Wafer yield prediction method and device, electronic equipment and storage medium

Abstract

The invention relates to a wafer yield prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining defect density data and a defect distribution diagram of a to-be-predicted wafer, the defect density data comprises the area defect density of a plurality of areas of the wafer, and the defect distribution diagram indicates the position distribution of defects on the wafer; a first prediction model is utilized to determine a first prediction result according to the defect density data, and the first prediction result indicates the yield of the wafer predicted by the first prediction model; a second prediction model is utilized to determine a second prediction result according to the defect distribution diagram, and the second prediction result indicates the yield of the wafer predicted by the second prediction model; and determining the target yield of the wafer according to the first prediction result and the second prediction result. Therefore, the accuracy and efficiency of wafer yield prediction can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of semiconductor manufacturing technology, and in particular to a wafer yield prediction method and device, an electronic device and a storage medium. Background Art

[0002] In the integrated semiconductor manufacturing process, the prediction of wafer yield is crucial to reducing costs and improving production efficiency. Wafer yield refers to the ratio of the number of qualified chips on the whole wafer after completing all process steps to the total number of chips on the whole wafer. Wafer yield is usually affected by many factors, including equipment performance, process control, material quality, design factors, and defect density. Among them, with the advancement of semiconductor manufacturing technology, the size of wafers has gradually decreased, and defects on the wafer surface (such as tiny cracks, particles, contamination, etc.) have become the main factors affecting the yield. Both micro and macro defects on the wafer may cause chip failure, thereby reducing the yield.

[0003] Existing defect detection technologies, especially high-precision detection methods such as optical detection equipment such as scanning electron microscopes, have become key quality control tools. Although current defect detection equipment can effectively identify defects on wafers and output data such as defect type, quantity, location and distribution map, how to quickly and accurately predict wafer yield remains a challenge that needs to be solved. Summary of the invention

[0004] In view of this, the present disclosure proposes a wafer yield prediction method and device, an electronic device and a storage medium, which can improve the accuracy and efficiency of wafer yield prediction.

[0005] According to one aspect of the present disclosure, a wafer yield prediction method is provided, including: obtaining defect density data and a defect distribution map of a wafer to be predicted, the defect density data including regional defect densities of multiple regions of the wafer, and the defect distribution map indicating the position distribution of defects on the wafer; using a first prediction model to determine a first prediction result based on the defect density data, the first prediction result indicating the yield of the wafer predicted by the first prediction model; using a second prediction model to determine a second prediction result based on the defect distribution map, the second prediction result indicating the yield of the wafer predicted by the second prediction model; and determining a target yield of the wafer based on the first prediction result and the second prediction result.

[0006] In one possible implementation, the defect density data and defect distribution map of the wafer to be predicted are obtained, including: determining multiple regions of the wafer and the number of defects in each region according to a preset area division method, and determining the ratio between the number of defects in each region and the area of ​​each region as the regional defect density of each region; generating a defect distribution map according to the position of the defects on the wafer.

[0007] In one possible implementation, the first prediction model includes a linear regression model, wherein the generation process of the first prediction model includes: obtaining a first sample data set, the first sample data set including sample defect density data of multiple sample wafers and the actual yield corresponding to each sample wafer; using the first sample data set to fit the parameters in a preset linear regression model to obtain the first prediction model.

[0008] In one possible implementation, the second prediction model includes a convolutional neural network model, wherein the training process of the second prediction model includes: obtaining a second sample data set, the second sample data set including sample defect distribution maps of multiple sample wafers and the actual yield corresponding to each sample wafer; using a preset convolutional neural network model to output sample prediction results of each sample wafer according to the sample defect distribution map of each sample wafer, the sample prediction results indicating the predicted yield of the sample wafer output by the convolutional neural network model; optimizing the parameters in the convolutional neural network model according to the loss between the predicted yield of each sample wafer and the actual yield of each sample wafer to obtain the trained second preset model.

[0009] In one possible implementation, determining the target yield of the wafer based on the first prediction result and the second prediction result includes: based on a first weight coefficient corresponding to the first prediction model and a second weight coefficient corresponding to the second prediction model, performing weighted summation on the first prediction result and the second prediction result to obtain the target yield of the wafer, the sum of the first weight coefficient and the second weight coefficient being 1.

[0010] In one possible implementation, determining the target yield of the wafer based on the first prediction result and the second prediction result includes: using a target regression model to output the target yield of the wafer based on the first prediction result and the second prediction result, wherein the parameters in the target regression model include a weight coefficient for weighting the first prediction result output by the first prediction model and a weight coefficient for weighting the second prediction result output by the second prediction model.

[0011] In one possible implementation, the generation process of the target regression model includes: obtaining a third sample data set, the third sample data set including first sample prediction results, second sample prediction results and actual yields of multiple sample wafers, wherein the first sample prediction results include the predicted yield of the sample wafer determined by using the first prediction model, and the second sample prediction results include the predicted yield of the sample wafer determined by using the second prediction model; using the initial regression model to output the sample predicted yield of each sample wafer according to the first sample prediction results and the second sample prediction results of each sample crystal; optimizing the parameters in the initial regression model according to the loss between the sample predicted yield of each sample wafer output by the initial regression model and the actual yield of each sample wafer to obtain the target regression model.

[0012] According to another aspect of the present disclosure, a wafer yield prediction device is provided, including: an acquisition module, used to acquire defect density data and a defect distribution map of a wafer to be predicted, the defect density data including regional defect densities of multiple regions of the wafer, and the defect distribution map indicating the position distribution of defects on the wafer; a first prediction module, used to determine a first prediction result based on the defect density data using a first prediction model, the first prediction result indicating the yield of the wafer predicted by the first prediction model; a second prediction module, used to determine a second prediction result based on the defect distribution map using a second prediction model, the second prediction result indicating the yield of the wafer predicted by the second prediction model; and a target determination module, used to determine a target yield of the wafer based on the first prediction result and the second prediction result.

[0013] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0014] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions implement the above method when executed by a processor.

[0015] According to another aspect of the present disclosure, a computer program product is provided, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0016] According to various aspects of the present disclosure, by using a first prediction model to predict the yield of a wafer based on defect density data (that is, predicting the wafer yield based on defect density distribution characteristics), and using a second prediction model to predict the yield of a wafer based on a defect distribution map (that is, predicting the wafer yield based on defect position distribution characteristics), it is equivalent to predicting the yield of the wafer respectively from the defect density distribution and the defect position distribution, and then determining the target yield of the wafer based on the yields predicted by the two models. This can achieve the combination of the respective advantages of the two prediction models (that is, combining the two aspects of defect density distribution and defect position distribution data), and efficiently predict a more accurate target yield of the wafer, which improves the accuracy and efficiency of wafer yield prediction.

[0017] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0019] Figure 1 A flowchart of a wafer yield prediction method according to an embodiment of the present disclosure is shown.

[0020] Figure 2 A schematic diagram showing a wafer yield prediction process according to an embodiment of the present disclosure is shown.

[0021] Figure 3 A block diagram of a wafer yield prediction device according to an embodiment of the present disclosure is shown.

[0022] Figure 4 A block diagram of an electronic device 1900 according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0023] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0024] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0025] The term "and / or" herein is only a description of the association relationship of associated objects, indicating that there may 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. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent any one or more elements selected from the set consisting of A, B, and C. In the description of the present disclosure, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0026] It should be understood that the terms "first", "second", etc. in the claims, specifications and drawings of the present disclosure are used to distinguish different objects rather than to describe a specific order. The terms "include" and "comprise" used in the specifications and claims of the present disclosure indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.

[0027] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0028] The wafer yield prediction method of the embodiment of the present disclosure can be deployed on various terminal devices through software or hardware modification. The terminal device involved in the embodiment of the present disclosure may refer to a device with a wireless connection function and / or a wired connection function. The wireless connection function refers to the ability to connect to other devices through wireless connection methods such as wifi and Bluetooth. The terminal device involved in the embodiment of the present disclosure may also communicate with other devices through a wired connection function. The terminal device involved in the embodiment of the present disclosure may be a touch screen, a non-touch screen, or a screenless terminal. The touch screen can be controlled by clicking and sliding on the display screen with a finger or a stylus. The non-touch screen device can be connected to an input device such as a mouse, a keyboard, a touch panel, and the terminal device is controlled through the input device. For example, the device without a screen may be a Bluetooth speaker without a screen. For example, the terminal device of the present application may include but is not limited to a user equipment (UE), a mobile device, a mobile terminal, a handheld device, a tablet computer, a laptop, a PDA, a computing device, etc.

[0029] The wafer yield prediction method of the embodiment of the present disclosure can also be deployed on a server, which can be located in the cloud or locally, and can be a physical device or a virtual device, such as a virtual machine, a container, etc., with a wireless communication function, wherein the wireless communication function can be set in the chip (system) or other parts or components of the server. It can refer to a device with a wireless connection function, and the function of wireless connection refers to the ability to connect to other servers or terminal devices through wireless connection methods such as Wi-Fi and Bluetooth. The server involved in the embodiment of the present disclosure may also have the function of communicating with a wired connection. For example, the server of the embodiment of the present disclosure can be located in the cloud, communicate with the terminal device, receive the defect density data and defect distribution map of the wafer to be predicted sent by the terminal device, and use the wafer yield prediction method deployed on the server based on the defect density data and defect distribution map to obtain the predicted target yield, and return it to the terminal device to display the predicted target yield to the user in the terminal device.

[0030] Figure 1 FIG. 1 is a flow chart showing a method for predicting wafer yield according to an embodiment of the present disclosure. Figure 1 As shown, the method includes: step S11 to step S14.

[0031] In step S11 , defect density data and a defect distribution map of a wafer to be predicted are acquired, wherein the defect density data includes regional defect densities of multiple regions of the wafer, and the defect distribution map indicates position distribution of defects on the wafer.

[0032] It can be seen that there is a significant correlation between the defect density of the wafer and the wafer yield. Generally speaking, the higher the defect density on the wafer, the lower the yield of the wafer. This correlation is particularly evident in semiconductor manufacturing, because defects on the wafer will directly affect the performance and quality of a single chip (die), thereby affecting the yield of the entire wafer. Therefore, defect density is a standard indicator for measuring the number of defects on a wafer, and is usually used to evaluate the defects that occur during the production process and their impact on the yield. There are many ways to calculate defect density. The commonly used defect density calculation method is: Defect density = total number of defects / area. The total number of defects refers to the number of defects detected within a specified area. The area is usually expressed in square centimeters. For example, if 100 defects are detected on a wafer and the area of ​​the wafer is 200 centimeters 2 , then the defect density of the entire wafer = 100 / 200 = 0.5 defects / cm 2 .

[0033] Considering that the defect distribution in different areas on the same wafer is usually different, for example, the outer ring area of ​​the wafer is usually more prone to defects due to factors such as material stress and temperature gradient. Therefore, the defects in the edge area may have a greater impact on the yield, especially under uneven conditions during the production process. The central area of ​​the wafer is generally affected by a more uniform process, and the defect density and yield change relatively smoothly, but in some cases, the defects in the central area may also have a greater impact on the overall yield, especially when there are local problems with the process equipment. Therefore, the embodiment of the present disclosure proposes to divide the wafer into multiple areas, and determine the regional defect density of each area respectively, so as to predict the wafer yield based on the regional defect density distribution of multiple areas, so as to capture the different defect distribution characteristics of different areas of the wafer (such as the edge area and the central area), and more accurately reflect the impact of the defect density of different areas on the overall yield of the wafer; and it can also be combined with the defect distribution map (i.e., the position distribution of defects on the wafer) to achieve a more accurate yield prediction.

[0034] Therefore, in a possible implementation manner, the above-mentioned obtaining the defect density data and defect distribution map of the wafer to be predicted may include:

[0035] Determine multiple regions of the wafer and the number of defects in each region according to a preset region division method, and determine the ratio between the number of defects in each region and the area of ​​each region as the regional defect density of each region;

[0036] A defect distribution map is generated according to the locations of defects on the wafer.

[0037] The area division method may include, for example, dividing the wafer into a plurality of areas according to the radius of the wafer. For example, assuming that the radius of the wafer is r, the wafer may be divided into areas with a radius of The circular area (i.e., the center area, the inner radius area), and the area from the radius The two regions are the ring region from r to r (i.e., the edge region and the outer radius region); of course, for example, it can also be divided into the region with a radius of The circular area and radius arrive The ring area and radius The area division method may also include dividing the wafer into multiple areas according to a preset fixed ratio, for example, the 10% area along the outer edge of the wafer inwardly can be used as the edge area, and the remaining 90% area can be used as the center area. Of course, for example, the 10% area, 50% area, and 60% area may be divided inwardly from the outer edge of the wafer, and the embodiments of the present disclosure are not limited to this.

[0038] It should be understood that the sizes of the regions divided into the wafer can be evenly divided or unevenly divided. For example, the wafer can be divided into regions with a radius of The circular area and radius arrive The ring area and radius In fact, those skilled in the art can customize the number and size of regions into which the wafer is divided according to actual needs. The embodiments of the present disclosure do not limit the way the wafer is divided into regions.

[0039] In practical applications, for example, defect detection equipment known in the art (such as optical detection equipment such as scanning electron microscopes) can be used to detect the location of defects on the wafer and the number of defects in each area on the wafer, which is not limited to the embodiments of the present disclosure.

[0040] It should be understood that if the total area of ​​the wafer is known, the area of ​​each area divided on the wafer is also known. After detecting the number of defects in each area on the wafer, the ratio between the number of defects in each area and the area of ​​each area can be calculated to obtain the regional defect density of each area. For example, if the wafer is divided into an edge area and a central area, the regional defect density of the edge area can be expressed as: The difference defect density in the center area can be expressed as:

[0041] Among them, after detecting the position of the defect on the wafer, a defect distribution map can be generated based on the position of the defect on the wafer, and the defect distribution map can indicate the position with defects and the position without defects on the wafer (that is, indicating the position distribution of defects on the wafer), or, in other words, each pixel on the defect distribution map can indicate whether the position corresponding to each pixel on the wafer is defective. In practical applications, the defect distribution map can be represented as a heat map (that is, using color to indicate the position distribution of defects on the wafer), a grayscale map (using grayscale values ​​to indicate the position distribution of defects on the wafer), a matrix (using elements in the matrix to indicate the position distribution of defects on the circle), etc., and the embodiments of the present disclosure are not limited to this.

[0042] In step S12, a first prediction result is determined based on the defect density data using a first prediction model, where the first prediction result indicates the yield of the wafer predicted by the first prediction model.

[0043] In a possible implementation, the first prediction model may include a linear regression model, that is, the linear regression model may be used to predict the yield of the wafer according to defect density data (such as regional defect density of the central area and the edge area), wherein the generation process of the first prediction model may include:

[0044] Acquire a first sample data set, where the first sample data set includes sample defect density data of a plurality of sample wafers and a true yield corresponding to each sample wafer;

[0045] The first sample data set is used to fit the parameters in the preset linear regression model to obtain a first prediction model.

[0046] Among them, the sample defect density data of each sample wafer in the first sample data set can be generated by referring to the method of obtaining the defect density data in the above-mentioned step S11, that is, the sample wafer can be divided into multiple regions (such as edge regions and center regions), and the regional defect density of each region can be calculated; and the actual yield of each sample wafer can be detected by wafer yield detection equipment or other detection technologies known in the art, so as to obtain the above-mentioned first sample data set.

[0047] In practical applications, the first sample data set actually obtained may also contain noise data (such as invalid regional defect density) and abnormal data (such as the actual yield is too large or too small), missing values ​​(such as missing regional defect density, missing actual yield, etc.), inconsistent data formats (such as regional defect density may be density, number, or defects / cm 2 、Some are defects / micrometer 2 Therefore, the first sample data set can also be cleaned, that is, noise data and abnormal data can be removed, missing values ​​can be supplemented, data format can be unified, etc., to obtain the first sample data set after data cleaning to generate the first prediction model.

[0048] Among them, those skilled in the art can customize the linear regression model. The embodiment of the present disclosure does not limit the specific expression form of the linear regression model. For example, assuming that the areas divided into the wafer are the central area and the edge area, the linear regression model can be expressed as formula (1):

[0049] y=w1*Dedge+w2* Dcenter+b (1)

[0050] Among them, y represents the predicted yield, Dedge represents the regional defect density in the edge area, Dcenter represents the regional defect density in the center area, w1 and w2 are weight parameters in the linear regression model, and b is the bias parameter.

[0051] It should be understood that the sample defect density data of multiple sample wafers and the actual yield corresponding to each sample wafer are known, which is equivalent to Dedge, Dcenter and y in the known formula (1). Based on this, a linear fitting method known in the art, such as the least squares method, the gradient descent method, the normal equation (Least Squares), etc., can be used to fit the parameters in the above-mentioned linear regression model, namely w1, w2 and b, based on the sample defect density data of multiple sample wafers and the actual yield corresponding to each sample wafer, that is, to obtain the parameter values ​​of the parameters in the linear regression model. The values ​​of w1, w2 and b in formula (1) are known, and the first prediction model is obtained. The input data of the first prediction model can be the regional defect density of the central area and the regional defect density of the edge area, and the output is the predicted wafer yield.

[0052] It should be noted that using the linear regression model shown in the above formula (1) to generate the first prediction model is a possible implementation method provided by the embodiment of the present disclosure. In fact, those skilled in the art can customize the linear regression model that adapts to the conditions such as the number of regions divided on the wafer to generate the corresponding first prediction model, and the embodiment of the present disclosure does not limit this.

[0053] In step S13, a second prediction result is determined according to the defect distribution map using a second prediction model, where the second prediction result indicates the yield of the wafer predicted by the second prediction model.

[0054] It is known that the convolutional neural network (CNN) model is usually used for image processing and can also be used to process data with spatial structure. Therefore, the second prediction model can adopt a convolutional neural network model. For example, for a defect distribution map, a convolutional neural network model can be used to automatically extract defect distribution features from the defect distribution map and predict the wafer yield. Based on this, the training process of the second prediction model may include:

[0055] Acquire a second sample data set, where the second sample data set includes sample defect distribution maps of a plurality of sample wafers and a true yield corresponding to each sample wafer;

[0056] Outputting a sample prediction result of each sample wafer according to a sample defect distribution map of each sample wafer using a preset convolutional neural network model, wherein the sample prediction result indicates a predicted yield of the sample wafer output by the convolutional neural network model;

[0057] According to the loss between the predicted yield of each sample wafer and the actual yield of each sample wafer, the parameters in the convolutional neural network model are optimized to obtain the second preset model after training.

[0058] Among them, the sample defect distribution map of each sample wafer in the multiple sample wafers can be generated by referring to the method of obtaining the defect distribution map in the above step S11, that is, the sample defect distribution map corresponding to each sample wafer can be generated according to the position of the defect on each sample wafer; and the actual yield of each sample wafer can be detected by detection technologies such as wafer yield detection equipment known in the art, so as to obtain the above-mentioned second sample data set, wherein the actual yield of each sample wafer is also the label of the sample defect distribution map of each sample wafer. It should be understood that the sample wafers used in the second sample data set and the sample wafers used in the above-mentioned first sample data set can be sample wafers from the same batch, or sample wafers from different batches, that is, the sample wafers used in the second sample data set and the sample wafers used in the above-mentioned first sample data set may or may not have an intersection, and the embodiments of the present disclosure are not limited to this.

[0059] Among them, those skilled in the art can customize the specific network structure of the convolutional neural network model according to actual needs. For example, the convolutional neural network model may include multiple convolutional layers, pooling layers and fully connected layers to achieve the extraction of image features and prediction. Among them, the convolutional layer (Conv2D) can be used to extract image features (such as defect distribution features) of the input image, the pooling layer (MaxPooling2D) can be used to reduce the dimension of the input image and reduce the amount of calculation. The fully connected layer (Dense) can be used to map the features extracted by the convolutional layer to the predicted value of the yield. It should be understood that those skilled in the art can customize the operator size (such as the size of the convolution operator) and operator type (such as the type of pooling operator used in the pooling layer can be a maximum pooling operator, an average pooling operator, etc., and an activation operator can also be set in the convolutional layer) used by the convolutional layer, the pooling layer, and the fully connected layer, and the embodiment of the present disclosure is not limited to this.

[0060] Among them, according to the loss between the predicted yield of each sample wafer and the actual yield of each sample wafer, the parameters in the convolutional neural network model are optimized, for example, it may include: calculating the deviation between the predicted yield of each sample wafer and the actual yield of each sample wafer, and calculating the sum, mean or mean square value of the deviation between the predicted yield and the actual yield of each sample wafer in multiple sample wafers as the total loss, and then, optimizing the parameters in the convolutional neural network model based on the total loss through optimization algorithms such as gradient descent and back propagation to obtain a second prediction model after training.

[0061] It should be understood that the above-mentioned training process for the convolutional neural network model can be iteratively executed for multiple rounds until the preset training end conditions are reached. For example, the training end conditions may include that the training rounds reach the specified rounds, the total loss has converged or is set to 0, etc., and the convolutional neural network model that meets the above-mentioned training end conditions is determined as the second prediction model; wherein, a test set can also be set to test the model performance (such as accuracy, recall rate and other performance indicators) of the trained convolutional neural network model with the test set. If the model test result does not meet the preset standard, a second sample data set with a larger data volume can be constructed to continue training the convolutional neural network model until the model performance of the trained convolutional neural network model reaches the preset standard to obtain a second prediction model.

[0062] In step S14, a target yield of the wafer is determined according to the first prediction result and the second prediction result.

[0063] Among them, the results predicted by the first prediction model and the second prediction model can be integrated to make a final prediction in a weighted distribution manner, for example, a weighted average method or a model weighted integration method can be used. That is, a weighted sum can be performed on the first prediction result and the second prediction result output by the first prediction model and the second prediction model (such as a linear regression model and a convolutional neural network model) to obtain a final prediction result, that is, a predicted target yield of the wafer.

[0064] Therefore, the above-mentioned determination of the target yield of the wafer based on the first prediction result and the second prediction result may include: based on the first weight coefficient corresponding to the preset first prediction model and the second weight coefficient corresponding to the second prediction model, weighted summing the first prediction result and the second prediction result to obtain the target yield of the wafer, and the sum of the first weight coefficient and the second weight coefficient is 1. This method can be understood as multiplying the prediction result of each model by a corresponding weight, and weighted summing them to obtain the final yield prediction value. This weighted method can be expressed as formula (2):

[0065] Y final =W1·y linear +W2·y cnn (2)

[0066] Where W1 and W2 represent the first weight coefficient corresponding to the first prediction model and the second weight coefficient corresponding to the second prediction model, respectively. linear Represents the first prediction result output by the first prediction model, y cnn represents the second prediction result output by the second prediction model, Y final Represents the predicted target yield. By satisfying W1+W2=1, it can be ensured that the final prediction result is a reasonable weighted average.

[0067] Among them, by combining the first prediction model (linear regression model) and the second prediction model (convolutional neural network model) and assigning different weight coefficients to them, the weighted integrated model can be used to predict the yield. In this way, the advantages of the two prediction models can be combined (i.e., the first prediction model extracts the defect density distribution and the second prediction model extracts the defect location distribution) to obtain a more accurate yield prediction value (i.e., a more accurate target yield), thereby improving the accuracy and robustness of the prediction.

[0068] Optionally, the first weight coefficient and the second weight coefficient can be set according to the importance of the two prediction models combined with manual experience. For example, the first weight coefficient corresponding to the first prediction model can be set to 0.6, and the second weight coefficient corresponding to the second prediction model can be set to 0.4.

[0069] Optionally, in order to obtain more accurate first weight coefficients and second weight coefficients, a new linear regression model may be trained to optimize the first weight coefficients and second weight coefficients, so that the final predicted target yield is as close to the actual yield value as possible. In other words, a target regression model may be directly trained to perform weighted summation of the first prediction result and the second prediction result to obtain the predicted target yield. Thus, the above-mentioned determination of the target yield of the wafer based on the first prediction result and the second prediction result may include:

[0070] The target regression model is used to output the target yield of the wafer according to the first prediction result and the second prediction result, wherein the parameters in the target regression model include a weight coefficient for weighting the first prediction result output by the first prediction model and a weight coefficient for weighting the second prediction result output by the second prediction model. The target regression model can be expressed as a linear regression model form shown in the above formula (2) (i.e., without a bias term), or can also be expressed as a linear regression model form shown in the above formula (1) (i.e., with a bias term), and this is not limited to the embodiments of the present disclosure.

[0071] Wherein, using the target regression model to output the target yield of the wafer according to the first prediction result and the second prediction result is equivalent to using the parameters in the target regression model to weighted sum the first prediction result and the second prediction result to obtain the target yield of the wafer. It should be understood that when the target regression model is expressed in the form of the above formula (2), the parameters in the target regression model (that is, the weight coefficient for weighting the first prediction result output by the first prediction model and the weight coefficient for weighting the second prediction result output by the second prediction model) can be determined as the above first weight coefficient and the above second weight coefficient, and this embodiment of the present disclosure is not limited to this. Wherein, for example, common regression methods (such as ridge regression, Lasso regression) can be used to train the above target regression model, and this embodiment of the present disclosure is not limited to this.

[0072] In a possible implementation, the generation process of the target regression model may include:

[0073] Acquire a third sample data set, wherein the third sample data set includes a first sample prediction result, a second sample prediction result, and an actual yield of a plurality of sample wafers, wherein the first sample prediction result includes a predicted yield of the sample wafer determined using the first prediction model, and the second sample prediction result includes a predicted yield of the sample wafer determined using the second prediction model;

[0074] Outputting a sample prediction yield of each sample wafer according to a first sample prediction result and a second sample prediction result of each sample crystal using an initial regression model;

[0075] According to the loss between the sample predicted yield of each sample wafer output by the initial regression model and the actual yield of each sample wafer, the parameters in the initial regression model are optimized to obtain the target regression model.

[0076] It should be understood that the sample wafers used in the third sample data set may be from the same batch of sample wafers as the sample wafers used in the first sample data set and the second sample data set, or may be from different batches of sample wafers, that is, the sample wafers used in the third sample data set may include all or part of the sample wafers used in the first sample data set and the second sample data set, or may be different from the sample wafers used in the first sample data set and the second sample data set, and the embodiments of the present disclosure are not limited to this.

[0077] After fitting the first prediction model using the above-mentioned first sample data set and training the second prediction model using the second sample data set, the first prediction model can be used to output the first sample prediction results of each sample wafer based on the sample defect density data of each sample wafer used in the third sample data set, and the second prediction model can be used to output the second sample prediction results of each sample wafer based on the sample defect distribution map of each sample wafer used in the third sample data set, and the actual yield of each sample wafer used in the third sample data set can be detected by detection technologies such as wafer yield detection equipment known in the art to obtain the above-mentioned third sample data set. It should be understood that if the sample wafers used in the third sample data set include the sample wafers used in the above-mentioned first sample data set and the second sample data set, the first sample prediction results and the second sample prediction results generated in the process of generating the first prediction model and in the process of training the second prediction model, as well as the corresponding actual yields, the embodiments of the present disclosure do not limit the method for obtaining the third sample data set.

[0078] Using the third sample data set to train the target regression model is equivalent to using the prediction results of the two prediction models as new features and combining them with the actual yield of the sample wafer to train the target regression model. That is, the outputs of the two prediction models can be used as a new training set to train an initial regression model (such as a linear regression model) to learn the optimal weight coefficients so that the prediction results (i.e., the target yield) output by the trained target regression model are as close to the actual yield as possible. Among them, the initial regression model can be expressed as the linear regression model form shown in the above formula (2) (i.e., without a bias term), or it can also be expressed as the linear regression model form shown in the above formula (1) (i.e., with a bias term), and this is not limited to the embodiments of the present disclosure.

[0079] The initial regression model is used to output the sample predicted yield of each sample wafer according to the first sample prediction result and the second sample prediction result of each sample crystal, that is, the first sample prediction result and the second sample prediction result of each sample crystal are input into the initial regression model to obtain the sample predicted yield of each sample crystal output by the initial regression model; then, the loss between the sample predicted yield of each sample wafer output by the initial regression model and the actual yield of each sample wafer can be calculated to optimize the parameters in the initial regression model, that is, to optimize the weight coefficient for weighting the first prediction result output by the first prediction model and the weight coefficient for weighting the second prediction result output by the second prediction model in the initial regression model; or, when the initial regression model also includes a bias item parameter, the bias item parameter in the initial regression model can also be optimized at the same time.

[0080] Among them, according to the loss between the sample predicted yield of each sample wafer output by the initial regression model and the actual yield of each sample wafer, the parameters in the initial regression model are optimized, for example, it can include: calculating the deviation between the sample predicted yield of each sample wafer and the actual yield of each sample wafer, and calculating the sum, mean or mean square value of the deviation between the sample predicted yield and the actual yield of each sample wafer in multiple sample wafers as the total loss, and then, using the optimization algorithm known in the art to optimize the parameters in the initial regression model based on the total loss to obtain the above-mentioned target regression model. It should be understood that the above-mentioned optimization process for the initial regression model can be iteratively executed for multiple rounds until a preset end condition is reached. For example, the end condition can include that the iteration rounds reach a specified round, the total loss has converged or is set to 0, etc., so that the initial regression model that reaches the above-mentioned end condition is determined as the target regression model, so as to use the target regression model to achieve the output of the target yield of the wafer according to the first prediction result and the second prediction result.

[0081] According to the wafer yield prediction method of the embodiment of the present disclosure, the first prediction model is used to predict the wafer yield based on the defect density data (that is, the wafer yield is predicted based on the defect density distribution characteristics), and the second prediction model is used to predict the wafer yield based on the defect distribution map (that is, the wafer yield is predicted based on the defect position distribution characteristics). This is equivalent to predicting the wafer yield from the defect density distribution and the defect position distribution respectively, and then determining the target yield of the wafer based on the yields predicted by the two models. This can achieve the combination of the respective advantages of the two prediction models (that is, combining the two aspects of defect density distribution and defect position distribution data) to efficiently predict a more accurate target yield of the wafer, which improves the accuracy and efficiency of wafer yield prediction.

[0082] Based on the above steps S11 to S14 of the embodiment of the present disclosure, the embodiment of the present disclosure also provides Figure 2 A schematic diagram of a wafer yield prediction process is shown in FIG. Figure 2As shown, the prediction process includes: a data input stage, including inputting defect density data and defect distribution image data of the wafer (a two-dimensional image of wafer defects, that is, a defect distribution map); a linear regression model training stage, including: performing linear regression modeling on the defect density data, inputting the defect density data, and outputting the linear regression prediction result (that is, the first prediction result y_linear_pred); a CNN model training stage, including: performing convolutional neural network modeling on the defect image data, inputting the defect distribution image data, and outputting the CNN prediction result (that is, the second prediction result y_cnn_pred); an integration stage, including: weighted integration of the prediction results output by the two models, and optimizing the weight coefficients W1 and W2 through the regression model (that is, using the target regression model to determine the first weight coefficient and the second weight coefficient), input the first prediction result y_linear_pred and the second prediction result y_cnn_pred, and output the weighted integrated prediction result (i.e., the predicted target yield y_final_pred); the prediction and result output stage, including: using the final integrated model (i.e., the model obtained by integrating the first prediction model and the second prediction model using the first weight coefficient and the second weight coefficient) to predict the new data, and output the yield prediction result, input the trained integrated model and test data (i.e., the new defect density data and the defect distribution map), output the final yield loss prediction result (i.e., output the predicted target yield y_final_pred), and then, the model can be deployed in the actual production environment to predict the wafer yield online and in real time.

[0083] In practical applications, after obtaining the trained first prediction model, the second prediction model and the above-mentioned target regression model, the trained models can be deployed in the production environment and integrated into the yield management system to obtain the defect density data and defect distribution map of the wafer in the production process in real time, so as to achieve online prediction of the target yield of the wafer. Among them, an early warning mechanism can also be added to the yield management system, that is, if the predicted target yield is lower than a certain threshold, the system can trigger an early warning to prompt the production line to adjust the process parameters or check the equipment status.

[0084] In actual application, new sample data sets can also be collected regularly to incrementally train each model to ensure that the model can adapt to the ever-changing production environment and process conditions, and each model can also be continuously adjusted and optimized according to the actual production results to ensure that its prediction accuracy improves over time. In addition, each prediction model can also be integrated with other production management systems to form a complete intelligent manufacturing system to achieve comprehensive optimization and intelligence of the production process, which is not limited in the embodiments of the present disclosure.

[0085] The key to wafer yield prediction lies in how to efficiently and accurately analyze defect data and its impact on yield. The prediction method of the disclosed embodiment can significantly improve prediction efficiency and accuracy by introducing machine learning and deep learning, combined with regional analysis and visualization methods, and provide effective support for subsequent defect analysis and yield improvement.

[0086] According to the prediction method of the embodiment of the present disclosure, the efficiency and accuracy of yield loss prediction can be greatly improved by using machine learning and deep learning methods. By constructing a prediction model, the yield can be predicted directly based on historical data and current defect data without analyzing each area one by one. By using a training set containing a large amount of historical wafer defect data (including defect type, location, distribution, density and other information), a deep learning model (such as a convolutional neural network, CNN) is trained to identify defect patterns and predict the yield of each wafer. The defect concentration area analysis is realized, and the defect distribution is analyzed by a convolutional neural network (CNN), and the defect concentration areas on the wafer are automatically detected and marked, reducing manual intervention and improving analysis efficiency.

[0087] According to the prediction method of the embodiment of the present disclosure, the calculation can be simplified by using a regional analysis method based on the characteristics of different regions on the wafer. For the defect density and yield in each region, a prediction model is constructed to analyze the regional defect density, and finally summarized into the prediction of the entire wafer. By dividing the wafer into different regions, according to the defect density of each region, a linear regression model is used to predict the yield of the entire wafer based on the regional defect density of each region, which can reduce the global calculation amount. In addition, the defect distribution map is combined with the yield to generate a heat map, which intuitively displays the failure rate and yield loss of each region on the wafer, which is convenient for subsequent defect analysis.

[0088] According to the prediction method of the embodiment of the present disclosure, it is possible to reduce the computational cost. On large-scale data sets, the amount of computation required for each prediction can be reduced by adopting incremental learning and online learning methods, thereby avoiding loading all data for training at one time and reducing the computational cost.

[0089] According to the prediction method of the embodiment of the present disclosure, in order to facilitate defect analysis, the defect distribution map, regional defect density and prediction results on the wafer can be intuitively displayed through integrated visualization tools. For example, the yield of different areas can be quickly displayed by means of heat maps, 3D graphics or interactive charts, helping engineers to quickly find the concentrated areas of defects and potential sources of problems. In addition, it is also possible to visualize the defect concentration areas, and use heat maps to show the defect concentration areas in color depth, and combine actual production data and yield heat maps to quickly find the key areas that affect the yield. In addition, it is also possible to visualize the prediction results, and display the yield loss prediction results in a graphical way to clarify the contribution of each area to the overall yield, and help make more accurate adjustments and optimizations.

[0090] As mentioned above, there is a significant negative correlation between wafer defect density and wafer yield. By effectively controlling and reducing defect density, the wafer yield can be significantly improved, which is also an important goal in the semiconductor manufacturing process. By establishing a model between defect density and yield, the wafer yield can be predicted based on the defect density, helping manufacturers to adjust process parameters in a timely manner and optimize the production process, thereby improving the wafer yield. In turn, the wafer yield can be predicted in advance, reducing the workload of subsequent CP testing, shortening the CP testing cycle, and thus improving the production efficiency of memory chips.

[0091] The prediction method of the disclosed embodiment realizes the innovative application of integrated linear regression and convolutional neural network. In the field of semiconductor yield prediction, traditional methods usually use linear regression or deep learning (CNN) methods alone, while the disclosed embodiment proposes a unique integration method that combines linear regression with convolutional neural network (CNN) to output the final yield prediction result in a weighted manner. By integrating linear regression and CNN models, the advantages of both are utilized to improve the prediction effect and improve the accuracy of yield loss prediction, especially in the processing of complex defect data. The disclosed embodiment proposes a weighted decision-making mechanism for the integrated method, that is, the prediction results of the two models are synthesized by allocating specific weights.

[0092] The prediction method of the disclosed embodiment implements a training method for weight optimization, by using the prediction results of historical data (i.e., the prediction results generated by linear regression and CNN, respectively) as new feature inputs, and further training a regression model to optimize the final weight coefficients, thereby determining the importance of each model in the final prediction. This optimization of weight coefficients through regression models, rather than simple manual settings, improves the accuracy and adaptability of the model, and can dynamically adjust weights according to different data sets. The disclosed embodiment proposes a weight optimization algorithm and regression training steps, that is, optimizing the weights in the integrated model through the training set to improve prediction accuracy.

[0093] The prediction method of the disclosed embodiment realizes the prediction input feature design based on defect density, and innovatively uses defect density (including defect density in edge areas and center areas) as the key input feature for predicting yield, rather than relying solely on traditional single defect data (such as defect number, location, etc.). By using the defect density in different areas on the wafer as an independent feature, the impact of defects in different areas on the yield can be more accurately reflected, thereby enhancing the accuracy and precision of the prediction model. The disclosed embodiment proposes a defect density area division and regional feature extraction method, that is, a prediction model is constructed by calculating defect density by region.

[0094] The prediction method of the disclosed embodiment realizes the application of a deep learning model based on a defect distribution image. In traditional defect detection and yield prediction, image data is rarely used for prediction. The present technology processes the defect distribution map of the wafer (such as a heat map, a two-dimensional defect coordinate map) through a convolutional neural network model, automatically learns defect patterns and predicts yield loss. The convolutional neural network model can automatically extract features from complex defect distribution images, and has a powerful spatial information processing capability, which can more effectively identify potential influencing factors than traditional numerical methods. The disclosed embodiment proposes a method for processing defect distribution maps, that is, a technique for inputting image data into CNN and predicting yield loss.

[0095] The prediction method of the disclosed embodiment realizes an integrated method for improving prediction efficiency. It is not only innovative in improving prediction accuracy, but also proposes a method for improving prediction efficiency. By combining traditional linear regression with deep learning, the high computational cost and low efficiency of a single model on complex problems are avoided. By integrating two different prediction models (linear regression model and CNN model), efficient processing of complex defect data is achieved, especially in large-scale production data, which can significantly improve the prediction efficiency. The disclosed embodiment proposes the optimization of the computational efficiency of the integrated model, that is, how to improve the computational efficiency of the prediction process by combining different models and methods.

[0096] The prediction method of the disclosed embodiment implements a weighted mechanism for dynamically adjusting the model output. Unlike the traditional static model weight setting, it can adapt to changes in different data sets and different process conditions by dynamically adjusting the weights of linear regression and CNN outputs. Dynamically adjusting weights means that the model can automatically adjust according to actual conditions, data changes, and actual prediction errors, increasing the influence of a certain prediction model on the final prediction results, thereby improving the flexibility and accuracy of the prediction. The disclosed embodiment proposes a dynamic adjustment method for the weighted mechanism, that is, automatically adjusting the weights of linear regression and CNN outputs according to different data and process conditions.

[0097] Figure 3 A block diagram of a wafer yield prediction device according to an embodiment of the present disclosure is shown. Figure 3 As shown, the device comprises:

[0098] An acquisition module 301 is used to acquire defect density data and a defect distribution map of a wafer to be predicted, wherein the defect density data includes regional defect densities of multiple regions of the wafer, and the defect distribution map indicates the position distribution of defects on the wafer;

[0099] A first prediction module 302, configured to determine a first prediction result according to the defect density data using a first prediction model, wherein the first prediction result indicates a yield of the wafer predicted by the first prediction model;

[0100] A second prediction module 303, configured to determine a second prediction result according to the defect distribution map using a second prediction model, wherein the second prediction result indicates a yield of the wafer predicted by the second prediction model;

[0101] The target determination module 304 is used to determine the target yield of the wafer according to the first prediction result and the second prediction result.

[0102] In one possible implementation, the defect density data and defect distribution map of the wafer to be predicted are obtained, including: determining multiple regions of the wafer and the number of defects in each region according to a preset area division method, and determining the ratio between the number of defects in each region and the area of ​​each region as the regional defect density of each region; generating a defect distribution map according to the position of the defects on the wafer.

[0103] In one possible implementation, the first prediction model includes a linear regression model, wherein the generation process of the first prediction model includes: obtaining a first sample data set, the first sample data set including sample defect density data of multiple sample wafers and the actual yield corresponding to each sample wafer; using the first sample data set to fit the parameters in a preset linear regression model to obtain the first prediction model.

[0104] In one possible implementation, the second prediction model includes a convolutional neural network model, wherein the training process of the second prediction model includes: obtaining a second sample data set, the second sample data set including sample defect distribution maps of multiple sample wafers and the actual yield corresponding to each sample wafer; using a preset convolutional neural network model to output sample prediction results of each sample wafer according to the sample defect distribution map of each sample wafer, the sample prediction results indicating the predicted yield of the sample wafer output by the convolutional neural network model; optimizing the parameters in the convolutional neural network model according to the loss between the predicted yield of each sample wafer and the actual yield of each sample wafer to obtain the trained second preset model.

[0105] In one possible implementation, determining the target yield of the wafer based on the first prediction result and the second prediction result includes: based on a first weight coefficient corresponding to the first prediction model and a second weight coefficient corresponding to the second prediction model, performing weighted summation on the first prediction result and the second prediction result to obtain the target yield of the wafer, the sum of the first weight coefficient and the second weight coefficient being 1.

[0106] In one possible implementation, determining the target yield of the wafer based on the first prediction result and the second prediction result includes: using a target regression model to output the target yield of the wafer based on the first prediction result and the second prediction result, wherein the parameters in the target regression model include a weight coefficient for weighting the first prediction result output by the first prediction model and a weight coefficient for weighting the second prediction result output by the second prediction model.

[0107] In one possible implementation, the generation process of the target regression model includes: obtaining a third sample data set, the third sample data set including first sample prediction results, second sample prediction results and actual yields of multiple sample wafers, wherein the first sample prediction results include the predicted yield of the sample wafer determined by using the first prediction model, and the second sample prediction results include the predicted yield of the sample wafer determined by using the second prediction model; using the initial regression model to output the sample predicted yield of each sample wafer according to the first sample prediction results and the second sample prediction results of each sample crystal; optimizing the parameters in the initial regression model according to the loss between the sample predicted yield of each sample wafer output by the initial regression model and the actual yield of each sample wafer to obtain the target regression model.

[0108] According to the wafer yield prediction device of the embodiment of the present disclosure, the first prediction model is used to predict the wafer yield according to the defect density data (that is, the wafer yield is predicted based on the defect density distribution characteristics), and the second prediction model is used to predict the wafer yield based on the defect distribution map (that is, the wafer yield is predicted based on the defect position distribution characteristics). This is equivalent to predicting the wafer yield respectively from the defect density distribution and the defect position distribution. In this way, the target yield of the wafer is determined according to the yields predicted by the two models, which can combine the respective advantages of the two prediction models (that is, combine the two aspects of defect density distribution and defect position distribution data) to efficiently predict a more accurate target yield of the wafer, thereby improving the accuracy and efficiency of wafer yield prediction.

[0109] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0110] The embodiment of the present disclosure also provides a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor. The computer-readable storage medium can be a volatile or non-volatile computer-readable storage medium.

[0111] An embodiment of the present disclosure further proposes an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0112] The embodiments of the present disclosure also provide a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0113] Figure 4 1 is a block diagram of an electronic device 1900 according to an embodiment of the present disclosure. For example, the electronic device 1900 may be provided as a server or a terminal device. Figure 4 , the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions executable by the processing component 1922, such as an application. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0114] The electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input / output interface 1958 (I / O interface). The electronic device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server 2003. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM or similar.

[0115] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the electronic device 1900 to perform the above method.

[0116] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0117] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0118] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0119] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0120] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0121] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0122] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0123] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0124] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A wafer yield prediction method, characterized in that: include: Acquire defect density data and a defect distribution map of a wafer to be predicted, wherein the defect density data includes regional defect densities of multiple regions of the wafer, and the defect distribution map indicates position distribution of defects on the wafer; Determine a first prediction result according to the defect density data using a first prediction model, wherein the first prediction result indicates a yield of the wafer predicted by the first prediction model; Determine a second prediction result according to the defect distribution map using a second prediction model, wherein the second prediction result indicates a yield of the wafer predicted by the second prediction model; A target yield of the wafer is determined according to the first prediction result and the second prediction result.

2. The method according to claim 1, characterized in that The step of obtaining defect density data and defect distribution map of a wafer to be predicted includes: Determine multiple regions of the wafer and the number of defects in each region according to a preset region division method, and determine the ratio between the number of defects in each region and the area of ​​each region as the regional defect density of each region; A defect distribution map is generated according to the locations of defects on the wafer.

3. The method according to claim 1, characterized in that The first prediction model includes a linear regression model, wherein the generation process of the first prediction model includes: Acquire a first sample data set, wherein the first sample data set includes sample defect density data of a plurality of sample wafers and a true yield corresponding to each sample wafer; The first sample data set is used to fit the parameters in a preset linear regression model to obtain the first prediction model.

4. The method according to claim 1, characterized in that: The second prediction model includes a convolutional neural network model, wherein the training process of the second prediction model includes: Acquire a second sample data set, where the second sample data set includes sample defect distribution maps of a plurality of sample wafers and a true yield corresponding to each sample wafer; Outputting a sample prediction result of each sample wafer according to a sample defect distribution map of each sample wafer using a preset convolutional neural network model, wherein the sample prediction result indicates a predicted yield of the sample wafer output by the convolutional neural network model; According to the loss between the predicted yield of each sample wafer and the actual yield of each sample wafer, the parameters in the convolutional neural network model are optimized to obtain the second preset model after training.

5. The method according to claim 1, characterized in that Determining a target yield of the wafer according to the first prediction result and the second prediction result includes: Based on the preset first weight coefficient corresponding to the first prediction model and the second weight coefficient corresponding to the second prediction model, the first prediction result and the second prediction result are weightedly summed to obtain the target yield of the wafer, and the sum of the first weight coefficient and the second weight coefficient is 1.

6. The method according to claim 1, characterized in that Determining a target yield of the wafer according to the first prediction result and the second prediction result includes: A target regression model is used to output a target yield of the wafer based on the first prediction result and the second prediction result, wherein the parameters in the target regression model include a weight coefficient for weighting the first prediction result output by the first prediction model and a weight coefficient for weighting the second prediction result output by the second prediction model.

7. The method according to claim 6, characterized in that The generation process of the target regression model includes: Acquire a third sample data set, the third sample data set comprising a first sample prediction result, a second sample prediction result, and an actual yield of a plurality of sample wafers, wherein the first sample prediction result comprises a predicted yield of the sample wafer determined using the first prediction model, and the second sample prediction result comprises a predicted yield of the sample wafer determined using the second prediction model; Outputting a sample prediction yield of each sample wafer according to a first sample prediction result and a second sample prediction result of each sample crystal using an initial regression model; According to the loss between the sample predicted yield of each sample wafer output by the initial regression model and the actual yield of each sample wafer, the parameters in the initial regression model are optimized to obtain the target regression model.

8. A wafer yield prediction device, characterized in that: include: An acquisition module, used to acquire defect density data and a defect distribution map of a wafer to be predicted, wherein the defect density data includes regional defect densities of multiple regions of the wafer, and the defect distribution map indicates the position distribution of defects on the wafer; A first prediction module, configured to determine a first prediction result according to the defect density data using a first prediction model, wherein the first prediction result indicates a yield of the wafer predicted by the first prediction model; A second prediction module, configured to determine a second prediction result according to the defect distribution map using a second prediction model, wherein the second prediction result indicates a yield of the wafer predicted by the second prediction model; A target determination module is used to determine a target yield of the wafer according to the first prediction result and the second prediction result.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to implement the method described in any one of claims 1 to 7 when executing the instructions stored in the memory.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images