Manufacturing method of measurement template

Abstract

The invention provides a method for manufacturing a measurement template. The method comprises the following steps: taking a layout file of an existing product and an image acquired on the existing product as a data set; forming a neural network model by using the data set through a machine learning algorithm; and inputting a layout file of a to-be-measured product into the neural network model, outputting a simulation image, and taking the simulation image as a standard measurement template image of the measurement machine. The layout file of the to-be-measured product is directly generated into the template image by using the neural network model, so that the wafer preparation and image acquisition process can be omitted, the manufacturing process of the measurement template is simplified, wafers, machine hours and human resources are greatly saved, the cost is effectively reduced, the process difficulty is also reduced, and the production efficiency is improved. Meanwhile, the available time of the measuring equipment is prolonged, and the research and development efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor integrated circuits, and in particular to a method for manufacturing a measurement template. Background Art

[0002] Most of the mainstream measurement equipment in Fab (factory) uses image comparison system for precise positioning. The principle is to find images with certain features on the surface of the wafer and compare them with the standard images set in the recipe (menu). When the similarity between the two images exceeds the threshold setting, the wafer coordinates of the actual image are used as reference points to continuously correct the reference system, thereby improving the positioning accuracy and locating the specific measurement structure. This working principle requires that the actual wafer be used as a template for image acquisition when establishing the measurement recipe, that is, a set of images of the actual wafer are taken as a template for image comparison.

[0003] However, in actual applications, there is often a situation where there is no standard wafer to measure and establish recipes due to the tape-out schedule requirements. In order to get the measurement data as quickly as possible, a short loop (short process wafer) is used to establish a measurement recipe. Due to the difference between the images of the short loop and the full loop (standard wafer), the measurement recipe is not universal. When the standard wafer actually arrives, it is necessary to change the wafer and establish a new set of recipes, which causes a great waste of machine time and manpower, and also delays the wafer tape-out pi-run (trial run) process. Summary of the invention

[0004] The object of the present invention is to provide a method for making a measurement template, which uses a layout file to directly generate a template image, saves the process of wafer preparation and image acquisition, improves measurement efficiency, and saves costs.

[0005] In order to solve the above technical problems, the present invention provides a method for making a measurement template, comprising the following steps:

[0006] Using a layout file of an existing product and images collected on the existing product as a data set;

[0007] Using the data set to form a neural network model through a machine learning algorithm; and

[0008] The layout file of the product to be measured is input into the neural network model, a simulation image is output, and the simulation image is used as a standard measurement template image of the measurement machine.

[0009] Optionally, the data set includes a training set and a validation set, in which the input is a layout file of the existing product, and the output is an image on the existing product corresponding to the position of the layout file.

[0010] Optionally, the layout files of the existing product include a first layer layout file, a combination of a first layer layout file and a second layer layout file, a combination of a first layer layout file, a second layer layout file and a third layer layout file, a combination of successively increasing one layer of layout files, and a combination of all layout files.

[0011] Optionally, the layout file of the product to be measured is input into the neural network model, and simulation images corresponding to different process steps are obtained by selecting different layer combinations of the input layout file.

[0012] Optionally, the simulated image includes a simulated image of a finished product after all process steps are completed, or / and a simulated image of a semi-finished product after one or more process steps are completed.

[0013] Optionally, the layout file of the existing product is located in a structure-intensive area.

[0014] Optionally, the method further includes: comparing the image of the product to be inspected actually produced with the image of the standard measurement template to quickly identify defective spots.

[0015] Optionally, the layout file includes a file in a layout file format and / or an image file format; and / or the captured image includes a file in an image file format.

[0016] Optionally, the layout file format includes GDS or JB, and the image file format includes JPG or PNG.

[0017] Optionally, the measuring machine includes a CD measuring device, a film thickness measuring device or an atomic force microscope; and / or.

[0018] In summary, in the method for making a measurement template provided by the present invention, the layout file of an existing product and the image collected on the existing product are first collected as a data set, and then the data set is formed into a neural network model through a machine learning algorithm, and then the layout file of the product to be measured is input into the neural network model, and a simulation image is output, and the simulation image is used as a standard measurement template image of the measurement machine. The present invention uses a neural network model to directly generate a template image from the layout file of the product to be measured, which can save the process of wafer preparation and image acquisition, simplify the production process of the measurement template, greatly save wafers, machine time and human resources, effectively reduce costs, and reduce process difficulty, while increasing the available time of the measurement equipment and improving R&D efficiency.

[0019] Furthermore, when the layout file of the product to be measured is input into the neural network model, by selecting different layer combinations of the input layout file, simulation images corresponding to different process steps are obtained, and morphological images of intermediate semi-finished products with various combinations are generated, which can be used for the measurement of actual products to be measured in multiple stages, thereby improving the measurement efficiency and greatly reducing the time and labor costs of the measurement process. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Those skilled in the art will appreciate that the accompanying drawings are provided for a better understanding of the present invention and do not constitute any limitation on the scope of the present invention.

[0021] Figure 1 It is a flow chart of a method for making a measurement template provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, advantages and features of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the drawings are all in a very simplified form and are not drawn to scale, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In addition, the structure shown in the drawings is often a part of the actual structure. In particular, the emphasis of each drawing is different, and sometimes different scales are used.

[0023] As used in the present invention, the singular forms "one", "an", and "the" include plural objects, unless the content clearly indicates otherwise. As used in the present invention, the term "or" is generally used in a sense that includes "and / or", unless the content clearly indicates otherwise. As used in the present invention, the term "several" is generally used in a sense that includes "at least one", unless the content clearly indicates otherwise. As used in the present invention, the term "at least two" is generally used in a sense that includes "two or more", unless the content clearly indicates otherwise. In addition, the terms "first", "second", and "third" are used only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first", "second", and "third" may explicitly or implicitly include one or at least two of the features.

[0024] In the prior art, when establishing a measurement recipe, it is necessary to use an actual wafer as a template for image acquisition to obtain a standard image, which is easily limited by the wafer tape-out schedule. In view of this problem, the present invention provides a method for making a measurement template, comprising: using the layout file of an existing product and the image acquired on the existing product as a data set; using the data set to form a neural network model through a machine learning algorithm; inputting the layout file of the product to be measured into the neural network model, outputting a simulation image, and using the simulation image as a standard measurement template image of the measurement machine.

[0025] The present invention uses a neural network model to directly generate a template image from the layout file of the product to be measured, which can eliminate the process of wafer preparation and image acquisition, simplify the production process of the measurement template, greatly save wafers, machine time and human resources, effectively reduce costs, and reduce process difficulty. At the same time, it increases the available time of the measurement equipment and improves R&D efficiency.

[0026] Figure 1 is a flow chart of a method for making a measurement template provided by an embodiment of the present invention. Figure 1 As shown, the method for making a measurement template provided by an embodiment of the present invention includes the following steps:

[0027] S1: Using the layout file of an existing product and the images collected on the existing product as a data set;

[0028] S2: using the data set to form a neural network model through a machine learning algorithm;

[0029] S3: Input the layout file of the product to be measured into the neural network model, output a simulation image; and use the simulation image as a standard measurement template image of the measurement machine.

[0030] In step S1, the layout file of the existing product and the image collected on the existing product are used as a data set. In this embodiment, the input is the layout file of the existing product, and the output is the image on the existing product corresponding to the position of the layout file. The existing product can be a finished product, a finished product or a semi-finished product (i.e., a product that has only completed one or more process steps and has not been finally made into a finished product), which has a layout file, and the image on the existing product can be directly collected.

[0031] The data set may include a training set and a validation set. The training set is used to train the model, that is, the training set is used to train different models, and then the validation set is used to select the optimal model using cross-validation, and the performance of the model on the validation set is improved through continuous iteration. The data set may also include a test set, which is used to evaluate the performance of the model. The layout file of the existing product and the corresponding image are used as a set of data sets to obtain multiple sets of data sets of the existing product, and the multiple sets of data sets are divided into training sets and validation sets, or into training sets, validation sets and test sets. For example, 70% of the data sets in the multiple sets of the data sets can be used as training sets, and 30% of the data sets can be used as validation sets. Alternatively, 60% of the data sets in the multiple sets of the data sets can be used as training sets, 20% of the data sets can be used as validation sets, and 20% of the data sets can be used as test sets. It should be noted that the division percentage is not limited to this, and the division method is not limited to this.

[0032] In one embodiment of the present invention, when selecting the layout file and image of the data set, in order to obtain the best use effect, it is preferred to use a structure-intensive area, that is, the selected layout file is located in a structure-intensive area, such as the SRAM area of ​​a logic device. The neural network model trained in the structure-intensive area has a stronger use ability.

[0033] In another embodiment of the present invention, an existing product with a structure as similar as possible to the new product to be developed can be selected, which will be more conducive to producing a good neural network model. For example, the new product of interest has a structure with trenches and vias clustered.

[0034] In this embodiment, the layout file of the existing product includes the first-layer layout file, the combination of the first-layer layout file and the second-layer layout file, the combination of the first-layer layout file, the second-layer layout file and the third-layer layout file, and so on, and the combination of all layout files. The corresponding image includes the first layer of the product corresponding to the layout file, the combination of the first layer and the second layer, the combination of the first layer, the second layer and the third layer, and so on, and the combination of all layers, so as to train the neural network model to derive the image corresponding to the semi-finished product of the product.

[0035] In this embodiment, the input layout file includes a file in a layout file format or / and an image file format, the layout file format includes GDS (Graphic Design System, image database system file) or JB (JavaBean), and the image file format includes JPG (Joint Photographic Experts Group, Joint Photographic Experts Group) or PNG (Portable Network Graphics, portable network image), but not limited thereto. The output image includes a file in an image file format, and the image file format includes JPG or PNG, but not limited thereto.

[0036] In step S2, a neural network model is formed using the data set through a machine learning algorithm.

[0037] The input of the neural network model is a layout file, and the output is an image. That is, the layout file is input into the neural network model, and the output is an image corresponding to the layout file.

[0038] The neural network model may be composed of any structure, any depth and any number of nodes known to those skilled in the art, and may be specifically implemented in any programming language known to those skilled in the art. For example, the neural network model may be a feedforward neural network model, a perceptron, a Hopfield network, a radial basis function network, etc., or may be a convolutional neural network model, a recurrent neural network model, a generative adversarial network, etc.

[0039] Exemplarily, a back propagation algorithm can be used to form a neural network model, and the learning process can be composed of two parts: forward propagation and reverse propagation of errors. During forward propagation, the input sample enters from the input layer, is processed layer by layer by each hidden layer, and is transmitted to the output layer. If the output layer does not meet the expectations, the error is returned layer by layer as an adjustment signal, and the connection weight matrix between neurons is processed to reduce the error. After repeated learning, the error is eventually reduced to an acceptable range. Specific steps include, for example: 1. Take a sample from the training set and input the information into the network. 2. After forward processing layer by layer through the connection conditions between each node, the actual output of the neural network model is obtained. 3. Calculate the error between the actual output of the network and the expected output. 4. Backwardly propagate the error layer by layer to the previous layers, and load the error information to the connection weight according to certain principles, so that the connection weight of the entire neural network model is transformed in the direction of reducing the error. 5. Repeat the above steps for each input-output sample pair in the training set until the error of the entire training sample set is reduced to a position that meets the requirements. Of course, it is not limited to this.

[0040] This embodiment uses the current relatively mature machine learning algorithm, takes the layout file as the input signal, and the corresponding wafer structure image as the output information, to train a neural network model specifically for generating image templates. Since the input layout file and the output wafer image are mostly abstract geometric figures, the size standard and color information can be almost ignored, and the training cost and accuracy requirements of this neural network model are much lower than the existing AI models (such as beauty, face replacement or film and television post-production, etc.).

[0041] In this embodiment, the trained neural network model corresponds to a specific process, such as a platform corresponding to a node. Since the process information corresponding to different nodes is quite different, for example: the shrinkage ratio from the layout to the real product is different, the CD BIAS (etching CD minus the yellow light CD before etching) is different, the angle of the product structure is different, and the morphology presented after CMP is also very different. Therefore, if the platform is changed (for example, the node is changed), retraining is required, and the nodes include, for example, 90nm or 45nm, etc., but are not limited to this.

[0042] In one embodiment of the present invention, neural network models of multiple different platforms can be trained simultaneously and placed in a measuring machine, and the measuring machine calls the appropriate neural network model according to demand. In another embodiment of the present invention, when a relatively large change occurs in the product process, retraining can be performed to replace the neural network model in the measuring machine.

[0043] In step S3, the layout file of the product to be measured is input into the neural network model, a simulation image is output, and the simulation image is used as a standard measurement template image of the measurement machine.

[0044] The neural network model is applied to the measurement process of the product to be measured. The measurement process applied to the product to be measured is: inputting the layout file of the product to be measured into the neural network model, outputting a simulation image, and importing the obtained simulation image into the measurement machine as a database image of the measurement machine, which is used for measuring the surface of the wafer in actual production to detect defective points.

[0045] In the data set in step S1, the layout file includes the first energy layout of the existing product, the combination of the first layer layout and the second layer layout, the combination of the first layer layout, the second layer layout and the third layer layout, the combination of one layer of layout files added in sequence, and the combination of all layout files. In this step, you can input a layout file of any layer combination or a single layer layout file, and export simulation images of several arbitrary layer combinations or single layer simulation images. The arbitrary layer can be an adjacent film layer in the middle or a non-adjacent film layer, so as to shorten the tape-out cycle, thereby greatly reducing the time and labor cost of the measurement process.

[0046] Exemplarily, by selecting different layer combinations of the layout file to be input, simulation images corresponding to different process steps are obtained. For example: if all layer combinations are selected to be input, a simulation graphic of a finished product that completes the process steps corresponding to all layers is obtained. When the actual image of the product to be measured is subsequently compared with the simulation graphic, the product to be measured that is actually produced is a finished product that completes all process steps. If some layers are selected to be input, for example, the first layer layout file and the second layer layout file are selected, the simulation image obtained is a simulation graphic of a semi-finished product that completes the first process step corresponding to the first layer layout file and the second process step corresponding to the second layer layout file. The actual image of the product to be measured used for subsequent comparison is the semi-finished product after the first process step and the second process step are completed.

[0047] Of course, the input layout file can also jump layers. For example, a certain product requires a first-layer layout file, a second-layer layout file, a third-layer layout file, and a fourth-layer layout file to complete production. The input can select a combination of the first-layer layout file and the third-layer layout file. The obtained simulation image is a simulation graphic of a semi-finished product that completes the first process step corresponding to the first-layer layout file and the third process step corresponding to the third-layer layout file. The actual image of the product to be measured for subsequent comparison is the semi-finished product after completing the first process step and the third process step. There are many ways to match. According to research and development needs, the layout information of the second and fifth layers may also be imported, and a simulation image composed of the second and fifth layers may be obtained for measurement, so as to achieve a more purposeful measurement information collection in the research and development process and shorten the tape-out cycle.

[0048] In this way, the measurement of wafers produced in the intermediate process is realized with low cost, and there is no need to use short loop wafers to establish the measurement parameter database. It also avoids the disadvantages of the prior art that the short loop and full loop graphics are different and the measurement parameters cannot be universal, that is, after the full loop wafer (standard wafer) actually arrives at the station, the previous measurement parameters cannot be used and need to be re-established.

[0049] In this embodiment, the process steps include any steps of wafer processing or processes such as mask manufacturing that require fine patterning.

[0050] According to the neural network model, it can be applied to the measurement of semi-finished products in the research and development process of various products. Compared with the prior art, it saves the process of generating intermediate semi-finished products when making measurement templates, effectively reduces costs, and because the production of intermediate semi-finished products is relatively difficult, it also reduces the process difficulty. In addition, it is also possible to obtain simulation images of semi-finished products in different stages by arbitrarily matching different layers of layouts, and generate morphological images of intermediate semi-finished products with a variety of matching combinations, which can be used for the measurement of actual products to be measured at multiple stages, thereby improving the measurement efficiency and greatly reducing the time and labor costs of the measurement process.

[0051] The simulation image is used as the standard measurement template image of the measurement machine, and then the actual product to be measured is measured and inspected, and the image of the actual product to be inspected is specifically compared with the standard measurement template image to quickly identify defective points. Since the data set contains the actual layer definition and process information, the neural network model is formed using the data set, and the simulation graphics corresponding to the layout file of the product to be measured generated according to the neural network model will have a very high accuracy.

[0052] The measuring equipment includes but is not limited to CD measuring equipment, film thickness measuring equipment, atomic force microscope and other defect scanning and observation equipment and all other equipment using image comparison and positioning.

[0053] The present invention applies machine learning algorithms and neural network models to the measurement process of the Fab factory, uses the existing layout files and images of the Fab factory to train a dedicated neural network model, and then uses the neural network model to generate the standard measurement template image required by the measurement process from the layout file of the product to be measured. This process reduces the dependence on the actual wafer and has great advantages in improving equipment utilization. At the same time, since the output image can contain the process characteristics of the Fab factory, such as optical proximity correction (OPC) and / or process deviation, etc., it can be made into a dedicated application for the Fab factory, and can also be made into a general application for surface measurement equipment manufacturers, and has a wide range of applications.

[0054] In summary, in the method for making a measurement template provided by the present invention, firstly, the existing layout file of the product and the image collected from the existing product are collected as a data set, and then the data set is formed into a neural network model through a machine learning algorithm, and then the layout file of the product to be measured is input into the neural network model, and a simulation image is output, and then the simulation image is used as a standard measurement template image of the measurement machine. The present invention uses a neural network model to directly generate a template image from the layout file of the product to be measured, which can save the process of wafer preparation and image acquisition, simplify the production process of the measurement template, greatly save wafers, machine time and human resources, effectively reduce costs, and reduce process difficulty, while increasing the available time of the measurement equipment and improving R&D efficiency.

[0055] Furthermore, when the layout file of the product to be measured is input into the neural network model, by selecting different layer combinations of the input layout file, simulation images corresponding to different process steps are obtained, and morphological images of intermediate semi-finished products with various combinations are generated, which can be used for the measurement of actual products to be measured in multiple stages, thereby improving the measurement efficiency and greatly reducing the time and labor costs of the measurement process.

[0056] The above description is only a description of the preferred embodiments of the present invention, and is not intended to limit the scope of the present invention. Any changes or modifications made by a person skilled in the art in the field of the present invention based on the above disclosure shall fall within the scope of protection of the claims.

Claims

1. A method for making a measurement template, characterized in that: The following steps are involved: Using a layout file of an existing product and images collected on the existing product as a data set; Using the data set to form a neural network model through a machine learning algorithm; as well as The layout file of the product to be measured is input into the neural network model, a simulation image is output, and the simulation image is used as a standard measurement template image of the measurement machine.

2. The method for making a measurement template according to claim 1, characterized in that: The data set includes a training set and a verification set. In the data set, the input is the layout file of the existing product, and the output is an image on the existing product corresponding to the position of the layout file.

3. The method for making a measurement template according to claim 1, characterized in that: The layout files of the existing products include the first layer layout file, the combination of the first layer layout file and the second layer layout file, the combination of the first layer layout file, the second layer layout file and the third layer layout file, the combination of adding one layer of layout files in sequence and the combination of all layout files.

4. The method for making a measurement template according to claim 3, characterized in that: The layout file of the product to be measured is input into the neural network model, and simulation images corresponding to different process steps are obtained by selecting different layer combinations of the input layout file.

5. The method for making a measurement template according to claim 4, characterized in that: The simulated image includes a simulated image of a finished product after all process steps are completed, or / and a simulated image of a semi-finished product after one or more process steps are completed.

6. The method for making a measurement template according to any one of claims 1 to 5, characterized in that: The layout file of the existing product is located in a structure-intensive area.

7. The method for making a measurement template according to claim 1 or 5, characterized in that: Also includes: The image of the product to be inspected that is actually produced is compared with the image of the standard measurement template to quickly identify defective spots.

8. The method for making a measurement template according to claim 1, characterized in that: The layout file includes a file in a layout file format and / or an image file format; and / or the captured image includes a file in an image file format.

9. The method for making a measurement template according to claim 8, characterized in that: The layout file format includes GDS or JB, and the image file format includes JPG or PNG.

10. The method for making a measurement template according to claim 1, characterized in that: The measuring machine includes a CD measuring device, a film thickness measuring device or an atomic force microscope.

Images