310285A (en) |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104460590A (en) * | 2014-10-23 | 2015-03-25 | 北京化工大学 | Semiconductor production line multi-product workpiece combining method |
| CN105005693A (en) * | 2015-07-08 | 2015-10-28 | 中国科学院合肥物质科学研究院 | Genetic material specificity based tumor cell drug sensitivity evaluation method |
| CN105045243A (en) * | 2015-08-05 | 2015-11-11 | 同济大学 | Semiconductor production line dynamic scheduling device |
| WO2016101181A1 (en) * | 2014-12-23 | 2016-06-30 | 清华大学 | Photoetching procedure dynamic scheduling method based on indicator forecasting and solution similarity analysis |
| CN105913180A (en) * | 2016-04-11 | 2016-08-31 | 西北工业大学 | Multiple fault mode assembly line performance assessing method |
| CN103745273B (en) * | 2014-01-06 | 2017-01-18 | 北京化工大学 | Semiconductor fabrication process multi-performance prediction method |
| CN107256453A (en) * | 2016-12-05 | 2017-10-17 | 东北大学 | A kind of quality of hollow billet forecasting procedure based on improvement ELM algorithms |
| CN109507961A (en) * | 2018-11-07 | 2019-03-22 | 同济大学 | A kind of semiconductor production line dynamic load uniform feeding control method |
| CN109523136A (en) * | 2018-10-29 | 2019-03-26 | 同济大学 | A kind of scheduling knowledge management system towards intelligence manufacture |
| CN109791627A (en) * | 2018-06-19 | 2019-05-21 | 香港应用科技研究院有限公司 | Using input pretreatment and switch target for training the semiconductor devices of deep neural network to model |
| CN110070248A (en) * | 2018-01-19 | 2019-07-30 | 财团法人工业技术研究院 | Dynamic and intelligent dispatching method and device |
| CN110298578A (en) * | 2019-06-24 | 2019-10-01 | 电子科技大学 | A kind of mix flow bottleneck prediction technique |
| WO2019241937A1 (en) * | 2018-06-19 | 2019-12-26 | Hong Kong Applied Science and Technology Research Institute Company Limited | Semiconductor device modeling using input pre-processing and transformed targets for training a deep neural network |
| CN111356988A (en) * | 2020-02-21 | 2020-06-30 | 长江存储科技有限责任公司 | Method and system for scheduling semiconductor processing |
| CN111539556A (en) * | 2020-04-02 | 2020-08-14 | 北京北方华创微电子装备有限公司 | Wafer scheduling method and device |
| CN112631216A (en) * | 2020-12-11 | 2021-04-09 | 江苏晶度半导体科技有限公司 | Semiconductor test packaging production line performance prediction control system based on DQN and DNN twin neural network algorithm |
| WO2021069984A1 (en) * | 2019-10-08 | 2021-04-15 | International Business Machines Corporation | Tool control using multistage lstm for predicting on-wafer measurements |
| CN112947339A (en) * | 2021-02-22 | 2021-06-11 | 同济大学 | Semiconductor production line dynamic scheduling method based on performance index |
| CN113031542A (en) * | 2021-02-22 | 2021-06-25 | 同济大学 | Semiconductor production line dynamic scheduling method based on load balancing |
| CN113031543A (en) * | 2021-02-24 | 2021-06-25 | 同济大学 | Control scheduling method and device for semiconductor production line |
| CN113064388A (en) * | 2021-02-24 | 2021-07-02 | 同济大学 | Scheduling optimization method and device for semiconductor production line |
| CN113139702A (en) * | 2021-05-28 | 2021-07-20 | 上海华力微电子有限公司 | Method and system for predicting work-in-process quantity |
| CN114115155A (en) * | 2021-12-17 | 2022-03-01 | 吉林大学 | Industrial Internet of things multithreading intelligent production scheduling method and system |
| WO2022227018A1 (en) * | 2021-04-30 | 2022-11-03 | 华为数字能源技术有限公司 | Testing system for power semiconductor device, cloud server and testing machine |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002052365A2 (en) * | 2000-12-26 | 2002-07-04 | Insyst Ltd. | Model predictive control (mpc) system using doe based model |
| US20040039556A1 (en) * | 2002-08-22 | 2004-02-26 | Ibex Process Technology, Inc. | Filter models for dynamic control of complex processes |
| CN1734382A (en) * | 2005-06-10 | 2006-02-15 | 同济大学 | The dynamic dispatching method that is used for semiconductor production line based on pheromones |
| CN102735760A (en) * | 2012-06-26 | 2012-10-17 | 河海大学 | Method for predicting transformer oil chromatographic data based on extreme learning machine |
-
2013
- 2013-06-17 CN CN2013102395010A patent/CN103310285A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002052365A2 (en) * | 2000-12-26 | 2002-07-04 | Insyst Ltd. | Model predictive control (mpc) system using doe based model |
| US20040039556A1 (en) * | 2002-08-22 | 2004-02-26 | Ibex Process Technology, Inc. | Filter models for dynamic control of complex processes |
| CN1734382A (en) * | 2005-06-10 | 2006-02-15 | 同济大学 | The dynamic dispatching method that is used for semiconductor production line based on pheromones |
| CN102735760A (en) * | 2012-06-26 | 2012-10-17 | 河海大学 | Method for predicting transformer oil chromatographic data based on extreme learning machine |
Non-Patent Citations (7)
| Title |
|---|
| GUANG-BIN HUANG 等: "Etreme learning machine:Theory and applications", 《NEUROCOMPUTING》 * |
| 乔非 等: "半导体晶圆生产线调度的性能指标体系研究", 《同济大学学报》 * |
| 刘士荣 等: "基于极端学习机的光伏发电功率短期预测", 《控制工程》 * |
| 周品: "《MATLAB 神经网络设计与应用》", 31 March 2013 * |
| 徐国祥: "《统计预测和决策》", 31 August 2012 * |
| 曹政才 等: "基于遗传算法的半导体生产线调度研究进展", 《同济大学学报（自然科学版）》 * |
| 陈媛媛 等: "基于极限学习机的混合气体FTIR光谱定量分析", 《中北大学学报（自然科学版）》 * |
Cited By (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745273B (en) * | 2014-01-06 | 2017-01-18 | 北京化工大学 | Semiconductor fabrication process multi-performance prediction method |
| CN104460590A (en) * | 2014-10-23 | 2015-03-25 | 北京化工大学 | Semiconductor production line multi-product workpiece combining method |
| CN104460590B (en) * | 2014-10-23 | 2017-02-15 | 北京化工大学 | Semiconductor production line multi-product workpiece combining method |
| WO2016101181A1 (en) * | 2014-12-23 | 2016-06-30 | 清华大学 | Photoetching procedure dynamic scheduling method based on indicator forecasting and solution similarity analysis |
| CN105005693B (en) * | 2015-07-08 | 2017-12-01 | 中国科学院合肥物质科学研究院 | One kind is based on the specific tumour cell drug susceptibility appraisal procedure of inhereditary material |
| CN105005693A (en) * | 2015-07-08 | 2015-10-28 | 中国科学院合肥物质科学研究院 | Genetic material specificity based tumor cell drug sensitivity evaluation method |
| CN105045243A (en) * | 2015-08-05 | 2015-11-11 | 同济大学 | Semiconductor production line dynamic scheduling device |
| CN105045243B (en) * | 2015-08-05 | 2017-08-25 | 同济大学 | A kind of Dynamic Schedule of Semiconductor Fabrication Line device |
| CN105913180B (en) * | 2016-04-11 | 2019-04-19 | 西北工业大学 | Multiple faults mode assembly line performance estimating method |
| CN105913180A (en) * | 2016-04-11 | 2016-08-31 | 西北工业大学 | Multiple fault mode assembly line performance assessing method |
| CN107256453A (en) * | 2016-12-05 | 2017-10-17 | 东北大学 | A kind of quality of hollow billet forecasting procedure based on improvement ELM algorithms |
| CN107256453B (en) * | 2016-12-05 | 2020-10-23 | 东北大学 | Capillary quality forecasting method based on improved ELM algorithm |
| CN110070248A (en) * | 2018-01-19 | 2019-07-30 | 财团法人工业技术研究院 | Dynamic and intelligent dispatching method and device |
| CN110070248B (en) * | 2018-01-19 | 2022-03-29 | 财团法人工业技术研究院 | Dynamic intelligent scheduling method and device |
| CN109791627A (en) * | 2018-06-19 | 2019-05-21 | 香港应用科技研究院有限公司 | Using input pretreatment and switch target for training the semiconductor devices of deep neural network to model |
| CN109791627B (en) * | 2018-06-19 | 2022-10-21 | 香港应用科技研究院有限公司 | Semiconductor device modeling for training deep neural networks using input preprocessing and conversion targets |
| WO2019241937A1 (en) * | 2018-06-19 | 2019-12-26 | Hong Kong Applied Science and Technology Research Institute Company Limited | Semiconductor device modeling using input pre-processing and transformed targets for training a deep neural network |
| US11176447B2 (en) | 2018-06-19 | 2021-11-16 | Hong Kong Applied Science and Technology Research Institute Company Limited | Semiconductor device modeling using input pre-processing and transformed targets for training a deep neural network |
| CN109523136A (en) * | 2018-10-29 | 2019-03-26 | 同济大学 | A kind of scheduling knowledge management system towards intelligence manufacture |
| CN109507961A (en) * | 2018-11-07 | 2019-03-22 | 同济大学 | A kind of semiconductor production line dynamic load uniform feeding control method |
| CN109507961B (en) * | 2018-11-07 | 2021-03-26 | 同济大学 | Semiconductor production line dynamic load balancing feeding control method |
| CN110298578B (en) * | 2019-06-24 | 2022-11-29 | 电子科技大学 | Dynamic bottleneck prediction method for mixed flow production line |
| CN110298578A (en) * | 2019-06-24 | 2019-10-01 | 电子科技大学 | A kind of mix flow bottleneck prediction technique |
| WO2021069984A1 (en) * | 2019-10-08 | 2021-04-15 | International Business Machines Corporation | Tool control using multistage lstm for predicting on-wafer measurements |
| GB2603400B (en) * | 2019-10-08 | 2024-05-22 | Ibm | Tool control using multistage LSTM for predicting on-wafer measurements |
| GB2603400A (en) * | 2019-10-08 | 2022-08-03 | Ibm | Tool control using multistage LSTM for predicting on-wafer measurements |
| US11385627B2 (en) | 2020-02-21 | 2022-07-12 | Yangtze Memory Technologies Co., Ltd. | Method and system for scheduling semiconductor fabrication |
| CN111356988A (en) * | 2020-02-21 | 2020-06-30 | 长江存储科技有限责任公司 | Method and system for scheduling semiconductor processing |
| CN111539556A (en) * | 2020-04-02 | 2020-08-14 | 北京北方华创微电子装备有限公司 | Wafer scheduling method and device |
| CN112631216A (en) * | 2020-12-11 | 2021-04-09 | 江苏晶度半导体科技有限公司 | Semiconductor test packaging production line performance prediction control system based on DQN and DNN twin neural network algorithm |
| CN112947339A (en) * | 2021-02-22 | 2021-06-11 | 同济大学 | Semiconductor production line dynamic scheduling method based on performance index |
| CN113031542A (en) * | 2021-02-22 | 2021-06-25 | 同济大学 | Semiconductor production line dynamic scheduling method based on load balancing |
| CN113031543A (en) * | 2021-02-24 | 2021-06-25 | 同济大学 | Control scheduling method and device for semiconductor production line |
| CN113064388B (en) * | 2021-02-24 | 2022-05-03 | 同济大学 | Scheduling optimization method and device for semiconductor production line |
| CN113064388A (en) * | 2021-02-24 | 2021-07-02 | 同济大学 | Scheduling optimization method and device for semiconductor production line |
| WO2022227018A1 (en) * | 2021-04-30 | 2022-11-03 | 华为数字能源技术有限公司 | Testing system for power semiconductor device, cloud server and testing machine |
| CN113139702A (en) * | 2021-05-28 | 2021-07-20 | 上海华力微电子有限公司 | Method and system for predicting work-in-process quantity |
| CN113139702B (en) * | 2021-05-28 | 2024-04-30 | 上海华力微电子有限公司 | Method and system for predicting number of products in process |
| CN114115155A (en) * | 2021-12-17 | 2022-03-01 | 吉林大学 | Industrial Internet of things multithreading intelligent production scheduling method and system |
| CN114115155B (en) * | 2021-12-17 | 2023-11-03 | 吉林大学 | Industrial Internet of things multithreading intelligent production scheduling method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103310285A (en) | Performance prediction method applicable to dynamic scheduling for semiconductor production line | |
| CN105427021A (en) | Intelligent clothes production scheduling method | |
| CN103676881B (en) | A kind of dynamic bottleneck analytical method of semiconductor production line | |
| CN109902954B (en) | Flexible job shop dynamic scheduling method based on industrial big data | |
| CN103745273B (en) | Semiconductor fabrication process multi-performance prediction method | |
| CN105045243A (en) | Semiconductor production line dynamic scheduling device | |
| CN106094757B (en) | A kind of dynamic flexible solving job shop scheduling problem control method based on data-driven | |
| Yao et al. | A multi-objective dynamic scheduling approach using multiple attribute decision making in semiconductor manufacturing | |
| Li et al. | Data-based scheduling framework and adaptive dispatching rule of complex manufacturing systems | |
| Manupati et al. | Adaptive production control system for a flexible manufacturing cell using support vector machine-based approach | |
| Kłos et al. | Improving manufacturing processes using simulation methods | |
| CN106327053B (en) | Construction method of weaving process recommendation model based on multi-mode set | |
| Chen | Estimating job cycle time in a wafer fabrication factory: A novel and effective approach based on post-classification | |
| Aksyonov et al. | Rules for construction of simulation models for production processes optimization | |
| Wang et al. | Real-time decision support with reinforcement learning for dynamic flowshop scheduling | |
| CN112884321A (en) | Semiconductor production line scheduling method based on scheduling environment and tasks | |
| Bramhane et al. | Simulation of flexible manufacturing system using adaptive neuro fuzzy hybrid structure for efficient job sequencing and routing | |
| Chen | Intelligent scheduling approaches for a wafer fabrication factory | |
| Guneri et al. | The usage of artificial neural networks for finite capacity planning | |
| Rao et al. | On-line simulation for shop floor control in manufacturing execution system | |
| Chen | Combining statistical analysis and artificial neural network for classifying jobs and estimating the cycle times in wafer fabrication | |
| CN113050553A (en) | Scheduling modeling method of semiconductor production line based on federal learning mechanism | |
| Lee et al. | A sequential search framework for selecting weights of dispatching rules in manufacturing systems | |
| Ebrahimi et al. | Determining the optimal performance of flexible manufacturing systems using network analysis and simulation process | |
| Frye et al. | Production rescheduling through product quality prediction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130918 |
|
| RJ01 | Rejection of invention patent application after publication |