Overlay data analysis and prediction system based on feedforward simulation of feedback

Abstract

The application provides an Overlay data analysis and prediction system based on feedback feedforward simulation, relates to the field of semiconductor alignment optimization and intelligent control, and comprises a main body system, wherein the main body system comprises a data acquisition module, a data processing and storage module, an analysis simulation module and a prediction and correction module; the data acquisition module is arranged on sensors at key positions of a photoetching device and comprises 0.1nm resolution optical encoders, 64-channel MEMS vibration sensors and infrared thermal imaging units arranged on an exposure unit, a workpiece table and a lens group, and can acquire Overlay data of a production batch in real time; the system adopts a wavelet transform-independent component analysis joint noise reduction technology and a deep residual network algorithm, can accurately extract key features from massive photoetching data, can discover potential process problems in advance through deep analysis of the features and mining of potential rules hidden behind the data, and can make the analysis of Overlay no longer limited to the surface, thereby providing strong support for preventive maintenance and process optimization.

Description

Technical Field

[0001] This invention relates to the field of semiconductor alignment optimization and intelligent control, specifically to an overlay data analysis and prediction system based on feedback feedforward simulation. Background Technology

[0002] In the semiconductor manufacturing field, lithography technology is a key step in chip manufacturing, and its precision directly affects the chip's performance and production yield. Overlay accuracy is an important indicator for measuring the accuracy of the lithography process. Accurate analysis and prediction of it are crucial for optimizing the lithography process and improving chip manufacturing quality.

[0003] Currently, the following technical methods are mainly used in overlay data processing and photolithography process optimization:

[0004] Basic Data Acquisition and Simple Analysis: By installing basic sensors on the lithography equipment, key data from the lithography process, such as position information and exposure time, are collected. Simple statistical analysis is then performed on this data.

[0005] Traditional model prediction: This method uses classic physical or empirical models to predict the overlay based on the parameters of the lithography equipment and process conditions.

[0006] Single algorithm processing: Some companies use a single algorithm, such as a simple regression algorithm or neural network algorithm, to analyze and predict overlay data.

[0007] Manual experience adjustment: When problems occur in the photolithography process, the operation mainly relies on the experience of the operators to adjust the photolithography parameters in order to improve the overlay.

[0008] Although existing technologies can monitor and adjust the photolithography process to some extent, they still have many shortcomings:

[0009] Insufficient depth of data mining: Existing basic data collection and simple analysis methods are unable to extract key features and potential patterns from large amounts of lithography data, resulting in the analysis of overlay remaining superficial and making it difficult to detect potential process problems in advance.

[0010] Limited prediction accuracy: Due to the simplification of the lithography process and insufficient consideration of complex factors in traditional models, as well as the limitations of single algorithms in processing complex data, the prediction accuracy of overlay is difficult to meet the requirements of high-precision chip manufacturing, which can easily lead to lithography deviations and affect chip performance and production yield.

[0011] Lack of real-time feedback and adaptive adjustment: Most existing methods lack real-time feedback mechanisms, making it impossible to dynamically adjust the process based on real-time data during the lithography process. Furthermore, they struggle to achieve adaptive adjustment in the face of changes in process parameters and fluctuations in equipment status, thus reducing the stability and reliability of the lithography process.

[0012] Reliance on manual experience: The method of adjusting lithography parameters by manual experience lacks scientific basis and systematicity. The adjustment effect varies greatly among different operators, and it is difficult to respond quickly to changes in the production process, which is not conducive to improving production efficiency and product quality.

[0013] Therefore, an overlay data analysis and prediction system based on feedback feedforward simulation is needed to solve the above problems. Summary of the Invention

[0014] Technical problems to be solved

[0015] To address the shortcomings of existing technologies, this invention provides an Overlay data analysis and prediction system based on feedback feedforward simulation, which solves the problems mentioned in the background technology.

[0016] Technical solution

[0017] To achieve the above objectives, the present invention is implemented through the following technical solution: an Overlay data analysis and prediction system based on feedback feedforward simulation, comprising a main system, wherein the main system includes a data acquisition module, a data processing and storage module, an analysis and simulation module, and a prediction and correction module;

[0018] The data acquisition module is equipped with sensors located at key positions in the lithography equipment, including a 0.1nm resolution optical encoder, a 64-channel MEMS vibration sensor, and an infrared thermal imaging unit deployed in the exposure unit, workpiece stage, and lens group, to collect overlay data of production batches in real time.

[0019] The exposure unit, workpiece stage, and lens group here are all main components of existing lithography equipment.

[0020] The data processing and storage module adopts a distributed storage architecture based on the IPFS protocol. It performs wavelet transform-independent component analysis joint noise reduction on the collected data, removes noise interference, and then stores it in the blockchain database through SM4 national cryptographic encryption unit.

[0021] The analysis and simulation module employs a bidirectional coupling system of a physics-driven model and a deep residual network data-driven model; it dynamically allocates model weights of 0.3-0.7 through an attention mechanism to simulate the production process under different parameters.

[0022] The prediction and correction module predicts overlay deviations based on simulation results and adjusts lithography parameters through positive compensation and reverse correction functions. A bidirectional coupling system of a physics-driven model and a deep residual network data-driven model is used, dynamically allocating model weights of 0.3-0.7 through an attention mechanism to simulate the production process under different parameters.

[0023] Preferably, in the analysis and simulation module, the feedback algorithm establishes a thermo-mechanical coupling model based on the Navier-Stokes equations and dynamically compensates for the difference between the actual data and the expected data in the production process. The feedforward algorithm extracts high-order features through a deep residual network and performs predictive simulation of the production process based on pre-set parameters and models. An incremental learning strategy is introduced, which triggers model retraining every 50 batches of data to dynamically optimize algorithm parameters.

[0024] Preferably, in the prediction and correction module, forward compensation reduces overlay deviation by adjusting the exposure energy and alignment offset parameters of the lithography equipment, while reverse correction employs the XGBoost intelligent decision tree model to automatically select a correction strategy based on historical process scenarios and reverse-correct over-adjustments during the correction process. Simultaneously, a dynamic time warping algorithm is used to achieve spatiotemporal synchronization of multi-sensor data, ensuring the accuracy of lithography alignment.

[0025] Preferably, the data acquisition module employs a fiber Bragg grating strain measurement unit to acquire overlay data in real time; it is also equipped with an adaptive acquisition frequency adjustment function, automatically switching the sampling rate within the range of 50Hz-10kHz according to the equipment vibration amplitude, and integrating a temperature compensation circuit to eliminate the influence of ±0.5℃ temperature drift on sensor accuracy.

[0026] Preferably, the data processing and storage module adopts a data sharding storage architecture, storing process data in blocks according to timestamps; it uses the LZMA compression algorithm to reduce storage space, and achieves blockchain traceability management of key data through Merkle tree version tracing mechanism and attribute-based encrypted access control.

[0027] Preferably, the main system also includes a digital twin visualization module, which can display the analysis results in intuitive graphical and chart forms, including real-time trend charts and comparative analysis charts, to facilitate operators to quickly understand the production status. Through CFD thermal field distribution cloud maps, it dynamically displays temperature changes of ±0.1℃, utilizes an AR-assisted calibration interface compatible with HoloLens2 devices to achieve virtual-real overlay alignment, and establishes a three-level deviation early warning system, setting alarm thresholds for yellow (>3nm), orange (>5nm), and red (>8nm).

[0028] Preferably, the main system has a multi-device collaboration function, which is realized through a quantum-secure communication unit. It integrates NTRU lattice-based encryption to ensure data transmission between lithography devices, monitors channel security in real time through a BB84 protocol monitoring module, and automatically refreshes the 256-bit quantum key every hour.

[0029] Preferably, the main system introduces a fault diagnosis module, which predicts faults of the lithography equipment in advance by analyzing the anomalies of the collected data. The faults include a process knowledge graph engine: a fault entity relationship network is constructed using the BERT-BiLSTM-CRF model, historical maintenance plans are matched through a case reasoning mechanism, and the fault features are adaptively updated using a dynamic node embedding algorithm.

[0030] Preferably, the main system has an online learning function, which is specifically manifested as follows: equipped with a model fusion unit to automatically adjust the weights of the physical-data model according to the prediction error; in emergency situations, a 10kHz high-frequency acquisition mode is activated to update compensation parameters in real time; and incremental retraining is triggered every 50 batches of data.

[0031] Beneficial effects

[0032] This invention provides an overlay data analysis and prediction system based on feedback feedforward simulation. It has the following beneficial effects:

[0033] 1. This system employs wavelet transform-independent component analysis (ICA) combined denoising technology and a deep residual network algorithm to accurately extract key features from massive amounts of lithography data. Through in-depth analysis of these features, it uncovers hidden patterns within the data, enabling overlay analysis to move beyond surface-level analysis and identify potential process issues early, providing strong support for preventative maintenance and process optimization. In the data acquisition module, rich data is acquired through various high-precision sensors. This data is then processed by complex algorithms in the data processing and storage module, allowing subsequent analysis and simulation modules to perform in-depth analysis based on high-quality data, promptly detecting subtle factors that may affect lithography quality.

[0034] 2. This system employs a bidirectional coupling of a physics-driven model and a deep residual network data-driven model, dynamically allocating model weights through an attention mechanism. This innovative model architecture fully considers various complex factors in the lithography process, overcoming the simplification drawbacks of traditional models and the limitations of single algorithms. This significantly improves the accuracy of overlay prediction, effectively reduces lithography deviations, and ensures chip performance and production yield. In the analysis and simulation module, through continuous model training and parameter optimization, the system becomes more accurate in predicting overlay under different process parameters and equipment conditions, providing reliable prediction results for the lithography process.

[0035] 3. This invention is based on a feedback-feedforward simulation mechanism, giving the system real-time feedback capabilities and enabling dynamic adjustments to the process based on real-time data during the lithography process. When process parameters change or equipment status fluctuates, the system can respond quickly and automatically adjust relevant parameters to achieve adaptive adjustment. In the prediction and correction module, overlay deviations are predicted based on real-time acquired data, and lithography parameters are adjusted promptly through positive compensation and reverse correction functions to ensure that the lithography process remains in a stable and reliable state, greatly enhancing the stability and reliability of the lithography process.

[0036] 4. This system achieves scientific adjustment of lithography parameters through intelligent algorithms and automated processes. It no longer relies on the personal experience of operators, avoiding inconsistencies in adjustment results caused by differences in operator experience. Simultaneously, the system can quickly respond to changes in the production process, making timely and accurate parameter adjustments, effectively improving production efficiency and product quality. The fault diagnosis module predicts lithography equipment failures in advance through anomaly analysis of collected data, and the online learning module automatically adjusts model weights based on prediction errors. All of these features make the entire lithography production process more scientific and efficient, reducing interference from human factors. Attached Figure Description

[0037] Figure 1 This is a flowchart of the main system framework of the present invention;

[0038] Figure 2 This is a simulation diagram of data acquisition for the present invention;

[0039] Figure 3 This is a real-time trend simulation graph of the present invention;

[0040] Figure 4 This is a comparative analysis simulation diagram of the present invention;

[0041] Figure 5 This is a CFD thermal field distribution cloud map of the present invention;

[0042] Figure 6 This is a simulation diagram of the AR-assisted calibration interface of the present invention;

[0043] Figure 7 This is a simulation diagram of the three-level deviation early warning system of the present invention.

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0046] like Figure 1-7 As shown, the Overlay data analysis and prediction system based on feedback feedforward simulation includes a main system, which contains a data acquisition module, a data processing and storage module, an analysis and simulation module, and a prediction and correction module.

[0047] The data acquisition module is equipped with sensors located at key positions in the lithography equipment, including a 0.1nm resolution optical encoder, a 64-channel MEMS vibration sensor, and an infrared thermal imaging unit deployed in the exposure unit, workpiece stage, and lens group, to collect overlay data of production batches in real time.

[0048] The data processing and storage module adopts a distributed storage architecture based on the IPFS protocol. It performs wavelet transform-independent component analysis joint noise reduction on the collected data, removes noise interference, and then stores it in the blockchain database through SM4 national cryptographic encryption unit.

[0049] The analysis and simulation module employs a bidirectional coupled system of a physics-driven model and a deep residual network data-driven model; it dynamically allocates model weights ranging from 0.3 to 0.7 using an attention mechanism to simulate the production process under different parameters.

[0050] The prediction and correction module predicts overlay deviations based on simulation results and adjusts lithography parameters through positive compensation and reverse correction functions. It employs a bidirectional coupling system of a physics-driven model and a deep residual network data-driven model, dynamically allocating model weights from 0.3 to 0.7 using an attention mechanism to simulate the production process under different parameters.

[0051] In the analysis and simulation module, the feedback algorithm establishes a thermo-mechanical coupling model based on the Navier-Stokes equations and dynamically compensates for the difference between the actual data and the expected data in the production process. The feedforward algorithm extracts high-order features through a deep residual network and performs predictive simulation of the production process based on pre-set parameters and models. An incremental learning strategy is introduced, which triggers model retraining every 50 batches of data to dynamically optimize algorithm parameters.

[0052] In the prediction and correction module, forward compensation reduces overlay deviation by adjusting the exposure energy and alignment offset parameters of the lithography equipment. Reverse correction employs the XGBoost intelligent decision tree model, automatically selecting a correction strategy based on historical process scenarios and correcting over-adjustments during the correction process. It also incorporates a dynamic time warping algorithm to achieve spatiotemporal synchronization of multi-sensor data, ensuring the accuracy of lithography alignment.

[0053] The data acquisition module employs a fiber Bragg grating strain measurement unit to acquire overlay data in real time. It also features an adaptive acquisition frequency adjustment function, automatically switching the sampling rate within the 50Hz-10kHz range based on the equipment vibration amplitude, and integrates a temperature compensation circuit to eliminate the impact of ±0.5℃ temperature drift on sensor accuracy.

[0054] The data processing and storage module adopts a data sharding storage architecture, storing process data in blocks according to timestamps; it uses the LZMA compression algorithm to reduce storage space, and achieves blockchain traceability management of key data through Merkle tree version tracing mechanism and attribute-based encrypted access control.

[0055] The main system also includes a digital twin visualization module, which can display the analysis results in intuitive graphical and chart formats, including real-time trend graphs and comparative analysis charts, allowing operators to quickly understand the production status. Through CFD thermal field distribution cloud maps, it dynamically displays temperature changes of ±0.1℃. Utilizing an AR-assisted calibration interface, it is compatible with HoloLens2 devices to achieve virtual-real overlay alignment and establishes a three-level deviation early warning system, setting alarm thresholds for yellow (>3nm), orange (>5nm), and red (>8nm).

[0056] The main system has a multi-device collaboration function, which is realized through a quantum-safe communication unit. It integrates NTRU lattice-based encryption to ensure data transmission between lithography devices, and uses a BB84 protocol monitoring module to monitor channel security in real time. At the same time, it automatically refreshes the 256-bit quantum key every hour.

[0057] The main system introduces a fault diagnosis module, which predicts faults in the lithography equipment in advance by analyzing the anomalies in the collected data. The faults include a process knowledge graph engine: a fault entity relationship network is constructed using the BERT-BiLSTM-CRF model, historical maintenance plans are matched through a case reasoning mechanism, and the fault features are adaptively updated using a dynamic node embedding algorithm.

[0058] The main system has an online learning function, which is specifically manifested in the following ways: equipped with a model fusion unit to automatically adjust the weights of the physical-data model according to the prediction error; in emergency situations, it can start a 10kHz high-frequency acquisition mode to update compensation parameters in real time; and it can trigger incremental retraining every 50 batches of data.

[0059] This system achieves precise control of the lithography overlay through closed-loop control of multi-source sensing, intelligent analysis, and dynamic correction. High-precision sensors (0.1nm optical encoder, 64-channel MEMS vibration array, and infrared thermal imager) deployed at key workstations of the lithography machine collect real-time data on exposure position, equipment vibration, and temperature field. The adaptive sampling frequency dynamically switches between 50Hz and 10kHz according to operating conditions, and temperature compensation is embedded to eliminate environmental interference. After wavelet-ICA joint noise reduction processing, the collected data is stored in a blockchain database through an IPFS distributed architecture, and Merkle tree tracing and SM4 encryption are used to ensure data security. The analysis module integrates the physical model (Navier-Stokes thermodynamic coupling equation) and the data model (improved ResNet network), and dynamically allocates the weights of the two (0.3-0.7) through an attention mechanism. Incremental learning is triggered to update the model parameters every 50 batches of data processed. The predictive correction module combines fuzzy PID forward compensation (adjusting exposure dose ±0.5mJcm² and alignment offset ±2nm) with XGBoost reverse correction (decision tree matching 20 process scenarios), enabling a six-degree-of-freedom nanometer platform to perform closed-loop calibration with ≤1.5nm accuracy. The digital twin interface visualizes the CFD thermal field, AR-assisted alignment (Hololens2 integration), and three-level early warning (yellow-orange-red). A quantum communication unit (NTRU encryption + BB84 protocol) ensures secure data collaboration across multiple devices. The fault diagnosis engine dynamically links historical cases through a knowledge graph, and the online learning system continuously optimizes model weights and acquisition strategies, forming a fully autonomous optimization process encompassing "perception-analysis-decision-verification." Specific Implementation Example 2:

[0061] like Figure 1-7 As shown, the key algorithm mentioned in Example 1 will be analyzed in detail below, including its core mathematical formulas and explanations:

[0062] Wavelet transform-independent component analysis combined noise reduction:

[0063] Wavelet transform:

[0064] The formula for Discrete Wavelet Transform (DWT) is:

[0065]

[0066] These are wavelet coefficients, which are part of the original signal. After wavelet transform, at scale Peaceful relocation The coefficient values ​​represent the characteristic information of the original signal at different scales and locations.

[0067] The raw signal is the overlay data collected from the sensors of the lithography equipment. Represents discrete time or space sampling points.

[0068] The conjugate of the wavelet basis functions, It is a wavelet basis function, which is scaled. Scalable and translational The function after that, It is a scaling parameter that controls the scaling degree of the wavelet function; a larger scaling parameter... Corresponding to low-frequency components, smaller Corresponding to high-frequency components; It is a translation parameter that controls the position of the wavelet function in time or space.

[0069] Wavelet transform is a time-frequency analysis method that transforms the original signal... Wavelet basis functions of different scales and translations Perform inner product operations (i.e.) This method decomposes the signal into different frequency components and observes the signal at different scales and locations, thereby enabling time-frequency analysis of the signal.

[0070] In the data processing and storage module, wavelet transform is used to decompose the acquired overlay data. Because noise and useful signals have different frequency characteristics, wavelet transform can highlight the details and trends in the signal, distinguishing noise and useful signals at different scales to prepare for subsequent independent component analysis. This helps remove noise interference, improves data quality, and ensures the accuracy and reliability of the data for subsequent storage and analysis.

[0071] Independent component analysis (ICA):

[0072] Assuming the observed signal It consists of independent components It is obtained through linear mixing, i.e. ,in It is a mixing matrix. The goal of ICA is to find a demixing matrix. , making To approximate the independent component s as closely as possible, the solution is typically found by maximizing non-Gaussianity, using negative entropy as a metric. The formula for negative entropy is:

[0073]

[0074] in The observed signal vector is a mixed signal acquired from the photolithography equipment and after preliminary processing. Indicates the first One observed signal;

[0075] Independent component vectors (ICVs) are the independent original signal components that we want to separate from the mixed signal. Indicates the first Each independent component;

[0076] The mixing matrix describes how the independent components are linearly mixed to form the observed signal; the elements in the matrix... Indicates the first The independent component is related to the first The degree of contribution of each observed signal;

[0077] To unmix the matrix, by finding a suitable , for observed signals Perform a linear transformation (y=Wx) to make the transformed signal As close as possible to independent components ;

[0078] Negative entropy is used to measure signals. The greater the negative entropy, the stronger the signal, indicating a difference from the Gaussian distribution. The more it deviates from the Gaussian distribution;

[0079] To and The entropy of a Gaussian distribution with the same variance is a measure of the uncertainty of a random variable. The entropy of a Gaussian distribution is a fixed value that depends only on the variance.

[0080] For signal The entropy reflects the signal The uncertainty or amount of information.

[0081] ICA assumes that the observed signal is a linear mixture of independent components, and seeks the unmixing matrix by maximizing the non-Gaussianity (measured here by negative entropy). Because independent components are usually non-Gaussian distributed, while Gaussian distribution is the most "random" distribution, when the negative entropy of the separated signals is the largest, it means that they are least like Gaussian distributions and are the closest to independent components.

[0082] In the data processing and storage module, wavelet transform is used to further remove noise components from the data after wavelet transform. ICA is used to find the unmixing matrix, separating the noise and useful signals from the mixed signal to obtain the true useful signal. This improves data reliability, facilitates subsequent storage and analysis, and ensures that the stored data accurately reflects the working status of the lithography equipment.

[0083] Feedback Algorithm - Based on the Thermo-Mechanical Coupling Model of the Navier-Stokes Equations:

[0084] Navier-Stokes equations (for incompressible fluids):

[0085] Continuity equation:

[0086] Hamiltonian operator It is a vector differential operator used to perform spatial differentiation operations on functions.

[0087] Velocity vector , , They are in , , The velocity component in the direction describes the velocity of the fluid in space. This equation represents the conservation of fluid mass, meaning that the mass of fluid flowing into a certain infinitesimal element per unit time is equal to the mass of fluid flowing out of that infinitesimal element, implying that there is no source or sink of mass in the flow field.

[0088] Momentum equation:

[0089] in Fluid density is the mass of a unit volume of fluid, reflecting the degree of compactness or aloofness of the fluid. For time, it is used to describe the change of fluid motion over time. Let u be the velocity vector in time. The partial derivative of represents the rate of change of velocity with time. For convection, it describes the flow caused by the macroscopic motion (velocity) of the fluid. The resulting momentum transfer It is a vector operator, representing the directional derivative along the velocity direction. For pressure gradient, It's pressure. This represents the rate of change of pressure in space; the force it generates propels the fluid. Dynamic viscosity reflects the viscous properties of a fluid, that is, its ability to resist deformation. The Laplacian operator acts on the velocity vector It describes the second-order spatial variation of velocity, which is related to viscous forces. External forces, such as gravity and electromagnetic forces, act on a fluid.

[0090] The continuity equation guarantees the conservation of mass in a fluid during flow, representing the application of the law of conservation of mass in fluid mechanics. The momentum equation describes how the rate of change of momentum of a fluid element equals the sum of all forces acting on that element, including the pressure gradient force. ), viscosity And external force (f), and momentum change caused by the fluid's own motion. This is an application of Newton's second law in fluid mechanics.

[0091] The analysis and simulation module is used to establish a thermo-mechanical coupling model. During the operation of lithography equipment, issues related to fluid flow (such as cooling media), heat transfer, and stress distribution arise. Using the Navier-Stokes equations, fluid flow patterns can be calculated based on temperature and pressure data from the production process, thereby analyzing stress distribution. Dynamic compensation is performed based on the discrepancies between actual and expected data, making the simulation results more consistent with actual production conditions, improving simulation accuracy, better predicting the performance of lithography equipment under different operating conditions, and providing a basis for optimizing the production process.

[0092] Feedforward Algorithm - Deep Residual Network:

[0093] Basic unit of deep residual network (ResNet)

[0094] Assume the input is After a series of convolution operations (denoted as...) After that, the output is:

[0095]

[0096] in The input data, in this scheme, is pre-processed (such as data acquisition and preliminary noise reduction) lithography equipment production process data, which includes information about equipment status, process parameters, etc. The residual function, typically composed of multiple convolutional layers and activation functions, represents the sum of the input and the residual function. Feature extraction and transformation operations are performed to extract high-order features from the input data. The output data is the result after residual connection, which includes the features extracted through convolution operations. ), while retaining the original input data Information.

[0097] Traditional neural networks are prone to vanishing or exploding gradients when increasing the number of layers, making training difficult. ResNet addresses this by introducing residual connections, allowing the network to learn residuals. Instead of directly learning complex input-output mappings, this allows gradients to propagate more easily during backpropagation, thus enabling more efficient training of deep networks.

[0098] In the analysis and simulation module, high-order features are extracted from production process data. By training a deep residual network with a large amount of historical data, it can learn complex feature patterns within the data. Based on pre-set parameters and a model, it performs predictive simulations of the production process, using the learned features to predict future production processes, improving the accuracy of future production predictions, identifying potential problems in advance, and providing support for production decision-making.

[0099] Prediction and Correction Module - Positive Compensation:

[0100] Adjustment of positive compensation parameters:

[0101] Assuming the exposure energy of the original photolithography equipment is Alignment offset is Based on the predicted overlay deviation The adjusted exposure energy is:

[0102]

[0103] The adjusted alignment offset is:

[0104]

[0105] in The exposure energy of the original photolithography equipment is a crucial parameter in the photolithography process, affecting the exposure level of the photoresist and the quality of pattern transfer. This is the original alignment offset, i.e., the initial deviation of the photolithography equipment during the alignment process. and They represent in and Offset in direction; The predicted overlay deviation is the deviation value between the lithographic pattern predicted by the simulation module and the expected pattern. According to the coefficient For the original exposure energy Adjustments were made to compensate for any potential deviations. The adjusted alignment offset is calculated separately at... and Direction based on predicted overlay deviation According to the coefficient and For the original alignment offset Make adjustments. These adjustment coefficients, determined experimentally or empirically, dictate the magnitude of the adjustment of exposure energy and alignment offset based on overlay deviation.

[0106] Based on the predicted overlay deviation, the exposure energy and alignment offset are adjusted by a certain proportion. When an overlay deviation is predicted, the exposure effect of the photoresist can be changed by adjusting the exposure energy, and the alignment offset can be adjusted to make the photolithographic pattern more accurately aligned, thereby reducing the deviation.

[0107] In the prediction and correction module, the predicted overlay deviation is directly compensated by adjusting the parameters of the lithography equipment. During the lithography process, overlay deviation affects the manufacturing precision and performance of the chip. Positive compensation can improve the accuracy of lithography alignment, reduce product quality problems caused by deviation, and increase the yield of chip manufacturing.

[0108] Prediction and Correction Module - Backward Correction - XGBoost Intelligent Decision Tree Model:

[0109] XGBoost objective function:

[0110]

[0111] in The objective function is used to measure the performance of the model. These are model parameters, including the structure of the decision tree and the node splitting conditions. For the loss function part, It is the sample size. It is the first The true value of each sample, in this scheme, may be real information such as actual overlay deviation or lithography process scenario category. The model is for the first The predicted value for each sample, Indicates predicted value Compared with the true value The differences between these two loss functions can be addressed by using mean squared error (MSE) for regression problems and cross-entropy (CEE) for classification problems. Minimizing these loss functions allows the model's predictions to be closer to the true values. For regularization terms, It is the number of trees. It is the first A tree, Used to measure the The complexity of a tree includes factors such as its depth and the number of leaf nodes. Regularization prevents overfitting, improves generalization ability, and avoids a model that performs well on training data but poorly on test data or in real-world applications.

[0112] XGBoost trains its model by minimizing an objective function. The loss function measures the model's prediction error, prompting the model to continuously adjust its parameters to improve prediction accuracy; the regularization term limits the model's complexity, prevents the model from overfitting the training data, and enables the model to better adapt to different real-world situations.

[0113] In the prediction and correction module, correction strategies are automatically selected based on historical process scenarios, and over-adjustments during the correction process are reverse-corrected. The XGBoost model can learn the features and patterns in historical process scenario data, and when encountering a new process scenario, it can automatically select an appropriate correction strategy based on the learned knowledge. Furthermore, after positive compensation adjustment of lithography parameters, over-adjustment may occur. XGBoost can reverse-correct over-adjustment based on historical data and the current situation, leveraging its powerful classification and regression capabilities to improve the accuracy and effectiveness of correction strategy selection, further enhancing the precision of lithography alignment. Specific Implementation Example 3:

[0115] like Figure 1-7 As shown below, the specific application logic steps of each module and algorithm in the Overlay data analysis and prediction system based on feedback feedforward simulation are explained:

[0116] Data acquisition module:

[0117] First, overlay data for each production batch is collected in real time using sensors positioned at key locations on the lithography equipment. The sampling rate is dynamically adjusted based on the equipment vibration amplitude. When the vibration amplitude is small, a lower sampling rate (50Hz) is used to save resources; when the vibration amplitude is large, a higher sampling rate (10kHz) is switched to ensure that the data accurately reflects the equipment status.

[0118] Hardware components and descriptions:

[0119] 0.1nm resolution optical encoder: Deployed in the exposure unit, workpiece stage, and lens group, it is used to accurately measure the position information of these key components with a resolution of up to 0.1nm, which can provide high-precision positional basis data for overlay data.

[0120] 64-channel MEMS vibration sensor: Real-time monitoring of equipment vibration. The 64-channel design can sense vibrations in different directions from all angles, providing vibration amplitude data for adaptive acquisition frequency adjustment.

[0121] Infrared thermal imaging unit: Used to monitor the temperature distribution of key parts of the equipment and help analyze the impact of the equipment's thermal state on the overlay.

[0122] Fiber Bragg grating strain measurement unit: Real-time acquisition of overlay data. Utilizing the characteristics of optical fiber, it is sensitive to temperature and strain changes and can accurately acquire relevant data.

[0123] Temperature compensation circuit: Integrated into the sensor, it eliminates the impact of ±0.5℃ temperature drift on sensor accuracy and ensures the accuracy of sensor data under different temperature environments.

[0124] Data processing and storage module:

[0125] First, wavelet transform and independent component analysis are performed on the collected data for noise reduction to remove noise interference. Next, the data is encrypted using the SM4 national cryptographic encryption unit. Then, a distributed storage architecture or data sharding storage architecture based on the IPFS protocol is adopted to store the process data in blocks according to timestamps, and the LZMA compression algorithm is used to reduce storage space. Finally, the blockchain traceability management of key data is achieved through Merkle tree version tracing mechanism and attribute-based encrypted access control.

[0126] Hardware components and descriptions:

[0127] SM4 National Cryptographic Encryption Unit: A hardware encryption module that uses the SM4 encryption algorithm to encrypt data, ensuring data security during storage and transmission.

[0128] Blockchain database storage hardware: used to build a distributed storage architecture, based on the IPFS protocol to realize distributed data storage, and improve the reliability and scalability of data storage.

[0129] Analysis and simulation module:

[0130] A bidirectional coupled system employing a physics-driven model and a deep residual network data-driven model dynamically allocates model weights ranging from 0.3 to 0.7 using an attention mechanism to simulate production processes under different parameters. The feedback algorithm establishes a thermo-mechanical coupling model based on the Navier-Stokes equations, dynamically compensating for discrepancies between actual and expected data during production. The feedforward algorithm extracts high-order features through a deep residual network and performs predictive simulations of the production process based on pre-defined parameters and the model. Model retraining is triggered every 50 batches of data processed to dynamically optimize algorithm parameters.

[0131] Hardware composition and description: It mainly relies on computer hardware resources, such as high-performance CPUs and GPUs, to run complex models and algorithms and perform large-scale data computation and analysis.

[0132] Prediction and Correction Module:

[0133] Based on the simulation results from the analysis and simulation module, overlay deviation is predicted. Forward compensation reduces overlay deviation by adjusting the exposure energy and alignment offset parameters of the lithography equipment. Reverse correction adopts the XGBoost intelligent decision tree model, which automatically selects the correction strategy according to the historical process scenario types and reverses the over-adjustment in the correction process. At the same time, it is equipped with a dynamic time warping algorithm to realize the spatiotemporal synchronization of multi-sensor data and ensure the accuracy of lithography alignment.

[0134] Hardware composition and description: It also relies on computer hardware resources to run prediction and correction algorithms, and may need to interact with the control hardware of the lithography equipment to adjust the lithography parameters.

[0135] Digital twin visualization module:

[0136] The analysis results are presented in intuitive graphical and chart formats, including real-time trend graphs and comparative analyses. A CFD thermal field distribution cloud map dynamically displays temperature changes of ±0.1℃. An AR-assisted calibration interface, compatible with HoloLens2 devices, enables virtual-real overlay alignment. A three-level deviation early warning system is established, with yellow (>3nm), orange (>5nm), and red (>8nm) alarm thresholds set to facilitate quick understanding of production status by operators.

[0137] Hardware components and descriptions:

[0138] Display devices: such as high-resolution displays, used to show various graphics, charts and cloud maps.

[0139] AR device (Hololens2): Enables alignment of virtual and real objects, assisting operators in calibration work.

[0140] Multi-device collaboration function:

[0141] Multi-device collaboration is achieved through a quantum-secure communication unit, and NTRU lattice-based encryption is integrated to ensure data transmission between lithography devices. The BB84 protocol monitoring module detects channel security in real time and automatically refreshes the 256-bit quantum key every hour to ensure the security and stability of data transmission between multiple devices.

[0142] Hardware components and descriptions:

[0143] Quantum-secure communication unit: Includes quantum key distribution device hardware to achieve secure communication based on quantum technology.

[0144] NTRU lattice encryption hardware: used to encrypt transmitted data to ensure the security of data transmission.

[0145] Fault Diagnosis Module: By analyzing anomalies in the collected data, faults in the lithography equipment can be predicted in advance. A fault entity relationship network is constructed using a BERT-BiLSTM-CRF model. Historical maintenance plans are matched through a case reasoning mechanism, and adaptive updates of fault features are achieved using a dynamic node embedding algorithm.

[0146] Hardware composition and description: It relies on computer hardware resources for data processing and model calculation, and is used to run fault diagnosis algorithms and store historical fault data and repair plans.

[0147] Online learning features:

[0148] Equipped with a model fusion unit, it automatically adjusts the weights of the physical-data model based on the prediction error. In emergency situations, it activates a 10kHz high-frequency acquisition mode to update compensation parameters in real time. Incremental retraining is triggered every 50 batches of data to continuously optimize model performance.

[0149] Hardware composition and description: It mainly relies on computer hardware resources to run the model fusion algorithm and the data acquisition mode switching control program. Specific Implementation Example 4:

[0151] like Figure 1-7 As shown, the following is an explanation based on the generated simulation diagrams:

[0152] Real-time trend chart:

[0153] This graph shows the trends of actual overlay data (solid blue line) and expected overlay data (dashed red line) over time. The X-axis represents time, and the Y-axis represents the numerical value of the overlay data. By observing the two curves, operators can visually see how the difference between the actual and expected overlay data evolves over time during production. If the actual data curve deviates from the expected data curve, it may indicate an anomaly in the production process, requiring further analysis and adjustments.

[0154] Comparative analysis chart:

[0155] This chart presents the difference between actual and expected overlay data in bar chart format. The X-axis represents time, and the Y-axis represents the numerical difference in overlay data. Each bar represents the difference between the actual and expected data at a given point in time. A bar on the positive half of the Y-axis indicates that the actual data is greater than the expected data, while a bar on the negative half indicates that the actual data is less than the expected data. This comparison clearly shows at which points in time the differences are significant, thus aiding in the analysis of fluctuations and problems in the production process.

[0156] CFD thermal field distribution cloud map:

[0157] This graph simulates the temperature distribution within a two-dimensional region, specifically a CFD (Computational Fluid Dynamics) thermodynamic field distribution. The X and Y axes represent the coordinates of the simulated region. Different colors in the graph represent different temperature values; the color variations visually indicate the temperature distribution within the region, such as the presence of areas with higher or lower temperatures, and the trend of temperature changes. Color bars correspond to specific temperature values, aiding in the accurate understanding of the thermodynamic field's distribution characteristics.

[0158] AR-assisted calibration interface simulation:

[0159] This image provides a simplified simulation of the interface for AR (Augmented Reality) assisted calibration. The background is a simulated virtual image, possibly representing a scene or view within a lithography device. The red "+" markers in the image simulate actual alignment points, indicating the locations within the virtual image that require calibration. This combination of virtual and real elements allows AR devices (such as HoloLens 2) to overlay virtual information onto the real-world scene, helping operators perform calibration operations more accurately.

[0160] Level 3 Deviation Early Warning System:

[0161] This scatter plot illustrates the operation of the three-level deviation warning system. The X-axis represents time, and the Y-axis represents the overlay data. Blue scatter points represent normal data points, where the difference between the actual and expected overlay data is within the normal range. Yellow, orange, and red scatter points correspond to different levels of warnings (yellow, orange, and red), indicating that the difference between the actual and expected data exceeds the corresponding warning threshold (yellow for >3nm, orange for >5nm, and red for >8nm). This plot clearly shows at a glance when different levels of deviation warnings occurred, allowing for timely adjustments and adjustments.

[0162] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0163] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An overlay data analysis and prediction system based on feedback feedforward simulation, comprising a main system, characterized in that: The main system includes a data acquisition module, a data processing and storage module, an analysis and simulation module, and a prediction and correction module. The data acquisition module is equipped with sensors located at key positions of the lithography equipment, including a 0.1nm resolution optical encoder, vibration sensor, and infrared thermal imaging unit deployed on the exposure unit, workpiece stage, and lens group, to collect overlay data of production batches in real time. The data processing and storage module adopts a distributed storage architecture based on the IPFS protocol. It performs wavelet transform-independent component analysis joint noise reduction on the collected data, removes noise interference, and then stores it in the blockchain database through SM4 national cryptographic encryption unit. The analysis and simulation module employs a bidirectional coupling system of a physics-driven model and a deep residual network data-driven model. The prediction and correction module predicts the overlay deviation based on the simulation results and adjusts the lithography parameters through positive compensation and reverse correction functions. In the analysis and simulation module, the feedback algorithm establishes a thermo-mechanical coupling model based on the Navier-Stokes equations and performs dynamic compensation based on the difference between the actual data and the expected data in the production process. The feedforward algorithm extracts high-order features through a deep residual network and performs predictive simulation of the production process based on pre-set parameters and models. An incremental learning strategy is introduced, which triggers model retraining every 50 batches of data to dynamically optimize algorithm parameters.

2. The Overlay Data Analysis and Prediction System Based on Feedforward Simulation according to claim 1, characterized in that: In the prediction and correction module, forward compensation reduces overlay deviation by adjusting the exposure energy and alignment offset parameters of the lithography equipment, while reverse correction adopts the XGBoost intelligent decision tree model, which automatically selects the correction strategy based on the historical process scenario types and reverses the over-adjustment in the correction process. At the same time, it is equipped with a dynamic time warping algorithm to realize the spatiotemporal synchronization of multi-sensor data.

3. The Overlay Data Analysis and Prediction System Based on Feedforward Simulation according to claim 1, characterized in that: The data acquisition module uses a fiber Bragg grating strain measurement unit to acquire overlay data in real time during the application process; it is also equipped with an adaptive acquisition frequency adjustment function.

4. The Overlay Data Analysis and Prediction System Based on Feedforward Simulation according to claim 1, characterized in that: The data processing and storage module adopts a data sharding storage architecture, storing process data in blocks according to timestamps; it uses the LZMA compression algorithm to reduce storage space, and achieves blockchain traceability management of key data through Merkle tree version tracing mechanism and attribute-based encrypted access control.

5. The Overlay Data Analysis and Prediction System Based on Feedforward Simulation according to claim 1, characterized in that: The main system also includes a digital twin visualization module, which dynamically displays temperature changes through a CFD thermal field distribution cloud map and uses an AR-assisted calibration interface to achieve virtual-real overlay alignment compatible with HoloLens2 devices.

6. The Overlay Data Analysis and Prediction System Based on Feedforward Simulation according to claim 1, characterized in that: The main system has a multi-device collaboration function, which is realized through a quantum-secure communication unit. It integrates NTRU lattice-based encryption to ensure data transmission between lithography devices, monitors channel security in real time through a BB84 protocol monitoring module, and automatically refreshes the 256-bit quantum key every hour.

7. The Overlay Data Analysis and Prediction System Based on Feedforward Simulation according to claim 1, characterized in that: The main system introduces a fault diagnosis module, which predicts faults in the lithography equipment in advance by analyzing the anomalies in the collected data. The faults include a process knowledge graph engine: a fault entity relationship network is constructed using the BERT-BiLSTM-CRF model, historical maintenance plans are matched through a case reasoning mechanism, and the fault features are adaptively updated using a dynamic node embedding algorithm.

8. The Overlay Data Analysis and Prediction System Based on Feedforward Simulation according to claim 1, characterized in that: The main system has an online learning function, which is specifically manifested in the following ways: equipped with a model fusion unit to automatically adjust the weights of the physical-data model according to the prediction error; in emergency situations, it can start a 10kHz high-frequency acquisition mode to update compensation parameters in real time; and trigger incremental retraining every 50 batches of data.

Images