Apparatus, method and system for transforming measurement data

Abstract

The invention relates to an apparatus, method and system for transforming measurement data. An example apparatus for transforming measurement data includes: a communicator configured to receive first measurement data including a step height value on a semiconductor chip on which a chemical mechanical polishing (CMP) process is performed, and receive layout data including a layout included in the semiconductor chip; and a processor configured to transform the first measurement data into second measurement data based on the layout data, the second measurement data including a step height value of the semiconductor chip on which the metal is deposited.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the priority of Korean Patent Application No. 10-2023-0094650 filed on July 20, 2023, and No. 10-2023-0120489 filed on September 11, 2023, in the Korean Intellectual Property Office, the disclosures of which are incorporated herein by reference in their entirety. Technical Field

[0003] The present application relates to an apparatus, method and system for transforming measurement data. Background Art

[0004] A metrology system for verifying and monitoring semiconductor processes and controlling verification operations and monitoring operations is being developed. The metrology system can use various metrology methods to measure the step height of the pattern included in the semiconductor chip. The step height data can be used to find the weaknesses (e.g., weak points) of the semiconductor chip. For example, the metrology method of the metrology system may include a scanning electron microscope (SEM), a transmission electron microscope (TEM), an atomic force microscope (AFM), a white light interferometer (WLI), a patterned wafer geometry (PWG), etc. Because each metrology method has its own measurement method and unique characteristics, each metrology method has advantages and disadvantages in terms of accuracy, cost, metrology speed, range, etc. Compared with TEM, etc., optical metrology methods such as PWG measure a relatively wide range, so the time required to measure the entire semiconductor chip using the PWG metrology method is relatively less than other metrology methods. However, because this type of optical metrology method is susceptible to optical interference occurring in the lower layer (or pattern) of the semiconductor chip, there may be a situation where it is difficult to detect hidden dangers in the semiconductor chip.

[0005] In order to remove optical interference while obtaining step height data in the entire large-area semiconductor chip, an additional process of depositing metal on the semiconductor chip that has completed the chemical mechanical polishing (CMP) process may be required. The requirement to find the optimal thickness (or amount) of metal deposition results in wafers no longer being available, relatively large additional process time consumption, and destructive analysis, and thus problems such as increased manufacturing costs occur. Therefore, a technology for obtaining optical metrology data of step heights of an entire semiconductor chip without actually depositing metal is being studied. Summary of the invention

[0006] The present disclosure relates to apparatus, methods and systems for predicting data using artificial intelligence models to remove optical interference effects on raw data without destructive analysis.

[0007] In some implementations, a device for transforming measurement data includes: a communicator configured to receive first measurement data, the first measurement data including step height values ​​optically measured in units of target areas on a first semiconductor chip when a chemical mechanical polishing (CMP) process is performed on the first semiconductor chip, and receive layout data, the layout data including a layout included in the first semiconductor chip; and a processor configured to generate density data including an image based on the layout data, to which a layout density indicating a ratio of an area occupied by a layout in a unit resolution area is applied, input the density data and the first measurement data into an artificial intelligence model, and predict second measurement data as an output of the artificial intelligence model, the second measurement data including a step height per target area of ​​a second semiconductor chip, the second semiconductor chip including the first semiconductor chip and a metal deposited on the first semiconductor chip.

[0008] In some implementations, a method for transforming measurement data includes: receiving first measurement data, the first measurement data including step height values ​​optically measured in units of target areas on a first semiconductor chip when a chemical mechanical polishing (CMP) process is performed on the first semiconductor chip, and receiving layout data including a layout included in the first semiconductor chip; generating density data including an image, to which a layout density indicating a ratio of an area occupied by a layout in a unit resolution area is applied; and using an artificial intelligence model, predicting second measurement data from the density data and the first measurement data, the second measurement data including step height values ​​per target area of ​​a second semiconductor chip, the second semiconductor chip including the first semiconductor chip and metal deposited on the first semiconductor chip.

[0009] In some implementations, a system for transforming measurement data includes: an optical measurement device configured to generate first measurement data including step height values ​​of a first semiconductor chip by optically measuring the first semiconductor chip on which a chemical mechanical polishing (CMP) process is performed in units of target areas; a database storing layout data including a layout included in the first semiconductor chip; and a measurement data transformation device configured to transform the first measurement data into second measurement data based on the first measurement data, layout data, and an artificial intelligence model, the second measurement data including step height values ​​of a second semiconductor chip including the first semiconductor chip and metal deposited on the first semiconductor chip.

[0010] In some implementations, a method for transforming measurement data includes: performing a chemical mechanical polishing (CMP) process on a first semiconductor chip; generating first measurement data including a step height value of the first semiconductor chip by optically measuring the first semiconductor chip in units of a target area; transforming the first measurement data into second measurement data based on layout data, the first measurement data, and an artificial intelligence model stored in a database, the second measurement data including the step height value of a second semiconductor chip, the second semiconductor chip including the first semiconductor chip and a metal deposited on the first semiconductor chip; and detecting a weak point in the first semiconductor based on the second measurement data. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The implementation will be more clearly understood from the following detailed description taken in conjunction with the accompanying drawings.

[0012] Figure 1 is a block diagram of an example of a system.

[0013] Figure 2 is based on Figure 1 Example graphs of the inspection range and weak point detection capabilities of the metrology method for the metrology device in FIG.

[0014] Figure 3 An example graph showing first measurement data and second measurement data.

[0015] Figure 4 is an example diagram of the step heights of the first measurement data and the second measurement data.

[0016] Figure 5 An example graph showing step information and measurement accuracy of a semiconductor chip according to a metal deposition amount.

[0017] Figure 6 is a block diagram of an example of a measurement data transformation device.

[0018] Figure 7 is a flowchart illustrating an example of a measurement data transformation method.

[0019] Figure 8 A diagram illustrating an example implementation for removing noise from measurement data.

[0020] Fig. 9 Example diagram showing layout data and density data.

[0021] Fig.10 An example graph illustrating layout density of each of a first layout and a second layout.

[0022] Fig.11 is an example diagram of the input and output of an AI model.

[0023] Fig.12 is a diagram illustrating an example of an artificial intelligence model.

[0024] Fig.13 is an example graph of step heights for a semiconductor chip with metal deposited thereon.

[0025] Fig.14 is an example graph of step heights of semiconductor chips deposited with metal and the predicted values ​​of the artificial intelligence model.

[0026] Fig.15 A diagram showing an example of a learning process for an artificial intelligence model.

[0027] Fig.16 is based on Fig.15 Example plots of the root mean square error (RMSE) and r-squared (R2) score of the learning process of an AI model.

[0028] Fig.17 Shown for illustration Fig.15 An example diagram of the implementation of removing noise from measurement data during the learning process of an artificial intelligence model.

[0029] Fig.18 and Fig.19 is an example graph used to illustrate the test results of the artificial intelligence model.

[0030] Fig. 20 Example diagram showing weak points. DETAILED DESCRIPTION

[0031] Hereinafter, implementations of the present disclosure will be described in detail with reference to the accompanying drawings.

[0032] Figure 1 is a block diagram of an example of system 1.

[0033] Reference Figure 1 The system 1 may include a chemical mechanical polishing (CMP) process device 10 , a metrology device 20 , a database 30 , a measurement data conversion device 40 , and a weak point detection device 50 .

[0034] The CMP process device 10 may perform a CMP process on a semiconductor chip. The CMP process may include a process for flattening a wafer by polishing a film surface of an uneven or curved wafer using chemical elements and mechanical elements. To this end, the CMP process device 10 may include a polishing head, a CMP slurry polishing pad, a rotating plate, a rotating head, an adjuster, a workbench, and the like.

[0035] The metrology device 20 may measure a wafer on which a CMP process has been completed. The wafer may include a plurality of semiconductor chips. In some implementations, the metrology device 20 may optically measure each semiconductor chip on which a CMP process has been performed. For example, the metrology device 20 may generate first measurement data by optically measuring a first semiconductor chip on which a CMP process has been performed in units of target areas. The time required for optically measuring the first semiconductor chip (e.g., measurement turnaround time (TAT)) may be less than about 1 minute, but is not limited thereto. In this case, the target area may be an optical measurement unit of the metrology device 20. For example, the shape of the target area may be a rectangle, and the size of the target area may be about 40 μm*about 40 μm, but is not limited thereto. In this case, the length of one side of the target area (e.g., about 40*μm) may be referred to as resolution. Because the size of the first semiconductor chip is larger than the size of the target area, the first semiconductor chip may include a plurality of target areas. The first measurement data may include scalar data and image data. The scalar data may include a step height of a target area on the first semiconductor chip. For example, the scalar data may include a step height of each target area of ​​the entire first semiconductor chip, that is, a step height of the target area. The image data may include an image representing step heights of target regions in the first semiconductor chip in different colors. The number of step heights included in one semiconductor chip may be about 500,000, but is not limited thereto. The metrology device 20 may provide the first measurement data to the measurement data conversion device 40.

[0036] In some implementations, the database 30 may store layout data including layouts included in respective semiconductor chips included in a wafer. The database 30 may provide the layout data to the measurement data transformation device 40.

[0037] The measurement data transformation device 40 may transform the first measurement data into the second measurement data based on the first measurement data, the layout data and the artificial intelligence model. The second measurement data may include data predicted by the artificial intelligence model. The second measurement data may include scalar data and image data. The scalar data included in the second measurement data may include a step height optically measured in units of a target area on the second semiconductor chip, and for example, the scalar data included in the second measurement data may include the step height of the second semiconductor chip. The image data included in the second measurement data may include an image in which the step heights of the target area in the second semiconductor chip are represented in different colors respectively. The second semiconductor chip may include a first semiconductor chip on which a metal is deposited on the first semiconductor chip.

[0038] In some implementations, the measurement data transformation device 40 may generate a layout density indicating a ratio of areas occupied by layouts in a unit resolution region, and predict the second measurement data from the density data and the first measurement data using an artificial intelligence model.

[0039] In some implementations, by removing noise from raw data of the first measurement data using a denoising algorithm, the measurement data transformation device 40 may input the first measurement data and density data from which the noise has been removed into the artificial intelligence model.

[0040] In some implementations, the measurement data transformation device 40 may input first measurement data including a step height optically measured in a target area and a plurality of density data including an image of each of the different layouts into the artificial intelligence model.

[0041] In some implementations, the measurement data transformation device 40 may train the artificial intelligence model using a learning data set including the first measurement data, the density data, and the second measurement data through supervised learning.

[0042] The weak point detection device 50 may detect a weak point in the first semiconductor chip based on the second measurement data. The weak point may be referred to as a weak location, a hot spot, or the like.

[0043] The method performed by the system 1 may include: performing a CMP process on a first semiconductor chip by a CMP process device 10; generating first measurement data including a step height of the first semiconductor chip by optically measuring the first semiconductor chip in units of a target area by a metrology device 20; transforming the first measurement data into second measurement data by a measurement data transformation device 40 based on layout data, the first measurement data, and an artificial intelligence model stored in a database 30; and detecting a weak point from the first semiconductor chip based on the second measurement data by a weak point detection device 50.

[0044] In some implementations, by using an artificial intelligence model to generate data with optical interference effects removed without the need for destructive analysis of semiconductor chips, the cost and time required for destructive analysis can be reduced.

[0045] In addition, in some implementations, by using an artificial intelligence model that has learned information about semiconductor chip layers that affect optical interference, information about the step height of the semiconductor chip can be obtained without actually depositing metal on the semiconductor chip, process operations can be simplified, and the time and cost of the model can be reduced.

[0046] Additionally, in some implementations, by obtaining information about the step height of the semiconductor chip without actually depositing metal on the semiconductor chip, weak points of the semiconductor chip can be quickly detected.

[0047] Figure 2 is based on Figure 1 An example graph of the inspection range and weak point detection capability of the metrology method of the metrology device 20 in FIG.

[0048] Reference Figure 1 and Figure 2 , the metrology device 20 may use any of a variety of metrology methods to measure the wafer. For example, the various metrology methods may include transmission electron microscopy (TEM), vertical scanning electron microscopy (VSEM), atomic force microscopy (AFM), white light interferometry (WLI), patterned wafer geometry (PWG), density analysis, digital holographic microscopy (DHM), scanning electron microscopy (SEM), etc. However, the implementation is not limited to this. These metrology methods may have the ability to perform weak point detection for the inspection range. For example, in Figure 2 In the case of TEM among the metrology methods shown, the inspection range can be minimized and the weak point detection capability can be maximized. Figure 2 In the case of PWG among the metrology methods shown, the inspection range may be the largest and the weak point detection capability may be the smallest. Since the PWG metrology method can measure not only the wafer displacement (called warpage) but also the entire displacement by measuring two surfaces of the wafer, the PWG metrology method can measure parameters such as displacement in a plan view, local curvature, and surface changes in a short time (e.g., less than about 1 minute).

[0049] The PWG metrology method may include a method of optically measuring the geometry of a patterned wafer. Compared with other metrology methods (e.g., TEM methods), the PWG metrology method may measure the entire semiconductor chip. In some implementations, when the metrology method of the metrology device 20 is the PWG method, the metrology device 20 may generate measurement data (e.g., first measurement data) including all measurement values ​​of each semiconductor chip.

[0050] Figure 3 An example diagram showing first measurement data OMD1 and second measurement data OMD2 . Figure 4 is an example diagram of step heights of the first measurement data OMD1 and the second measurement data OMD2.

[0051] Figure 5 An example graph showing step information and measurement accuracy of a semiconductor chip according to a metal deposition amount.

[0052] Reference Figure 1 , Figure 3 and Figure 4 , the first measurement data OMD1 may be generated according to the PWG metrology method performed by the metrology device 20. The first measurement data OMD1 may include an image indicating the layout of the entire semiconductor chip, in which the step height is represented by color. In this case, each step may appear in one semiconductor chip on which the CMP process has been performed. Figure 4 , for example, the step height included in the first measurement data OMD1 (eg, reference Figure 4 The "PWG CMP step height of OMD1" in the first measurement data may have high variability. According to the PWG metrology method, optical interference of a silicon (Si) film included in a semiconductor chip due to an underlying film may occur. In other words, because optical interference occurs in the first measurement data OMD1 generated by the PWG metrology method, when a user performs quantitative analysis on the step height of a wafer using the first measurement data OMD1, it may be difficult to perform the quantitative analysis due to the optical interference.

[0053] The method of removing optical interference may be performed by depositing metal on a wafer and optically measuring a semiconductor chip included in the wafer on which the metal is deposited. In other words, the second measurement data OMD2 obtained by optically measuring the semiconductor chip on which the metal is deposited may be available. Figure 4 For example, assuming that the metal includes titanium nitride (TiN) and the metal deposition amount is about 60 nm. The step height included in the second measurement data OMD2 (e.g., reference to "PWG CMP step height of OMD2") may have less variability. However, in order to consistently measure wafers with metal deposited thereon, there may be a problem of requiring destructive analysis.

[0054] On the other hand, refer to Figure 5 , a first case Case 1, a second case Case 2, a third case Case 3, and a fourth case Case 4 may illustrate example cross sections of semiconductor chips having different metal deposition amounts for each case. Figure 5The number of cases is shown as four, but this is an example. The first case Case1 may show a portion of a cross section of a semiconductor chip on which no metal is deposited. In the present disclosure, the semiconductor chip on which no metal is deposited may be referred to as the first semiconductor chip SEMICON1. ​​The second case Case2, the third case Case3, and the fourth case Case4 may show a portion of a cross section of a semiconductor chip on which metal is deposited. In the present disclosure, the semiconductor chip on which metal is deposited may be referred to as the second semiconductor chip SEMICON2_1, SEMICON2_2, or SEMICON2_3. When metal is not deposited on the semiconductor chip as in the first case Case 1, the step information of the semiconductor chip may be relatively accurate, but the measurement accuracy of the PWG metrology method may be relatively low. On the other hand, when metal is over-deposited on the semiconductor chip as in the fourth case Case4, the measurement accuracy of the PWG metrology method may be relatively high, but although steps actually occur in the semiconductor chip, the step information about the semiconductor chip may be measured relatively low and inaccurate. Therefore, it may be necessary to analyze the optimal value, just like in the third case Case3. In other words, it may be necessary to use an artificial intelligence model to predict the second measurement data OMD2 method instead of actually depositing metal to avoid destructive analysis. In addition, it may be necessary for the second measurement data OMD2 predicted by the artificial intelligence model to have a step measured in a semiconductor chip on which metal is deposited with an optimal metal deposition amount.

[0055] Figure 6 is a block diagram of an example of the measurement data transformation device 40 .

[0056] Reference Figure 6 , the measurement data transformation device 40 in the present disclosure may include all kinds of devices capable of providing the result of executing the calculation process to the user. For example, the measurement data transformation device 40 may include a computer, a server device and / or a mobile terminal. In this case, the computer may include, for example, a notebook, a desktop computer, etc., which is installed with a web browser. The server device may include a server that processes information by communicating with an external device, and may include an application server, a computing server, a web server, etc. The mobile terminal may include all types of handheld wireless communication devices, such as a personal communication system (PCS), a personal digital cellular (PDC), and a smart phone.

[0057] The measurement data transformation device 40 may include a device for transforming measurement data. The measurement data transformation device 40 may include a processor 100 , a memory 200 , and a communicator 300 .

[0058] The processor 100 may generate density data based on the layout data. The layout data may include data stored in Figure 1 The processor 100 may obtain data from the database 30 in Figure 3The density data may include an image to which the density of the layout indicating the ratio of the area occupied by the layout in the unit resolution area is applied. The density data is referred to below. Fig. 9 and Fig.10 Description. The processor 100 may input the density data and the first measurement data into the artificial intelligence model, and predict the second measurement data as the output of the artificial intelligence model. The second measurement data may include at least one step height optically measured in units of a target area on the second semiconductor chip. The second semiconductor chip may include a first semiconductor chip on which a metal is deposited.

[0059] The functions related to artificial intelligence according to the present disclosure may be operated using the processor 100 and the memory 200. One or more processors 100 may include a general-purpose processor such as a central processing unit (CPU), an application processor (AP), and a digital signal processor (DSP), a graphics-specific processor such as a graphics processing unit (GPU) and a visual processing unit (VPU), or an artificial intelligence-specific processor such as a neural processing unit (NPU). One or more processors 100 may process input data based on predefined operating rules or artificial intelligence models stored in the memory 200. When one or more processors 100 include an artificial intelligence-specific processor, the artificial intelligence-specific processor may be designed with a hardware structure dedicated to processing a specific artificial intelligence model.

[0060] The predefined operating rules or artificial intelligence models can be obtained through learning. In this case, obtaining through learning may mean that the predefined operating rules or artificial intelligence models are obtained through learning by the basic artificial intelligence model using various learning data by applying a learning algorithm to perform the desired characteristics (or purposes). The above-mentioned learning can be performed in the device itself that performs artificial intelligence according to the present disclosure, or it can be performed using a separate server and / or system. Examples of learning algorithms may include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited thereto.

[0061] The artificial intelligence model may include one artificial intelligence model or multiple artificial intelligence models. The artificial intelligence model may be included in a neural network (or artificial neural network) and may include a statistical learning algorithm that imitates a biological neural network in machine learning and recognition science. A neural network may be referred to as an entire model in which artificial neurons (nodes) that form a network by combining synapses change the binding strength of the synapses through learning to obtain problem-solving capabilities. The neurons of a neural network may include a combination of weights or biases. A neural network may include one or more layers, which include one or more neurons or nodes. As an example, a neural network may include an input layer, a hidden layer, and an output layer. A neural network can infer (or predict) the results to be predicted from any input by changing the weights of neurons through learning.

[0062] The processor 100 may generate a neural network, train (or learn) a neural network, perform calculations based on received input data, generate information signals based on the operation results, or retrain the neural network. The model of the neural network may include various types of models, such as convolutional (C) neural networks (NN) (CNN) such as GoogleNet, AlexNet, and VGG networks, regional (R) CNN (R-CNN), R proposal network (RPN), recursive (R) NN (RNN), stacked (S) based deep NN (DNN) (SDNN), state space (SS) DNN (S-SDNN), deconvolution network, deep belief network (DBN), restricted Boltzmann machine (RBM), full convolution network, long short-term memory (LSTM) network and classification network, but is not limited thereto. The neural network may also include DNN.

[0063] NN may include CNN, RNN, perceptron, multi-layer perceptron, feedforward (FF), radial basis function network (RBFN), deep feedforward (DFF), long short-term memory (LSTM), gated recurrent unit (GRU), autoencoder (AE), variational autoencoder (VAE), denoising autoencoder (DAE), sparse autoencoder (SAE), Markov chain (MC), Hopfield network (HN), Boltzmann machine (BM), restricted Boltzmann machine (RBM), deep belief network (DBN), deep convolutional network (DCN), deconvolution network (DN), deep convolutional inverse graphic neural network (DCIGN), generative adversarial network (GAN), liquid state machine (LSM), extreme learning machine (ELM), echo state network (ESN), deep residual network (DRN), differentiable neural computer (DNC), neural steering machine (NTM), capsule network (CN), Kohonen network (KN) and attention network (AN), but it should be understood by ordinary technicians in this field that NN is not limited to this and may include any neural network.

[0064] The memory 200 may store data supporting various functions of the measurement data transformation device 40 and programs for the operation of the processor 100, store input / output data (e.g., music files, still images, videos, etc.), and store a plurality of application programs (or applications) operated by the measurement data transformation device 40, data and commands for the operation of the measurement data transformation device 40. At least some of these applications may be downloaded from an external server via wireless communication. The plurality of applications may include various algorithms, such as artificial intelligence models, denoising algorithms, etc.

[0065] The memory 200 may include at least one type of storage medium among a flash memory type, a hard disk type, a solid state disk (SSD) type, a silicon disk drive (SDD) type, a multimedia card micro, a card type memory (for example, a scan disk (SD), an extreme digital (XD) memory, etc.), a random access memory (RAM), a static RAM (SRAM), a read-only memory (ROM), an electrically erasable programmable ROM (EEPROM), a programmable ROM (PROM), a magnetic memory, a magnetic disk, and an optical disk.

[0066] The communicator 300 may receive the first measurement data and the layout data including the layout included in the first semiconductor chip, and provide the first measurement data and the layout data to the processor 100. In some implementations, the communicator 300 may function as a communication interface to perform a path between various types of external devices connected to the measurement data transforming device 40.

[0067] Although not shown, the measurement data transformation device 40 may further include a display unit for displaying the processing result and the processed information of the processor 100. The display unit may display screen information about executing the application driven by the measurement data transformation device 40 or a user interface (UI) / graphic user interface (GUI) according to the screen information being executed.

[0068] Figure 7 is a flowchart illustrating an example of a measurement data transformation method.

[0069] Reference Figure 6 and Figure 7 , the measurement data transformation method may include a method performed by the measurement data transformation device 40.

[0070] In operation S10, the measurement data transformation device 40 may receive the first measurement data and the layout data. In some implementations, the measurement data transformation device 40 may receive raw data of the first measurement data and the layout data including the layout included in the first semiconductor chip.

[0071] In operation S20, the measurement data transformation device 40 may remove noise from the first measurement data. In some implementations, the measurement data transformation device 40 may remove noise from the raw data of the first measurement data. In some implementations, the measurement data transformation device 40 may generate the first measurement data from which noise has been removed by removing noise from the raw data of the first measurement data using a denoising algorithm. The first measurement data from which noise has been removed may be learned by an artificial intelligence model. In addition, the first measurement data from which noise has been removed may be input to a learned artificial intelligence model. The implementation of noise removal is described below with reference to Figure 8 to describe.

[0072] In operation S30, the measurement data transformation device 40 may generate density data based on the layout data. The density data may include an image of the layout to which the layout density is applied. The layout density may represent a ratio of an area occupied by the layout in a unit resolution area. In some implementations, the measurement data transformation device 40 may generate a plurality of density data including images of different layouts.

[0073] In operation S40, the measurement data transformation device 40 may use the learning data set to train the artificial intelligence model. In some implementations, the measurement data transformation device 40 may train the artificial intelligence model to learn the learning data set including the first measurement data, the density data, and the second measurement data. The second measurement data may include the output of the artificial intelligence model. In some implementations, operation S40 may include: using a denoising algorithm to remove noise from the first original data of the first measurement data and noise from the second original data of the second measurement data, setting the first measurement data and the second measurement data from which the noise is removed and the density data as the learning data set, and training the artificial intelligence model to learn the learning data set through supervised learning. The implementation method of training the artificial intelligence model is described below with reference to Fig.13 , Fig.14 , Fig.15 and Fig.16 to describe.

[0074] In operation S50, the measurement data transformation device 40 may use an artificial intelligence model to predict the second measurement data from the density data and the first measurement data. The implementation of operation S50 is described below with reference to Fig.11 and Fig.12 to describe.

[0075] In some implementations, by using an artificial intelligence model to generate data with optical interference effects removed without the need for destructive analysis of semiconductor chips, the cost and time required for destructive analysis can be reduced.

[0076] In addition, in some implementations, by using an artificial intelligence model that has learned information about semiconductor chip layers that affect optical interference, information about the step height of the semiconductor chip can be obtained without actually depositing metal on the semiconductor chip, process operations can be simplified, and the time and cost of the model can be reduced.

[0077] Additionally, in some implementations, by obtaining information about the step height of the semiconductor chip without actually depositing metal on the semiconductor chip, weak points of the semiconductor chip can be quickly detected.

[0078] Figure 8 A diagram illustrating an example implementation for removing noise from measurement data.

[0079] Reference Figure 6 and Figure 8 , the processor 100 may receive the raw data 800 of the first measurement data via the communicator 300. Figure 8 As shown, the raw data 800 may include an entire image of the first semiconductor chip.

[0080] The processor 100 may use a denoising algorithm to remove noise from the raw data 800. In some implementations, the denoising algorithm may include a fast Fourier transform (FFT). However, the implementation is not limited thereto. The processor 100 according to the implementation may use FFT to change the domain of the raw data 800 to the frequency domain of FFT. The FFT spectrum 801 of the raw data 800 may be as follows: Figure 8 The processor 100 according to the implementation can remove noise from the original data 800 by reflecting at least one mask in the FFT spectrum 801. The FFT spectrum 802 included in the at least one mask can be as shown in FIG. Figure 8 Generated as shown. Figure 8 The FFT spectrum 802 in may include a first mask MSK1 and a second mask MSK2. In this case, the mask may correspond to zero padding with a value of about 0. The processor 100 according to the implementation may restore the frequency domain of the FFT to the domain. Figure 8 For example, before applying FFT to the domain, portion 803 of the original data 800 may include a noise pattern NP having a diagonal shape. In portion 804 of the original data 800 after applying FFT, the noise pattern NP may be removed. The R-squared R2 score of the steps of the measurement data from which the noise is removed may be increased compared to the steps of the original data.

[0081] The processor 100 may input the first measurement data from which noise has been removed into the artificial intelligence model.

[0082] According to the above implementation, by removing noise from the original data, the predictability of the artificial intelligence model can be improved more accurately.

[0083] Fig. 9 Examples of layout data LAYOUT DATA and density data are shown. Fig.10 An example graph illustrating layout density of each of a first layout and a second layout.

[0084] Reference Fig. 9 and Fig.10, the layout data may include an image of a specific layout included in a portion of a semiconductor chip FULL CHIP. For example, any one piece of layout data may include an image of a first layout included in a portion of a semiconductor chip FULL CHIP. The first layout may include at least one layer. Another piece of layout data may include an image of a second layout included in another portion of the semiconductor chip FULL CHIP. The second layout may include at least one other layer of the layers included in the first layout. The layout density may represent a ratio of an area occupied by a layout in a unit resolution area. Fig.10 The layout density of each of the first layout and the second layout is shown in FIG. Fig.10 In, the value of the layout density may be referred to as the layer density. For example, the size of the unit resolution area may be about 40μm*about 40μm, about 8μm*about 8μm, about 4μm*about 4μm, about 2μm*about 2μm, or about 1μm*about 1μm. However, the implementation is not limited thereto. The unit resolution area may be referred to as the area of ​​the unit domain. The unit resolution or unit domain may be about 40μm, about 8μm, about 4μm, about 2μm, or about 1μm. The size of the image or the size of the layout of the layout data is assumed to be about 40μm*about 40μm. When the unit resolution is about 40μm resolution, because the unit resolution area is about 40μm*about 40μm, the density data may include an image of a 1*1 unit resolution area. When the unit resolution is about 8μm resolution, because the unit resolution area is about 8μm*about 8μm, the density data may include an image of a layout with a 5*5 unit resolution area. When the unit resolution is about 4 μm resolution, the density data may include an image of a layout having a 10*10 unit resolution area. When the unit resolution is about 2 μm resolution, the density data may include an image of a layout having a 20*20 unit resolution area. When the unit resolution is about 2 μm resolution, the density data may include an image of a layout having a 40*40 unit resolution area.

[0085] As the unit resolution decreases, the time required for the artificial intelligence model to learn may increase. As the resolution decreases, the prediction accuracy of the artificial intelligence model may generally increase, but when a specific unit resolution increases above a specific value, the prediction accuracy of the artificial intelligence model may decrease. Therefore, there may be an optimal value among the unit resolutions. The optimal value among the unit resolutions may be set during the learning process of the artificial intelligence model or when the learning of the artificial intelligence model is completed. For example, the optimal unit resolution may be about 2 μm resolution, but is not limited thereto.

[0086] In some implementations, as a data preparation process prior to learning and applying an artificial intelligence model, normalization, outlier data augmentation, input layout domain selection (e.g., the domain of density data), unit size selection (e.g., unit resolution), etc. may be performed.

[0087] In some implementations, the time used to train an artificial intelligence model can be optimized and the performance of the artificial intelligence model can be improved.

[0088] Fig.11 It is an example diagram of the input and output of the artificial intelligence model AIM.

[0089] Reference Fig.11 , the input of the artificial intelligence model AIM may include first measurement data and at least one density data. The first measurement data may include a measurement result of a first semiconductor chip that has completed a CMP process measured by the metrology device 20 using the PWG metrology method, that is, a step height CMP step height value of a target area. The number of scalar data representing the step height CMP step height value may be approximately 500,000, and the first measurement data may include approximately 500,000 scalar data. The size of the target area of ​​the first measurement data may be referred to as the input unit domain size. For example, the size of the target area may be approximately 40μm*approximately 40μm. Specific density data may include an image of a specific layout to which layout density is applied among a plurality of layouts included in the first semiconductor chip. As Fig.11Each of the first density data DENSITY DATA 1 to the sixth density data DENSITY DATA 6 shown in the example of may include the layout density of the lower layer of each module of the first semiconductor layer on the surface processed by the CMP process. For example, the first density data DENSITY DATA 1 may include an image of the first layout of the first semiconductor chip to which the layout density is applied (or an image of the layer included in the first layout). The second density data DENSITY DATA2 may include an image of the second layout of the first semiconductor chip to which the layout density is applied (or an image of the layer included in the second layout). Similarly, each of the third density data DENSITY DATA 3 to the sixth density data DENSITY DATA 6 may include an image of each of the third layout to the sixth layout of the fourth semiconductor chip to which the layout density is applied. However, implementations are not limited thereto. The first layout to the sixth layout may have different types of layers or different critical dimension (CD) ranges included in the layout. In some implementations, the density data included in the input of the artificial intelligence model AIM may correspond to the first measurement data. For example, the layout included in the image of the density data may correspond to the layout of the first semiconductor chip to which the CMP process has been performed. At least one piece of density data may be included in the input of the artificial intelligence model AIM. For example, at least one density data of the first density data DENSITY DATA 1 to the sixth density data DENSITY DATA 6 and the first measurement data may be input to the artificial intelligence model AIM. In some implementations, the size of the image included in each density data may be greater than or equal to the size of the target area (e.g., about 40 μm*about 40 μm). For example, the domain of each density data may be about 40 μm to about 120 μm, but is not limited thereto.

[0090] The output of the artificial intelligence model AIM may include second measurement data. The second measurement data may include a step height CMP step height value with metal deposition of a target area in the second semiconductor chip. The number of pieces of scalar data representing the step height CMP step height value with metal deposition may be approximately 500,000. The size of the target area of ​​the second measurement data may be referred to as an output unit domain size.

[0091] According to the above implementation, by learning and applying density data by considering the specific layout periphery of the artificial intelligence model, the accuracy of prediction of the artificial intelligence model can be improved.

[0092] Fig.12 is a diagram showing an example of an artificial intelligence model AIM.

[0093] Reference Fig.11 and Fig.12, the artificial intelligence model AIM may include a neural network. The neural network may include an input layer, a hidden layer, and an output layer. The input layer may include a layer to which the input of the artificial intelligence model of the AI ​​model is input. The input of the artificial intelligence model of the AI ​​model may include the first measurement data of the full chip and the density data of the layout. For example, the full chip may mean the entire first semiconductor chip. The density data of the layout may include an image of the layout in which the layout density is applied in the first semiconductor chip. The output layer may include a layer to which the output of the artificial intelligence model of the AI ​​model is input. The output of the artificial intelligence model of the AI ​​model may include the second measurement data of the full chip. The second measurement data may include the step height optically measured in the full chip from which optical interference has been removed. With respect to the second measurement data of the full chip, the full chip may mean the entire semiconductor chip on which metal is deposited. The hidden layer may include a plurality of layers included between the input layer and the output layer. In some implementations, the hidden layer may include a fully connected layer FCL and a regression layer RL. The convolutional neural network CNN may include one or more convolutional layers and pooling layers.

[0094] In some implementations, the artificial intelligence model AIM may be implemented as a convolutional neural network CNN, a fully connected layer FCL, an exponential linear unit (ELU) activation, L2 regularization, dropout, and a deep regression model including a deeper dense layer.

[0095] Fig.13 is an example graph of step heights for a semiconductor chip having metal deposited thereon. Fig.14 is an example graph of step heights of a semiconductor chip on which metal is deposited and the predicted values ​​of an artificial intelligence model.

[0096] Reference Fig.13 , assuming that the metal includes TiN. The change value of the step caused by the CMP process on the semiconductor chip on which TiN is not deposited (for example, refer to Fig.13 ) and the variation value of the step caused by the CMP process on each of the semiconductor chips on which TiN is deposited (e.g., reference Fig.13 The distribution of “CMP variation (TiN 60 nm)” in FIG. 5 may have a nonlinear shape.

[0097] Reference Fig.14 , a change value of a step caused by a CMP process on each of the semiconductor chips on which a metal is deposited (for example, referring to Fig.14 “CMP variation (TiN60nm)” in ) and the predicted values ​​of the AI ​​model (e.g., Ref. Fig.14 The distribution of “CMP prediction” in can have a linear shape.

[0098] Fig.15A diagram showing an example of a learning process for an artificial intelligence model. Fig.16 is based on Fig.15 Example plots of the root mean square error (RMSE) and r-squared (R2) score of the learning process of an AI model.

[0099] Reference Figures 13 to 16 , the processor 100 may train the artificial intelligence model to learn a learning data set including the first measurement data OMD1, the density data, and the second measurement data OMD2. About 500,000 learning data sets may be set for each semiconductor chip. For example, the learning method may include supervised learning, but is not limited thereto. The second measurement data OMD2 may include a step height optically measured using a PWG metrology method on a semiconductor chip on which a metal is deposited. Fig.15 In FIG. 1 , a reference based on first measurement data OMD1 and second measurement data OMD2 is shown as an example. The reference may have Fig.13 A similar nonlinear shape is shown.

[0100] When performing the learning process of an AI model (e.g., Fig.15 ) in the "Training in Progress" window, similar to Fig.14 As shown, the step height of the semiconductor chip on which the metal is deposited and the predicted value of the artificial intelligence model may correspond linearly to each other.

[0101] On the other hand, the processor 100 may train the artificial intelligence model to learn data, wherein the absolute value of the CMP step is equal to or greater than a specific value during the learning process of the artificial intelligence model, so as to enhance the prediction capability regarding the weak points.

[0102] While the AI ​​model is learning (e.g., Fig.16 On the other hand, while executing the learning process of the artificial intelligence model (e.g., referring to Fig.16 The R-squared R2 measured in "Training in Progress" can gradually increase.

[0103] Fig.17 Shown for illustration Fig.15 An example diagram of how noise removal from measurement data is implemented during the learning process of an artificial intelligence model in .

[0104] Reference Figure 8 , Fig. 9 , Fig.11 , Fig.15 and Fig.17, when the artificial intelligence model AIM learns according to supervised learning, the processor 100 may use a denoising algorithm to remove noise from the first raw data of the first measurement data and noise from the second raw data of the second measurement data. An implementation method for removing noise from the first raw data of the first measurement data (e.g., referring to Fig.17 "OMD1 noise removal" in the above reference Figure 8 The implementation method described above is the same as that of the second measurement data, and the implementation method of removing noise from the second raw data (for example, referring to Fig.17 "OMD2 Noise Removal" in the above reference Figure 8 The described implementation operates similarly. The R-squared R2 score of the steps of the noise-removed measurement data may be increased compared to the steps of the original data.

[0105] The processor 100 according to an implementation may set the first and second measurement data from which noise has been removed, and the density data as a learning data set.

[0106] In some implementations, the processor 100 may optimize the size of the unit resolution area based on the learning result of the artificial intelligence model. For example, the size of the optimized unit resolution area may be about 2 μm*about 2 μm, but is not limited thereto.

[0107] Fig.18 and Fig.19 is an example graph used to illustrate the test results of the artificial intelligence model.

[0108] Reference Fig.18 and Fig.19 , it is assumed that the number of test data used to test the artificial intelligence model of each implementation is approximately 200,000.

[0109] Reference Fig.18 , after applying the test data to the implemented AI model, the overall RMSE was reduced by about 33.2%. The outlier RSME was reduced by about 59.2%. For example, in the entire region (e.g., reference Fig.18 ), the RMSE of the predicted values ​​(e.g., Fig.18 ) relative to the original data (e.g., Fig.18 The "original value of CMP step height" in the reference Fig.18 The RSME of the predicted value of the AI ​​model (e.g., reference Fig.18 ) relative to the original data (e.g., Fig.18 The “original value of CMP step height” in the reference Fig.18 The RSME of the predicted value of the AI ​​model (e.g., reference Fig.18 ) relative to the original data (e.g., Fig.18 The “original value of CMP step height” in FIG. 1 is reduced by about 59.2%.

[0110] Reference Fig.19 As a result of applying the test data to the artificial intelligence model of the present disclosure, the consistency of outliers can be improved, and the outlier RSME can be reduced by about 78.6%, and the R-squared R2 score can be increased. For example, in the entire area (e.g., reference Fig.19 In the “whole area” in the reference, the RMSE of the prediction value of the artificial intelligence model can be reduced by about 23%. Fig.19 In the “normal area” in the reference, the RMSE of the predicted value of the artificial intelligence model can be reduced by about 13% relative to the original data. Fig.19 In the "hot spot areas" in the dataset, the RMSE of the predicted values ​​of the artificial intelligence model can be reduced by about 79% relative to the original data.

[0111] Fig. 20 Example diagram showing weak points.

[0112] Reference Fig. 20 , in which an image of a semiconductor chip is optically measured (e.g., reference Fig. 20 A weak point in the “chip morphology” in the semiconductor chip may appear on a certain surface. For example, the weak point may include metal thinning, metal pooling, a concave surface formed into a concave-convex shape to cause defocusing, etc. The weak point detection device 50 can use the second measurement data predicted by the artificial intelligence model AIM to detect the weak point of the semiconductor chip.

[0113] The above terms such as first and second are used to distinguish one component from another component, and the components are not limited by the above terms.

[0114] Unless clearly and exceptionally specified in the context, a singular expression includes a plural expression.

[0115] In various operations, reference numbers are used for convenience of explanation and do not describe the order of operations. Unless the context clearly indicates a specific order, the various operations may be performed differently from the specified order.

[0116] Although this specification contains many specific implementation details, these details should not be interpreted as limitations on the scope of any invention or the scope of protection that can be claimed, but should be interpreted as descriptions of features specific to a particular implementation of a particular invention. Certain features described in this specification in the context of separate implementations may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented in multiple implementations individually or in any suitable sub-combination. In addition, although features may be described above as acting in certain combinations, one or more features from a combination may be removed from the combination in some cases, and a combination may involve a sub-combination or a variation of a sub-combination.

[0117] Those skilled in the art will clearly understand that the structure of the present disclosure may be modified or changed in various ways without departing from the scope or technical ideas of the present disclosure. In view of the description given above, when the modifications and changes to the present disclosure fall within the scope of the attached claims and equivalents, the present disclosure will be deemed to include these modifications and changes to the present disclosure.

[0118] While the present disclosure has been particularly shown and described with reference to implementations thereof, it will be understood that various changes in form and details may be made therein without departing from the spirit and scope of the appended claims.

Claims

1. A device for transforming measurement data, the device comprising: communicator, which is configured as receiving first measurement data, the first measurement data comprising a plurality of step height values, wherein the plurality of step height values ​​are optically measured in units of target regions on a first semiconductor chip on which a chemical mechanical polishing process is performed, and receiving layout data, the layout data including a layout in the first semiconductor chip; and a processor configured to generate density data based on the layout data, the density data including an image to which a layout density is applied, the layout density indicating a ratio of an area occupied by the layout in a unit resolution area, inputting the density data and the first measurement data into an artificial intelligence model, and Second measurement data is predicted as an output of the artificial intelligence model, the second measurement data including a step height value per target region of a second semiconductor chip including the first semiconductor chip and a metal deposited on the first semiconductor chip.

2. The device according to claim 1, wherein: The processor is configured to receiving raw data of the first measurement data via the communicator, removing noise from the raw data based on a denoising algorithm, and The first measurement data from which noise has been removed is input into the artificial intelligence model.

3. The device according to claim 2, wherein: The denoising algorithm includes a fast Fourier transform, and The processor is configured to transform the domain of the original data into a frequency domain of a fast Fourier transform based on a fast Fourier transform, removing noise from the raw data based on reflecting at least one mask in a fast Fourier transform spectrum relative to the raw data, and The frequency domain of the Fast Fourier Transform is transformed to the domain of the original data.

4. The device according to claim 1, wherein: The first measurement data includes a step height value optically measured in a target area, and the density data includes a plurality of images of a plurality of different layouts in the first semiconductor chip.

5. The device according to claim 1, wherein: The size of the image included in the density data is greater than or equal to the size of the target area.

6. The device according to claim 1, wherein: The artificial intelligence model includes a neural network, which includes an input layer, a hidden layer and an output layer.

7. The device according to claim 6, wherein: The hidden layers include convolutional layers, pooling layers, fully connected layers and regression layers.

8. The device according to claim 1, wherein: The processor is configured to train the artificial intelligence model based on supervised learning using a learning data set, wherein the learning data set includes the first measurement data, the density data, and the second measurement data.

9. The device according to claim 8, wherein: The processor is configured to Based on the artificial intelligence model learned according to the supervised learning, noise is removed from first raw data of the first measurement data based on a denoising algorithm and noise is removed from second raw data of the second measurement data based on the denoising algorithm, and The first measurement data from which noise is removed, the density data, and the second measurement data from which noise is removed are set as the learning data set.

10. The device according to claim 8, wherein: The processor is configured to optimize the size of the unit resolution area based on the learning result of the artificial intelligence model.

11. A method for transforming measurement data, the method comprising: receiving first measurement data, the first measurement data comprising a plurality of step height values, wherein the plurality of step height values ​​are optically measured in units of target regions on a first semiconductor chip on which a chemical mechanical polishing process is performed; receiving layout data, the layout data including a layout in the first semiconductor chip; generating density data including an image to which a layout density is applied, the layout density indicating a ratio of an area occupied by the layout in a unit resolution area; and Second measurement data is predicted from the density data and the first measurement data based on an artificial intelligence model, the second measurement data including a step height value per target region of a second semiconductor chip including the first semiconductor chip and a metal deposited on the first semiconductor chip.

12. The method according to claim 11, further comprising: removing noise from raw data of the first measurement data based on a denoising algorithm; as well as The first measurement data from which noise is removed is generated.

13. The method according to claim 11, wherein: The density data includes a plurality of images of a plurality of different layouts in a first semiconductor chip.

14. The method according to claim 11, wherein: The artificial intelligence model is implemented as a neural network, which includes an input layer, a convolution layer, a pooling layer, a fully connected layer, a regression layer and an output layer.

15. The method according to claim 14, further comprising: The artificial intelligence model is trained using a learning data set based on supervised learning, wherein the learning data set includes the first measurement data, the density data, and the second measurement data.

16. A system for transforming measurement data, comprising: an optical measuring device configured to generate first measurement data including a step height value of a first semiconductor chip by optically measuring a first semiconductor chip on which a chemical mechanical polishing (CMP) process has been performed; a database configured to store layout data, the layout data including a layout in the first semiconductor chip; as well as The measurement data conversion device is configured to generating density data based on the layout data, and The first measurement data is transformed into second measurement data based on the first measurement data, the density data and the artificial intelligence model, wherein the second measurement data includes multiple step height values ​​of a second semiconductor chip, and the second semiconductor chip includes the first semiconductor chip and metal deposited on the first semiconductor chip.

17. The system of claim 16, wherein: The measurement data conversion device is configured as follows: generating a layout density indicating a ratio of an area occupied by the layout in a unit resolution area, and The second measurement data is predicted from the density data and the first measurement data based on an artificial intelligence model as an output of the artificial intelligence model.

18. The system of claim 16, wherein: The measurement data conversion device is configured as follows: removing noise from raw data of the first measurement data based on a denoising algorithm, and The first measurement data and the density data from which noise has been removed are input into the artificial intelligence model.

19. The system of claim 16, wherein: The first measurement data includes a step height value optically measured in a target area, and the density data includes a plurality of images of a plurality of different layouts in the first semiconductor chip.

20. The system of claim 16, wherein: The measurement data transformation device is configured to train the artificial intelligence model using a learning data set based on supervised learning, and the learning data set includes the first measurement data, the density data and the second measurement data.

Images