Yield rate management method and device based on large language model and electronic equipment
By applying a large language model in the production and manufacturing process of electronic equipment and using the measurement data to generate yield management data, the problem of lack of intelligent management in the existing technology is solved, intelligent monitoring, prediction and process parameter optimization are achieved, and production is improved Efficiency and product quality.
Patent Information
- Application Number
- CN202411098490.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-26
AI Technical Summary
In the production and manufacturing process of electronic equipment, it is difficult for the existing technology to effectively use large language models for yield management, resulting in a lack of intelligent and automated management methods in the production process.
By obtaining the measurement data of the object to be tested, input it into the large language yield management model, and using the trained corpus model to generate yield management data, including yield evaluation reports, prediction reports and process parameter optimization suggestions, to realize the production of electronic equipment Intelligent monitoring, prediction and optimization of processes.
It realizes the application of large language models in the production and manufacturing process of electronic equipment, improves the intelligence and automation level of yield management, and improves production efficiency and product quality.
Smart Images

Figure CN119026984A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent production and manufacturing, and more specifically, to a yield management method, device and electronic equipment based on a large language model. Background Art
[0002] Large Language Model (LLM) refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. With the development of large language models, more and more large language models in vertical fields have emerged. In the production and manufacturing process of electronic equipment, how to use large language models for yield management is a technical problem that needs to be solved urgently in this field. Summary of the invention
[0003] In view of the defects of the prior art, the purpose of this application is to achieve yield management using a large language model during the production and manufacturing process of electronic equipment.
[0004] To achieve the above objectives, in a first aspect, the present application provides a yield management method based on a large language model, comprising:
[0005] Obtain the commissioning data of the object under test;
[0006] Input the debugging data into the big language yield management model to obtain the yield management data output by the big language yield management model;
[0007] Among them, the big language yield management model is used to determine the yield management data based on the commissioning data. The big language yield management model is obtained by training the big language model based on the corpus. The corpus includes multiple dialogue corpus samples. The dialogue corpus samples are generated based on the commissioning data samples and the yield management data samples.
[0008] In a possible implementation, the test data includes description information of the object under test and one or more of the following information: process parameters, test parameters, and test results;
[0009] The commissioning parameters are configuration parameters of the commissioning equipment, the commissioning equipment is used to commission the object under test, and the commissioning results are data obtained by commissioning the object under test through the commissioning equipment.
[0010] In a possible implementation, when the test data includes a test result, the yield management data includes a yield evaluation report, inputting the test data into a large language yield management model, and obtaining the yield management data output by the large language yield management model include:
[0011] Based on the description information of the object under test, the test results and the yield evaluation index, a first dialogue data is generated, and the first dialogue data is used to instruct the large language yield management model to determine a yield evaluation report;
[0012] Input the first dialogue data into the big language yield management model to obtain the yield evaluation report output by the big language yield management model.
[0013] In a possible implementation, when the test data includes process parameters and test parameters, the yield management data includes a yield prediction report, inputting the test data into a large language yield management model, and obtaining the yield management data output by the large language yield management model include:
[0014] Based on the description information of the object under test, process parameters, test parameters and yield evaluation indicators, a second dialogue data is generated, and the second dialogue data is used to instruct the large language yield management model to determine a yield prediction report;
[0015] Input the second dialogue data into the big language yield management model to obtain the yield prediction report output by the big language yield management model.
[0016] In a possible implementation, when the test data includes process parameters and test results, the yield management data includes process parameters to be optimized and / or process parameter recommendation information, inputting the test data into the large language yield management model, and obtaining the yield management data output by the large language yield management model include:
[0017] Based on the description information of the object under test, the process parameters, the test results and the yield evaluation index, a third dialogue data is generated, and the third dialogue data is used to instruct the large language yield management model to determine the process parameters to be optimized and / or the process parameter recommendation information;
[0018] The third dialogue data is input into the large language yield management model to obtain the process parameters to be optimized and / or the process parameter recommendation information output by the large language yield management model.
[0019] In a possible implementation, when the commissioning data includes a commissioning result, and the commissioning result is a defect detection result, obtaining the commissioning data of the object under test includes:
[0020] Acquire a test image of the object under test;
[0021] Input the test image into the image segmentation model and obtain the defect detection result output by the image segmentation model;
[0022] Among them, the image segmentation model is used to detect defects of the object under test based on the test image of the object under test.
[0023] In a possible implementation, the image segmentation model is obtained by performing the following steps:
[0024] Based on the real data set, a model is generated through image data to generate a simulated data set. The real data set is constructed based on the historical test images of the object under test;
[0025] Based on the real data set and the simulated data set, the initial image segmentation model is trained to obtain the trained image segmentation model.
[0026] In a second aspect, the present application provides a yield management device based on a large language model, comprising:
[0027] The measured object data acquisition module is used to obtain the test data of the measured object;
[0028] The yield data acquisition module is used to input the test data into the large language yield management model and obtain the yield management data output by the large language yield management model;
[0029] Among them, the big language yield management model is used to determine the yield management data based on the commissioning data. The big language yield management model is obtained by training the big language model based on the corpus. The corpus includes multiple dialogue corpus samples. The dialogue corpus samples are generated based on the commissioning data samples and the yield management data samples.
[0030] In a third aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.
[0031] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.
[0032] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the prior art:
[0033] The big language yield management model is trained based on a corpus, and the corpus includes multiple dialogue corpus samples. The dialogue corpus samples are generated based on the test data samples and the yield management data samples. The trained big language yield management model has the ability to determine (predict) the yield management data based on the test data. Furthermore, in the application stage of the big language yield management model, the test data of the object under test can be input into the big language yield management model to obtain the yield management data output by the big language yield management model, so as to realize the yield management by the big language yield management model in the production and manufacturing process of electronic equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a flowchart of a yield management method based on a large language model provided in an embodiment of the present application;
[0035] Figure 2 It is a schematic diagram of using a large language model for yield management provided in an embodiment of the present application;
[0036] Figure 3 It is a flowchart of model training and use provided in the embodiment of the present application;
[0037] Figure 4 It is a schematic diagram of the process of training an image segmentation model provided in an embodiment of the present application;
[0038] Figure 5 It is a schematic diagram of the process of training a large language model provided in an embodiment of the present application;
[0039] Figure 6 It is a structural schematic diagram of a yield management device based on a large language model provided in an embodiment of the present application;
[0040] Figure 7 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0042] The term "and / or" in this article is a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The symbol " / " in this article indicates that the associated objects are in an or relationship, for example, A / B means A or B.
[0043] The terms "first", "second", etc. in the specification and claims herein are used to distinguish different objects, rather than to describe a specific order of objects. For example, the first dialogue material and the second dialogue material are used to distinguish different dialogue materials, rather than to describe a specific order of dialogue materials.
[0044] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.
[0045] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more than two. For example, multiple processing units refer to two or more processing units, etc.; multiple elements refer to two or more elements, etc.
[0046] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.
[0047] Figure 1 is a flow chart of a yield management method based on a large language model provided in an embodiment of the present application, such as Figure 1 As shown, the execution subject of the method may be an electronic device, such as a server, etc. The method includes the following steps S101 and S102.
[0048] Step S101, obtaining the commissioning data of the object under test.
[0049] Step S102, inputting the test data into the large language yield management model, and obtaining the yield management data output by the large language yield management model;
[0050] Among them, the big language yield management model is used to determine the yield management data based on the commissioning data. The big language yield management model is obtained by training the big language model based on the corpus. The corpus includes multiple dialogue corpus samples. The dialogue corpus samples are generated based on the commissioning data samples and the yield management data samples.
[0051] The yield rate (or product yield rate) is explained here. It refers to the ratio of the number of good products that finally pass the test on the production line to the theoretical production quantity of the input materials.
[0052] The debugging data is explained here. The debugging data may be data collected during the debugging test of the object under test.
[0053] The yield management data is explained here. The yield management data may include one or more of the following information: yield evaluation report, yield prediction report, process parameters to be optimized and process parameter recommendation information (recommended parameters are given for the process parameters to be optimized). The yield evaluation report, process parameters to be optimized and process parameter recommendation information may be obtained by evaluating the current yield of the object under test based on the test data of the object under test by the large language yield management model. The yield prediction report may be obtained by predicting the future yield of the object under test based on the test data of the object under test by the large language yield management model.
[0054] Regarding the above-mentioned input of the test data to the large language yield management model, the yield management data output by the large language yield management model can be obtained by filling the test data into the corresponding position in the corpus template (the template is used to instruct the large language yield management model to determine the yield management data), generating a dialogue corpus (as a prompt), and then inputting the dialogue corpus into the large language yield management model, and then obtaining the yield management data output by the large language yield management model. Among them, the corpus template contains a task instruction statement (for example, "Please provide the yield management data of the object under test"), and the corpus template reserves a filling position corresponding to the test data.
[0055] For example, the corpus template may be “the test data of the object under test includes [fill position], please provide the yield management data of the object under test”, and accordingly, the test data may be filled into the fill position in the corpus template to generate a dialogue corpus.
[0056] It can be understood that the big language yield management model is trained based on a corpus, and the corpus includes multiple dialogue corpus samples, and the dialogue corpus samples are generated based on the test data samples and the yield management data samples. The trained big language yield management model has the ability to determine (predict) the yield management data based on the test data. Furthermore, in the application stage of the big language yield management model, the test data of the object under test can be input into the big language yield management model to obtain the yield management data output by the big language yield management model, thereby realizing yield management using the big language yield management model in the production and manufacturing process of electronic equipment.
[0057] In one possible implementation, the debugging data includes description information of the object under test and one or more of the following information: process parameters, debugging parameters and debugging results; the debugging parameters are configuration parameters of the debugging equipment, and the debugging equipment is used to debug the object under test; the debugging results are data obtained by debugging the object under test through the debugging equipment.
[0058] The description information of the object to be measured is used to describe the name, product number, product structure, product material and product function of the object to be measured.
[0059] Here, the process parameters are explained. In the manufacturing process of electronic devices (such as display panels), the object to be tested can be an electronic device. The production process of electronic devices is a complex and precise task involving numerous process parameters, which have a decisive influence on the quality, performance and reliability of the product. The process parameters of electronic devices may include welding temperature, curing temperature (temperature for curing adhesives or coatings), welding time (time for heating the solder joints), curing time (time for curing adhesives or coatings), patch pressure (pressure applied to the patch machine during the SMT process), conveyor speed (the speed of the conveyor in assembly line operations), workshop humidity, workshop cleanliness, positioning accuracy (the position accuracy of components during assembly) and solder composition (type and composition of solder), etc.
[0060] In a possible implementation, when the test data includes a test result, the yield management data includes a yield evaluation report, and the step of inputting the test data into the large language yield management model and obtaining the yield management data output by the large language yield management model includes:
[0061] Based on the description information of the object under test, the test results and the yield evaluation index, a first dialogue data is generated, and the first dialogue data is used to instruct the large language yield management model to determine a yield evaluation report;
[0062] Input the first dialogue data into the big language yield management model to obtain the yield evaluation report output by the big language yield management model.
[0063] For the yield evaluation index, it can describe the value range of the index. Through the adjustment result, it can be judged whether the index of the object under test is within the value range. If the index of an object under test is within the value range, it indicates that the object under test is a good product. If the index of an object under test is not within the value range, it indicates that the object under test is a defective product.
[0064] Exemplarily, the test results and yield evaluation indicators can be filled into the corresponding positions in the first corpus template (the template is used to instruct the large language yield management model to determine the yield evaluation report), generate the first dialogue corpus (as a prompt word), and then input the first dialogue corpus into the large language yield management model. The large language yield management model can use the description information of the object under test in the first dialogue corpus to obtain the basic information of the object under test, and then the large language yield management model can perform yield evaluation on the object under test in combination with the test results and yield evaluation indicators in the first dialogue corpus, and then obtain the yield evaluation report output by the large language yield management model. Among them, the first corpus template contains a task instruction statement (for example, "Please give a yield evaluation report of the object under test"), and the first corpus template reserves the filling position corresponding to the description information of the object under test, the filling position corresponding to the test result, and the filling position corresponding to the yield evaluation indicator.
[0065] In a possible implementation, when the test data includes process parameters and test parameters, the yield management data includes a yield prediction report, and the step of inputting the test data into the large language yield management model and obtaining the yield management data output by the large language yield management model includes:
[0066] Based on the description information of the object under test, process parameters, test parameters and yield evaluation indicators, a second dialogue data is generated, and the second dialogue data is used to instruct the large language yield management model to determine a yield prediction report;
[0067] Input the second dialogue data into the big language yield management model to obtain the yield prediction report output by the big language yield management model.
[0068] Exemplarily, process parameters, tuning parameters and yield evaluation indicators can be filled into corresponding positions in a second corpus template (the template is used to instruct the large language yield management model to determine a yield prediction report) to generate a second dialogue corpus (as a Prompt), and then the second dialogue corpus is input into the large language yield management model. The large language yield management model can use the description information of the object under test in the second dialogue corpus to obtain basic information about the object under test. The process parameters can characterize the production process of the object under test, and the tuning parameters can characterize the tuning process of the object under test. The process parameters and tuning parameters provide a basis for predicting the tuning results of the object under test. Then, the large language yield management model can predict the tuning results of the object under test by combining the process parameters and tuning parameters in the second dialogue corpus. Then, the large language yield management model can predict the yield of the object under test by combining the yield evaluation indicators and the predicted tuning results, and then the yield prediction report output by the large language yield management model can be obtained.
[0069] Among them, the second corpus template contains a task instruction statement (for example, "Please give a yield prediction report for the object under test"), and the second corpus template reserves filling positions corresponding to the description information of the object under test, filling positions corresponding to the process parameters, filling positions corresponding to the adjustment parameters, and filling positions corresponding to the yield evaluation indicators.
[0070] The second corpus template may also reserve filling positions corresponding to historical yield assessment reports. Accordingly, process parameters, adjustment parameters, yield assessment indicators and historical yield assessment reports may be added to the second corpus template. The generated second dialogue corpus includes historical yield assessment reports. In this way, the large language yield management model can use historical yield assessment data during the yield prediction process to improve the accuracy of yield prediction.
[0071] In a possible implementation, when the test data includes process parameters and test results, the yield management data includes process parameters to be optimized and / or process parameter recommendation information, and the step of inputting the test data into the large language yield management model and obtaining the yield management data output by the large language yield management model includes:
[0072] Based on the description information of the object under test, the process parameters, the test results and the yield evaluation index, a third dialogue data is generated, and the third dialogue data is used to instruct the large language yield management model to determine the process parameters to be optimized and / or the process parameter recommendation information;
[0073] The third dialogue data is input into the large language yield management model to obtain the process parameters to be optimized and / or the process parameter recommendation information output by the large language yield management model.
[0074] Exemplarily, the process parameters, debugging results and yield evaluation indicators can be filled into the corresponding positions in the third corpus template (the template is used to instruct the large language yield management model to determine one or more of the following information: process parameters to be optimized and process parameter recommendation information) to generate a third dialogue corpus (as a Prompt word), and then the third dialogue corpus is input into the large language yield management model.
[0075] The big language yield management model can obtain the basic information of the object under test by using the description information of the object under test in the third dialogue data, and then the big language yield management model can evaluate the yield of the object under test in combination with the test results and yield evaluation indicators in the third dialogue data to obtain a yield evaluation report, and then the big language yield management model can analyze which process parameters of the multiple process parameters of the object under test need to be further optimized based on the yield evaluation report to determine the process parameters to be optimized, and can use the mapping relationship between the process parameters, test results and yield evaluation indicators and the yield (the big language yield management model can learn this mapping relationship through training) to give process parameter recommendation information for the process parameters to be optimized. By replacing the process parameters to be optimized with the recommended parameters in the process parameter recommendation information, production test personnel can be assisted to efficiently improve product yield.
[0076] Among them, the third corpus template contains a task instruction statement (for example, "Please give recommended process parameter information for the object under test"), and the third corpus template reserves filling positions corresponding to the description information of the object under test, filling positions corresponding to the process parameters, filling positions corresponding to the adjustment results, and filling positions corresponding to the yield evaluation indicators.
[0077] For example, two or more of the first dialogue data, the second dialogue data, and the third dialogue data can be combined into one dialogue data, and the combined dialogue data can be input into the big language yield management model, and then the yield management data output by the big language yield management model can be obtained. In this case, the yield management data includes one or more of the following information: yield evaluation report, yield prediction report, process parameters to be optimized, and process parameter recommendation information.
[0078] For example, Figure 2 is a schematic diagram of using a large language model for yield management provided in an embodiment of the present application, such as Figure 2 As shown, the big language yield management model can be a model based on LLama2. After obtaining the debugging data, the debugging data can be input into the big language yield management model based on LLama2, and then the big language yield management model can be used to perform yield management analysis, including: intelligent monitoring, intelligent prediction, intelligent analysis and process parameter recommendation.
[0079] For intelligent monitoring, the above-mentioned first dialogue data can be generated based on the commissioning data (such as description information of the object under test, commissioning results and yield evaluation indicators), and then the first dialogue data can be input into the big language yield management model to obtain the yield evaluation report output by the big language yield management model. Then, based on the yield evaluation report, it can be determined whether the yield of the object under test meets expectations in the recent period of time. If the yield does not meet expectations, a warning prompt will be given.
[0080] For intelligent prediction, the above-mentioned second dialogue data can be generated based on the debugging data (such as description information of the object under test, process parameters, debugging parameters and yield evaluation indicators), and then the second dialogue data can be input into the large language yield management model to obtain the yield prediction report output by the large language yield management model.
[0081] For intelligent analysis, the third dialogue data mentioned above can be generated based on the test data (such as description information of the object under test, process parameters, test results and yield evaluation indicators), and then the third dialogue data can be input into the large language yield management model to obtain the process parameters to be optimized output by the large language yield management model.
[0082] For process parameter recommendation, the third dialogue data may be generated based on the test data, and then the third dialogue data may be input into the large language yield management model to obtain the process parameter recommendation information output by the large language yield management model.
[0083] It can be understood that the yield of the object under test can be monitored in real time through intelligent monitoring and intelligent prediction; the yield of the object under test can be effectively optimized through intelligent analysis and process parameter recommendations; combined with yield monitoring and optimization, efficient yield management can be achieved.
[0084] In a possible implementation, when the commissioning data includes a commissioning result, and the commissioning result is a defect detection result, obtaining the commissioning data of the object under test includes:
[0085] Obtain a test image (2D image, 3D image or spectral image, etc.) of the object under test;
[0086] Input the test image into the image segmentation model and obtain the defect detection result output by the image segmentation model;
[0087] Among them, the image segmentation model is used to detect defects of the object under test (such as defect type, defect location, defect morphology, defect confidence score, contour size, defect area and contrast, etc.) based on the test image of the object under test.
[0088] It should be noted that the quality inspection of existing products is generally achieved through machine vision methods. Through various types of detectors, such as 2D cameras, 3D cameras, and hyperspectral cameras, image data of the object under test can be obtained. Then, deep learning defect detection and analysis are performed on the image data to obtain the quality and defect type of the object under test. Although the existing large language model has multimodal characteristics and can directly process image data, the amount of image data corresponding to each object under test in industrial visual inspection is large (generally more than several hundred megabytes), and industrial visual scenes generally have real-time requirements. Therefore, the efficiency of directly using a large language model to process the image data of the object under test is low.
[0089] In order to improve the efficiency of processing the image data of the object under test during the yield management process, after obtaining the test image of the object under test, the image segmentation model can be used to detect the test image to obtain a structured defect detection result with controllable data volume. The data volume of the defect detection result is significantly smaller than the data volume of the image data. The defect detection result can then be input into the large language yield management model to avoid directly using the large language model to process the image data of the object under test.
[0090] It is understandable that the complexity of the image segmentation model is significantly lower than that of the big language model. The sum of the time it takes for the image segmentation model to process the test images and the time it takes for the big language yield management model to process the defect detection results is also significantly shorter than the time it takes to directly use the big language yield management model to process the test images. It can be seen that in industrial vision scenarios, combining the image segmentation model with the big language yield management model can effectively improve the efficiency of processing image data of the object under test during the yield management process.
[0091] Figure 3 is a flow chart of model training and use provided in the embodiment of the present application, such as Figure 3As shown, the image segmentation model can be a Segment Anything Model (SAM), and the big language model used can be LLama2. In the model training stage, by training the SAM model, a SAM model adapted to a specific visual scene can be obtained, and by training LLama2, a big language yield management model based on LLama2 can be obtained. In the model application (use) stage, the big language yield management model based on LLama2 can be used to analyze the commissioning data (the defect detection results in the commissioning data can be obtained through the trained SAM model) to obtain yield management data.
[0092] In a possible implementation, the method further includes obtaining an image segmentation model by the following steps:
[0093] Based on the real data set, a simulated data set is generated through an image data generation model (Diffusion model or generative adversarial network, etc.). The real data set is constructed based on the historical test images of the object under test.
[0094] Based on the real data set and the simulated data set, the initial image segmentation model is trained to obtain the trained image segmentation model.
[0095] For example, Figure 4 is a flow chart of the training image segmentation model provided in the embodiment of the present application, such as Figure 4 As shown in the figure, taking the training of the SAM model as an example, the open source general SAM is mainly aimed at natural scenes, and its effect is very limited in some industrial specific scenarios, and it cannot meet the needs of a general defect detection model. Therefore, data from specific visual scenes can be used to fine-tune the SAM training.
[0096] Specifically, if Figure 4 As shown, visual data, such as 2D image data, 3D image data, and spectral image data, can be extracted from the historical test data of the equipment test machine (test equipment); for the visual data, some data deduplication schemes are used to remove similar redundant data in the data to ensure the diversity of the data set, and at the same time, the pre-trained model of SAM is used to quickly mark the data (after the pre-trained model of SAM gives the defect information, it is manually adjusted) to obtain the location, type, contour morphology and other information of the defect, thereby obtaining a high-quality real data set; using the real data set, a Diffusion model for data generation is trained, and a large amount of high-quality simulated data is generated based on the model; then based on the simulated data and the real data, a high-quality large-scale defect data set is obtained; based on this data set, the general segmentation model based on SAM is fine-tuned to obtain a SAM model adapted to specific visual scenes.
[0097] In a possible implementation, the method further includes obtaining a large language yield management model by following the steps:
[0098] Based on multiple groups of test samples and corpus sample templates, multiple dialogue corpus samples are generated, where a group of test samples includes: commissioning data samples and yield management data samples;
[0099] Build a corpus based on multiple dialogue corpus samples;
[0100] Based on the corpus, the initial large language model is trained (or fine-tuned) to obtain a trained large language model. The trained large language model is used as the large language yield management model.
[0101] Optionally, the debugging data sample includes process parameters, debugging parameters and debugging results, etc. The yield management data sample includes: a yield evaluation report, process parameters to be optimized and process parameter recommendation information.
[0102] The corpus sample template includes a task instruction statement (eg, "please provide the yield management data of the object under test"), and the corpus sample template reserves a filling position corresponding to the test data sample and a filling position corresponding to the yield management data sample.
[0103] For example, a corpus sample template may be "Question: The test data of the object under test includes [fill position 1]. Please provide the yield management data of the object under test. Answer: The yield management data of the object under test includes [fill position 2]". Accordingly, the test data sample can be filled into filling position 1 in the corpus sample template, and the yield management data sample can be filled into filling position 2 in the corpus sample template to generate a dialogue corpus sample. The dialogue corpus sample contains question content and answer content. In the process of training the large language model, the dialogue corpus sample can effectively guide the large language model to learn the ability to determine (predict) yield management data based on the test data.
[0104] Figure 5 is a flow chart of training a large language model provided in an embodiment of the present application, such as Figure 5As shown, during the debugging process, the product to be inspected (the object to be tested) will go through N debugging processes, and each debugging process will have corresponding debugging equipment. For any of the debugging processes 1-N, the debugging equipment in the debugging process will debug the product to be inspected according to the debugging parameters, obtain the test image, and input the test image into the SAM-based visual inspection model (a SAM model that has been trained to adapt to specific visual scenes), and the debugging results (that is, defect detection results) can be obtained. The debugging data may include process parameters, debugging parameters, and debugging results. Through the manual analysis of the debugging data by experts (that is, expert analysis), the expert analysis results can be obtained, including yield evaluation reports, process parameters to be optimized, and process parameter recommendation information.
[0105] The commissioning data obtained in the historical commissioning process can be used as commissioning data samples, and the expert analysis results obtained in the historical commissioning process can be used as yield management data samples. Figure 5 The data structuring process specifically includes: a debugging data sample and a yield management data sample can form a group of test samples. Based on multiple groups of test samples and corpus sample templates, multiple dialogue corpus samples can be generated. Based on multiple dialogue corpus samples, a corpus is constructed.
[0106] like Figure 5 As shown, after the data is structured (that is, after the corpus is built), the LLama2 model can be trained (or fine-tuned) based on the corpus, making full use of the expert knowledge in the sample, realizing the injection of information in specific fields, and obtaining a large language yield management model based on LLama2.
[0107] The yield management device based on a large language model provided in the present application is described below. The yield management device based on a large language model described below and the yield management method based on a large language model described above can refer to each other.
[0108] Figure 6 is a schematic diagram of the structure of a yield management device based on a large language model provided in an embodiment of the present application, such as Figure 6 As shown, the device comprises: a test object data acquisition module 10 and a yield data acquisition module 20. Wherein:
[0109] The measured object data acquisition module 10 is used to obtain the test data of the measured object;
[0110] The yield data acquisition module 20 is used to input the test data into the large language yield management model and obtain the yield management data output by the large language yield management model;
[0111] Among them, the big language yield management model is used to determine the yield management data based on the commissioning data. The big language yield management model is obtained by training the big language model based on the corpus. The corpus includes multiple dialogue corpus samples. The dialogue corpus samples are generated based on the commissioning data samples and the yield management data samples.
[0112] It can be understood that the detailed functional implementation of each of the above-mentioned units / modules can be found in the introduction of the aforementioned method embodiment, and will not be repeated here.
[0113] It should be understood that the above-mentioned device is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method, which will not be repeated here.
[0114] Based on the method in the above embodiment, an embodiment of the present application provides an electronic device, Figure 7 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, such as Figure 7 As shown, the electronic device may include: a processor (Processor) 810, a communication interface (Communications Interface) 820, a memory (Memory) 830 and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logic instructions in the memory 830 to execute the method in the above embodiment.
[0115] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.
[0116] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.
[0117] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.
[0118] It is understandable that the processor in the embodiment of the present application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.
[0119] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.
[0120] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions may be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.
[0121] It should be understood that the various numerical numbers involved in the embodiments of the present application are only used for the convenience of description and are not used to limit the scope of the embodiments of the present application.
[0122] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.
Claims
1. A yield management method based on a large language model, characterized in that: include: Obtain the commissioning data of the object under test; Inputting the test data into the large language yield management model to obtain the yield management data output by the large language yield management model; Among them, the big language yield management model is used to determine the yield management data based on the commissioning data, and the big language yield management model is obtained by training a big language model based on a corpus, and the corpus includes multiple dialogue corpus samples, and the dialogue corpus samples are generated based on the commissioning data samples and the yield management data samples.
2. The yield management method based on a large language model according to claim 1, characterized in that: The test data includes description information of the object under test and one or more of the following information: process parameters, test parameters and test results; The test parameters are configuration parameters of a test device, the test device is used to test the object under test, and the test results are data obtained by testing the object under test through the test device.
3. The yield management method based on a large language model according to claim 2, characterized in that: In the case where the test data includes the test result, the yield management data includes a yield evaluation report, and the inputting of the test data into the large language yield management model to obtain the yield management data output by the large language yield management model includes: Based on the description information of the object under test, the test result and the yield evaluation index, a first dialogue corpus is generated, wherein the first dialogue corpus is used to instruct the large language yield management model to determine a yield evaluation report; The first dialogue data is input into the large language yield management model to obtain a yield evaluation report output by the large language yield management model.
4. The yield management method based on a large language model according to claim 2, characterized in that: In the case where the test data includes the process parameters and the test parameters, the yield management data includes a yield prediction report, and the inputting of the test data into the large language yield management model to obtain the yield management data output by the large language yield management model includes: Based on the description information of the object under test, the process parameters, the test parameters and the yield evaluation index, a second dialogue data is generated, where the second dialogue data is used to instruct the large language yield management model to determine a yield prediction report; The second dialogue data is input into the large language yield management model to obtain a yield prediction report output by the large language yield management model.
5. The yield management method based on a large language model according to claim 2, characterized in that: In the case where the test data includes the process parameters and the test results, the yield management data includes the process parameters to be optimized and / or process parameter recommendation information, and the inputting of the test data into the large language yield management model to obtain the yield management data output by the large language yield management model includes: Based on the description information of the object under test, the process parameters, the test results and the yield evaluation index, a third dialogue corpus is generated, wherein the third dialogue corpus is used to instruct the large language yield management model to determine the process parameters to be optimized and / or process parameter recommendation information; The third dialogue data is input into the large language yield management model to obtain the process parameters to be optimized and / or process parameter recommendation information output by the large language yield management model.
6. The yield management method based on a large language model according to claim 2, characterized in that: In a case where the test data includes the test result, and the test result is a defect detection result, obtaining the test data of the object under test includes: Acquire a test image of the object under test; Inputting the test image into an image segmentation model to obtain a defect detection result output by the image segmentation model; The image segmentation model is used to detect defects of the object under test based on the test image of the object under test.
7. The yield management method based on a large language model according to claim 6, characterized in that: The image segmentation model is obtained by the following steps: Based on a real data set, a model is generated by image data to generate a simulated data set, wherein the real data set is constructed based on historical test images of the object under test; Based on the real data set and the simulated data set, an initial image segmentation model is trained to obtain a trained image segmentation model.
8. A yield management device based on a large language model, characterized in that: include: The measured object data acquisition module is used to obtain the test data of the measured object; A yield data acquisition module, used for inputting the test data into the large language yield management model, and acquiring the yield management data output by the large language yield management model; Among them, the big language yield management model is used to determine the yield management data based on the commissioning data, and the big language yield management model is obtained by training a big language model based on a corpus, and the corpus includes multiple dialogue corpus samples, and the dialogue corpus samples are generated based on the commissioning data samples and the yield management data samples.
9. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program runs on a processor, the processor is caused to execute the method according to any one of claims 1 to 7.
Citation Information
Cited By
Product yield control method and device based on large model
CN119904149A
