Yield prediction system based on machine learning model

Abstract

The invention discloses a yield prediction system based on a machine learning model, and relates to the field of yield prediction systems for semiconductor test equipment. The yield prediction system based on the machine learning model comprises a data acquisition module used for acquiring historical lot data of a product; the data calculation module is used for calculating the total yield according to the collected data; and the model training module is used for training a plurality of yield prediction models based on artificial intelligence based on the historical lot data to obtain a yield prediction model, the input of the yield prediction model is production data, and the output of the yield prediction model is predicted yield. According to the method, the yield prediction model can be obtained on the basis of existing historical lot data through a machine learning algorithm model in combination with a large amount of production data, dominant factors influencing the yield are found through training and feature extraction of multiple groups of data sets, meanwhile, the model of the data sets is adjusted for accuracy and regression values, and it is guaranteed that the yield prediction model is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of yield prediction systems for semiconductor testing equipment, and in particular to a yield prediction system based on a machine learning model. Background Art

[0002] Currently, the production of large-scale integrated circuits is a high-tech manufacturing process. Due to the wide variety of products and complex processing procedures, semiconductor companies are required to implement more lean management of the production process in order to better control the product yield in the processing flow. The yield rate in the production process is crucial to the production of advanced wafer fabs.

[0003] Regarding the above-mentioned related technologies, it is found that the current yield observation is mainly through checking whether the yield of each batch reaches the index, and then taking corresponding measures, but there is a certain lag in production efficiency, and the yield loss that has been caused cannot be recovered. Summary of the invention

[0004] In order to achieve the effect of predicting product yield in advance and taking corresponding measures in advance, the present application provides a yield prediction system based on a machine learning model.

[0005] The present invention is achieved in that:

[0006] A yield prediction system based on a machine learning model, comprising:

[0007] Data collection module, used to collect historical lot data of products;

[0008] A data calculation module calculates the total yield rate based on the collected data;

[0009] A model training module, which trains a plurality of yield prediction models based on artificial intelligence based on the historical lot data to obtain a yield prediction model, wherein the input of the yield prediction model is the production data and the output is the predicted yield;

[0010] The data recognition module automatically recognizes the production data of the product to be predicted in real time and inputs it into the model training module;

[0011] The prediction display module receives the predicted yield from the output of the model training module and displays the predicted yield in real time.

[0012] By adopting the above technical solution, the data acquisition module and the data calculation module are used to provide a large amount of data and algorithms for the model training module, so that the model training module can process a large amount of collected historical lot data based on artificial intelligence, and then obtain multiple yield prediction models, and determine the input and output of the yield prediction model, so as to output the predicted yield through the input production data, and provide data for the input of the yield prediction model through the setting of the data recognition module, and display the output results through the setting of the prediction display module, so as to facilitate more intuitive observation.

[0013] Furthermore, the historical lot data includes the coordinates of each die in the lot and test yield data.

[0014] By adopting the above technical solution, the coordinates of each die and the test yield data are provided to the model training module as historical lot data.

[0015] Furthermore, the test yield data includes the yield during CP testing and the yield during FT testing.

[0016] By adopting the above technical solution, the CP test yield and the FT test yield are used as the product test yield.

[0017] Furthermore, the method for calculating the total yield is to obtain the total yield of a wafer by taking the weighted average of the yields of all dies.

[0018] By adopting the above technical solution, the total yield is calculated according to the yield of each die.

[0019] Furthermore, it also includes an alarm module. When the predicted yield displayed by the prediction display module has a downward trend, the alarm module starts to perform an alarm operation.

[0020] By adopting the above technical solution, the alarm module is set to ensure that when the yield is predicted to have a downward trend, an alarm reminder can be given in time, which is convenient for operators to take measures in advance, thereby avoiding the loss of yield.

[0021] Furthermore, the alarm module includes a first warning unit, a second warning unit and a third warning unit, the second warning units are in two groups and are respectively arranged on both sides of the first warning unit, the third warning unit is installed on the upper end surface of the first warning unit, and the second warning unit and the third warning unit are both fixedly connected to the first warning unit.

[0022] By adopting the above technical solution, by designing the alarm module into a three-part structure of a first warning unit, a second warning unit and a third warning unit, different alarm effects can be achieved by cooperating the three parts when it is convenient to use, and different prediction models can be reversely judged according to the number of alarms and the accuracy rate, so as to better select the applicable yield prediction model.

[0023] Furthermore, the first warning unit is a screen display module for displaying the yield prediction model model used; the second warning unit is a sound alarm module; and the third warning unit is a multi-level light alarm module for displaying different lights according to the yield prediction results.

[0024] By adopting the above technical solution, by setting the first warning unit as a screen display module, the model of the current prediction model can be displayed through the screen display module during use, the model training module trains multiple yield prediction models based on artificial intelligence based on the historical lot data, and the system can number the trained multiple yield prediction models, which is convenient for distinguishing and marking different models, and also convenient for the screen display module to better correspond to the display, by setting the second warning unit as a sound alarm module, and two groups of sound alarm modules are arranged on both sides of the screen display module, so that when a sound alarm is needed, the two groups of sound alarm modules can alarm in sequence at intervals, thereby increasing the three-dimensionality of the alarm sound, making it easier for outsiders to hear the alarm sound, and by setting the third warning unit as a multi-level light alarm module, so that when the alarm is used, the sound alarm module of the second warning unit can be used to realize the use of sound and light alarm.

[0025] Furthermore, the third warning unit includes a red light group, a green light group and a yellow light group, and the red light group, green light group and yellow light group are arranged in sequence from top to bottom, and the red light group, green light group and yellow light group correspond to different thresholds of the predicted yield decline trend in sequence.

[0026] By adopting the above technical solution, by designing the multi-level light alarm module into a structure that cooperates with a red light group, a green light group and a yellow light group, different light groups can be used to correspond to different yield decline trend thresholds for convenient use. When different thresholds are reached, the corresponding alarm lights can be turned on. When an alarm is triggered, external personnel can observe the color of the light to timely understand the yield decline trend.

[0027] Furthermore, the alarm module includes a control unit, which is used to stop the alarm, and the control unit includes a manual switch and an electric switch.

[0028] By adopting the above technical solution and providing a control unit in the alarm module, it is convenient for the operator to take measures according to the alarm situation and then close the alarm manually or automatically.

[0029] Furthermore, the alarm module includes a wireless transmission unit for communicating between the alarm module and the system.

[0030] By adopting the above technical solution, the wireless transmission unit can ensure data transmission between the alarm module and the system host in the entire production environment, making it easier to better record the alarm information and for the host to better control the alarm module.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: the present application uses a machine learning algorithm model in combination with a large amount of production data to obtain a yield prediction model based on existing historical lot data, finds the dominant factors affecting the yield through training and feature extraction of multiple sets of data sets, and adjusts the model of the data set according to the accuracy and regression value to ensure that the yield prediction model is more accurate. When in use, the yield prediction model is used in conjunction with the data recognition module and the prediction display module to display the predicted yield, ensuring that under the specified hardware conditions such as the test machine, sorting machine, LB and socket, the yield of the next batch of test lots can also be predicted. If it is determined that the yield has a downward trend, corresponding measures can be taken in advance to keep the yield stable. It has the advantages of early prediction, avoiding product losses caused by lags, reducing the number of defective products, and improving the production efficiency of the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 Schematic diagram of the structure of the yield prediction system in the embodiment of the present invention.

[0034] Figure 2 It is a structural diagram of an alarm module in an embodiment of the present invention.

[0035] In the figure: 1, first warning unit; 2, second warning unit; 3, third warning unit; 31, red light group; 32, green light group; 33, yellow light group. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0038] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, the numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The technology, methods and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, methods and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0039] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, perpendicular, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the devices or elements referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention. The directional words "inside and outside" refer to the inside and outside relative to the contours of each component itself.

[0040] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" may include both "above" and "below". The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0041] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.

[0042] The following is combined with Figure 1 and attached Figure 2 This application is described in further detail.

[0043] In the semiconductor manufacturing process, the "wafer" stage refers to the state of the wafer after all key production steps have been completed. This stage indicates that the wafer has completed all transistor manufacturing, lithography, etching, doping and other processes, and the chip pattern on it has been fully formed. Wafer Out is an important term in the semiconductor manufacturing process, which usually refers to the state of the wafer coming out of the production line after all manufacturing steps and processing. The process after Wafer Out includes:

[0044] Dicing: The wafer is cut into individual chips.

[0045] Packaging: The cut chips will be packaged to protect the chips and enable them to be connected to external circuits.

[0046] Testing: After packaging, the chips will undergo a series of tests, including wafer-level testing and final testing. Only chips that pass the tests will be shipped.

[0047] A single chip after wafer cutting, scientifically known as die, becomes a chip after packaging: for example, 3,000 dies can be manufactured on a 6-inch wafer. Due to the influence of factors such as process, not all dies are qualified during the production process, so different dies have different definitions, and the yield reflects the process level and cost competitiveness of the wafer manufacturing plant. Therefore, the qualified die on each wafer is an important parameter that needs to be counted in production. GoodDie means the qualified die after each wafer is tested, BadDie: the qualified die after each wafer is tested, and the proportion of GoodDie in the total number can be the yield of the product. The yield determines the production efficiency of the production line, so yield observation and prediction is a very necessary task.

[0048] Example 1

[0049] Reference Figure 1 As shown, a yield prediction system based on machine learning model is proposed, including: data acquisition module, data calculation module, model training module, data recognition module and prediction display module, wherein the data acquisition module and the data calculation module provide training data and calculation method for the model training module, and the data acquisition module is used to collect the existing historical lot data and send it to the model training module as a training basis, and each set of data is based on the specified hardware conditions such as test machine, sorting machine, LB and socket.

[0050] Reference Figure 1As shown, the chip mass production test requires a test machine, a test socket, a chip sorter (handler), a change kit, and a test load board (LB) to form a complete final test (Final Test, FT) mass production test environment. The change kit includes a shuttle, a hotplate, a dock plate (also known as a platform plate), and a nest head. The dock plate is fixed on the socket and the loadboard. The chip is replaced by multiple mechanical arms of the handler to complete the chip FT mass production test. LB, or Load Board, is a printed circuit board (PCB) designed specifically for carrying the device under test (DUT) in the ATE test environment. The circuit board integrates the test socket, test pads, and load circuit elements that simulate actual working situations, aiming to construct a suitable hardware test environment during ATE testing. A batch of several wafers is called a lot, such as 8 wafers in a lot and 25 wafers in a lot. Usually, this number has a regular value based on the product. It is usually identified by an ID.

[0051] Among them, the historical lot data includes the coordinates of each die in the lot and the test yield data. The test yield data includes the yield during the CP test and the yield during the FT test. By corresponding the coordinates of each die to the CP test yield value and the FT test yield value, the model training module can train an artificial intelligence-based yield prediction model through the data correspondence to realize a yield prediction model that can be predicted. CP refers to the chip in the wafer stage, which is to test the performance and function of the chip by piercing the chip pins with probes. Sometimes this process is also called WS (Wafer Sort). The FT test is mainly carried out after the chip is packaged, and it is the final test of the packaged chip. Its purpose is to verify whether the reliability and performance of the chip under actual working conditions meet the design specifications, including its functions, power consumption and reliability. The obtained yield prediction model can be used in conjunction with the data recognition module and the prediction display module for prediction on the production line. During the production process, the data recognition module identifies the product data on the production line and supplies it to the yield prediction model for predicted yield output. After the predicted yield is output, it is intuitively displayed through the prediction display module to achieve the purpose of accurate yield prediction.

[0052] The data acquisition module is used to collect the historical lot data of the product, and the data calculation module can calculate the total yield of a wafer by weighted average of the yield of all dies in the collected data. The calculation of the total yield of a wafer can provide more accurate data information for the model training module and make it easier to combine a large amount of data for prediction.

[0053] The model training module can train multiple AI-based yield prediction models based on historical lot data to obtain a yield prediction model. The input of the yield prediction model is production data, and the output is the predicted yield. The data recognition module can automatically identify the production data of the product to be predicted in real time and input it into the model training module. The prediction display module is used to receive the predicted yield at the output of the model training module and display the predicted yield in real time. The yield prediction model here cooperates with the data recognition module and the prediction display module to serve as the prediction operation part of the production line. The product production data is supplied to the input of the yield prediction model through the data recognition module, and then the yield prediction model can display the predicted yield from the prediction display module through the output according to the input information.

[0054] Example 2

[0055] Reference Figure 1 As shown, a yield prediction system based on a machine learning model includes: a data acquisition module for collecting historical lot data of a product. A data calculation module for calculating the total yield based on the collected data. A model training module for training multiple yield prediction models based on artificial intelligence based on the historical lot data to obtain a yield prediction model, wherein the input of the yield prediction model is production data and the output is the predicted yield. A data recognition module for automatically identifying the production data of the product to be predicted in real time and inputting it into the model training module. A prediction display module for receiving the predicted yield at the output end of the model training module and displaying the predicted yield in real time. It also includes an alarm module, and when the predicted yield displayed by the prediction display module has a downward trend, the alarm module is started to perform an alarm operation. The setting of the alarm module is used to cooperate with the prediction system to interact with external operators. When the predicted yield displayed by the prediction display module has a downward trend, the alarm module is started to perform an alarm operation, so that the operator can take corresponding measures in advance according to the prediction to keep the yield stable.

[0056] Reference Figure 2As shown, the alarm module includes a first alarm unit 1, a second alarm unit 2 and a third alarm unit 3. The number of the second alarm unit 2 is two groups, and they are respectively arranged on both sides of the first alarm unit 1. The third alarm unit 3 is installed on the upper end surface of the first alarm unit 1, and the second alarm unit 2 and the third alarm unit 3 are fixedly connected to the first alarm unit 1. The alarm module is designed to be a three-part structure of the first alarm unit 1, the second alarm unit 2 and the third alarm unit 3. When it is convenient to use, different alarm effects can be achieved by cooperating with the three parts, and different prediction models can be judged in reverse according to the number of alarms and the accuracy rate, so as to better select the applicable yield prediction model. The first alarm unit 1 is a screen display module, which is used to display the model of the yield prediction model used. By setting the first alarm unit 1 as a screen display module, the model of the current prediction model can be displayed through the screen display module during use. The model training module trains multiple yield prediction models based on artificial intelligence based on the historical lot data, and the system can number the trained multiple yield prediction models, which is convenient for distinguishing and marking different models, and also convenient for the screen display module to display more correspondingly. The second alarm unit 2 is a sound alarm module. By setting the second warning unit 2 as a sound alarm module, and two groups of sound alarm modules are set on both sides of the screen display module, when a sound alarm is needed, the two groups of sound alarm modules can alarm in sequence at intervals, thereby increasing the three-dimensionality of the alarm sound and making it easier for outsiders to hear the alarm sound.

[0057] Reference Figure 2 As shown, the third warning unit 3 is a multi-level light alarm module, which is used to display different lights according to the yield prediction result. The third warning unit 3 includes a red light group 31, a green light group 32 and a yellow light group 33, which are arranged in sequence from top to bottom, and the red light group 31, the green light group 32 and the yellow light group 33 correspond to different thresholds of the predicted yield decline trend in sequence. By making the third warning unit 3 a multi-level light alarm module, it can be used in conjunction with the sound alarm module of the second warning unit 2 to realize the use of sound and light alarm when used for alarm, and by designing the multi-level light alarm module into a structure in which the red light group 31, the green light group 32 and the yellow light group 33 cooperate, different light groups can be used to correspond to different yield decline trend thresholds when used conveniently, for example, the yield decline trend can be represented by the slope, and different slope values ​​can be set to correspond to the red light group 31, the green light group 32 and the yellow light group 33, so that when the alarm is convenient, external personnel can timely understand the yield decline trend by observing the color of the light.

[0058] The alarm module includes a control unit, which is used to stop the alarm, and the control unit includes a manual switch and an electric switch. By setting the control unit in the alarm module, it is convenient for the operator to take measures according to the alarm situation and then manually or automatically turn off the alarm. The alarm module includes a wireless transmission unit for the alarm module to communicate with the system. The wireless transmission unit here can use short-range wireless transmission to ensure data transmission between the alarm module and the system host in the entire production environment, making it easier to record the alarm information and better control the alarm module.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A yield prediction system based on a machine learning model, characterized in that: include: Data collection module, used to collect historical lot data of products; A data calculation module calculates the total yield rate based on the collected data; A model training module, which trains a plurality of yield prediction models based on artificial intelligence based on the historical lot data to obtain a yield prediction model, wherein the input of the yield prediction model is the production data and the output is the predicted yield; The data recognition module automatically recognizes the production data of the product to be predicted in real time and inputs it into the model training module; The prediction display module receives the predicted yield from the output of the model training module and displays the predicted yield in real time.

2. The yield prediction system based on a machine learning model according to claim 1, characterized in that: The historical lot data includes the coordinates of each die in the lot and test yield data.

3. A yield prediction system based on a machine learning model according to claim 2, characterized in that: The test yield data includes the yield during CP testing and the yield during FT testing.

4. The yield prediction system based on a machine learning model according to claim 3, characterized in that: The method for calculating the total yield is to obtain the total yield of a wafer by taking the weighted average of the yields of all dies.

5. The yield prediction system based on a machine learning model according to claim 1, characterized in that: It also includes an alarm module. When the predicted yield rate displayed by the prediction display module shows a downward trend, the alarm module is activated to perform an alarm operation.

6. The yield prediction system based on a machine learning model according to claim 5, characterized in that: The alarm module comprises a first alarm unit (1), a second alarm unit (2) and a third alarm unit (3); the second alarm units (2) are in two groups and are respectively arranged on both sides of the first alarm unit (1); the third alarm unit (3) is installed on the upper end surface of the first alarm unit (1); and the second alarm unit (2) and the third alarm unit (3) are both fixedly connected to the first alarm unit (1).

7. The yield prediction system based on a machine learning model according to claim 6, characterized in that: The first warning unit (1) is a screen display module, used for displaying the yield prediction model type used; the second warning unit (2) is a sound alarm module; the third warning unit (3) is a multi-level light alarm module, used for displaying different lights according to the yield prediction result.

8. The yield prediction system based on a machine learning model according to claim 7, characterized in that: The third warning unit (3) comprises a red light group (31), a green light group (32) and a yellow light group (33), wherein the red light group (31), the green light group (32) and the yellow light group (33) are arranged in sequence from top to bottom, and the red light group (31), the green light group (32) and the yellow light group (33) correspond in sequence to different thresholds for predicting a yield decline trend.

9. The yield prediction system based on a machine learning model according to claim 8, characterized in that: The alarm module comprises a control unit, the control unit is used to stop the alarm, and the control unit comprises a manual switch and an electric switch.

10. The yield prediction system based on a machine learning model according to claim 9, characterized in that: The alarm module comprises a wireless transmission unit, which is used for the alarm module to communicate with the system.

Images