Wafer and surface nanotopography prediction method and apparatus therefor, device, and medium

Abstract

The present disclosure relates to the technical field of semiconductor manufacturing, and provides a wafer and a surface nanotopography prediction method and apparatus therefor, a device, and a medium. The method may comprise: during processing of a wafer, detecting surface topography measurement data of the wafer; and on the basis of a trained convolutional neural network (CNN) model and the surface topography measurement data, obtaining a predicted surface topography value of the wafer corresponding to a set wavelength range after subsequent processing procedures.

Description

Methods, apparatus, equipment and media for predicting nanoscale morphology of wafers and their surfaces

[0001] Cross-references to related applications

[0002] This application claims priority to Chinese Patent Application No. 202410956702.0, filed in China on July 17, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure relates to the field of semiconductor manufacturing technology, and in particular to a method, apparatus, device and medium for predicting the nanomorphology of a wafer and its surface. Background Technology

[0004] In wafer manufacturing, after a single-crystal silicon rod is prepared using the Czochralski method, it undergoes a series of processing steps including wire cutting, grinding, etching, polishing, and chemical mechanical polishing (CMP) to ultimately obtain a single-crystal silicon wafer. For single-crystal silicon wafers, their surface morphology is a key parameter for evaluating their quality. Nanotopography (NT) is an important quality parameter among the surface morphology parameters of wafers.

[0005] In related solutions, NT (non-metallic) inspection of wafers is typically performed after the final surface treatment process, CMP (chemical mechanical polishing). However, for short-wavelength-related quality parameters such as nano-morphology, the process is largely determined by the pre-CMP processing. This means that if problems occur in the pre-CMP processing, they can only be detected after CMP. Consequently, wafers with pre-CMP defects still undergo the complete wafer manufacturing process, resulting in a waste of production resources. Summary of the Invention

[0006] In view of this, the present disclosure aims to provide a method, apparatus, device and medium for predicting the nano-morphology of a wafer and its surface, which can predict the nano-morphology parameters of the wafer surface based on the product after the front-end processing step in the wafer manufacturing process, so as to monitor the product performance of the front-end processing step based on the predicted nano-morphology data, avoid defective wafers obtained from the front-end processing step from flowing into subsequent processing steps, and avoid waste of production resources.

[0007] The technical solution disclosed herein is implemented as follows:

[0008] In a first aspect, this disclosure provides a method for predicting the nano-morphology of a wafer surface, the method comprising:

[0009] During the wafer fabrication process, the surface morphology measurement data of the wafer is detected;

[0010] Based on the trained Convolutional Neural Networks (CNN) model, the predicted surface morphology value of the wafer after subsequent processing steps is predicted according to the surface morphology measurement data, corresponding to a set wavelength range.

[0011] Secondly, this disclosure provides a device for predicting the nano-morphology of a wafer surface, the device comprising: a detection section and a prediction section; wherein,

[0012] The detection section is configured to detect surface morphology measurement data of the wafer during the wafer processing.

[0013] The prediction part is configured to predict the surface morphology value of the wafer after subsequent processing steps, corresponding to a set wavelength range, based on the surface morphology measurement data and a trained convolutional neural network (CNN) model.

[0014] Thirdly, this disclosure provides a computing device comprising: a processor and a memory; the processor being configured to execute instructions stored in the memory to implement the method for predicting the nanomorphic topography of a wafer surface as described in the first aspect.

[0015] Fourthly, this disclosure provides a computer-readable storage medium storing at least one instruction that is executed by a processor to implement the method for predicting the nanomorphic topography of a wafer surface as described in the first aspect.

[0016] Fifthly, this disclosure provides a wafer whose predicted nanomorphic value is less than 5 nm in a 2 mm * 2 mm size and / or less than 10 nm in a 10 mm * 10 mm size.

[0017] This disclosure provides a method, apparatus, device, and medium for predicting the nanoscale morphology of a wafer and its surface. It utilizes a trained CNN model and surface morphology measurement data of the wafer after completing the current processing step to predict the predicted surface morphology value of the wafer after completing subsequent processing steps. This predicted surface morphology value not only allows for monitoring of each processing step to ensure the stability and consistency of the processing equipment used, but also enables monitoring of the products from each processing step to ensure that qualified products proceed to the next processing step. This improves the microscopic morphology quality of the wafer surface, avoids producing products with low future qualification rates in subsequent processing steps, thereby reducing processing costs. Furthermore, it can be used to determine if there are any problems with the completed processes, assisting in process adjustments. Attached Figure Description

[0018] Figure 1 is a flowchart of a method for predicting nanomorphic features on a wafer surface provided in this disclosure.

[0019] Figure 2 is a schematic diagram of the CNN model architecture provided in this disclosure.

[0020] Figure 3 is a schematic diagram of the transformation of the regression task into a classification task provided in this disclosure.

[0021] Figure 4 is a schematic diagram of the loss error curve provided in this disclosure.

[0022] Figure 5 is a schematic diagram of the composition of a wafer surface nanomorphology prediction device provided in this disclosure.

[0023] Figure 6 is a schematic diagram of another wafer surface nanomorphology prediction device provided in this disclosure.

[0024] Figure 7 is a schematic diagram of the structure of a computing device provided in this disclosure. Detailed Implementation

[0025] The technical solutions in this disclosure will now be clearly and completely described with reference to the accompanying drawings.

[0026] In the relevant scheme, after the wafer surface that has undergone the CMP process is divided into analysis areas according to a set size, the peak and valley values ​​in the filtered measurement data at each sampling point in each analysis area are sorted in ascending order, and the value at the corresponding position in the ascending order is selected as the nano-topography (NT) value of the wafer according to a set percentile (e.g., 99.5%).

[0027] From the wafer surface perspective, the filtered measurement data at each sampling point within each region exhibits overall fluctuations. From a wave perspective, these fluctuations can be considered as the superposition of wave signals of different wavelengths. Different wavelengths correspond to the size of the region where the waveform occurs; that is, the smaller the wavelength of the wave signal, the smaller the region representing the measurement data fluctuation; conversely, the larger the wavelength of the wave signal, the larger the region representing the measurement data fluctuation. Among the fluctuations in the measurement data, longer wavelength signals may include parameters such as bend and warp. These parameters are largely determined after the wire cutting process, and subsequent processing steps have little impact on them. For shorter wavelength signals, such as nanoscale morphology values, all front-end processing steps before the CMP process will affect them.

[0028] Furthermore, as the linewidth of semiconductor devices continues to shrink and device structures evolve towards multi-layer stacked structures, the requirements for the surface roughness of silicon wafers are becoming increasingly stringent, particularly for the roughness of even smaller regions. It is highly likely that surface topography parameters corresponding to signal data with even shorter wavelengths will emerge in the future.

[0029] Taking the NT value as an example, the NT value of the wafer obtained by the relevant scheme is the NT value of the wafer after the CMP process. This data can only represent the performance of the entire wafer processing process and cannot evaluate the performance of each processing step. Moreover, as advanced processes place increasingly stringent requirements on the surface microstructure of wafers, it is necessary to monitor each processing step to ensure the stability and consistency of the processing equipment used in each step, and to monitor the products of each processing step to ensure that qualified products enter the next processing step, thereby improving the microstructure quality of the wafer surface.

[0030] Based on this, Figure 1 illustrates a method for predicting the nanomorphic features of a wafer surface provided in this disclosure, the method comprising steps S101 to S102.

[0031] In step S101, during the wafer processing, the surface morphology measurement data of the wafer is detected.

[0032] In this disclosure, the surface morphology measurement data of the wafer is obtained after any step in the wafer fabrication process.

[0033] In some examples, after any processing step (such as wire cutting) is completed, the surface height of the sampling points on the surface of the wire-cut wafer is measured using a single-point measurement scheme to obtain the current surface topography measurement data of the wafer.

[0034] In the above example, a single-measurement scheme means that a measurement process can only measure the measurement data at one sampling point. For example, a contact measurement scheme can be used, such as using a probe to contact the surface of the wafer under test and move it horizontally on the wafer surface. As this horizontal movement occurs, the height difference of the wafer surface under test will cause the probe to generate a longitudinal displacement. This longitudinal displacement is sensed by a displacement sensor and converted into height data of the wafer surface under test, that is, historical measurement data about the historical wafer surface height. For example, non-contact measurement schemes such as capacitance measurement and laser focusing measurement can also be used.

[0035] In some examples, surface morphology measurement data of a wafer after the completion of the current processing step can be obtained in a single measurement process, such as using optical methods (e.g., Fizeau interferometry, differential interferometry, etc.). Although these measurement schemes have high resolution and can obtain uniformly sampled data, they have high requirements for the wafer surface, requiring a relatively smooth surface to ensure that the incident light is reflected rather than scattered. They are not suitable for wafer surfaces after all processing steps. For example, the surface of a wafer after wire cutting has many wire marks and high roughness, and the surface of a wafer after grinding does not show crystal planes. Therefore, in practical implementation, a single-point measurement scheme is more suitable for obtaining surface morphology measurement data.

[0036] In step S102, based on the trained convolutional neural network (CNN) model, the predicted surface morphology value of the wafer after subsequent processing steps, corresponding to a set wavelength range, is predicted according to the surface morphology measurement data.

[0037] In this disclosure, subsequent processing steps refer to processing steps that follow the wafer fabrication process after the step of obtaining the surface morphology measurement data of the wafer described in step S101. For example, if step S101 obtains the surface morphology measurement data of the wafer after the wire cutting process, the subsequent processing steps described in step S102 can be processes that follow the wire cutting process in the wafer fabrication process, such as grinding, etching, polishing, CMP, etc.

[0038] In this disclosure, the surface morphology measurement data of all sampling points can be viewed as a three-dimensional fluctuation phenomenon across the entire wafer surface. This fluctuation phenomenon can be composed of superimposed wave signals of different wavelengths. Based on this understanding, the surface morphology value corresponding to a set wavelength range, such as the nano-morphology (NT) value corresponding to a wavelength range of 22 micrometers to 20 millimeters, can be regarded as a statistical parameter for that corresponding wavelength range, which can be obtained based on filtering. In this disclosure, the NT value is used as an example of a surface morphology value corresponding to a set wavelength range for illustration. It is understood that surface morphology values ​​corresponding to other set wavelength ranges are also applicable to the technical solution of this disclosure, and will not be elaborated further here.

[0039] It is worth noting that since filtering and convolution are equivalent computations, the process of obtaining the NT value is highly similar to that of computer vision tasks in deep learning. Based on this understanding, this disclosure employs a CNN model category suitable for computer vision tasks to handle the NT value prediction task.

[0040] After determining the CNN model, the surface topography measurement data of the historical wafer after completing the current processing step (such as wire cutting) and the NT value of the historical wafer after completing the subsequent processing step (such as CMP) can be used as a dataset to train the initialized CNN model, thereby obtaining a trained CNN model. The trained CNN model can predict the predicted NT value after completing the subsequent processing step based on the surface topography measurement data after completing the current processing step.

[0041] In this disclosure, taking the predicted surface morphology value corresponding to the set wavelength range as the predicted NT value as an example, after obtaining the predicted nanomorphology (NT) value of the wafer through step S102, the currently completed processing steps can be evaluated.

[0042] In some examples, the operating status of the equipment performing the currently completed processing steps can be determined based on the statistical measures of the predicted nanostructure values ​​of the wafer, ensuring equipment stability and consistency and avoiding machine errors. For example, monitoring the predicted NT values ​​of all products processed by each machine daily can indicate that the equipment is unstable and requires maintenance if the predicted NT values ​​are discrete, have too many outliers, or have an excessively large mean.

[0043] In some examples, the comparison between the predicted nano-morphology value of the wafer and the set evaluation index can determine whether the wafer should continue to the subsequent processing steps. This allows for the screening of qualified products for subsequent processing, avoiding the production of products with low future qualification probability in subsequent processing steps, thereby reducing processing costs.

[0044] In some examples, the process parameters of the currently completed processing steps are adjusted based on the predicted nanomorphic values ​​of the wafer. For instance, the predicted nanomorphic values ​​of the products from each processing step are used to determine if there are any problems with the completed process, thus assisting in more targeted process adjustments. Furthermore, these predicted nanomorphic values ​​can also be referenced during process adjustments to provide feedback and adjust the process parameters of the processing steps.

[0045] The technical solution shown in Figure 1 uses a trained CNN model and surface morphology measurement data of the wafer after completing the current processing step to predict the predicted surface morphology value of the wafer after completing subsequent processing steps. This predicted surface morphology value not only allows for monitoring of each processing step to ensure the stability and consistency of the processing equipment used, but also for monitoring the products of each processing step to ensure that qualified products proceed to the next processing step. This improves the microscopic morphology quality of the wafer surface, avoids producing products with low future qualification rates in subsequent processing steps, thereby reducing processing costs. Furthermore, it can be used to determine if there are any problems with the completed processes, assisting in process adjustments.

[0046] For the technical solution shown in Figure 1, in some examples, the measurement data of the surface morphology of the wafer includes:

[0047] After the wafer completes the current processing step, for each sampling point on the wafer surface, the original surface morphology measurement data of the wafer surface height at each sampling point is obtained through a single-point measurement scheme;

[0048] The surface morphology measurement data is generated based on the original surface morphology measurement data.

[0049] Regarding the above example, it's important to note that due to the large number of layers in CNN models, as the network deepens, there's often a design principle where the feature size of the input image is halved while the number of data channels doubles. However, for single-pass measurement schemes on the wafer surface, there are typically not many sampling points. For example, taking a wire-cut bare wafer with a radius of 150mm, after removing the edges with an edge exclusion (EE) of 4mm, the surface height data is measured using capacitance sampling to obtain the raw measurement data of the bare wafer surface. During the sampling measurement process, uniform sampling is performed in a polar coordinate system with the wafer center as the pole, i.e., measuring one diameter direction every 45°, with a sampling interval of 4mm in each diameter direction. In this polar coordinate system, 8*37 sampling points of raw measurement data can be obtained. This amount of raw measurement data, after being input into the CNN model, will decay to a level that is no longer effective for subsequent convolutional layers after the first few layers of forward propagation.

[0050] To address the issue of the small amount of original measurement data, this disclosure performs height interpolation based on the original measurement data, and uses the interpolated height data to expand the amount of original measurement data, thereby finally obtaining the surface morphology measurement data of the wafer after completing the current processing step.

[0051] Based on this, in some examples, the step of forming the surface topography measurement data based on the original surface topography measurement data includes:

[0052] Transform all sampling points on the wafer surface from polar coordinates to Cartesian coordinates.

[0053] Based on the original surface topography measurement data of all sampling points in the rectangular coordinate system, interpolation is performed by cubic spline interpolation to obtain the interpolation points and the height interpolation data at the interpolation points.

[0054] The original surface topography measurement data at all sampling points and the height interpolation data at all interpolation points are combined to form the surface topography measurement data.

[0055] Specifically, the original surface morphology measurement data obtained by sampling in the polar coordinate system is not uniform in the rectangular coordinate system. To address this non-uniformity, this disclosure performs uniform sampling based on the location of the sampling points, and then uses cubic spline interpolation to perform interpolation, thereby reshaping the final surface morphology measurement data from the aforementioned 8*37 shape to a 448*448 shape.

[0056] For the scheme shown in Figure 1, some possible implementations may also include a training process for the CNN model, which may include:

[0057] Based on the surface morphology measurement data of the historical wafer after completing the current processing step and the measured surface morphology values ​​of the historical wafer after completing the subsequent processing step corresponding to a set wavelength range, an initialized convolutional neural network (CNN) model is trained to obtain a trained CNN model.

[0058] For the above implementation method, the detection process, implementation method and example of the surface morphology measurement data of the historical wafer after the completion of the current processing step can be referred to the aforementioned detection process of surface morphology measurement data, and will not be repeated here.

[0059] The CNN model described in the technical solution shown in Figure 1, as shown in Figure 2, includes convolutional layers, max pooling layers, global average pooling layers connected to the input layer, and fully connected layers connected to the output layer. It also includes multiple residual blocks and skip connections between the convolutional layers within each residual block, as shown by the arc arrows in Figure 2. Each residual block contains multiple convolutional layers with the same number of kernels, kernel size, and stride. In Figure 2, CNN model 2 is defined as having two residual blocks, labeled R1 and R2. Each residual block includes four convolutional layers; for example, R1 includes convolutional layers R1-1, R1-2, R1-3, and R1-4, and R2 includes convolutional layers R2-1, R2-2, R2-3, and R2-4.

[0060] It should be noted that the goals of computer vision tasks are usually image classification, object detection, instance segmentation, etc. Therefore, computer vision tasks are usually classification tasks, while NT value prediction tasks are regression tasks, which requires modifications to the CNN model.

[0061] In this disclosure, the output dimension of the last fully connected layer in the CNN model is modified accordingly and represented as FC-N, where FC is short for Fully Connected Layer (FC) and N represents the number of output dimensions.

[0062] In some examples, N can be set to the number of intervals into which the predicted surface topography value is divided, so that the CNN model outputs the interval in which the predicted surface topography value is located; accordingly, the loss function used to train the CNN model is cross-entropy loss.

[0063] For example, replacing the last fully connected layer in a CNN model with an FC-50 divides the predicted NT value into 50 equally wide intervals, each interval corresponding to a class label. The final result is the conclusion that the predicted NT value belongs to a specific interval; in other words, the regression task of NT value prediction is treated as a classification task. Based on this modification, the softmax classifier is still used to calculate the cross-entropy loss as the loss function during the training phase. As shown in Figure 3, the horizontal axis of the coordinate system represents the predicted value, which is specifically a box within the coordinate system representing the 50 intervals to which the predicted NT value can belong. The dashed line represents the ideal regression curve. It can be seen that the regression task of NT value prediction can be treated as a classification task. Furthermore, this modification enhances the robustness of the model, avoiding the risk of overfitting, which means avoiding unknown changes caused by instability in later processing techniques.

[0064] In some examples, as shown in Figure 2, the fully connected layer that is the last layer in the CNN model is replaced with an FC-1 to linearly output the predicted surface topography value; accordingly, the loss function used to train the CNN model is the mean squared error (MSE).

[0065] For the example above, an appropriate value of N can be selected during implementation according to actual needs, thereby achieving interval division with different precision or linear output.

[0066] After determining the specific model and corresponding modifications, in some examples, training an initial CNN model based on the surface topography measurement data of the historical wafer after completing the current processing step and the measured surface topography values ​​of the historical wafer after completing the subsequent processing step, corresponding to a set wavelength range, to obtain the trained CNN model includes:

[0067] The surface morphology measurement data of the historical wafers and the measured surface morphology values ​​of the historical wafers are used to form a dataset;

[0068] The dataset is divided into a training set and a validation set;

[0069] The training set is input into the initialized CNN model to update the network parameters of the CNN model, thereby obtaining a CNN model that has been initially trained.

[0070] The validation set is input into the initially trained CNN model for validation to evaluate its performance.

[0071] It should be noted that, still using the NT value as an example, the surface morphology measurement data of historical wafers after completing the current processing step is used as input data, and the NT value of historical wafers after completing the subsequent processing step is used as output data for training until the loss function is minimized. From the input-output dataset, 90% is selected as the training set, and the remaining 10% is used as the validation set. Combining the aforementioned implementation method, when the last fully connected layer in the CNN model is FC-50, the loss function is cross-entropy loss. When the output dimension of the last fully connected layer in the CNN model is set to 1, the loss function is MSE.

[0072] Based on the aforementioned technical solution, this disclosure is illustrated with a specific embodiment. In this embodiment, taking a bare wafer with a radius of 150mm after wire cutting as an example, after removing the edge of each bare wafer with an edge removal amount (EE) of 4mm, the height data of its surface is sampled and measured by capacitance method and interpolated to obtain the surface morphology measurement data of each bare wafer. During the sampling and measurement process, uniform sampling is performed using a polar coordinate system with the wafer center as the pole. That is, a diameter direction is measured every 45°, and the sampling interval in each diameter direction is 4mm. Under this polar coordinate system, the original surface topography measurement data of 8*37 sampling points can be obtained, which is to obtain a feature map size of 1*8*37. Based on this feature map size, it is reshaped into a shape of 1*448*448 through interpolation, forming a tensor of [-1,1,448,448]. The CNN model modified as described above is trained, as shown in Figure 4. During the training process, the error loss of the training set and the error loss of the validation set tend to converge with the increase of the number of training iterations (Epoch). Furthermore, for the trained models, namely the regression task model (with the output dimension of the fully connected layer set to 1) and the classification task model (with the output dimension of the fully connected layer set to the number of intervals divided by the predicted surface morphology value), see Table 1. Taking 16023 wafers as an example, both the regression task model and the classification task model achieved good prediction results under the generalization of the model.

[0073] Table 1

[0074] As shown in Table 1, taking 16023 wafers as an example, in the classification task mode, the precision and recall are 0.72 and 0.6, respectively. In the regression task mode, the aforementioned indicator values ​​are not involved, and the results are based on the R-value of the predicted NT value and the actual NT value. 2 The correlation between the predicted NT value and the actual NT value is 0.71, indicating that the model provided in this disclosure can predict the processed NT with high accuracy.

[0075] Based on the wafer surface nanomorphology prediction method provided by the aforementioned technical solution, the predicted nanomorphology value of the wafer is less than 5nm in a 2mm*2mm size and / or less than 10nm in a 10mm*10mm size.

[0076] It should be noted that the processing steps for obtaining the predicted nano-topography values ​​occur before obtaining the nano-topography values ​​of the wafer. For example, the predicted nano-topography values ​​can be obtained after the wire dicing process, while the nano-topography values ​​within the wafer can be obtained after the CMP process. Alternatively, the predicted nano-topography values ​​can be obtained after the CMP process, while the nano-topography values ​​within the wafer can be obtained when the wafer flows into the downstream semiconductor manufacturing process; this disclosure will not elaborate on these aspects.

[0077] Based on the same concept as the aforementioned technical solution, referring to Figure 5, it illustrates a wafer surface nanomorphology prediction device 50 provided in this disclosure. The device 50 includes: a detection section 501 and a prediction section 502; wherein...

[0078] The detection section 501 is configured to detect surface morphology measurement data of the wafer during the wafer processing.

[0079] The prediction part 502 is configured to predict the surface morphology value of the wafer after subsequent processing steps and the corresponding wavelength range based on the surface morphology measurement data, according to the trained convolutional neural network (CNN) model.

[0080] In some examples, the output dimension N of the last fully connected layer in the CNN model is at least 1.

[0081] In some examples, when the output dimension of the fully connected layer is 1, the predicted surface topography value is output linearly; correspondingly, the loss function used for the CNN model is the minimum mean square error (MSE).

[0082] When the output dimension of the fully connected layer is greater than 1, the output dimension of the fully connected layer represents the number of intervals into which the predicted surface morphology value is divided; correspondingly, the loss function used to train the CNN model is the cross-entropy loss.

[0083] In some examples, the detection section 501 is configured to:

[0084] After the wafer completes the current processing step, for each sampling point on the surface of the wafer, the original surface morphology measurement data of the height of the currently processed wafer surface at each sampling point is obtained through a single-point measurement scheme;

[0085] The surface morphology measurement data is generated based on the original surface morphology measurement data.

[0086] In some examples, the detection section 501 is configured to:

[0087] Transform all sampling points on the wafer surface from polar coordinates to Cartesian coordinates.

[0088] Based on the original surface topography measurement data of all sampling points in the rectangular coordinate system, interpolation is performed by cubic spline interpolation to obtain the interpolation points and the height interpolation data at the interpolation points.

[0089] The original surface topography measurement data at all sampling points and the height interpolation data at all interpolation points are combined to form the surface topography measurement data.

[0090] In some examples, referring to Figure 6, the wafer surface nano-topography prediction device 50 further includes a training section 503, configured to: train an initialized CNN model based on the surface topography measurement data of the historical wafer after completing the current processing step and the measured surface topography values ​​of the historical wafer after completing the subsequent processing step corresponding to a set wavelength range, to obtain the trained CNN model.

[0091] In some examples, the training part 503 is configured as follows:

[0092] The surface morphology measurement data of the historical wafers and the measured surface morphology values ​​of the historical wafers are used to form a dataset;

[0093] The dataset is divided into a training set and a validation set;

[0094] The training set is input into the initialized CNN model to update the network parameters of the CNN model, thereby obtaining a CNN model that has been initially trained.

[0095] The validation set is input into the initially trained CNN model for validation to evaluate its performance.

[0096] In some examples, referring to Figure 6, the wafer surface nanomorphology prediction device 50 further includes a feedback section 504 configured as follows:

[0097] The operating status of the equipment performing the currently completed processing step is determined based on the statistical values ​​of the predicted surface morphology of the wafer; or,

[0098] The decision on whether to proceed with subsequent processing steps for the wafer is determined based on a comparison between the predicted surface morphology values ​​of the wafer and the set evaluation indicators; or,

[0099] The process parameters of the currently completed processing step are adjusted based on the predicted surface morphology values ​​of the wafer.

[0100] It should be noted that for the specific implementation of the functions configured in each "part" of the above device, please refer to the implementation method and examples of the corresponding steps in the aforementioned method for predicting the nano-morphology of wafer surfaces, which will not be repeated here.

[0101] Please refer to Figure 7, which shows a structural block diagram of a computing device provided in an exemplary embodiment of this disclosure. In some examples, the computing device 70 can be at least one of devices such as a smartphone, smartwatch, desktop computer, laptop, virtual reality terminal, augmented reality terminal, wireless terminal, and laptop computer. The computing device 70 has communication functions and can access wired or wireless networks. The computing device 70 can refer to one of a plurality of terminals, and those skilled in the art will understand that the number of such terminals can be more or less. In some examples, the computing device 70 can receive data based on the accessed wired or wireless network. It is understood that the computing device 70 undertakes the calculation and processing work of the technical solution of this disclosure, and this disclosure does not limit it in this respect.

[0102] As shown in Figure 7, the computing device of this disclosure may include one or more of the following components: processor 710 and memory 720.

[0103] Optionally, the processor 710 connects various parts within the computing device using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 720, and by calling data stored in the memory 720. Optionally, the processor 710 can be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 710 can integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), Neural-network Processing Unit (NPU), and baseband chip. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required to be displayed on the touch screen; the NPU is used to implement Artificial Intelligence (AI) functions; and the baseband chip is used to handle wireless communication. It is understandable that the aforementioned baseband chip may not be integrated into the processor 710, but may be implemented using a separate chip.

[0104] The memory 720 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 720 may include a non-transitory computer-readable storage medium. The memory 720 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 720 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data created according to the use of the computing device, etc.

[0105] In addition, those skilled in the art will understand that the structure of the computing device shown in the above figures does not constitute a limitation on the computing device. The computing device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the computing device may also include a display screen, camera assembly, microphone, speaker, radio frequency circuit, input unit, sensors (such as accelerometer, angular velocity sensor, light sensor, etc.), audio circuit, WiFi module, power supply, Bluetooth module, etc., which will not be described in detail here.

[0106] This disclosure also provides a computer-readable storage medium storing at least one instruction that is executed by a processor to implement the wafer surface nanomorphology prediction method as described in the various embodiments above.

[0107] This disclosure also provides a computer program product including computer instructions stored in a computer-readable storage medium; a processor of a computing device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computing device to perform the method for predicting the nanomorphic topography of the wafer surface as described in the above embodiments.

[0108] This disclosure also provides a wafer whose predicted nanomorphic value obtained by the wafer surface nanomorphic prediction method described in the above embodiments is less than 5 nm in a 2 mm * 2 mm size, and / or less than 10 nm in a 10 mm * 10 mm size, the nanomorphic value of the wafer is less than 5 nm in a 2 mm * 2 mm size and less than 10 nm in a 10 mm * 10 mm size.

[0109] Those skilled in the art will recognize that the functions described in this disclosure in one or more of the examples above can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transfer of a computer program from one place to another. Storage media can be any available medium accessible to a general-purpose or special-purpose computer.

[0110] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for predicting wafer surface nanotopography, comprising: detecting surface profile measurement data of a wafer during processing of the wafer; predicting, based on a trained convolutional neural network (CNN) model, a predicted surface profile value corresponding to a set wavelength range of the wafer after a subsequent processing step according to the surface profile measurement data.

2. The method of claim 1, wherein, an output dimension N of a fully connected layer in the last layer of the CNN model is at least 1.

3. The method of claim 2, wherein, when the output dimension of the fully connected layer is 1, the predicted surface profile value is output by linear output; accordingly, a loss function of the CNN model is a minimum mean square error (MSE) ; when the output dimension of the fully connected layer is greater than 1, the output dimension of the fully connected layer represents a number of intervals into which the predicted surface profile value is divided; accordingly, the loss function of the CNN model is a cross-entropy loss.

4. The method of claim 1, wherein, the detecting of the surface profile measurement data of the wafer comprises: after the wafer completes a current processing step, obtaining, for each sampling point on the wafer surface, original surface profile measurement data of the sampling point regarding a height of the wafer surface by a single-point measurement scheme; forming the surface profile measurement data according to the original surface profile measurement data.

5. The method of claim 4, wherein, the forming of the surface profile measurement data according to the original surface profile measurement data comprises: converting all sampling points on the wafer surface from a polar coordinate system to a rectangular coordinate system; based on the original surface profile measurement data of all sampling points in the rectangular coordinate system, interpolating by a cubic spline interpolation method to obtain interpolation points and height interpolation data of the interpolation points; forming the surface profile measurement data from the original surface profile measurement data of all sampling points and the height interpolation data of all interpolation points. 6.The method of claim 1, further comprising: training an initialized CNN model according to surface profile measurement data of historical wafers after a current processing step and measured surface profile values of the historical wafers corresponding to a set wavelength range after a subsequent processing step to obtain the trained CNN model.

7. The method of claim 6, wherein, the training of the initialized CNN model according to the surface profile measurement data of the historical wafers after the current processing step and the measured surface profile values of the historical wafers corresponding to the set wavelength range after the subsequent processing step to obtain the trained CNN model comprises: forming a data set from the surface profile measurement data of the historical wafers and the measured surface profile values of the historical wafers; dividing the data set into a training set and a validation set; inputting the training set into the initialized CNN model to update network parameters of the CNN model to obtain a CNN model that is preliminarily trained; inputting the validation set into the CNN model that is preliminarily trained to verify and evaluate performance.

8. An apparatus for predicting a nanotopography of a wafer surface, comprising: a detection part and a prediction part; wherein the detection part is configured to detect surface profile measurement data of a wafer during processing of the wafer. The prediction part is configured to predict, according to the surface topography measurement data, a predicted surface topography value corresponding to a set wavelength range of the wafer after a subsequent processing procedure based on a trained convolutional neural network (CNN) model.

9. A computing device comprising: a processor and a memory; The processor is configured to execute instructions stored in the memory to implement the wafer surface nanotopography prediction method according to any one of claims 1 to 7. 10.A computer readable storage medium storing at least one instruction for being executed by a processor to implement the wafer surface nanotopography prediction method according to any one of claims 1 to 7. 11.A wafer, wherein a nanotopography value of the wafer is less than 5nm in a 2mm*2mm specification and / or less than 10nm in a 10mm*10mm specification, and a predicted nanotopography value of the wafer is less than 5nm in a 2mm*2mm specification and / or less than 10nm in a 10mm*10mm specification.

Images