CN110648935A - Semiconductor manufacturing defect dynamic random sampling method using AI model - Google Patents
Semiconductor manufacturing defect dynamic random sampling method using AI model Download PDFInfo
- Publication number
- CN110648935A CN110648935A CN201910910470.4A CN201910910470A CN110648935A CN 110648935 A CN110648935 A CN 110648935A CN 201910910470 A CN201910910470 A CN 201910910470A CN 110648935 A CN110648935 A CN 110648935A
- Authority
- CN
- China
- Prior art keywords
- model
- model training
- defect
- target product
- sampling
- 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
Classifications
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
- H01L22/00—Testing or measuring during manufacture or treatment; Reliability measurements, i.e. testing of parts without further processing to modify the parts as such; Structural arrangements therefor
- H01L22/10—Measuring as part of the manufacturing process
- H01L22/12—Measuring as part of the manufacturing process for structural parameters, e.g. thickness, line width, refractive index, temperature, warp, bond strength, defects, optical inspection, electrical measurement of structural dimensions, metallurgic measurement of diffusions
Abstract
The invention discloses a semiconductor manufacturing defect dynamic random sampling method using an AI model, which relates to the technical field of semiconductor defect random sampling, and comprises the following steps: collecting information of a target product; performing AI modeling on the information of the target product to generate a simulation AI model training-1; collecting main defect problems and associated data of target products; according to the collected data, carrying out AI model training establishment of the defect root cause, and generating AI model training-2; deploying AI model training-1 and AI model training-2 on an AI model operation platform for operation; establishing a butt joint between the step AI model operation platform and the existing sampling system; the sampling result in the sampling system is executed. The invention has the advantages.
Description
Technical Field
The invention relates to the technical field of semiconductor defect random sampling, in particular to a semiconductor manufacturing defect dynamic random sampling method using an AI model.
Background
In the manufacturing process of semiconductors, there is a certain defect problem on semiconductors. Defects are the most significant cause of reduced chip manufacturing yield and reliability, and therefore, management and control of defects is particularly important in semiconductor manufacturing processes. The main control method for the defects at present is as follows: 1) in the manufacturing process, selecting a sampling detection station according to experience; 2) through the experience of engineering personnel, a corresponding static sampling rule is set for each detection station, namely, the mantissa number of the product batch number is used as the basis for sampling.
However, the current method is a static sampling method and depends heavily on the experience of engineers, so that the coverage rate of defect problems is greatly insufficient, and limited detection resources are greatly wasted. Aiming at the selected detection site, the selected basis is basically the reproduction of the prior generation products, and the distribution characteristics of the defect problems of the products are not researched, so that the specific problems of the products can be easily ignored in the initial stage; aiming at the static sampling rule, as the sampling is only carried out at random without considering any other factors, only systematic and commonality defect problems can be found, and the detection rate of sporadic time caused by burst of production equipment, personnel errors and the like is extremely low.
In view of the above problems, it is desirable to provide a dynamic random sampling method for semiconductor manufacturing defects, which can obtain a relatively large problem coverage rate by using limited detection resources.
Disclosure of Invention
Aiming at the problem in practical application, the invention aims to provide a semiconductor manufacturing defect dynamic random sampling method using an AI model, and the specific scheme is as follows:
a dynamic random sampling method for semiconductor manufacturing defects by using an AI model comprises the following steps:
1) collecting information of a target product;
2) performing AI modeling on the information of the target product to generate a simulation AI model training-1;
3) collecting major defect issues and associated data for the target product;
4) according to the data collected in the step 3), carrying out AI model training establishment of the defect root cause, and generating AI model training-2;
5) deploying the AI model training-1 and the AI model training-2 in the step 2) and the step 4) on an AI model operation platform for operation;
6) establishing butt joint between the AI model operation platform in the step 5) and the existing sampling system;
7) performing sampling results in the sampling system in the step 6).
Further, the information of the target product in step 1) includes the attribute, problem distribution, target, historical similar product, capacity, and time defect of historical similar product of the target product.
Further, the main defect problem and associated data of the target product in step 4) include the equipment failure characteristic parameter, the production material/supplier failure characteristic parameter, the plant service system characteristic parameter and the environment change characteristic parameter,
the characteristic parameters of the plant service system comprise characteristic parameters of a hydroelectric system, and the characteristic parameters of the environmental change comprise characteristic parameters of temperature and humidity change.
Further, the AI model training-1 is used for detecting the setting of the station.
Further, the AI model training-2 is used to predict the likelihood of the same defect problem occurring in subsequent production lots.
Further, the AI model operation platform comprises storage management, operation control and real-time extraction and storage of required data of the AI model training-1 and the AI model training-2.
Compared with the prior art, the invention has the following beneficial effects:
(1) the AI counting is used for modeling historical data, deep analysis and research are carried out on defect problems, a reliable early warning model is established for factors such as process sites, time and environment which possibly cause defects, and therefore AI model training-1 and AI model training-2 are generated, the defect problems can be accurately predicted, and the problem that the coverage rate of the defect problems is insufficient due to the fact that a sampling scheme in the prior art is made by means of manual experience is solved;
(2) the problems are accurately predicted through AI model training-1 and AI model training-2, detection is only carried out on high-risk product batches, the detection efficiency is improved, and the problem that detection resources cannot be fully utilized under the prior art is solved;
(3) the AI model training-1 and the AI model training-2 are deployed on the AI model operation platform, and the AI model operation platform is in butt joint with the existing system, so that the scheme in the invention is easy to implement, and relevant knowledge experience can be extracted from the obtained AI model to serve other management services.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited to these examples.
Examples
As shown in fig. 1, a method for dynamic random sampling of semiconductor manufacturing defects using an AI model includes the steps of:
1) collecting information of a target product;
2) performing AI modeling on the information of the target product to generate a simulation AI model training-1;
3) collecting main defect problems and associated data of target products;
4) according to the data collected in the step 3), carrying out AI model training establishment of the defect root cause, and generating AI model training-2;
5) deploying the AI model training-1 and the AI model training-2 in the step 2) and the step 4) on an AI model operation platform for operation;
6) establishing butt joint between the AI model operation platform in the step 5) and the existing sampling system;
7) the sampling result in the sampling system in the step 6) is executed.
The information of the target product in the step 1) comprises the attribute of the target product, problem distribution, target, historical like products, capacity and time defects of the historical like products.
The main defect problems and associated data of the target product in the step 4) comprise failure characteristic parameters of used equipment, failure characteristic parameters of produced materials/suppliers, characteristic parameters of a factory service system and characteristic parameters of environmental change,
the characteristic parameters of the plant service system comprise characteristic parameters of a hydroelectric system, and the characteristic parameters of environmental change comprise characteristic parameters of temperature and humidity change.
AI model training-1 is used to detect station settings. The AI model training-1 establishes a set of simulation prediction models for the distribution situation of defects, defects and product types, process complexity and correlation situation of equipment conditions in the life cycle of a product based on historical product data, can predict the positions of key defect problems of a target product in the production cycle at a high probability, and intelligently plans which sites are detected by combining the production capacity allocation situation of the target.
AI model training-2 is used to predict the likelihood of the same defect problem occurring in subsequent production lots. The AI model training-2 aims at the prediction of defects, the prediction is based on historical data, the main defect problems which are common at present are mined for the relevance of production process parameters, the relevance comprises the sensor parameters of a machine, machine equipment failure parameters (such as aging period of parts), failure models of production materials (such as batch quality of gas acid, alkali and the like), failure parameters of a plant system (stability of water and electricity and the like), environment failure parameters (temperature and humidity caused by seasons) and the like, and perfect defect and key parameter models are established, so that the AI prediction model can cover the production defect problems in a large range, especially the abnormal sudden defect problems.
The AI model operation platform comprises storage management, operation control and real-time extraction and storage of required data of AI model training-1 and AI model training-2. Therefore, the AI model operation platform is in interface link with the existing sampling control system, and the AI model calculation result is output to the current control system for integration.
And finally, performing final sampling execution through the current sampling execution module, namely sending a command to the production management system and the machine for automatic detection.
The invention provides a perfect solution from the establishment of a sampling model to the application and development of a model result to the implementation of a final result, provides an intelligent sampling scheme for the current semiconductor manufacturing defect management and control, effectively utilizes limited detection resources and increases the coverage rate of problems.
The following will be described, taking the defect of short circuit caused by insufficient etching of semiconductor as an example:
according to the experience of engineering personnel, etching equipment of Lam manufacturers has a certain probability of causing insufficient etching after encountering error information of IEMS parameter ESC current leak max. However, the above parameter information cannot fully characterize the transmission of the defect, that is, when the signal occurs, the signal has an under-etching defect in about 10%, i.e., the false alarm rate is greater than 90%, and conversely, when the under-etching occurs, the signal has no abnormality in greater than 50%, i.e., the detection rate of the signal for the defect is less than 50%.
By utilizing the method, AI model training-1 and AI model training-2 are established, Defect and IEMS parameter correlation analysis is carried out, key characteristic parameters influencing short circuit caused by insufficient etching are found out, the occurrence of the defects is accurately predicted, namely, the detection rate is improved, and the false alarm rate of the current univariate monitoring is reduced.
And (3) performing feature screening (total 300+ variables, screening 61) by using a machine learning mode through a mixed feature screening model, performing feature classification by using a support vector machine to obtain a multivariable binary-classification algorithm model, performing secondary screening on wafer, and reducing the false alarm rate on the basis of keeping the original detection rate.
Finally, the detection rate is more than or equal to 95 percent, and the false alarm rate is less than or equal to 50 percent.
Through the AI model training-1 and the AI model training-2, when a product is produced in the equipment production process, the scheme provided by the invention collects 61 required characteristic variables of the etching machine in real time and outputs the possibility of generating the defect of insufficient etching.
By making a sampling strategy (for example, detecting the probability score above 0.9) for the probability result, the traditional static sampling scheme is replaced, namely, the tail number of the fixed product batch is detected in the traditional technology, so that the defect detection rate is greatly improved, the detection times are greatly reduced, and the resource saving is realized.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may occur to those skilled in the art without departing from the principle of the invention, and are considered to be within the scope of the invention.
Claims (6)
1. A semiconductor manufacturing defect dynamic random sampling method using AI model is characterized in that the method comprises the following steps:
1) collecting information of a target product;
2) performing AI modeling on the information of the target product to generate a simulation AI model training-1;
3) collecting major defect issues and associated data for the target product;
4) according to the data collected in the step 3), carrying out AI model training establishment of the defect root cause, and generating AI model training-2;
5) deploying the AI model training-1 and the AI model training-2 in the step 2) and the step 4) on an AI model operation platform for operation;
6) establishing butt joint between the AI model operation platform in the step 5) and the existing sampling system;
7) performing sampling results in the sampling system in the step 6).
2. The AI-model-based dynamic random sampling method for defects in semiconductor manufacturing as recited in claim 1, wherein the information about the target product in step 1) includes attributes, problem distributions, targets, historical peers, production capacity, and time defects of defects in historical peers.
3. The AI-model based dynamic random sampling method of defects in semiconductor manufacturing as recited in claim 1, wherein the main defect problem and associated data of the target product in step 4) comprises equipment failure characterization parameters, production material/supplier failure characterization parameters, plant system characterization parameters and environment variation characterization parameters,
the characteristic parameters of the plant service system comprise characteristic parameters of a hydroelectric system, and the characteristic parameters of the environmental change comprise characteristic parameters of temperature and humidity change.
4. The method as claimed in claim 1, wherein the AI model training-1 is used to detect the site settings.
5. The method of claim 1, wherein the AI model training-2 is used to predict the probability of the same defect problem occurring in subsequent production lots.
6. The method as claimed in claim 1, wherein the AI model operation platform comprises storage management, operation control and real-time extraction and storage of required data of the AI model training-1 and the AI model training-2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910910470.4A CN110648935A (en) | 2019-09-25 | 2019-09-25 | Semiconductor manufacturing defect dynamic random sampling method using AI model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910910470.4A CN110648935A (en) | 2019-09-25 | 2019-09-25 | Semiconductor manufacturing defect dynamic random sampling method using AI model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110648935A true CN110648935A (en) | 2020-01-03 |
Family
ID=68992637
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910910470.4A Pending CN110648935A (en) | 2019-09-25 | 2019-09-25 | Semiconductor manufacturing defect dynamic random sampling method using AI model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110648935A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11429091B2 (en) | 2020-10-29 | 2022-08-30 | Kla Corporation | Method of manufacturing a semiconductor device and process control system for a semiconductor manufacturing assembly |
| WO2022266833A1 (en) * | 2021-06-22 | 2022-12-29 | 华为技术有限公司 | Root cause identification method and related device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436062B (en) * | 2007-11-16 | 2015-09-16 | 台湾积体电路制造股份有限公司 | The method of the wafer result of prediction batch tool |
| US20170147909A1 (en) * | 2015-11-25 | 2017-05-25 | Canon Kabushiki Kaisha | Information processing apparatus, information processing method, and storage medium |
| CN108108848A (en) * | 2017-12-29 | 2018-06-01 | 英特尔产品（成都）有限公司 | The training method of ratio of defects prediction model, apparatus and system |
| CN108475351A (en) * | 2015-12-31 | 2018-08-31 | 科磊股份有限公司 | The acceleration training of the model based on machine learning for semiconductor application |
| CN109636026A (en) * | 2018-12-07 | 2019-04-16 | 东华大学 | A kind of wafer yield prediction technique based on deep learning model |
| CN110024097A (en) * | 2016-11-30 | 2019-07-16 | Sk 株式会社 | Semiconductors manufacture yield forecasting system and method based on machine learning |
-
2019
- 2019-09-25 CN CN201910910470.4A patent/CN110648935A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436062B (en) * | 2007-11-16 | 2015-09-16 | 台湾积体电路制造股份有限公司 | The method of the wafer result of prediction batch tool |
| US20170147909A1 (en) * | 2015-11-25 | 2017-05-25 | Canon Kabushiki Kaisha | Information processing apparatus, information processing method, and storage medium |
| CN108475351A (en) * | 2015-12-31 | 2018-08-31 | 科磊股份有限公司 | The acceleration training of the model based on machine learning for semiconductor application |
| CN110024097A (en) * | 2016-11-30 | 2019-07-16 | Sk 株式会社 | Semiconductors manufacture yield forecasting system and method based on machine learning |
| CN108108848A (en) * | 2017-12-29 | 2018-06-01 | 英特尔产品（成都）有限公司 | The training method of ratio of defects prediction model, apparatus and system |
| CN109636026A (en) * | 2018-12-07 | 2019-04-16 | 东华大学 | A kind of wafer yield prediction technique based on deep learning model |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11429091B2 (en) | 2020-10-29 | 2022-08-30 | Kla Corporation | Method of manufacturing a semiconductor device and process control system for a semiconductor manufacturing assembly |
| WO2022266833A1 (en) * | 2021-06-22 | 2022-12-29 | 华为技术有限公司 | Root cause identification method and related device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111596604B (en) | Intelligent fault diagnosis and self-healing control system and method for engineering equipment based on digital twinning | |
| CN111047082A (en) | Early warning method and device for equipment, storage medium and electronic device | |
| US9141915B2 (en) | Method and apparatus for deriving diagnostic data about a technical system | |
| CN109977624A (en) | Photovoltaic plant soft fault monitoring method based on deep neural network | |
| CN106100579B (en) | A kind of photovoltaic plant method for diagnosing faults based on data analysis | |
| CN104732276A (en) | On-line diagnosing method for faults of metering production facility | |
| CN110648935A (en) | Semiconductor manufacturing defect dynamic random sampling method using AI model | |
| CN116308304A (en) | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection | |
| CN113740666B (en) | Method for positioning root fault of storm alarm in power system of data center | |
| CN115270491A (en) | Offshore wind power operation and maintenance platform design method based on multivariate information fusion | |
| CN115980531A (en) | GIS switch cabinet quality detection method and system under specific environment | |
| CN111736574B (en) | Universal thermal power plant fault diagnosis system and diagnosis method thereof | |
| CN111894798B (en) | Method and system for determining running state of fan | |
| CN201666923U (en) | Device for processing DSP failure diagnosis of electric power system by using immune genetic algorithm | |
| CN112731022A (en) | Photovoltaic inverter fault detection method, device and medium | |
| KR102523701B1 (en) | Solar module quality prediction apparatus and method using ai | |
| CN116961215A (en) | Rapid fault response processing method for power system | |
| CN112240267B (en) | Fan monitoring method based on wind speed correlation and wind power curve | |
| CN114679133A (en) | Photovoltaic array abnormity judgment method, medium and equipment based on power generation prediction | |
| CN114266427A (en) | Topology analysis-based new equipment commissioning starting scheme generation method | |
| CN112381130A (en) | Cluster analysis-based power distribution room multivariate data anomaly detection method | |
| Ristić et al. | A short overview on Industry 4.0 in maintenance of hydropower plants | |
| CN106709158B (en) | Performance improvement method of power grid feeder automation system | |
| CN111650444A (en) | Regression analysis-based TCU failure early warning method in electromagnetic environment | |
| CN114414938B (en) | Dynamic response method and system for power distribution network faults |
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 |