Simulation method for semiconductor device, electronic equipment and storage medium

Abstract

The embodiment of the invention relates to a simulation method for a semiconductor device, electronic equipment and a storage medium. The simulation method is executed by a simulation model, the simulation model comprises a CMP physical model, the CMP physical model is configured to determine a simulation value of the surface height of a polished wafer based on input data, and the input data comprises graphic feature data and physical information; the neural network model is generated by taking the simulation value as a training sample for training, and is configured to perform parallel prediction on multiple groups of input data in the simulation process of the CMP physical model so as to generate multiple predicted values of the surface height of the polished wafer corresponding to the multiple groups of input data, and respectively taking the simulation values as corresponding simulation values of the surface heights. According to the technical scheme, the establishment and simulation process of the CMP physical model can be remarkably accelerated, the time consumption is reduced, and the simulation precision can be improved.

Description

Technical Field

[0001] Embodiments of the present disclosure generally relate to integrated circuits, and more particularly, to simulation methods, electronic devices, and storage media for semiconductor devices. Background Art

[0002] The current prediction of metal thickness after the chemical mechanical polishing (CMP) back-end process usually relies heavily on the CMP model.

[0003] The underlying layer of the traditional CMP model is mainly a time-step physical model based on the physical process of CMP itself. The characteristics of this model are that it fully simulates the real physical process of CMP, has high modeling accuracy, and can output all relevant variables of the intermediate process, helping users to better understand and adjust the CMP process table information (recipe) and predict the results after CMP grinding. However, its disadvantage is that the calculation speed is slow. Summary of the invention

[0004] According to example embodiments of the present disclosure, a simulation scheme for a semiconductor device and a training method for a neural network model are provided to at least partially overcome the above or other potential defects.

[0005] According to one aspect of the present disclosure, a method for generating training samples is provided. The method includes: generating a first number of feature vectors based on data points for representing various graphic features on a wafer, wherein the feature vectors are a combination of data representing different types of graphic features in the data points; generating a second number of physical vectors based on physical information related to CMP of the wafer, wherein the physical vectors are a combination of data representing different types of physical information; and performing a second number of simulations based on the feature vectors and the physical vectors by a CMP physical model, so as to generate simulated values ​​of the surface height of the wafer corresponding to the first number of feature vectors as training samples in each simulation.

[0006] In some embodiments, generating a first number of feature vectors based on data points used to represent various graphic features on a wafer includes: determining the data within each step range based on the value range of the data points of each graphic feature and the corresponding step size; and orthogonalizing the data within each step range to generate a feature data table, wherein the feature data table includes data of the first number of feature vectors.

[0007] In some embodiments, generating a second number of physical vectors based on physical information related to CMP of the wafer includes: determining data in each step range based on the value range of the data points of each physical vector and the corresponding step; and orthogonalizing the data in each step range to generate a physical information data table, wherein the physical information data table includes data of the second number of physical vectors.

[0008] In some embodiments, performing a second number of simulations based on the characteristic vector and the physical vector by the CMP physical model includes: generating a first simulation value of the surface height of the non-groove structure and a second simulation value of the surface height of the groove structure, respectively; and determining a simulation value of the wafer surface height based on the first simulation value and the second simulation value.

[0009] In some embodiments, the pattern characteristics include at least: the density of the pattern within the grid on the wafer; the width of the pattern; and the spacing between the patterns.

[0010] In some embodiments, the physical information includes at least the following items: information of deposited materials; information of maximum dishing defects and minimum dishing defects in each step; information of polishing liquid; information of long-tail effect parameters; duration information; and nominal pressure information.

[0011] According to a second aspect of the present disclosure, a method for training a neural network model is provided. The method includes: a method for training a neural network model, including: training the neural network model using training samples to determine the gradient of the loss function of the neural network model, wherein the training samples are generated by a CMP physical model based on graphic features and physical information for determining the surface height of a wafer after grinding; updating the parameters of the neural network model based on the gradient of the loss function; and in response to the training satisfying a predetermined condition, determining the neural network model corresponding to the corresponding updated parameters as a trained neural network model, wherein the trained neural network model outputs a predicted value of the surface height.

[0012] In a third aspect of the present disclosure, an electronic device is provided. The electronic device includes a processor; and a memory coupled to the processor, the memory having instructions stored therein, and when the instructions are executed by the processor, the device performs an action, the action including: training a neural network model using training samples to determine the gradient of a loss function of the neural network model, wherein the training samples are generated by a CMP physical model based on graphic features and physical information for determining the surface height of a polished wafer; updating parameters of the neural network model based on the gradient of the loss function; and in response to the training satisfying a predetermined condition, determining the neural network model corresponding to the corresponding updated parameters as a trained neural network model, wherein the trained neural network model outputs a predicted value of the surface height.

[0013] In some embodiments, training the neural network model using training samples to determine the gradient of the loss function of the neural network model includes: calculating the gradient of the loss function using a stochastic gradient descent algorithm.

[0014] In some embodiments, updating the parameters of the neural network model based on the gradient of the loss function includes: performing a gradient update on the gradient of the loss function in response to a predetermined number of trainings; and updating the parameters of the neural network model based on the gradient update.

[0015] In some embodiments, training a neural network model using training samples to determine the gradient of a loss function of the neural network model includes: training the neural network model using a portion of the training samples to perform gradient descent back propagation to update parameters; predicting the remaining portion of the training samples using the trained neural network model to generate a predicted value of the surface height of the wafer after grinding; and in response to a comparison result between the predicted value and a simulation value of the CMP physical model satisfying a predetermined condition, determining the corresponding trained neural network model as the final neural network model.

[0016] In some embodiments, the comparison result between the predicted value and the simulated value satisfies a predetermined condition, including: a root mean square error between the predicted value and the simulated value is less than a predetermined threshold.

[0017] In some embodiments, the method further comprises: determining a loss function value of the neural network model; and wherein the training satisfies a predetermined condition comprises: a difference between the loss function values ​​of two iterations is less than a predetermined difference threshold; or an absolute value of a ratio of the loss function values ​​of two iterations is greater than a predetermined ratio threshold. In some embodiments, the training samples are generated in parallel on multiple servers so that the training of the neural network model is performed in parallel.

[0018] In a fourth aspect of the present disclosure, a neural network model generated according to the method of the third aspect is provided.

[0019] In a fifth aspect of the present disclosure, a simulation model is provided, which includes: a CMP physical model, the CMP physical model is configured to determine a simulation value of the surface height of a wafer after grinding based on input data, wherein the input data includes graphic feature data and physical information; and a neural network model according to the fifth aspect, the neural network model is generated by training with the simulation value as a training sample, and is configured to perform parallel prediction on multiple sets of input data during the simulation process of the CMP physical model to generate multiple predicted values ​​of the surface height of the wafer after grinding corresponding to the multiple sets of input data, respectively serving as corresponding simulation values ​​of the surface height.

[0020] In some embodiments, performing parallel prediction on multiple sets of input data during the simulation of the CMP physical model includes: utilizing multiple graphics processors to perform parallel acceleration on the computation process of the neural network model.

[0021] In the sixth aspect of the present disclosure, a simulation method for a semiconductor device is provided, which is performed using the simulation model of the fifth aspect, and the method includes: a neural network model in the simulation model predicts a simulation value of the surface height of the wafer after grinding based on input data, wherein the input data includes graphic feature data and physical information.

[0022] In some embodiments, predicting the simulation value of the surface height of the wafer after grinding based on input data by a neural network model in the simulation model includes: the neural network model performs parallel prediction on multiple sets of input data to generate multiple predicted values ​​of the surface height of the wafer after grinding corresponding to the multiple sets of input data, respectively serving as the corresponding simulation values ​​of the surface height.

[0023] In some embodiments, the simulation value of the surface height of the wafer after grinding is predicted by the neural network model in the simulation model based on the input data, including: generating a first prediction value of the surface height of the non-groove structure and a second prediction value of the surface height of the groove structure by the neural network model respectively; and determining the simulation value of the wafer surface height based on the first prediction value and the second prediction value by the neural network model.

[0024] In some embodiments, the method further includes: determining the surface condition of the wafer during the grinding process by using a CMP physical model in the simulation model.

[0025] In the seventh aspect of the present disclosure, a neural network model is provided, comprising: an input layer, configured to receive graphic feature data on a wafer and physical information related to CMP of the wafer; an intermediate layer, configured to execute the method of the sixth aspect to determine the surface height value of the wafer after grinding; and an output layer, configured to output the surface height value.

[0026] In some embodiments, the method further includes: an embedding layer configured to map the input physical information to generate a physical vector.

[0027] In an eighth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above method according to the present disclosure is implemented.

[0028] It will be understood from the following description that the technical solution of the present disclosure can significantly accelerate the establishment and simulation process of the CMP physical model, reduce the related time consumption, and improve the simulation accuracy.

[0029] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram showing an example environment in which embodiments of the present disclosure can be implemented; Figure 2 A flowchart of a method for generating training samples according to some embodiments of the present disclosure is shown; Figure 3 A flowchart of a method for training a neural network model according to some embodiments of the present disclosure is shown; Figure 4 A schematic diagram showing a neural network architecture according to some embodiments of the present disclosure; Figure 5 A block diagram of a computing device capable of implementing various embodiments of the present disclosure is shown.

[0031] In the various drawings, the same or corresponding reference numerals represent the same or corresponding parts. DETAILED DESCRIPTION

[0032] The principles of the present disclosure will be described below with reference to the various exemplary embodiments shown in the accompanying drawings. It should be understood that the description of these embodiments is only to enable those skilled in the art to better understand and further implement the present disclosure, and is not intended to limit the scope of the present disclosure in any way. It should be noted that similar or identical reference numerals may be used in the figures where feasible, and similar or identical reference numerals may represent similar or identical functions. Those skilled in the art will readily recognize from the following description that alternative embodiments of the structures and methods described herein may be adopted without departing from the principles of the present invention described herein.

[0033] As used herein, the term "including" and its variations mean open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "based at least in part on". The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc. may refer to different or the same objects.

[0034] CMP is a method that combines chemical and physical methods to remove materials. It has become an indispensable step in the modern integrated circuit (IC) industry. Specifically, CMP uses chemical corrosion and mechanical force to flatten silicon wafers or other substrate materials during processing. CMP can achieve nanometer-level material removal, making the wafer surface flat.

[0035] The CMP model, as the name implies, is a physical model for simulating the CMP process. The model outputs some important physical indicators of each grid point of the wafer after the CMP process, such as the thickness of the trench structure, the thickness of the non-trench structure, and the thickness of the metal material after CMP. In simple terms, it is a physical model that predicts the surface morphology of the wafer after the CMP process. CMP is one of the eight major semiconductor processes, connecting the previous and the next. Other processes following CMP will be affected by the CMP process. In other words, when it is necessary to evaluate and control the process results after CMP, it is best to know the impact of the CMP process, which is of great significance to the full process control and yield improvement of the entire semiconductor manufacturing. In addition, the CMP process itself is an indispensable and necessary step, because the CMP process can be better adjusted through the CMP model. For example, the results of the CMP model can be used to infer which process parameter settings need to be modified and adjusted, and whether there are some unreasonable aspects in the design of the Graphic Data System (GDS), which may cause hotspots, that is, defects.

[0036] As mentioned above, the traditional CMP model has high modeling accuracy, but the disadvantage is that the calculation speed is slow. Because the time-step model is calculated step by step, the result of each time step depends on the result of the previous time step, and it cannot be accelerated in parallel using a graphics processing unit (GPU) or CPU. In the process of searching for the optimal solution for parameters, the running time of each attempt cannot be reduced. Therefore, building a better CMP model requires repeated adjustments and runs, which consumes a lot of time and energy. In addition, in the full chip simulation stage, if a larger chip is encountered, the amount of data increases significantly, and the simulation of the CMP physical model based on the time step will be very slow, usually taking several hours.

[0037] A CMP time-step physical model is known, which is a mixture of several physical and mathematical formulas. The core feature of the model is that it is necessary to set the grinding time of each step according to the actual CMP process, assuming it to be t1, t2, t3, ... and set the unit time dt, where dt is usually a relatively small number, and the smaller it is, the closer the CMP physical model is to the real CMP physical process, because the core assumption of the CMP physical model is that at a very short moment dt, the grinding rate of the base to the CMP material remains almost constant, and when dt infinitely approaches 0, the grinding rate at this moment infinitely approaches a constant value (the idea of ​​differential). Therefore, if you need to establish a more accurate physical model, you usually need to set dt very small, but a dt that is too small will cause a sharp increase in the amount of calculation. Assuming that there are several time steps in total, and each step lasts for t1, t2, ..., then the total time is their sum, assuming it is T. A complete CMP calculation requires calculating the results of T / dt time steps. Usually T will be hundreds of seconds (this is the case with real physical processes). If dt is set to 0.001s, hundreds of thousands of time steps need to be calculated. In the calculation of each time step, usually not one point is involved in the calculation. In the modeling stage, a doe test is required. The doe test algorithm generates Q test points for R iterations. The larger the number of Q and R is set, the closer the model parameters found in the end are to the global optimal solution. In the simulation stage, it depends on the scale of the whole chip. When it is relatively large, there will be tens of millions of points that need to be simulated. And a built model usually needs to be used several times, and there may be many full chips that need to be simulated using this model. All of the above illustrate the computational complexity of the current CMP physical model, so it is very necessary to accelerate it.

[0038] In view of this, the present disclosure provides an improved solution.

[0039] Some embodiments of the present disclosure provide an improved training method for a neural network model. The method includes: training the neural network model using training samples to determine the gradient of the loss function of the neural network model, wherein the training samples are generated by a CMP physical model based on graphic features and physical information used to determine the surface height of a wafer after grinding; updating the parameters of the neural network model based on the gradient of the loss function; and in response to the training satisfying a predetermined condition, determining the neural network model corresponding to the corresponding updated parameters as a trained neural network model, wherein the trained neural network model outputs a predicted value of the surface height.

[0040] Some embodiments of the present disclosure also provide improved simulation models and neural network models.

[0041] The embodiments of the present disclosure can significantly accelerate the establishment and simulation process of the CMP physical model, reduce the related time consumption, and improve the simulation accuracy.

[0042] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0043] Figure 1 1 is a schematic diagram of an example environment 100 in which embodiments according to the present disclosure can be implemented. Figure 1 As shown, the example environment 100 includes a computing device 110 and a client 120 .

[0044] In some embodiments, the computing device 110 may interact with the client 120. For example, the computing device 110 may receive an input message from the client 120 and output a feedback message to the client 120. In some embodiments, the input message from the client 120 may be design layout data. The computing device 110 may perform corresponding mathematical operations on the design layout data and output corresponding operation results to the client 120.

[0045] In some embodiments, computing device 110 may include, but is not limited to, a personal computer, a server computer, a handheld or laptop device, a mobile device (such as a mobile phone, a personal digital assistant (PDA), a media player, etc.), consumer electronics, a minicomputer, a mainframe computer, cloud computing resources, etc.

[0046] It should be understood that the structure and function of the example environment 100 are described for exemplary purposes only and are not intended to limit the scope of the subject matter described herein. The subject matter described herein can be implemented in different structures and / or functions. The environment is merely illustrative and is not intended to limit the application environment of the embodiments of the present disclosure.

[0047] In order to explain the principle of the present disclosure more clearly, the following will refer to Figure 2-5 Let's describe it in more detail.

[0048] First, let's introduce the CMP physical model. The essence of the CMP physical model is a mixture of several physical formulas, rather than a black box that cannot be described in mathematical language. As we all know, deep learning models have extremely strong fitting capabilities, and can even fit black box function relationships that cannot be described by explicit mathematical formulas. Therefore, based on deep learning, a suitable deep learning model can be established to learn and approximate the CMP physical model.

[0049] Next, we will explain in detail how to build the deep learning model. The CMP physical model is usually composed of three steps, because the actual CMP process can often be divided into three stages, namely: the bulk stage, which mainly grinds the top metal (usually copper); the touch down (TCHD) stage, which is defined as exposing the buffer layer (barrier) between the metal layer and the bottom oxide (oxide); the barrier removal (BR) stage, which is defined as grinding the buffer layer in this period of time, and most points touch the bottom oxide material. There are different physical parameters in each stage.

[0050] The bulk silicon process usually refers to the process from the beginning of grinding to the end of grinding in the bulk silicon process stage, when the top metal material is ground. In this stage, only metal is ground. CMP will perform a deposition process before grinding to deposit different materials. Usually, the top layer is metal (usually copper), the middle is the buffer layer, and the bottom layer is oxide. What is touched here is the buffer layer material, that is, in this stage, it involves grinding only metal to starting to touch the buffer layer material. Regarding buffer layer removal, as the name suggests, in this layer, the main thing is to remove the buffer layer. Ideally, the bulk silicon process only grinds metal. The Tchd stage will end just when the buffer layer is touched. The Br stage can completely grind the buffer layer material. But in fact, it is not so ideal. In the bulk silicon process stage, there may be some areas where the material has changed from metal to buffer layer, and the same is true for tchd. It is impossible to control all points to touch the buffer layer material at the same time, and it is even more so in the br stage. It is difficult to ensure that each point perfectly grinds off the buffer layer and does not touch the oxide at the lower layer.

[0051] In some embodiments, based on the above-mentioned real CMP process information, the data input into the deep learning neural network may include the following information: vectors of graphic data system (GDS) features, specifically density vectors, space vectors, width vectors, etc.; physical information vectors, which may specifically include deposition material information, specifically deposition material thickness information, nominal grinding rate information of deposition materials, maximum dishing defect (Dishing) information of each step, minimum Dishing information, slurry information, long tail effect (long range effect) parameter information, duration information, nominal pressure information, etc., parameters used in the CMP physical model. Dishing refers to the difference between the height of the groove structure and the height of the non-groove structure, and Erosion refers to the difference between the height of the non-groove structure of each grid point relative to the reference value thickness. The long tail effect can be understood as the density width space and other conditions of the current area not only affecting the grinding result of the current area, but also affecting the grinding result of the surrounding area.

[0052] The input of the neural network model can be determined based on the above information. Figure 4 Introduce the architecture of neural network models.

[0053] For the neural network model, it is necessary to train it through training samples. The following describes the method of generating training samples.

[0054] See first Figure 2 , Figure 2 A flowchart of a method 200 for generating training samples according to some embodiments of the present disclosure is shown.

[0055] At block 202 , a first number of feature vectors are generated based on data points representing various graphical features on a wafer, wherein the feature vectors are combinations of data representing different types of graphical features in the data points.

[0056] In some embodiments, generating a first number of feature vectors based on data points used to represent each graphic feature on a wafer may include: determining the data within each step range based on the value range of the data points of each graphic feature and the corresponding step length; and orthogonalizing the data within each step range to generate a feature data table, wherein the feature data table includes data of the first number of feature vectors. In some embodiments, the graphic features include at least: the density of the graphics within the grid on the wafer; the width of the graphics; and the spacing between the graphics. For example, the graphic features density, width, and space have a significant impact on the height after CMP grinding. They are significant independent variables that constitute the CMP grinding height prediction model, and these three variables have their own certain ranges. In addition, the step length can be divided within the range according to actual needs. This is described in detail later.

[0057] Data points are used to represent the data of measurement points (grid points) on the wafer. Data points can be provided by the user. For example, a data point represents features such as density, width, space, etc., such as (density, width, space), where, for example, density is 0.5, width is equal to 30nm, and space is equal to 40nm. Where density, width, and space each represent a graphic feature, namely density, width, and spacing. The combination of data used to represent these features can form a vector. For example, for density, width, and space, their respective values ​​combined together can be called a vector. For example, the previously mentioned (0.5, 30nm, 40nm). The combination of different values ​​of these variables can form multiple vectors. It can be combined in various ways.

[0058] One method is the orthogonal design verification method (Design of Experiments, abbreviated as DOE). The DOE design method, that is, DOE data orthogonalization, is an experimental design method used to explore and verify the impact of factors on results. In DOE, experiments are usually divided into multiple combinations, each of which controls a factor and measures its impact on the results. In this way, we can have a more comprehensive understanding of the impact of factors on the results and determine the best combination of factors.

[0059] In some embodiments, DOE design can be divided into two parts. One part is the GDS feature design, namely the density / width / space design mentioned above.

[0060] First, you can set the range and step size of density / width / space. Because these three variables have physical meanings, they usually have a relatively certain range in most processes. The common ranges of these three variables can be counted. For example, it is known that density is a number between 0 and 1, width and space must be positive numbers, usually between 0 and 50 (unit um), and the maximum value usually does not exceed hundreds. In this way, data is actually generated according to the actual situation (such as training resources, whether there are a large number of GPUs to accelerate training or not).

[0061] Assume there are k data points, and the densities are 0.1, 0.2, 0.3, ..., d_k, then the density vector is [0.1, 0.2, 0.3, ..., d_k]. The rest of the vectors can be obtained in the same way.

[0062] The DOE mentioned above is to train the neural network fully so as to track the CMP physical model more accurately. Specifically, orthogonal DOE data can be generated for density / width / space. The so-called orthogonality means that if density generates N possible value results, width generates M values, and space generates P values, then the data entries contained in the final data set are The characteristics of the data here are: for example, if you randomly check the generated data set, there are several values ​​of width, and the specific number of them is, of course, In fact, it is an exhaustive combination, ensuring that each width has all possible density and space value results. By orthogonalizing the data, the comprehensiveness of the data and the reliability of the results can be ensured.

[0063] At block 204 , a second number of physical vectors are generated based on physical information related to CMP of the wafer, the physical vectors being a combination of data representing different types of physical information.

[0064] In some embodiments, the physical information includes at least the following items: information of deposited materials; information of maximum dishing defects and minimum dishing defects in each step; information of polishing liquid; information of long-tail effect parameters; duration information; and nominal pressure information.

[0065] In some embodiments, generating a second number of physical vectors based on physical information related to CMP of the wafer may include: determining data in each step range based on the value range of the data points of each physical vector and the corresponding step; and orthogonalizing the data in each step range to generate a physical information data table, wherein the physical information data table includes data of the second number of physical vectors.

[0066] In some embodiments, the physical information can be expanded into vectors with lengths equal to density, space, and width. In other words, density, space, and width have equal lengths, and the physical information can be expanded into vectors with lengths equal to density, space, and width, respectively. These vectors can be combined to form a matrix, which is the input, denoted by X. This X can be used as the input of the neural network model.

[0067] Assume that the length of the generated GDS feature data table is k. Because physical parameters are also input variables, DOE design is also used to generate a DOE combination table based on actual process conditions. For example, pressure can be selected from several conditions such as low pressure, medium pressure, and high pressure. Assuming that there are L combination tables of DOE for physical parameters, the CMP physical simulation model needs to be run L times in total.

[0068] At block 206 , a second number of simulations are performed by the CMP physical model based on the feature vectors and the physical vectors to generate simulated values ​​of the surface height of the wafer corresponding to the first number of feature vectors as training samples in each simulation.

[0069] In some embodiments, based on the number of physical vectors, a corresponding number of simulations are performed.

[0070] In some embodiments, each time the designed GDS feature data and one of the L physical model combination tables are input, L result tables of length k (k is the number of data points) will be generated. Since these multiple tables are independent of each other, they can be generated in parallel, greatly speeding up the generation speed. Specifically, assuming that several features are involved in orthogonalization, an orthogonal table is obtained after orthogonalization. Each row of data in the orthogonal table contains the characteristics (density / space / width) and process parameters of the GDS, which are input into the physical model for calculation to obtain the calculation results, that is, the corresponding SThickNT, SThickT. Repeat several times until each row of data generates a corresponding SThickNT and SThickT.

[0071] In some embodiments, in order to facilitate operation and processing in the neural network, the data can be merged (or spliced) by embedding operation, or mapping operation. The specific method is as follows: Assuming that the value of process parameter a is 0.5, the vector of process parameter a is [0.5, 0.5, ...] (actually a copy). The vector with a length equal to density is similar to process vector b, so each process vector can be obtained. Similar to the previous operation of splicing density and width, these process vectors can also be spliced ​​to the back, thus forming a final input rectangle X. Assuming that density, space, and width are used as GDS features, and process parameters a, b, and c are used as process features, there are a total of k data points, then the final input matrix is ​​a matrix with k rows and 6 columns, of which there are 3 GDS features and 3 process parameters, so there are 6 columns in total. Splicing the data is more convenient in terms of engineering implementation. In addition, because the input matrix will perform matrix multiplication with the parameter matrix of the neural network, it is easy to integrate the process information into the input, which is efficient.

[0072] In some embodiments, performing a second number of simulations based on a characteristic vector and a physical vector by a CMP physical model may include: generating a first simulation value of a surface height of a non-groove structure and a second simulation value of a surface height of a groove structure, respectively; and determining a simulation value of a wafer surface height based on the first simulation value and the second simulation value.

[0073] It should be understood that the embodiments of the present disclosure are not limited to this. For example, in some embodiments, the vectors may not be concatenated but may be input into the neural network model as one-dimensional vectors.

[0074] It should be pointed out that the DOE orthogonal table only has input (neural network input), that is, there is no S-related column. The S-related column refers to the label data column required for training the neural network, so the DOE orthogonal table needs to be input into the physical model for simulation to obtain the physical model simulation result, so the simulation result of the physical model becomes the label data column. The function of the DOE orthogonal table (after the label data column is generated after the physical model simulation) is to provide neural network training and divide a part as a verification data set to verify whether the training is in place.

[0075] In some embodiments, the result table may be: SThickNT / SThickT / SDishing / SErosion, where SThickNT and SThickT are outputs of the neural network, and the difference between the two is dishing; the reference surface height (a certain constant) minus SThickNT is SErosion.

[0076] In addition, parallel acceleration was mentioned earlier. The so-called parallel acceleration can have two scenarios in this case.

[0077] One scenario is that we have a trained neural network, and this neural network replaces the physical model because neural networks can be accelerated in parallel. For example, there are k data points, so the vector length is k. Assuming that the genetic algorithm (GA) needs to iterate 100 times and the population size is 1000, it means that in each round of iteration, simulation results of k data points need to be generated. How many times do they need to be generated? 1000 times. Therefore, 1000 results of length k need to be generated in each round. When generating these 1000 results of length k in each round, because these 1000 populations are independent of each other, they can be accelerated in parallel using GPUs. However, the CMP physical model is difficult to deploy on the GPU due to its computational complexity. At most, it can be implemented by calling ordinary multi-threading in the computer, for example, using 20 threads for acceleration, that is, letting these 20 threads calculate the 1,000 k-length results (equivalent to each thread completing 50, and everyone completing it together). However, neural networks are naturally suitable for GPUs and are very easy to implement. The acceleration capability of GPUs is not comparable to that of ordinary multi-processes. GPUs can implement tens of thousands of threads or even more, while ordinary multi-threading depends on the power and scale of the CPU, but the CPU itself cannot have many threads. In addition, when the neural network generates the k-length result, because the neural network obtains the result in one step, k data points can be predicted independently, but the CMP physical model needs to consider environmental information and cannot be calculated independently. The neural network model can be deployed in parallel (on the GPU) to directly obtain the calculation results in full parallelism.

[0078] Another scenario is when generating samples for training neural networks. At this time, because the samples are independent of each other, many machines can be used to generate results and finally collect them. For example, if 10,000 samples need to be generated for training, one way is to generate these 10,000 samples one by one, and another way is to open 100 servers, each server generates part of the results, and then collect them (a total of 10,000) for training.

[0079] Combine the following Figure 4 Give a description. Figure 4A schematic diagram of a deep learning model according to some embodiments of the present disclosure is shown, specifically a schematic diagram of a neural network architecture 400. The architecture includes: an input layer 402; the input layer includes a physical parameter information table and a GDS feature vector (which may be called a graphic vector), including a Width vector, a Space vector, and a Density vector. An embedding layer 404; usually, the physical information parameters are all scalars, which are mapped to vectors of the same length as density / width / space through embedding, and then all these vectors can be concatenated to form an input matrix X. The first convolutional layer 406 is connected to the embedding layer 404; the pooling layer 408 is connected to the first convolutional layer 406; the second convolutional layer 410 is connected after the pooling layer; the first fully connected layer 412 is connected after the second convolutional layer 410; the first regularization layer 414 is connected after the first fully connected layer 412; the second fully connected layer 418 is connected after the first regularization layer 414; the second regularization layer 420 is connected after the second fully connected layer 418, and the third fully connected layer 422 is connected after the second regularization layer 420; the output layer of the third fully connected layer 422 is connected to two fully connected layers, namely, the fourth fully connected layer 424 and the fifth fully connected layer 426. The last two fully connected layers are each connected to an output layer, namely, the SThickNT output layer 428 and the SThickT output layer 430. The SThickNT output layer 428 outputs the simulated value of the surface height of the non-groove structure, and the SThickT output layer 430 outputs the simulated value of the surface height of the groove structure. The simulated value of the height of the wafer surface can be determined based on the simulated value of the surface height of the non-groove structure and the simulated value of the surface height of the groove structure.

[0080] In some embodiments, a simulation value H1 of the height of the non-trench structure on the wafer can be determined; a simulation value H2 of the height of the trench structure on the wafer can be determined; and a simulation value of the height of the circular surface can be determined based on the simulation value H1, the simulation value H2, and the density of the graphics in the grid area on the wafer. Specifically, the surface height is equal to the height of the non-trench structure multiplied by (1-density) plus the product of the height of the trench structure and the density. The formula is: ; where density represents the density of the graphics in the grid area on the wafer, and its range is between 0-1.

[0081] like Figure 4As shown, the architecture also includes a Relu activation function 416. The Relu activation function is the rectified linear unit (ReLU) activation function, which is one of the most commonly used activation functions in convolutional neural networks (CNNs) and many deep learning models. Its main function is to introduce nonlinearity so that the model can learn and express more complex features.

[0082] As mentioned above, multiple result tables of length k are generated. In addition, after the training samples are determined and the neural network model is constructed, the neural network model can be trained. Figure 3 Specifically describe the method of training the neural network model. Figure 3 A flowchart of a training method 300 for a neural network model according to some embodiments of the present disclosure is shown.

[0083] At block 302 , a neural network model is trained using training samples to determine a gradient of a loss function of the neural network model, wherein the training samples are generated by a CMP physical model based on graphic features and physical information for determining a surface height of a ground wafer.

[0084] In some embodiments, the training samples may be pre-generated, such as stored in a predetermined database, and may be used for training the neural network model.

[0085] In some embodiments, the training samples may be generated during the training process of the neural network model.

[0086] In some embodiments, training the neural network model using training samples to determine the gradient of the loss function of the neural network model includes: calculating the gradient of the loss function using a stochastic gradient descent algorithm.

[0087] In some embodiments, updating the parameters of the neural network model based on the gradient of the loss function includes: performing a gradient update on the gradient of the loss function after a predetermined number of trainings (e.g., 100 times); and updating the parameters of the neural network model based on the gradient update.

[0088] In some embodiments, using training samples to train a neural network model to determine the gradient of the loss function of the neural network model may include: using a portion of the training samples to train the neural network model to perform gradient descent back propagation to update parameters; using the trained neural network model to predict the remaining portion of the training samples to generate a predicted value of the surface height of the wafer after grinding; when the comparison result between the predicted value and the simulation value of the CMP physical model meets a predetermined condition, the corresponding trained neural network model may be determined as the final neural network model. That is, the degree of training of the neural network can be determined in this way. For example, assuming that DOE orthogonally generates 10 million data, the simulation results of the 10 million data are simulated using a physical model. Then 70% (or 80% or 90%) of them are taken out as training samples for training. When each of the 7 million training samples is input into the neural network for gradient descent back propagation to update the parameters, the training is completed. Next, use the trained neural network to predict the remaining 30% of the data, and then calculate the rmse of the neural network prediction value and the physical model simulation value. If this rmse meets a certain standard, the training can be considered in place. If it does not meet the standard, the DOE can be made more detailed, more data can be generated, and training can be continued, repeating the above steps.

[0089] For neural networks, in many cases, "input" and "training samples" are essentially the same. Training samples refer to some data that the neural network calculates, and then back-propagates to update the neural network parameters based on the calculation error. Compared with the training samples, the data set to be predicted (input) has no real corresponding values, and the neural network cannot update and adjust its own parameters. But the essence is the matrix X mentioned above. In addition, "input" is a broader concept than "training sample". If an input has feedback, it is a training sample, because it can be used to adjust the parameters of the neural network, which is also "training". If an input has no feedback, it is a data set to be predicted, because it is impossible to know whether the prediction result is good or bad, and it is impossible to know whether the result of the neural network fits accurately. It cannot be used for training. In general, the input describes a matrix or vector that meets the requirements of the input layer of the neural network, and the training sample refers to a set of data points, each of which has a standard reference value that can be used to evaluate the quality of the neural network prediction. The non-standard answer part of this data set can be extracted as input and input into the neural network. The neural network gives the corresponding output, and then back-propagates to update the parameters by comparing the standard answer and the output. So the input can also be said to be part of the training sample.

[0090] As mentioned earlier, after obtaining physical information, a matrix can be generated based on the physical information and graphic feature information as the input of the neural network model.

[0091] In some embodiments, an orthogonal table is generated based on DOE, and a CMP physical model is used for simulation to obtain a training sample. The training sample contains the information required for the neural network to be input (the results of the characteristics of the GDS and the process parameters) and the simulation results given by the physical model. As mentioned above, the GDS characteristics and process parameter columns can be extracted as the input X matrix and input into the neural network. The neural network will obtain the calculation results, and then the loss function is calculated and back-propagated by comparing the results of the neural network with the simulation results of the physical model, and the network parameters are updated to finally obtain a trained neural network. For example, there is a data point that contains GDS characteristics and process information. This point can be input into the neural network and the physical model. The output result of the neural network will be very close to the output result of the physical model, so the training is effective, and a neural network that has learned the physical model is obtained.

[0092] Back to Figure 2 Continuing the description, in some embodiments, training the neural network model using training samples to determine the gradient of the loss function of the neural network model includes: the gradient of the loss function can be calculated using a stochastic gradient descent algorithm.

[0093] In some embodiments, the gradient of the loss function may be calculated using a stochastic gradient descent algorithm.

[0094] In some embodiments, training a neural network model using training samples to determine the gradient of a loss function of the neural network model may include: training the neural network model using a portion of the training samples to perform gradient descent backpropagation to update parameters; and predicting the remaining portion of the training samples using the trained neural network model to generate a predicted value of the surface height of the wafer after grinding.

[0095] In fact, the essence of neural network training is to first use the neural network for simulation budgeting, then compare the labeled data, calculate the objective function loss, perform gradient descent, and backpropagate to update the parameters.

[0096] At block 304, parameters of the neural network model may be updated based on the gradient of the loss function.

[0097] In some embodiments, updating the parameters of the neural network model based on the gradient of the loss function may include: performing a gradient update on the gradient of the loss function after a predetermined number of trainings, such as 100 times; and updating the parameters of the neural network model based on the gradient update. It should be understood that the number of trainings for the prediction may vary according to actual needs, and the embodiments of the present disclosure are not limited thereto.

[0098] In some embodiments, a stochastic gradient descent algorithm is used to calculate gradients and continuously update neural network parameters.

[0099] At block 306 , in response to the training satisfying a predetermined condition, the neural network model corresponding to the corresponding updated parameters is determined as a trained neural network model, wherein the trained neural network model outputs a predicted value of the surface height.

[0100] In some embodiments, the method also includes: determining the loss function value of the neural network model; and wherein the training satisfies a predetermined condition including: the difference between the loss function values ​​of two previous iterations is less than a predetermined difference threshold; or the absolute value of the ratio of the loss function values ​​of two previous iterations is greater than a predetermined ratio threshold.

[0101] In some embodiments, the stochastic gradient descent algorithm is used to calculate the gradient and continuously update the neural network parameters until the loss curve approaches a stable state and the training is stopped.

[0102] In some embodiments, as mentioned above, the training sample may be used to train the neural network model to determine the gradient of the loss function of the neural network model, which may include: using a portion of the training sample to train the neural network model to perform gradient descent back propagation to update the parameters; using the trained neural network model to predict the remaining portion of the training sample to generate a predicted value of the surface height of the wafer after grinding. In the case where the comparison result between the predicted value and the simulation value of the CMP physical model meets the predetermined condition, the corresponding trained neural network model may be determined as the final neural network model. Wherein the comparison result between the predicted value and the simulation value meets the predetermined condition may include: the root mean square error between the predicted value and the simulation value is less than a predetermined threshold.

[0103] In some embodiments, it may also include: determining the loss function value of the neural network model; and wherein the training satisfies a predetermined condition including: the difference between the loss function values ​​of two previous iterations is less than a predetermined difference threshold; or the absolute value of the ratio of the loss function values ​​of two previous iterations is greater than a predetermined ratio threshold.

[0104] In some embodiments, training samples are generated in parallel on multiple servers so that the training of the neural network model is performed in parallel.

[0105] In some embodiments, a trained neural network model may be generated according to the above-mentioned neural network training method.

[0106] In some embodiments, a neural network model is also provided, including: an input layer, configured to receive graphic feature data on a wafer and physical information related to CMP of the wafer; an intermediate layer, configured to execute the above-mentioned simulation method of the semiconductor device to determine the surface height value of the wafer after grinding; and an output layer, configured to output the surface height value.

[0107] In some embodiments, the neural network model further includes: an embedding layer configured to map the input physical information to generate a physical vector.

[0108] The above embodiment provides a trained deep learning neural network model for tracking the CMP physical model, and the following describes how to use it. In some embodiments, the neural network model can be used together with the CMP model.

[0109] In some embodiments, a simulation model is also provided, which includes: a CMP physical model, which is configured to determine a simulation value of the surface height of the wafer after grinding based on input data, wherein the input data includes graphic feature data and physical information; and a neural network model as described above, which is generated by training with simulation values ​​as training samples, and is configured to perform parallel predictions on multiple sets of input data during the simulation process of the CMP physical model to generate multiple predicted values ​​of the surface height of the wafer after grinding corresponding to the multiple sets of input data, respectively serving as corresponding simulation values ​​of the surface height.

[0110] In some embodiments, performing parallel prediction on multiple sets of input data during the simulation of the CMP physical model includes: utilizing multiple graphics processors to perform parallel acceleration on the computation process of the neural network model.

[0111] In some embodiments, a simulation method is also provided, which is performed using the above-mentioned simulation model. The method includes: a neural network model in the simulation model predicts a simulation value of the surface height of the wafer after grinding based on input data, wherein the input data includes graphic feature data and physical information.

[0112] In some embodiments, predicting the simulation value of the surface height of the wafer after grinding based on input data by a neural network model in the simulation model may include: the neural network model performs parallel prediction on multiple sets of input data to generate multiple predicted values ​​of the surface height of the wafer after grinding corresponding to the multiple sets of input data, respectively serving as the corresponding simulation values ​​of the surface height.

[0113] In some embodiments, the simulation value of the surface height of the wafer after grinding is predicted by the neural network model in the simulation model based on the input data, including: generating a first prediction value of the surface height of the non-groove structure and a second prediction value of the surface height of the groove structure by the neural network model respectively; and determining the simulation value of the wafer surface height based on the first prediction value and the second prediction value by the neural network model.

[0114] In some embodiments, it also includes: determining the surface condition of the wafer during the grinding process through the CMP physical model in the simulation model. For users, they ultimately need a physical model, because sometimes they need to output the intermediate process to help them confirm some relevant details of the process. For example, if manufacturers need to change the process parameter settings of their base station, they may need to know the grinding conditions at some intermediate moments to assist them in adjusting the parameters. Neural networks cannot provide the intermediate process, so physical models can meet this need.

[0115] According to the above description, it can be known that the embodiment of the present disclosure also provides a simulation method. The method can be implemented by the above simulation model.

[0116] In some embodiments, simulation may be performed using a CMP physical model according to DOE design principles to obtain simulation results of SThickNT / SThickT / SDishing / SErosion of the CMP physical model.

[0117] In some embodiments, a neural network model in the simulation model predicts a simulated value of a surface height of the wafer after grinding based on input data, wherein the input data includes graphic feature data and physical information.

[0118] As can be seen from the above, the current CMP physical model will have a large loss of time and computing resources during training, mainly because when running a parameter combination for a certain test, due to the characteristics of the time step model, a lot of calculations need to be generated. The embodiment of the present disclosure provides a neural network model that has learned the complex physical relationship. The model obtains the same input as the CMP physical model, but the speed of giving the output result will be extremely fast. In fact, the time for the CMP physical model to calculate a test may be tens of seconds, while the time for the neural network model to give the output is only a few tenths of a second, which will greatly improve the computing speed and efficiency.

[0119] Similarly, for the CMP physical model, the GA genetic algorithm is used for fitting. The GA genetic algorithm needs to set two parameters: the population size (pop size) and the number of iterations (max iter). In the case of a long single calculation time, it cannot be set too large, otherwise the total time consumed will be very long. The GA algorithm is an algorithm used to solve the parameters in the CMP physical model. The GA algorithm needs to conduct experiments continuously, and each experiment requires the CMP physical model to give simulation results of many experimental individuals. The iteration of the GA algorithm only requires one objective function, where the objective function is set as the square error between the measured data point (real label data) and the simulation result given by the simulation model. The GA algorithm has a large amount of calculation (if you want to find a suitable solution). If you want to get a better solution, the GA algorithm needs to obtain a very large number of test results. The test result refers to the simulation value of all real data points given a parameter combination using the CMP physical model with this parameter combination, and then comparing the difference between the simulation value given by the model and the real label value. This is the result of an experiment.

[0120] Parameter combination refers to the parameter combination generated by the pending parameters in the CMP physical model. For example, 10 pending parameters are set in the CMP physical model. These 10 pending parameters are parameter combinations. Because each parameter has many possible values, there are many combinations. With each additional pending parameter, all possible parameter combinations increase exponentially. The total number of searches by the GA algorithm is the population size multiplied by the number of iterations. It takes 10 seconds to evaluate a CMP physical model, but only 0.1 seconds to evaluate a deep learning model. It can be seen that it is very necessary to use deep learning to accelerate.

[0121] The acceleration of solving the physical model parameters has been described before. Given a data set, the physical model obtains the simulation results of each point of the data set much slower than the neural network. In the iterative process of the GA algorithm, the data set needs to be simulated repeatedly to obtain the simulation results. If only one simulation is calculated, there is not much difference between 10s and 0.1s. If ten thousand simulations need to be calculated, the difference is very large. The same is true for accelerating the simulation. For a given data set, neural network calculations are much faster than physical models. This is no problem. In addition, as the size of the data set increases, the advantages of neural networks become more and more obvious. For example, if there are only 100 data points, the physical model needs 3s, and the neural network needs less than 0.1s. If 10 million data points need to be simulated, the physical model may need seconds, but the computing time of a neural network does not increase like this, it may only take 2 seconds.

[0122] As mentioned above, there are two main scenarios in the use of the CMP tool: parameter solution and global simulation. When solving parameters, the user provides dozens of collected real data points to solve the parameters of the physical model. When simulating globally, the user can provide tens of millions of data points to calculate the simulation value of each data point through simulation.

[0123] By minimizing the objective function, a set of parameter solutions can be found that can make the results given by the simulation model as close to the true value as possible. If deep learning acceleration is not introduced, the objective function here is the square error between the simulation results of the physical simulation model and the true results. If it is introduced, it becomes the square error between the predicted results of the deep learning network and the true results.

[0124] The physical model is a large model (very complex, and the calculation takes a long time. Solving the optimal parameters consumes a lot of computing resources and takes a long time, but it is more accurate and can output the simulation value of the CMP intermediate process. The neural network model operates very fast, and because the serial process is discarded, the calculation of the neural network can be completely accelerated in parallel. In short, even without parallel acceleration (using GPU), the pure neural network prediction calculation speed is much faster than the physical model. Now, precisely because of the characteristics of the two calculations, the neural network can be accelerated by GPU parallel again, which can greatly improve the speed. Therefore, the final model used for approximate substitution can greatly improve the efficiency of parameter solution.

[0125] In the embodiments of the present disclosure, the single operation time is greatly reduced through the approximate replacement of the neural network model. Therefore, a very large population size and number of iterations can be set to find the optimal parameter combination, and the total time has not increased compared to before, which greatly improves the probability of finding a better solution.

[0126] For simulation, the above embodiment provides a neural network model that can respond extremely quickly and is a model that is easy to accelerate using hardware such as GPU (compared to the CMP physical model). Therefore, even if the amount of full-chip data to be simulated is huge, even if it is tens of millions of rows, the original CMP physical model takes several hours, but the neural network model used in the present disclosure only takes a few minutes.

[0127] As we all know, deep learning neural network models have extremely strong fitting capabilities. However, one of the characteristics of deep learning neural networks is that they require a very large amount of data. Therefore, the defect is that the amount of training data required is large, and in many actual situations there is not so much labeled data available for training. In the CMP process, due to practical limitations (such as cost), manufacturers cannot provide the amount of data required to train a neural network. Usually, they can only provide 50-100 data points. For neural networks, hundreds of thousands or millions of data points are required, which is obviously impossible. In addition, as mentioned earlier, the CMP physical model is a serial calculation model, that is, the result at each moment also depends on the result of the previous step. This calculation feature makes it impossible to parallelize.

[0128] However, this application does not use the neural network model to directly fit the actual CMP process, and does not use real measurement data for fitting (the actual measurement points are often very few), but takes a different approach, allowing the neural network to track and imitate the CMP physical model. On the known CMP physical model, it can theoretically generate unlimited label data, thus solving the problem of neural network model training. In addition, this application does not directly replace the CMP physical model with a neural network, but only uses its extremely fast calculation speed to replace the return result requirements of the objective function in the GA algorithm. It does not directly use the CMP physical model to return the test results required by the GA algorithm, but uses a trained neural network model to give the test results. The ultimate goal has not changed, which is still to obtain a parameter combination that makes the CMP physical model fit better, but how to achieve this goal, this application cleverly uses neural networks for acceleration.

[0129] Some embodiments of the present disclosure provide a method for generating training samples, a method for training a neural network, and a method for simulating a semiconductor device. It should be noted that the examples given in the above embodiments are only for illustrating the solutions of the embodiments of the present disclosure and are not intended to limit the solutions of the present disclosure.

[0130] In some embodiments of the present disclosure, the idea of ​​approximate substitution is creatively used to skillfully introduce neural networks to accelerate the speed of parameter solution and simulation of physical models. The embodiments of the present disclosure can utilize the advantages of neural networks and are fully feasible in theory and practice.

[0131] The present disclosure creatively introduces deep learning / machine learning models to greatly accelerate the establishment and simulation process of CMP physical models, greatly reducing the related time and energy consumption, and is extremely practical in actual use.

[0132] The deep learning / machine learning model introduced in the present disclosure does not change the CMP physical model and cause it to lose its original accuracy and results. It only greatly helps its establishment and simulation through appropriate use, and can even further improve its accuracy. The present disclosure greatly reduces the simulation and establishment of the CMP physical model by using the deep learning model, reducing the time cost. The ingenious use of the deep learning model in the present disclosure can help improve the accuracy of the CMP physical model to a certain extent and help find a more suitable combination of model parameters.

[0133] It should be understood that the embodiments shown in the drawings are only for schematically illustrating the solutions of some embodiments of the present disclosure and are not intended to limit the present disclosure. The embodiments of the present disclosure may also have various other forms.

[0134] An electronic device is also disclosed in an embodiment of the present disclosure. The electronic device includes: a processor; and a memory coupled to the processor, the memory having instructions stored therein, and when the instructions are executed by the processor, the device performs actions, the actions including: training a neural network model using training samples to determine the gradient of a loss function of the neural network model, wherein the training samples are generated by a CMP physical model based on graphic features and physical information for determining the surface height of a wafer after grinding; updating parameters of the neural network model based on the gradient of the loss function; and in response to the training satisfying a predetermined condition, determining the neural network model corresponding to the corresponding updated parameters as a trained neural network model, wherein the trained neural network model outputs an enhanced formula for determining the surface height.

[0135] The embodiments of the present disclosure further disclose a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method of the embodiments of the present disclosure is implemented.

[0136] Figure 5 A schematic block diagram of an electronic device according to some exemplary embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0137] like Figure 5As shown, the device 500 includes a CPU 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The CPU 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0138] Multiple components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0139] The various processes and processing described above, such as methods 200, 300, may be executed by the CPU 501. For example, in some embodiments, the methods 200, 300 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the CPU 501, one or more steps in the methods 200, 300 described above may be performed.

[0140] The scheme according to the embodiment of the present disclosure may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing various aspects of the present disclosure are loaded. The computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. The computer-readable program instructions may be downloaded from the computer-readable storage medium to various computing / processing devices, 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.

[0141] Various embodiments of the present disclosure have been described above, and the above description is exemplary and is only an optional embodiment of the present disclosure, not exhaustive, and is not intended to limit the present disclosure. Although the claims in this application have been formulated for specific combinations of features, it should be understood that the scope of the present disclosure also includes any novel features or any novel combination of features disclosed herein, whether or not it relates to the same scheme in any claim currently claimed for protection. It should be understood that new claims may be formulated into these features and / or combinations of these features during the examination of this application or in any further application derived therefrom.

[0142] The terms used in this article are selected to best explain the principles of each embodiment, practical application or technical improvement in the market, or to enable other ordinary technicians in the field to understand the embodiments disclosed herein. For those skilled in the art, the present disclosure may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the scope of protection of the present disclosure.

Claims

1. A method for generating a training sample, comprising: Generating a first number of feature vectors based on data points for representing various graphic features on the wafer, wherein the feature vectors are combinations of data representing different types of graphic features in the data points; generating a second number of physical vectors based on physical information related to chemical mechanical polishing of the wafer, the physical vectors being a combination of data representing different types of physical information; as well as The second number of simulations are performed by the chemical mechanical polishing physical model based on the feature vectors and the physical vectors, so as to generate simulation values ​​of the surface height of the wafer corresponding to the first number of feature vectors as training samples in each simulation.

2. The method according to claim 1, wherein generating a first number of feature vectors based on data points for representing each graphic feature on the wafer comprises: Determine the data within each step length based on the value range of the data points of each of the graphic features and the corresponding step length; as well as Data orthogonalization is performed on the data within each step range to generate a feature data table, wherein the feature data table includes data of the first number of feature vectors.

3. The method according to claim 1, wherein generating the second quantity of physical vectors based on physical information related to chemical mechanical polishing of the wafer comprises: Determine data of each step range based on the value range of the data points of each physical vector and the corresponding step size; as well as Orthogonalization is performed on the data within each step range to generate a physical information data table, wherein the physical information data table includes the data of the second number of physical vectors.

4. The method according to claim 1, wherein performing the simulation of the second quantity based on the feature vector and the physical vector by a chemical mechanical polishing physical model comprises: respectively generating a first simulation value of the surface height of the non-groove structure and a second simulation value of the surface height of the groove structure; as well as A simulation value of the wafer surface height is determined based on the first simulation value and the second simulation value.

5. The method according to claim 1, wherein the graphic features at least include: The density of the pattern within the grid on the wafer; The width of the graphic; and The spacing between graphics.

6. The method according to claim 1, wherein the physical information comprises at least the following items: Information on deposited materials; Maximum dishing defect information and minimum dishing defect information for each step; Grinding fluid information; Long tail effect parameter information; Duration information; Nominal pressure information.

7. A method for training a neural network model, comprising: Training the neural network model using training samples to determine the gradient of the loss function of the neural network model, wherein the training samples are generated by a chemical mechanical polishing physical model based on graphic features and physical information for determining the surface height of the wafer after grinding; Updating the parameters of the neural network model based on the gradient of the loss function; as well as In response to the training satisfying a predetermined condition, the neural network model corresponding to the corresponding updated parameters is determined as a trained neural network model, wherein the trained neural network model outputs a predicted value of the surface height.

8. The method according to claim 7, wherein training the neural network model using the training samples to determine the gradient of the loss function of the neural network model comprises: The gradient of the loss function is calculated using a stochastic gradient descent algorithm.

9. The method of claim 8, wherein updating the parameters of the neural network model based on the gradient of the loss function comprises: In response to performing a predetermined number of trainings, performing a gradient update on the gradient of the loss function; as well as The parameters of the neural network model are updated based on the gradient update.

10. The method according to claim 7, wherein training the neural network model using the training samples to determine the gradient of the loss function of the neural network model comprises: Using a portion of the training samples to train the neural network model to perform gradient descent back propagation to update parameters; Using the trained neural network model to predict the remaining part of the training sample to generate a predicted value of the surface height of the wafer after grinding; as well as In response to a comparison result between the predicted value and the simulation value of the chemical mechanical polishing physical model satisfying a predetermined condition, the corresponding trained neural network model is determined as a final neural network model.

11. The method according to claim 10, wherein the comparison result between the predicted value and the simulated value satisfies a predetermined condition, comprising: A root mean square error between the predicted value and the simulated value is less than a predetermined threshold.

12. The method according to claim 7, further comprising: Determining a loss function value of the neural network model; And wherein, the training satisfies the predetermined conditions including: The difference between the loss function values ​​of two previous and subsequent iterations is less than a predetermined difference threshold; or An absolute value of a ratio of the loss function values ​​of two previous and subsequent iterations is greater than a predetermined ratio threshold.

13. The method according to any one of claims 7 to 12, wherein the training samples are generated in parallel on multiple servers so that the training of the neural network model is performed in parallel.

14. A neural network model generated by the method according to any one of claims 7 to 13.

15. A simulation model comprising: A chemical mechanical polishing physical model, wherein the chemical mechanical polishing physical model is configured to determine a simulation value of a surface height of a polished wafer based on input data, wherein the input data includes graphic feature data and physical information; as well as According to the neural network model of claim 14, the neural network model is generated by training with the simulation values ​​as training samples, and is configured to perform parallel predictions on multiple sets of input data during the simulation process of the chemical mechanical polishing physical model to generate multiple predicted values ​​of the surface height of the polished wafer corresponding to the multiple sets of input data, respectively serving as the corresponding simulation values ​​of the surface height.

16. The simulation model according to claim 15, wherein in the simulation process of the chemical mechanical polishing physical model, performing parallel prediction on multiple sets of input data comprises: Multiple graphics processors are used to accelerate the computation process of the neural network model in parallel.

17. A simulation method for a semiconductor device, the method being performed using the simulation model according to claim 15, the method comprising: The neural network model in the simulation model predicts a simulation value of the surface height of the wafer after grinding based on input data, wherein the input data includes graphic feature data and physical information.

18. The method according to claim 17, wherein predicting the simulation value of the surface height of the wafer after grinding based on the input data by the neural network model in the simulation model comprises: The neural network model performs parallel prediction on multiple sets of input data to generate multiple predicted values ​​of the surface height of the polished wafer corresponding to the multiple sets of input data, which are respectively used as corresponding simulation values ​​of the surface height.

19. The method according to claim 17, wherein predicting a simulation value of a surface height of a wafer after grinding based on input data by a neural network model in the simulation model comprises: The neural network model generates a first predicted value of the surface height of the non-groove structure and a second predicted value of the surface height of the groove structure respectively; as well as The neural network model determines a simulation value of the wafer surface height based on the first prediction value and the second prediction value.

20. The method according to claim 17, further comprising: The surface condition of the wafer during the grinding process is determined by the chemical mechanical polishing physical model in the simulation model.

21. A neural network model, comprising: An input layer, configured to receive pattern feature data on a wafer and physical information related to chemical mechanical polishing of the wafer; An intermediate layer configured to perform the method according to any one of claims 17 to 20 to determine a surface height value of a wafer after grinding; as well as The output layer is configured to output the surface height value.

22. The method according to claim 21, further comprising: The embedding layer is configured to map the input physical information to generate a physical vector.

23. An electronic device, comprising: processor; as well as A memory coupled to the processor, the memory having instructions stored therein, the instructions causing the device to perform actions when executed by the processor, the actions comprising: Training the neural network model using training samples to determine the gradient of the loss function of the neural network model, wherein the training samples are generated by a chemical mechanical polishing physical model based on graphic features and physical information for determining the surface height of the wafer after grinding; Updating the parameters of the neural network model based on the gradient of the loss function; as well as In response to the training satisfying a predetermined condition, the neural network model corresponding to the corresponding updated parameters is determined as a trained neural network model, wherein the trained neural network model outputs an enhanced formula for determining the surface height.

24. A computer-readable storage medium having machine-executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the method according to any one of claims 1 to 13 and claims 17 to 20.

Images