Wafer surface treatment process recommendation method and device, equipment and storage medium

Abstract

The invention provides a wafer surface treatment process recommendation method and device, equipment and a storage medium, and relates to the technical field of semiconductor manufacturing, and the method comprises the steps: obtaining a surface physicochemical characteristic parameter of a target wafer, and carrying out the preprocessing of the surface physicochemical characteristic parameter, and obtaining a target physicochemical characteristic parameter; inputting the target physicochemical characteristic parameter into a pre-trained single hidden layer neural network model to obtain a preliminary recommendation parameter of the target wafer in the aspect of each process parameter; and splicing the target physicochemical characteristic parameters and the preliminary recommendation parameters, then jointly inputting the spliced parameters into a Transform encoder based on an attention mechanism, modeling a complex coupling relationship between input characteristics by using a multi-head self-attention mechanism, and extracting a depth expression to output target recommendation parameters. Through the method, nonlinear mapping from the surface physicochemical characteristic parameters to the process parameters of the target wafer can be realized, so that efficient, accurate and automatic recommendation of the wafer plasma surface treatment process parameters can be realized.

Description

Technical Field

[0001] The present application relates to the field of semiconductor manufacturing technology, and in particular to a method, device, equipment and storage medium for recommending a wafer surface treatment process. Background Art

[0002] Heterogeneous wafer bonding is an advanced manufacturing technique that combines surface-treated semiconductor wafers made of different materials. This technology has broad applications in microelectronics manufacturing, microelectromechanical system (MEMS) packaging, multifunctional chip integration, and other emerging fields. Direct heterogeneous wafer bonding requires extremely high surface activation processes, including plasma treatment to adjust the surface chemistry to achieve high-quality bonding.

[0003] During wafer plasma surface treatment, a complex coupling relationship exists between various process parameters. For example, changes in gas type and power directly affect the activation effect, while activation time and vacuum level also present a delicate balance. Existing physical models and statistical methods often struggle to fully capture the complexity of this nonlinear, multivariate coupling, making it difficult to achieve ideal surface treatment results in actual production. Summary of the Invention

[0004] In view of this, the present application proposes a method, device, equipment and storage medium for recommending a wafer surface treatment process.

[0005] In a first aspect, the present application provides a method for recommending a wafer surface treatment process, comprising: Acquiring surface physicochemical characteristic parameters of a target wafer, and preprocessing the surface physicochemical characteristic parameters to obtain target physicochemical characteristic parameters; Inputting the target physicochemical characteristic parameters into a pre-trained single hidden layer neural network model to obtain preliminary recommended parameters for each process parameter of the target wafer; The target materialization characteristic parameters and the preliminary recommendation parameters are spliced ​​and input into the Transformer encoder based on the attention mechanism. The multi-head self-attention mechanism is used to model the complex coupling relationship between the input features, and the deep expression is extracted to output the target recommendation parameters.

[0006] In one embodiment, the wafer surface treatment process recommendation method further includes: Inputting the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network to obtain the predicted error rate of each process parameter of the target wafer; The target recommended parameters are calibrated according to the predicted error rates of the process parameters of the target wafer to generate calibrated recommended parameters.

[0007] In one embodiment, the wafer surface treatment process recommendation method further includes: Selecting a reference wafer with the most similar physical and chemical properties from a wafer database according to the target physical and chemical property parameters; Obtaining a target prediction error rate of each process parameter of the reference wafer based on the physicochemical characteristic parameters of the reference wafer, the single hidden layer neural network model, the Transformer encoder, and the error rate prediction model; Obtaining a first error rate difference between a target predicted error rate and an actual error rate of each process parameter of the reference wafer; The calibration recommended parameters are compensated according to the first error rate difference to generate compensated recommended parameters.

[0008] In one embodiment, the surface physicochemical characteristic parameters include material chemical characteristic parameters, roughness, and mechanical characteristic parameters; and the preprocessing of the surface physicochemical characteristic parameters to obtain target physicochemical characteristic parameters includes: Performing linear normalization processing on the chemical characteristic parameters of the material to obtain a first characteristic parameter; performing quantile normalization on the roughness and the mechanical characteristic parameter to obtain a second characteristic parameter and a third characteristic parameter; The first characteristic parameter, the second characteristic parameter and the third characteristic parameter are spliced ​​into an input characteristic matrix to obtain the target physicochemical characteristic parameter.

[0009] In one embodiment, the single hidden layer neural network model includes: an input layer, a hidden layer, and an output layer; The number of neurons in the input layer is the same as the number of parameter types included in the surface physical and chemical characteristic parameters; The hidden layer includes multiple neurons and uses a nonlinear function as an activation function; The number of neurons in the output layer is the same as the number of parameter types included in the process parameters, and the activation function of the output layer is a Sigmoid activation function or a linear activation function; The single hidden layer neural network model is trained using a sample set based on historical wafer processing experimental data to adjust the connection weights and bias parameters between the hidden layer and the output layer.

[0010] In one embodiment, the target materialization characteristic parameters and the preliminary recommendation parameters are concatenated and inputted into a Transformer encoder based on an attention mechanism, and a multi-head self-attention mechanism is used to model the complex coupling relationship between input features, extract deep expressions, and output target recommendation parameters, including: splicing the target physicochemical characteristic parameters and the preliminary recommended parameters to form a multi-dimensional joint input vector; Mapping the multidimensional joint input vector to the high-dimensional feature space required by the Transformer encoder through a linear embedding layer; The embedded features of the high-dimensional feature space are processed by the Transformer encoder to generate a high-expression hidden representation that integrates multi-source features, and the target parameters are recommended through regression head mapping.

[0011] In one embodiment, inputting the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network to obtain the predicted error rate of each process parameter of the target wafer includes: The error prediction of mature wafer samples is modeled as a time series process, which is defined as a continuous Ti sequence step error prediction task; Inputting the target recommended parameters and physicochemical characteristic parameters of the mature wafer sample into the error rate prediction model, obtaining the predicted error rate of each sequence step output by the error rate prediction model, and obtaining a second error rate difference between the predicted error rate and the actual error rate of each sequence step; A cumulative correction trend is formed according to the second error rate difference corresponding to each sequence step; Inputting the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network, and outputting the preliminary prediction error rate of each process parameter; The cumulative correction trend is used to correct the preliminary prediction error rate of each process parameter of the target wafer to obtain a target prediction error rate of each process parameter.

[0012] In a second aspect, the present application provides a wafer surface treatment process recommendation device, comprising: An acquisition module is used to acquire surface physicochemical characteristic parameters of a target wafer and pre-process the surface physicochemical characteristic parameters to obtain target physicochemical characteristic parameters; A first recommendation module is used to input the target physicochemical characteristic parameters into a pre-trained single hidden layer neural network model to obtain preliminary recommended parameters for process parameters; The second recommendation module is used to splice the target materialization characteristic parameters and the preliminary recommendation parameters and input them into the Transformer encoder based on the attention mechanism. It uses the multi-head self-attention mechanism to model the complex coupling relationship between input features, extract deep expressions and output target recommendation parameters.

[0013] In a third aspect, the present application provides an electronic device comprising a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the wafer surface treatment process recommendation method as described in the first aspect.

[0014] In a fourth aspect, the present application further provides a computer storage medium on which a computer program is stored, wherein when the computer program is executed by a processor, the wafer surface treatment process recommendation method as described in the first aspect is implemented.

[0015] The recommended wafer surface treatment process method of the present application has the following beneficial effects compared with the related art: 1. This application quickly generates preliminary recommended parameters for the target wafer in terms of various process parameters by inputting the target physicochemical characteristic parameters into a pre-trained single hidden layer neural network model; then the preliminary recommended parameters and the target physicochemical characteristic parameters are jointly input into a Transformer encoder based on an attention mechanism, and the multi-head self-attention mechanism of the Transformer encoder is used to model the complex coupling relationship between input features, extract the deep expression to complete the final recommendation, and generate the target recommended parameters, thereby capturing the complex coupling relationship between multiple process parameters through the neural network model, realizing the nonlinear mapping between the surface physicochemical characteristic parameters of the target wafer and the process parameters, and realizing efficient, accurate and automated recommendation of wafer plasma surface treatment process parameters.

[0016] 2. This application uses a single hidden layer neural network model for preliminary recommendation, and uses a Transformer encoder to optimize the preliminary recommendation results to obtain the target recommendation parameters, thereby combining the rapid response capability of a single hidden layer neural network with the deep modeling capability of the attention-based Transformer structure for complex dependencies of input features. While improving the prediction speed, it significantly enhances the model's adaptability to heterogeneous materials and multi-parameter coupling scenarios, and exhibits stronger adaptability and prediction accuracy when facing diverse wafer materials and complex process environments. Through the wafer surface treatment process recommendation method of this application, the wafer surface treatment process recommendation process has shifted from traditional experience-driven to data-driven, which not only reduces the trial and error cost and debugging time, but also improves the consistency, yield and stability of the process, and has good versatility, scalability and industrialization prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of a recommended method for wafer surface treatment process in one embodiment of the present application; Figure 2This is a flow chart of a recommended method for wafer surface treatment process in another embodiment of the present application; Figure 3 Schematic diagram of the process of the similar wafer difference correction strategy in one embodiment of the present application; Figure 4 This is a schematic diagram of the network structure of a single hidden layer neural network model in one embodiment of the present application; Figure 5 This is a structural diagram of a recommended device for wafer surface treatment process in one embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0020] Currently, wafer surface plasma treatment process parameters (such as vacuum level, gas type, power, treatment time, and power frequency) often require extensive experimentation and engineer experience to adjust. Optimal parameters vary significantly depending on wafer material and surface conditions. Validating treatment parameters for each wafer material requires multiple experiments, which is time-consuming and costly, leading to long debugging cycles and poor process stability. Existing empirical models for new materials (such as GaAs) suffer from a lack of rapid adaptability, insufficient accuracy, and difficulty accurately modeling the mapping between complex material physicochemical properties (such as roughness and crystal orientation deviation) and process parameters.

[0021] In related technologies, rule-based parameter recommendation systems cannot effectively handle multidimensional nonlinear relationships, resulting in recommendations that deviate from actual needs. Furthermore, traditional algorithms have limited responsiveness to dynamic environmental changes, making it difficult to meet the high-precision process parameter requirements of advanced manufacturing processes.

[0022] Based on this, in some embodiments, such as Figure 1 As shown, an embodiment of the present application provides a method for recommending a wafer surface treatment process, and the method for recommending a wafer surface treatment process includes the following steps S101 to S103.

[0023] S101: Acquire surface physicochemical characteristic parameters of a target wafer, and preprocess the surface physicochemical characteristic parameters to obtain target physicochemical characteristic parameters.

[0024] The surface physicochemical characteristic parameters of the target wafer may include material chemical characteristic parameters, roughness, and mechanical characteristic parameters. After obtaining the surface physicochemical characteristic parameters of the target wafer, the surface physicochemical characteristic parameters can be normalized for subsequent input into the neural network model or Transformer encoder.

[0025] S102: Input the target physicochemical characteristic parameters into a pre-trained single hidden layer neural network model to obtain preliminary recommended parameters for each process parameter of the target wafer.

[0026] It can be understood that a single hidden layer neural network model is used to construct a nonlinear mapping relationship between the physical and chemical characteristic parameters of the wafer and the plasma surface treatment process parameters. Therefore, after the target physical and chemical characteristic parameters are input into the pre-trained single hidden layer neural network model, the single hidden layer neural network model can be used to output the preliminary recommended parameters of the target wafer in terms of various process parameters.

[0027] It should be noted that the recommended parameters in the embodiments of the present application are characteristic parameters that have been uniformly normalized or standard-encoded, and the recommended parameters need to be restored to physical quantities that can be executed in actual engineering through inverse mapping processing.

[0028] In applications, process parameters may include vacuum parameters, medium gas type, power parameters, processing time parameters, and frequency parameters.

[0029] Vacuum degree parameter (Op1): Linear normalization is used to map the actual vacuum degree value (5Pa to 100Pa) to the interval [0,1] so that the neural network output is stable.

[0030] Medium gas type (Op2): Assign corresponding intervals according to the gas type (oxygen, nitrogen, argon), namely [0, 0.3], (0.3, 0.6], and (0.6, 1]), to achieve standardized coding of gas types.

[0031] Power parameter (Op3): The actual power (50W to 350W) is mapped to the interval [0, 1] through linear normalization to eliminate the scale inconsistency caused by different power levels.

[0032] Processing time parameter (Op4): The actual processing time (15 seconds to 150 seconds) is mapped to the interval [0, 1] through linear normalization to ensure the stability of the time feature value.

[0033] Frequency parameter (Op5): Based on the power supply operating frequency (13.56kHz or 40kHz), the frequency characteristic is discretized by assigning values ​​to the intervals [0, 0.5] and (0.5, 1], respectively.

[0034] For example, the reference range of the medium gas output value can be found in Table 1, and the reference range of the frequency parameter output value can be found in Table 2.

[0035] Table 1 is the reference range of medium gas output value

[0036] Table 2 is the reference range of frequency parameter output values

[0037] S103: The target materialized characteristic parameters and preliminary recommended parameters are concatenated and inputted into the attention-based Transformer encoder. The multi-head self-attention mechanism is used to model the complex coupling relationship between the input features, and the deep expression is extracted to output the target recommended parameters.

[0038] It can be understood that the Transformer encoder has the ability to model the complex coupling relationship between input features, and exhibits stronger adaptability and prediction accuracy when facing diverse wafer materials and complex process environments. Therefore, the target physical and chemical characteristic parameters and the preliminary recommendation parameters are spliced ​​and input into the Transformer encoder based on the attention mechanism. The effective fusion processing of the original input information and the rough recommendation results in the Transformer encoder is realized, which improves the contextual understanding ability of the recommendation parameters and the accuracy of the final output, thereby further improving the prediction accuracy.

[0039] The wafer surface treatment process recommendation method described above rapidly generates preliminary recommended parameters for each process parameter of the target wafer by inputting the target physicochemical characteristic parameters into a pre-trained single-hidden-layer neural network model. The preliminary recommended parameters and the target physicochemical characteristic parameters are then fed together into an attention-based Transformer encoder. The Transformer encoder's multi-head self-attention mechanism models the complex coupling relationships between input features, extracts deep representations, and ultimately recommends the target parameters. This neural network model captures the complex coupling relationships between multiple process parameters, achieving a nonlinear mapping between the target wafer's surface physicochemical characteristic parameters and process parameters, thus enabling efficient, accurate, and automated recommendation of wafer plasma surface treatment process parameters. Furthermore, by combining the rapid response capabilities of a single-hidden-layer neural network with the deep modeling capabilities of the attention-based Transformer architecture for complex input feature dependencies, the method significantly enhances the model's adaptability to heterogeneous materials and multi-parameter coupling scenarios while improving prediction speed. This method demonstrates stronger adaptability and prediction accuracy when dealing with diverse wafer materials and complex process environments.

[0040] In some embodiments, based on the above embodiments, Figure 2As shown, the recommended method for wafer surface treatment process also includes the following steps S201 and S202.

[0041] S104: Input the target physicochemical characteristic parameters and target recommended parameters into the error rate prediction model based on the long short-term memory neural network to obtain the predicted error rate of each process parameter of the target wafer.

[0042] It can be understood that the target physical and chemical characteristic parameters and target recommended parameters are input into a pre-built and trained error rate prediction model based on a long short-term memory neural network (LSTM). This error rate prediction model learns and analyzes the time series characteristics and complex correlation relationships in the input data, and outputs the predicted error rate corresponding to each process parameter of the target wafer, thereby realizing the prediction of the error of the wafer production process parameters.

[0043] S105: Calibrate the target recommended parameters according to the predicted error rate of each process parameter of the target wafer to generate calibrated recommended parameters.

[0044] It can be understood that after obtaining the predicted error rate of each process parameter of the target wafer, the target recommended parameters can be corrected in a targeted manner according to the predicted error rate of each process parameter of the target wafer, that is, the recommended values ​​corresponding to the process parameters with higher predicted error rates are adjusted to a large extent, and the parameters with lower error rates are fine-tuned or maintained, thereby generating calibration recommended parameters that take into account the prediction error feedback to optimize the accuracy of the process parameter settings.

[0045] In this embodiment, a dynamic error correction mechanism is introduced to achieve efficient, accurate, and automated recommendations for wafer plasma surface treatment process parameters. By combining the rapid response of a single-hidden-layer neural network with the deep modeling capabilities of the attention-based Transformer architecture for complex dependencies between input features, an LSTM error rate prediction model, coupled with time series error trend analysis, compensates for errors in target recommendations, effectively improving the accuracy of process parameter settings.

[0046] In some embodiments, based on the above embodiment, Figure 3 As shown, the recommended method for wafer surface treatment process also includes the following steps S301 to S304.

[0047] S301: Select a reference wafer with the most similar physical and chemical characteristics from a wafer database according to the target physical and chemical characteristic parameters.

[0048] For the target wafer, the surface physicochemical properties of the target wafer are used as a retrieval benchmark. The sample with the most similar physicochemical properties is retrieved from the established wafer database. The most similar wafer sample is used as the reference wafer, and the surface physicochemical properties and process parameters of the reference wafer can be obtained. It should be noted that the similarity assessment can use a nearest neighbor search method based on Euclidean distance to ensure that the reference wafer and the target wafer have highly similar physicochemical properties.

[0049] S302: Based on the physicochemical characteristic parameters of the reference wafer, the single hidden layer neural network model, the Transformer encoder and the error rate prediction model, the target prediction error rate of the reference wafer in each process parameter is obtained.

[0050] It can be understood that the physicochemical characteristic parameters of the reference wafer are input into a single hidden layer neural network model, and the single hidden layer neural network model can output preliminary recommended parameters for the reference wafer. The preliminary recommended parameters and physicochemical characteristic parameters of the reference wafer can then be input into the Transformer encoder, and the Transformer encoder can output the target recommended parameters for the reference wafer. After obtaining the target recommended parameters for the reference wafer, the physicochemical characteristic parameters and target recommended parameters of the reference wafer are input into the error rate prediction model, and the target predicted error rate of the reference wafer in each process parameter is obtained based on the parameters output by the error rate prediction model.

[0051] S303: Obtain a first error rate difference between a target predicted error rate and an actual error rate of each process parameter of a reference wafer.

[0052] It can be understood that the actual process parameters of the reference wafer are known. By comparing the target recommended parameters of the reference wafer with the actual process parameters, the actual error rate of each process parameter of the reference wafer can be determined. The first error rate difference can be obtained by subtracting the target predicted error rate from the actual error rate.

[0053] S304: Compensate the calibration recommended parameters according to the first error rate difference to generate compensated recommended parameters.

[0054] For example, the target recommended parameter of the reference wafer is G Op1_ref -G Op5_ref The target recommended parameters and physical and chemical characteristics of the reference wafer are used as input, and the LSTM model is used to predict the error rate of each process parameter of the reference wafer. pred_ref1 ~E pred_ref5 Then calculate the actual error rate of the reference wafer (E real_ref ), the difference between the actual error and the predicted error of each process parameter (ΔE i ), the corresponding formula is as follows: ; ; in, Indicates the actual process parameter Opi.

[0055] For the target wafer, enter its target recommended parameters (G Op1_tar -G Op5_tar ) and physical and chemical characteristics to the LSTM model, and predict its target prediction error rate (E pred_tar1 -E pred_tar5 ). The recommended calibration parameters are: .

[0056] Based on the difference ΔEi obtained from the reference wafer, the calibration recommended parameters are compensated according to the first error rate difference to obtain the compensation recommended parameters , the corresponding formula is as follows: .

[0057] It should be noted that, in the application, the target predicted error rate of the target wafer can be compensated according to the first error rate difference, and then the target recommended parameters of the target wafer can be corrected based on the compensated error rate, and finally the compensated and corrected wafer surface plasma treatment process parameters can be obtained as practical application recommendations. The corresponding formula can be as follows: ; .

[0058] In this embodiment, by introducing a similar wafer difference correction strategy, the first error rate difference between the target predicted error rate and the actual error rate of each process parameter of the reference wafer is obtained, and the calibration recommended parameters are compensated according to the first error rate difference to generate compensated recommended parameters, thereby performing secondary optimization on the target recommended parameters, thereby effectively controlling the error rate of the recommended parameters and improving the reliability of the final recommended parameters.

[0059] In some embodiments, the surface physicochemical characteristic parameters include material chemical characteristic parameters, roughness, and mechanical characteristic parameters. In step S101, the surface physicochemical characteristic parameters are preprocessed to obtain target physicochemical characteristic parameters, including: linear normalization of the material chemical characteristic parameters to obtain a first characteristic parameter; quantile normalization of the roughness and mechanical characteristic parameters to obtain a second characteristic parameter and a third characteristic parameter; and concatenation of the first characteristic parameter, the second characteristic parameter, and the third characteristic parameter into an input characteristic matrix to obtain the target physicochemical characteristic parameters.

[0060] X1 represents the material's chemical properties, X2 represents roughness (roughness refers to the roughness of the wafer surface being processed), and X3 represents the mechanical properties (a comprehensive consideration of local flatness, bow, and warpage). The chemical properties (X1) are relatively stable and can be processed using simple normalization. Roughness (X2) and mechanical properties (X3) are mapped to a uniform distribution using quantile normalization.

[0061] First, for the material chemical characteristic parameter X1, this application uses the Min-Max Normalization method based on the frequency of occurrence of wafer elements in the semiconductor industry as the contribution, linearly mapping it to the [0, 1] interval. For example, the contribution of common elements can be shown in Table 3.

[0062] Table 3 shows the contribution of common elements

[0063] That is, according to the minimum and maximum values ​​of the element contribution in the training data, normalization is performed according to the following formula:

[0064] The contribution of common elements can be recorded in Table 1. On this basis, the average value of the normalized contribution of the elements contained in the chemical formula is calculated according to the specific chemical formula of the wafer. The calculated value is the X1 value. The X1 value of common wafer materials can be found in Table 4.

[0065] Table 4 shows the X1 values ​​of common wafer materials

[0066] This process preserves the relative chemical activity differences between materials while unifying the numerical scale, helping to improve network convergence speed and training stability. For samples of extremely anomalous materials, if the X1 value exceeds the statistical range of the training set, interval truncation is also used during the inference phase to ensure that the normalized input value remains stable within the range of [0, 1].

[0067] Subsequently, for X2 (wafer surface roughness index) and X3 (mechanical characteristic parameter of weighted synthesis of local flatness, curvature, and warpage), this application introduces a preprocessing strategy based on quantile normalization. Specifically, the X2 and X3 features are fitted with a cumulative distribution function, the cumulative probability of each sample under the corresponding feature value sorting is calculated, and its cumulative probability is mapped to a uniform distribution in the interval [0,1]. This processing method can effectively suppress the skewed distribution, extreme outliers and scale inconsistency problems in the original features, making the processed X2 and X3 more uniform in numerical distribution, and further improving the adaptability of the neural network to the input features. For example, the parameters after roughness normalization can be found in Table 5, and the corresponding values ​​of the various indicators of X3 can be found in Table 6.

[0068] Table 5 shows the normalized roughness parameters

[0069] Table 6 shows the corresponding values ​​of various indicators of X3

[0070] After the normalization process is completed, the linearly normalized X1 is concatenated with the quantile-normalized X2 and X3 to form a unified and standardized new input feature matrix, which serves as the input layer feature of the neural network model.

[0071] During the model inference phase, in order to maintain consistency between the training and application environments, all new input data can be processed using the normalization parameters and quantile mapping rules fitted during the training phase, avoiding fluctuations in model performance due to differences in data distribution and ensuring the reliability and stability of the process parameters output by the recommendation system.

[0072] In this embodiment, by introducing the above-mentioned complete feature preprocessing process, the present application can systematically optimize the distribution form and scale relationship of each input feature, significantly accelerate the training convergence speed of the neural network model, improve the accuracy and generalization ability of the wafer surface treatment process parameter recommendation, and further enhance the application adaptability of the method of the present application in heterogeneous materials and complex process conditions.

[0073] In some embodiments, the single hidden layer neural network model includes: an input layer, a hidden layer, and an output layer.

[0074] Among them, such as Figure 4As shown, the number of neurons in the input layer is equal to the number of parameter types included in the surface physical and chemical characteristic parameters. The hidden layer includes multiple neurons and uses a nonlinear activation function. The number of neurons in the output layer is equal to the number of parameter types included in the process parameters, and the output layer activation function is either a sigmoid activation function or a linear activation function. The single-hidden-layer neural network model is trained using a sample set based on historical wafer processing experimental data to adjust the connection weights and bias parameters between the hidden and output layers.

[0075] In this embodiment, the specific network design of the single hidden layer neural network structure can be as follows: Input layer: Set 3 neurons, corresponding to the normalized material chemical property parameters (X1), surface roughness parameters (X2) and mechanical property parameters (X3).

[0076] Hidden layer: Set a single hidden layer with 4 neurons (Y11, Y12, Y13, Y14). The activation function uses a nonlinear function (such as ReLU or Tanh) to enhance the model's expressiveness.

[0077] Hidden layer calculation: ; in, 、 、 represents the weight parameter between the input layer and the hidden layer, Represents the bias parameter between the input layer and the hidden layer.

[0078] Output layer: Set 5 neurons, corresponding to five normalized output process parameters (Op1 to Op5). The output layer activation function is set to Sigmoid or linear activation as required to ensure that the output falls within the specified range.

[0079] Output layer calculation: ; in, represents the weight parameter between the output layer and the hidden layer, Represents the bias parameter between the input layer and the hidden layer.

[0080] The connection weights and bias parameters between the hidden and output layers are dynamically adjusted through a training process. Training utilizes a sample set based on historical wafer processing experimental data. Standard deep learning optimization algorithms, such as gradient descent (SGD) and the Adam optimizer, are used for optimization to minimize the mean squared error (MSE) between the output predictions and the target process parameters. This approach yields a pre-trained single-hidden-layer neural network model. This model can then be used to input the target physicochemical property parameters of the target wafer, yielding preliminary recommendations for each process parameter.

[0081] In some embodiments, in step S103, the target materialization characteristic parameters and the preliminary recommendation parameters are spliced ​​and input together into a Transformer encoder based on an attention mechanism, and the complex coupling relationship between the input features is modeled using a multi-head self-attention mechanism, and deep expressions are extracted to output the target recommendation parameters, including: splicing the target materialization characteristic parameters and the preliminary recommendation parameters to form a multi-dimensional joint input vector; mapping the multi-dimensional joint input vector to the high-dimensional feature space required by the Transformer encoder through a linear embedding layer; processing the embedded features of the high-dimensional feature space through the Transformer encoder to generate a high-expression hidden representation that integrates multi-source features, and mapping it to the target recommendation parameters through a regression head.

[0082] Specifically, firstly, the normalized original input feature vector X = [X1, X2, X3] is compared with the preliminary recommended parameter vector G output by the first stage single hidden layer neural network. Op =[G Op1 ′,G Op2 ′,G Op3 ′,G Op4 ′,G Op5 ′] to perform feature-level splicing to form an 8-dimensional joint input vector: Xcombined=[X1,X2,X3,G Op1 ′,G Op2 ′,G Op3 ′,G Op4 ′,G Op5 ′].

[0083] Subsequently, the present invention introduces a linear embedding layer to map the 8-dimensional joint vector to the high-dimensional feature space required by the Transformer encoder to enhance feature expression capabilities and unify the input scale: h0=We⋅Xcombined +be; Here, h0 represents the embedded feature, We∈Rd×8 is the embedding weight matrix, be∈Rd is the bias term, and d is the embedding dimension (128 dimensions) used within the Transformer module. This embedding layer can be viewed as a mapping mechanism from low-dimensional raw data to a high-dimensional context space, ensuring that the subsequent self-attention module can fully exploit the correlations between all input features.

[0084] After completing the linear embedding, h0 is passed into the Transformer encoder as a single token input, and is processed in sequence by sub-modules such as the multi-head self-attention mechanism, residual connection, layer normalization, and feedforward network. Finally, a highly expressive hidden representation that integrates multi-source features is generated, and is mapped to the optimized plasma treatment process parameter output [Op1, Op2, Op3, Op4, Op5] through the regression head.

[0085] Through the above-mentioned splicing, fusion and embedding strategies, this application achieves the effective fusion processing of original input information and rough recommendation results in the Transformer module without adding additional complex structures, thereby improving the contextual understanding ability of recommendation parameters and the accuracy of the final output.

[0086] In this embodiment, a simple, single-hidden-layer neural network is employed to rapidly map the input standardized wafer physical and chemical properties to generate preliminary recommended parameters. This stage offers advantages such as lightweight modeling, fast response, and adaptability to a wide range of material variations. A multi-layered Transformer encoder architecture is then employed to concatenate the preliminary recommended parameters with the original input features and feed them into the encoder. A multi-head self-attention mechanism models high-order dependencies between features, dynamically assigning the importance of each input dimension within a global context and extracting a more representative composite feature representation. The Transformer encoder, comprising a self-attention sublayer, a feedforward neural network sublayer, residual connections, and layer normalization, enables deep correction and optimization of the rough recommendation results, ultimately outputting the final optimized five plasma treatment process parameters (Op1 through Op5), i.e., the target recommended parameters, including vacuum level, gas type, process power, process time, and frequency.

[0087] It should be noted that this fusion structure can adopt a two-stage training approach: first, independently training the preliminary recommendation network (i.e., a single hidden layer neural network model), then introducing its output as a fixed input to the Transformer encoder for optimization training. Alternatively, an end-to-end joint training approach can be adopted, simultaneously optimizing the weight parameters of both stages for optimal overall performance. While retaining the responsiveness of traditional networks, this model further leverages the Transformer encoder's ability to model complex coupling relationships between input features, demonstrating enhanced adaptability and prediction accuracy when working with diverse wafer materials and complex process environments.

[0088] In some embodiments, in step S104, the target physicochemical characteristic parameters and the target recommended parameters are input into an error rate prediction model based on a long short-term memory neural network to obtain the predicted error rate of each process parameter of the target wafer, including: modeling the error prediction of the mature wafer sample as a time series process, which is defined as a continuous Ti sequence step error prediction task; inputting the target recommended parameters and the physicochemical characteristic parameters of the mature wafer sample into the error rate prediction model to obtain the predicted error rate of each sequence step output by the error rate prediction model, and obtaining the second error rate difference between the predicted error rate and the actual error rate of each sequence step; forming a cumulative correction trend according to the second error rate difference corresponding to each sequence step; inputting the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network to output the preliminary predicted error rate of each process parameter; and using the cumulative correction trend to correct the preliminary predicted error rate of each process parameter of the target wafer to obtain the target predicted error rate of each process parameter.

[0089] In the application, a large amount of historical data of mature wafer types can be collected. The historical data may include wafer physical and chemical characteristic parameters (X1, X2, X3), verified mature process parameters (Op1 to Op5), and target recommended parameters (G Op1_t to G Op5_t ).

[0090] After collecting a large amount of historical data on mature wafer types, we can build an error rate prediction model based on the historical data and complete the training. First, we can design the LSTM network structure, input the features at each moment t, and input the initially generated wafer process parameters (G Op1_t ~G Op5_t ) and the corresponding wafer physical and chemical characteristics (X1, X2, X3), a total of 8-dimensional features. The network architecture uses two layers of stacked LSTM units, with the number of hidden nodes in each layer set to 64 to fully capture the short-term and long-term dependency characteristics of the error evolution process. The output feature is the prediction error rate (E) of the five process parameters at each time t. pred1_t ~ E pred5_tThe internal structure of LSTM introduces a forget gate, an input gate, and an output gate. The forget gate determines which historical error information to retain or discard based on the historical memory state and the current input; the input gate determines which new error information to remember based on the current input error information; and the output gate controls the error rate prediction output at the current moment.

[0091] The error prediction of each mature wafer sample is modeled as a time series process, defined as a continuous Ti-step (T1, T2,..., Ti) error prediction task. For each step t, the initial recommended parameters and wafer physical and chemical characteristics are input, and the LSTM outputs the prediction error rate E for that step. pred_t , record the actual error rate E real_t (Calculated from recommended parameters and mature parameters, the Ti step stage is based on mature wafers. When targeting the target wafer in the Ti+1 step, the cumulative correction trend ΔEsum is used for compensation correction. After the subsequent target wafer is actually processed and verified, the ΔEsum is updated after obtaining the new real process parameters. avg ), calculate the difference ΔE between the predicted error rate and the actual error rate t = E real_t -E pred_t After each step, the internal state is updated so that the LSTM can dynamically adjust the memory content based on the historical prediction-actual difference at the next time step (t+1).

[0092] After the Ti-step historical error prediction is completed, based on the accumulated ΔE i-10 , ΔE i-9 , ..., ΔE i , forming a cumulative correction trend ΔEsum. When entering the Ti+1 step (i.e., when predicting the target wafer), the preliminary recommended parameters and physical and chemical properties of the target wafer are input, and the LSTM predicts the preliminary error rate E of the Ti+1 step pred _ Ti+1 , use the cumulative modified trend ΔEsum to calculate E pred_Ti+1 Compensation is performed to obtain the target prediction error rate, the formula is as follows: .

[0093] According to the compensated target prediction error rate E corrected_Ti+1 , the target recommended parameters of the target wafer can be corrected and the compensation recommended parameters can be calculated: .

[0094] In this embodiment, the target prediction error rate is calculated in the above manner. The error rate prediction model not only takes into account the current parameter state, but also combines historical error patterns, thereby achieving more accurate error prediction, which is conducive to obtaining more accurate compensation recommended parameters.

[0095] Based on the same inventive concept, in some embodiments, Figure 5 As shown, the present application also provides a wafer surface treatment process recommendation device 50, comprising: an acquisition module 51, a first recommendation module 52 and a second recommendation module 53; wherein, The acquisition module 51 is used to obtain the surface physicochemical characteristic parameters of the target wafer and pre-process the surface physicochemical characteristic parameters to obtain the target physicochemical characteristic parameters; The first recommendation module 52 is used to input the target physicochemical characteristic parameters into a pre-trained single hidden layer neural network model to obtain preliminary recommended parameters for the process parameters; The second recommendation module 53 is used to splice the target materialization characteristic parameters and the preliminary recommendation parameters and input them into the Transformer encoder based on the attention mechanism. It uses the multi-head self-attention mechanism to model the complex coupling relationship between the input features, extract the deep expression and output the target recommendation parameters.

[0096] In some embodiments, the wafer surface treatment process recommendation device 50 further includes a first error prediction module and a calibration module. The first error prediction module is configured to input the target physicochemical characteristic parameters and the target recommended parameters into an error rate prediction model based on a long short-term memory neural network to obtain the predicted error rates of the target wafer process parameters. The calibration module is configured to calibrate the target recommended parameters based on the predicted error rates of the target wafer process parameters to generate calibrated recommended parameters.

[0097] In some embodiments, the wafer surface treatment process recommendation method further includes: a retrieval module, a second error prediction module, a difference calculation module, and a compensation module. The retrieval module is used to select a reference wafer with the most similar physical and chemical characteristics from a wafer database based on the target physical and chemical characteristic parameters; the second error prediction module is used to obtain the target predicted error rate of each process parameter of the reference wafer based on the physical and chemical characteristic parameters of the reference wafer, a single hidden layer neural network model, a Transformer encoder, and an error rate prediction model; the difference calculation module is used to obtain a first error rate difference between the target predicted error rate and the actual error rate of each process parameter of the reference wafer; and the compensation module is used to compensate the calibration recommended parameters based on the first error rate difference to generate compensated recommended parameters.

[0098] In some embodiments, the acquisition module 51 is also used to perform linear normalization on the chemical characteristic parameters of the material to obtain the first characteristic parameter; perform quantile normalization on the roughness and mechanical characteristic parameters respectively to obtain the second characteristic parameter and the third characteristic parameter; and splice the first characteristic parameter, the second characteristic parameter and the third characteristic parameter into an input characteristic matrix to obtain the target physical and chemical characteristic parameters.

[0099] In some embodiments, the second recommendation module 53 is also used to splice the target materialized characteristic parameters and the preliminary recommended parameters to form a multi-dimensional joint input vector; map the multi-dimensional joint input vector to the high-dimensional feature space required by the Transformer encoder through a linear embedding layer; process the embedded features of the high-dimensional feature space through the Transformer encoder to generate a high-expression hidden representation that integrates multi-source features, and map it to the target recommended parameters through the regression head.

[0100] In some embodiments, the first error prediction module is also used to: model the error prediction of mature wafer samples as a time series process, defined as a continuous Ti sequence step error prediction task; input the target recommended parameters and physicochemical characteristic parameters of the mature wafer samples into the error rate prediction model, obtain the predicted error rate of each sequence step output by the error rate prediction model, and obtain the second error rate difference between the predicted error rate and the actual error rate of each sequence step; form a cumulative correction trend based on the second error rate difference corresponding to each sequence step; input the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network, and output the preliminary predicted error rate of each process parameter; use the cumulative correction trend to correct the preliminary predicted error rate of each process parameter of the target wafer to obtain the target prediction error rate of each process parameter.

[0101] It should be noted that the wafer surface treatment process recommendation device 50 provided in the embodiment of the present application and the wafer surface treatment process recommendation method provided in the embodiment of the present application are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned wafer surface treatment process recommendation method, and the repeated parts will not be repeated.

[0102] In some embodiments, the embodiments of the present application also provide an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the wafer surface treatment process recommendation method of any of the above-mentioned schemes.

[0103] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for executing different actions of the method flow according to the embodiments of the present application.

[0104] Memory, for example, can be any medium capable of containing, storing, conveying, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, components, or propagation media. Specific examples of memory include: magnetic storage devices, such as magnetic tape or hard disk drives (HDDs); optical storage devices, such as compact discs (CD-ROMs); random access memory (RAM) or flash memory; and / or wired or wireless communication links.

[0105] The present application also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the wafer surface treatment process recommendation method of any of the above-mentioned schemes. The computer-readable medium may be included in the device / apparatus / system described in the above-mentioned embodiments; or it may exist independently and not be assembled into the device / apparatus / system. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed, the method of the embodiment of the present application is implemented.

[0106] According to an embodiment of the present application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical cable, radio frequency signals, or any suitable combination thereof.

[0107] Those skilled in the art will understand that the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present application. Therefore, the scope of the present application should not be limited to the above-mentioned embodiments, but should be determined not only by the attached claims, but also by the equivalents of the attached claims. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for recommending a wafer surface treatment process, characterized in that: include: Acquiring surface physicochemical characteristic parameters of a target wafer, and preprocessing the surface physicochemical characteristic parameters to obtain target physicochemical characteristic parameters; Inputting the target physicochemical characteristic parameters into a pre-trained single hidden layer neural network model to obtain preliminary recommended parameters for each process parameter of the target wafer; The target materialization characteristic parameters and the preliminary recommendation parameters are spliced ​​and input into the Transformer encoder based on the attention mechanism. The multi-head self-attention mechanism is used to model the complex coupling relationship between the input features, and the deep expression is extracted to output the target recommendation parameters.

2. The wafer surface treatment process recommendation method according to claim 1, wherein: The recommended wafer surface treatment process also includes: Inputting the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network to obtain the predicted error rate of each process parameter of the target wafer; The target recommended parameters are calibrated according to the predicted error rates of the process parameters of the target wafer to generate calibrated recommended parameters.

3. The wafer surface treatment process recommendation method according to claim 2, wherein: The recommended wafer surface treatment process also includes: Selecting a reference wafer with the most similar physical and chemical properties from a wafer database according to the target physical and chemical property parameters; Obtaining a target prediction error rate of each process parameter of the reference wafer based on the physicochemical characteristic parameters of the reference wafer, the single hidden layer neural network model, the Transformer encoder, and the error rate prediction model; Obtaining a first error rate difference between a target predicted error rate and an actual error rate of each process parameter of the reference wafer; The calibration recommended parameters are compensated according to the first error rate difference to generate compensated recommended parameters.

4. The wafer surface treatment process recommendation method according to claim 1, wherein: The surface physicochemical characteristic parameters include material chemical characteristic parameters, roughness and mechanical characteristic parameters; the preprocessing of the surface physicochemical characteristic parameters to obtain target physicochemical characteristic parameters includes: Performing linear normalization processing on the chemical characteristic parameters of the material to obtain a first characteristic parameter; performing quantile normalization on the roughness and the mechanical characteristic parameter to obtain a second characteristic parameter and a third characteristic parameter; The first characteristic parameter, the second characteristic parameter and the third characteristic parameter are spliced ​​into an input characteristic matrix to obtain the target physicochemical characteristic parameter.

5. The wafer surface treatment process recommendation method according to claim 1, wherein: The single hidden layer neural network model includes: an input layer, a hidden layer and an output layer; The number of neurons in the input layer is the same as the number of parameter types included in the surface physical and chemical characteristic parameters; The hidden layer includes multiple neurons and uses a nonlinear function as an activation function; The number of neurons in the output layer is the same as the number of parameter types included in the process parameters, and the activation function of the output layer is a Sigmoid activation function or a linear activation function; The single hidden layer neural network model is trained using a sample set based on historical wafer processing experimental data to adjust the connection weights and bias parameters between the hidden layer and the output layer.

6. The wafer surface treatment process recommendation method according to any one of claims 1 to 5, wherein: The target materialization characteristic parameters and the preliminary recommendation parameters are concatenated and inputted into a Transformer encoder based on an attention mechanism. The multi-head self-attention mechanism is used to model the complex coupling relationship between the input features, extract the deep expression and output the target recommendation parameters, including: splicing the target physicochemical characteristic parameters and the preliminary recommended parameters to form a multi-dimensional joint input vector; Mapping the multidimensional joint input vector to the high-dimensional feature space required by the Transformer encoder through a linear embedding layer; The embedded features of the high-dimensional feature space are processed by the Transformer encoder to generate a high-expression hidden representation that integrates multi-source features, and the target parameters are recommended through regression head mapping.

7. The wafer surface treatment process recommendation method according to claim 2, wherein: Inputting the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network to obtain the predicted error rate of each process parameter of the target wafer includes: The error prediction of mature wafer samples is modeled as a time series process, which is defined as a continuous Ti sequence step error prediction task; Inputting the target recommended parameters and physicochemical characteristic parameters of the mature wafer sample into the error rate prediction model, obtaining the predicted error rate of each sequence step output by the error rate prediction model, and obtaining a second error rate difference between the predicted error rate and the actual error rate of each sequence step; A cumulative correction trend is formed according to the second error rate difference corresponding to each sequence step; Inputting the target physicochemical characteristic parameters and the target recommended parameters into the error rate prediction model based on the long short-term memory neural network, and outputting the preliminary prediction error rate of each process parameter; The cumulative correction trend is used to correct the preliminary prediction error rate of each process parameter of the target wafer to obtain a target prediction error rate of each process parameter.

8. A wafer surface treatment process recommendation device, characterized in that: include: An acquisition module is used to acquire surface physicochemical characteristic parameters of a target wafer and pre-process the surface physicochemical characteristic parameters to obtain target physicochemical characteristic parameters; A first recommendation module is used to input the target physicochemical characteristic parameters into a pre-trained single hidden layer neural network model to obtain preliminary recommended parameters for process parameters; The second recommendation module is used to splice the target materialization characteristic parameters and the preliminary recommendation parameters and input them into the Transformer encoder based on the attention mechanism. It uses the multi-head self-attention mechanism to model the complex coupling relationship between input features, extract deep expressions and output target recommendation parameters.

9. An electronic device comprising a processor and a memory; the memory stores a computer program, wherein: When the computer program is executed by the processor, the computer program implements the wafer surface treatment process recommendation method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored thereon, wherein when the computer program is executed by a processor, the wafer surface treatment process recommendation method according to any one of claims 1 to 7 is implemented.

Images