Intelligent wafer thermodynamic diagram analysis and abnormal region positioning system

Abstract

The invention provides an intelligent wafer thermodynamic diagram analysis and abnormal region positioning system, and relates to the technical field of intelligent manufacturing, and the system comprises a principal system which comprises a data collection subsystem, a data analysis subsystem, a thermodynamic diagram generation subsystem, an anomaly detection and positioning subsystem, a process optimization subsystem and an output subsystem. According to the invention, the temperature distribution, current density and material stress data of the wafer are collected in real time through the data collection subsystem by using an infrared sensor and optical imaging equipment, and the central control unit coordinates the subsystems to operate in sequence, so that the whole process automation from data collection to abnormal positioning is realized. Compared with traditional manual detection, the system does not need to depend on experience of technicians, large-scale and real-time monitoring can be carried out on wafers, the detection efficiency is remarkably improved, potential anomalies are found in time, defects are prevented from being generated, and an abnormal area is recognized through the anomaly detection and positioning subsystem by adopting a deep learning model.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, specifically to an intelligent wafer thermal mapping analysis and abnormal area location system. Background Technology

[0002] With the rapid development of the semiconductor industry, wafer manufacturing, as a core link in integrated circuit production, directly affects chip performance and production efficiency in terms of quality and stability. During wafer manufacturing, abnormalities in temperature distribution, current density, and material stress can lead to defects (such as hot spots, cracks, or circuit failures), causing not only economic losses but also impacting chip reliability and market competitiveness. Therefore, accurately and promptly detecting abnormal areas in wafer manufacturing and optimizing process parameters to reduce defect rates has become a key research direction in the semiconductor manufacturing field.

[0003] Currently, anomaly detection and process optimization in wafer manufacturing mainly rely on the following technical methods:

[0004] Manual inspection and microscopic observation: Traditional wafer defect detection methods involve manual observation of the wafer surface using a high-resolution microscope, relying on the experience of technicians to determine the presence of anomalies. This method can detect obvious defects to a certain extent, such as cracks or uneven deposition.

[0005] Infrared thermal imaging technology identifies hot spots or abnormal areas by detecting the temperature distribution on the wafer surface. This technology offers advantages such as non-contact operation and real-time performance, and can assist in identifying potential problems during processes such as photolithography and deposition.

[0006] Electrical parameter testing: By monitoring electrical parameters such as current and voltage during the wafer manufacturing process, it is possible to indirectly determine whether there are any abnormalities. For example, abnormal current density may indicate uneven etching depth or short circuits, thus affecting wafer quality.

[0007] Automated inspection equipment: With the development of automation technology, some wafer manufacturing plants have adopted automated optical inspection (AOI) equipment, which uses high-definition cameras and image processing technology to scan the wafer surface and identify defect areas.

[0008] While existing inspection and optimization technologies play a vital role in wafer manufacturing, they still have many shortcomings:

[0009] Problem 1: Traditional manual inspection and microscopic observation are time-consuming, inefficient, and rely heavily on the experience of technicians, making it difficult to achieve large-scale, real-time monitoring of wafers. This results in the difficulty in timely detection of potential abnormal areas, failing to effectively prevent the generation of defects.

[0010] Question 2: While infrared thermal imaging technology can detect temperature anomalies, its resolution and accuracy are limited, making it difficult to precisely locate sub-micron level anomaly boundaries. Furthermore, infrared imaging is significantly affected by environmental factors (such as humidity and air pressure), leading to reduced reliability in complex production environments.

[0011] Question 3: Existing automated testing equipment and electrical parameter testing mostly rely on simple threshold judgments, lacking the ability to comprehensively analyze multi-dimensional data (temperature, current density, stress). This single detection method is prone to false alarms or missed alarms, and cannot provide accurate anomaly classification and risk assessment.

[0012] Question 4: Current technology lacks real-time, continuous process optimization capabilities. Most detection methods only perform offline analysis at specific process stages (such as after photolithography), failing to achieve dynamic monitoring and feedback adjustments throughout the entire process. This results in potential problems not being detected in time when they first occur, missing the optimal opportunity for process optimization.

[0013] Question 5: Existing technologies lack an intelligent mechanism for generating process improvement suggestions after anomaly detection. Traditional optimization methods rely heavily on manual experience or simple rules, failing to fully utilize modern data analysis techniques (such as deep learning and time series analysis) to deeply mine anomaly trends and thus struggle to provide accurate remediation suggestions.

[0014] Therefore, an intelligent wafer thermal mapping analysis and abnormal area location system is needed to solve the above problems. Summary of the Invention

[0015] Technical problems to be solved

[0016] To address the shortcomings of existing technologies, this invention provides an intelligent wafer thermal mapping analysis and abnormal region location system, which solves the problems mentioned in the background technology.

[0017] Technical solution

[0018] To achieve the above objectives, the present invention provides the following technical solution: an intelligent wafer thermal mapping analysis and anomaly region location system, comprising a management system, which includes a data acquisition subsystem, a data analysis subsystem, a thermal mapping generation subsystem, an anomaly detection and location subsystem, a process optimization subsystem, and an output subsystem. The specific design of each subsystem is as follows:

[0019] The data acquisition subsystem is used to acquire raw data on temperature distribution, current density and material stress of the wafer through infrared sensors and optical imaging, and then transmits the raw data to the data analysis subsystem after being timestamped.

[0020] The data analysis subsystem receives raw data from the data acquisition subsystem, automatically analyzes data features using a convolutional neural network algorithm, generates multidimensional heat maps and statistical data, and transmits the analysis results to the heat map generation subsystem and the anomaly detection and localization subsystem.

[0021] The heatmap generation subsystem receives the analysis results from the data analysis subsystem, generates a multidimensional heatmap based on the interpolation algorithm, the heatmap supports custom division of circular, rectangular and annular regions, and transmits the generated heatmap data to the anomaly detection and localization subsystem and the output subsystem;

[0022] The anomaly detection and localization subsystem receives statistical data from the data analysis subsystem and heat map data from the heat map generation subsystem. It uses a deep learning model to identify abnormal areas, accurately locates anomaly boundaries, and transmits the anomaly localization results and statistical characteristics to the process optimization subsystem and the output subsystem.

[0023] The process optimization subsystem receives the anomaly location results from the anomaly detection and location subsystem and the statistical data from the data analysis subsystem. Based on the time series analysis algorithm, it calculates key parameters such as temperature uniformity and defect rate of the process window, generates process improvement suggestions, and transmits the suggested data to the output subsystem.

[0024] The output subsystem receives thermal map data from the thermal map generation subsystem, anomaly location results from the anomaly detection and location subsystem, and improvement suggestions from the process optimization subsystem. It presents the analysis results intuitively through a virtual wafer model and supports user interactive control.

[0025] Preferably, the management system is equipped with a central control unit. The management system coordinates the operation of each subsystem through the central control unit. The central control unit is triggered in the following order: data acquisition → data analysis → heat map generation → anomaly detection and location → process optimization → output. After anomaly detection is completed, a control signal is fed back to the data acquisition subsystem to adjust the acquisition frequency. The data analysis subsystem supports custom data statistics functions for circular, rectangular and annular areas. Users input area parameters (center coordinates, radius or side length) through the output subsystem. After receiving the area parameters, the data analysis subsystem calculates the temperature distribution, average current density and number of anomalies within the specified area based on the area division algorithm, and transmits the statistical results to the heat map generation subsystem to update the heat map.

[0026] Preferably, the heat map generation subsystem uses the Kriging interpolation algorithm to generate a multidimensional heat map, ensuring the spatial continuity of temperature and electrical performance data, and dynamically adjusting the display priority of circular, rectangular, or annular areas in the heat map based on feedback from the anomaly detection and location subsystem.

[0027] Preferably, the anomaly detection and localization subsystem identifies abnormal regions using a deep learning model, calculates the precise location of the abnormal boundary using the wafer 3D coordinate system, and transmits the localization data to the process optimization subsystem and the output subsystem in JSON format.

[0028] Preferably, the process optimization subsystem uses a time series analysis algorithm to analyze the evolution trend of abnormal regions, combines statistical data to calculate the temperature uniformity threshold and defect rate tolerance of the process window, generates optimization suggestions (such as adjusting etching time or photolithography parameters), and feeds them back to the data acquisition subsystem through control flow to optimize the next round of data acquisition.

[0029] Preferably, the output subsystem includes an augmented reality interactive interface. After receiving the heat map data from the heat map generation subsystem, the heat map is projected onto the real wafer via an AR device. Users can input control signals through gestures to trigger the data analysis subsystem to recalculate the statistical data of the specified area.

[0030] Preferably, the data acquisition subsystem integrates an environmental sensing module, which collects humidity, air pressure, and electromagnetic interference data in real time through built-in sensors, and transmits the environmental data to the data analysis subsystem. The data analysis subsystem uses an environmental variable self-calibration analysis algorithm to ensure the accuracy of statistical data.

[0031] Preferably, the anomaly detection and localization subsystem includes an adaptive anomaly "fingerprint" database. After receiving heatmap data from the heatmap generation subsystem, it updates the anomaly features in the fingerprint database using a clustering algorithm, and transmits the classification results to the output subsystem to generate a statistical report when a new anomaly is detected.

[0032] Preferably, the process optimization subsystem includes an anomaly repair suggestion generator. After receiving the anomaly location results from the anomaly detection and location subsystem, it generates repair suggestions (such as changing the mask or adjusting the deposition rate) based on expert rules and simulation algorithms, and transmits the predicted heatmap of the suggestion effect to the output subsystem.

[0033] Preferably, the control flow design supports anomaly priority management. When the anomaly detection and location subsystem identifies a serious anomaly, the central control unit interrupts the regular process, prioritizes triggering the process optimization subsystem to generate emergency repair suggestions, and notifies the user in real time through the output subsystem.

[0034] Beneficial effects

[0035] This invention provides an intelligent wafer thermal mapping analysis and anomaly region location system. It has the following beneficial effects:

[0036] 1. This invention utilizes an infrared sensor and optical imaging equipment in a data acquisition subsystem to collect real-time data on wafer temperature distribution, current density, and material stress. A central control unit coordinates the sequential operation of each subsystem, achieving full automation from data acquisition to anomaly localization. Compared to traditional manual inspection, this system eliminates the need for experienced technicians, enabling large-scale, real-time monitoring of wafers, significantly improving inspection efficiency, timely detection of potential anomalies, and prevention of defects.

[0037] 2. This invention employs a deep learning model to identify abnormal regions through an anomaly detection and localization subsystem, and combines this with a wafer 3D coordinate system to achieve sub-micron level anomaly boundary localization, significantly improving localization accuracy. Simultaneously, the data acquisition subsystem integrates an environmental sensing module to collect humidity, air pressure, and electromagnetic interference data in real time. The data analysis subsystem, based on an environmental variable self-calibration analysis algorithm, ensures the accuracy and reliability of statistical data in complex production environments, overcoming the limitations of traditional infrared thermal imaging technology that is susceptible to environmental influences.

[0038] 3. This invention employs a convolutional neural network algorithm in its data analysis subsystem to extract features from multidimensional data such as temperature, current density, and stress, generating comprehensive statistical data. The anomaly detection and localization subsystem utilizes an adaptive anomaly "fingerprint" database and K-means clustering algorithm to classify anomaly features, providing accurate anomaly type identification. This multidimensional data analysis and intelligent classification mechanism effectively reduces false positives and false negatives, improving the accuracy of anomaly detection and risk assessment capabilities.

[0039] 4. This invention achieves dynamic monitoring of the entire process through a central control unit that coordinates various subsystems, forming a closed loop from data acquisition to process optimization. After the anomaly detection and location subsystem completes its work, it feeds back control signals to the data acquisition subsystem to adjust the acquisition frequency. The process optimization subsystem uses time series analysis algorithms to predict anomaly trends, calculates the temperature uniformity and defect rate of the process window in real time, generates optimization suggestions, and feeds them back to the next round of data acquisition. This real-time feedback mechanism ensures that anomalies can be detected and handled in their early stages, significantly shortening response time and optimizing the timing of process adjustments.

[0040] 5. This invention achieves dynamic monitoring of the entire process through a central control unit that coordinates various subsystems, forming a closed loop from data acquisition to process optimization. After the anomaly detection and location subsystem completes its work, it feeds back control signals to the data acquisition subsystem to adjust the acquisition frequency. The process optimization subsystem uses time series analysis algorithms to predict anomaly trends, calculates the temperature uniformity and defect rate of the process window in real time, generates optimization suggestions, and feeds them back to the next round of data acquisition. This real-time feedback mechanism ensures that anomalies can be detected and handled in their early stages, significantly shortening response time and optimizing the timing of process adjustments.

[0041] 6. In this invention, the output subsystem projects a heat map onto a real wafer through an augmented reality interactive interface. Users can input control signals via gestures to trigger the data analysis subsystem to recalculate statistical data for a specified area. This intuitive visualization and interaction method not only allows users to quickly understand the wafer's status but also enables flexible adjustment of the detection area according to needs, improving operational convenience and user experience. Attached Figure Description

[0042] Figure 1 This is an overall framework diagram of the present invention;

[0043] Figure 2 This is a flowchart illustrating the operation of the present invention;

[0044] Figure 3 This is a simulation diagram of the overall system of the present invention. Detailed Implementation

[0045] 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:

[0047] like Figure 1-3 As shown, the Intelligent Wafer Thermal Map Analysis and Anomaly Location System aims to provide intelligent quality monitoring and improvement support for semiconductor wafer manufacturing through multi-dimensional thermal map generation, anomaly detection and location, and process optimization suggestions. The system includes a management system, which comprises a data acquisition subsystem, a data analysis subsystem, a thermal map generation subsystem, an anomaly detection and location subsystem, a process optimization subsystem, and an output subsystem. The management system coordinates the operation of each subsystem through a central control unit, ensuring smooth data and control flow. The following describes the system's operation, covering the entire process of initialization, normal operation, and anomaly handling.

[0048] Upon system startup, the main system powers on, and the central control unit initializes each subsystem, including hardware self-tests and communication connection confirmations. The infrared sensors, optical imaging equipment, and environmental sensing module of the data acquisition subsystem complete calibration. Users input initial parameters through the output subsystem's interactive interface (such as a touchscreen or AR gestures), such as the center coordinates, radius, or side length of the wafer inspection area, and anomaly detection thresholds (e.g., a temperature deviation limit of 10°C and a defect rate tolerance of 5%). These parameters are transmitted by the central control unit to the data analysis subsystem and the heatmap generation subsystem as the basis for subsequent processing.

[0049] During normal operation, the central control unit triggers each subsystem in the following order: data acquisition, data analysis, thermal map generation, anomaly detection and location, process optimization, and output. The data acquisition subsystem scans the wafer surface using an infrared sensor to collect temperature distribution data (resolution 0.01℃), and an optical imaging device records the current density (unit μA / cm²). 2 The environmental sensing module simultaneously collects humidity, air pressure, and electromagnetic interference data, along with material stress (in MPa). All raw data is timestamped and transmitted in real-time to the data analysis subsystem via a high-speed data bus. The data analysis subsystem receives the raw data and extracts spatial features using a convolutional neural network (CNN) algorithm. The CNN model processes multi-dimensional data (temperature, current density, stress) through multiple convolutional and pooling layers to generate feature maps. Simultaneously, based on user-defined regional parameters, it calculates the mean temperature, variance, and number of outliers within a specified region. Combined with an environmental data self-calibration analysis algorithm, the statistical data and feature maps are then transmitted to the heatmap generation subsystem and the anomaly detection and location subsystem.

[0050] The heatmap generation subsystem receives the analysis results and uses the Kriging interpolation algorithm to generate a multidimensional heatmap. The algorithm transforms discrete data points into a continuous distribution map (1 μm / pixel resolution), covering three dimensions: temperature, current density, and stress. Boundary markers are superimposed on user-defined circular, rectangular, or annular regions. The heatmap data is then transmitted to the anomaly detection and localization subsystem and the output subsystem. The anomaly detection and localization subsystem receives statistical data and heatmap data, uses a deep learning model (combining CNN and region proposal networks) to identify anomalous regions, outputs anomaly bounding boxes, calculates precise locations (0.1 μm accuracy) using the wafer 3D coordinate system, and inputs anomaly features into an adaptive anomaly "fingerprint" database. Anomaly categories (e.g., overheating, cracks) are updated using a clustering algorithm. Finally, the localization results and statistical features are transmitted to the process optimization subsystem and the output subsystem.

[0051] The process optimization subsystem receives anomaly location results and statistical data, uses time series analysis algorithms to analyze the anomaly evolution trend, calculates key parameters of the process window (such as temperature uniformity threshold ±2℃, defect rate tolerance 5%), generates optimization suggestions (such as "adjust etching time to 5 seconds" or "change lithography parameters"), and transmits the suggested data and predicted effects to the output subsystem. Simultaneously, it feeds back to the data acquisition subsystem via control flow to adjust the acquisition frequency (such as increasing from 1Hz to 2Hz). The output subsystem receives thermal map data, anomaly location results, and improvement suggestions, presents the analysis results through a virtual wafer model and statistical charts. Users can view the thermal map projection (anomaly areas are highlighted) via AR devices and trigger the data analysis subsystem to recalculate statistical data for a specified area via gesture input.

[0052] When the anomaly detection and location subsystem identifies a severe anomaly (such as a temperature deviation exceeding 10°C or a defect rate exceeding 5%), the central control unit interrupts the normal process, triggering the anomaly priority management mode. It prioritizes calling the process optimization subsystem to generate emergency repair suggestions (such as "replace the mask" or "adjust the deposition rate to 0.5nm / s"), and notifies the user in real time via the output subsystem (using pop-up alarms or voice prompts). Simultaneously, a heatmap predicting the repair effect is presented to the user. After the anomaly is resolved, the system resumes normal operation. The central control unit then feeds back the repair results to the data acquisition subsystem to optimize the parameters for the next round of acquisition.

[0053] The system achieves closed-loop operation from data acquisition to output through a central control unit, supporting user-defined regional statistics and anomaly priority management. The data acquisition subsystem ensures real-time acquisition of multiple parameters; the data analysis subsystem combines CNN and environmental self-calibration to provide accurate statistics; the heatmap generation subsystem generates high-resolution heatmaps using Kriging interpolation; the anomaly detection and localization subsystem achieves precise localization through deep learning and a fingerprint database; the process optimization subsystem optimizes the process window based on time series analysis; and the output subsystem enhances the user experience with AR interaction. The entire process is highly efficient and collaborative, providing intelligent support for wafer manufacturing. Specific Implementation Example 2:

[0055] like Figure 1-3 As shown, the key algorithm mentioned in Example 1 will be analyzed in detail below, including its core mathematical formulas and explanations:

[0056] Convolutional Neural Network (CNN) Algorithm:

[0057] The specific formula is as follows:

[0058]

[0059] Where O(i,j): the value of the output feature map at position (i,j), representing the feature intensity at a certain point in the wafer thermal analysis (such as abnormal temperature drop or current density change); I(i+m,j+n): the value of the input data matrix at position (i+m,j+n), corresponding to the raw wafer data (such as temperature, current density, or stress) acquired by the data acquisition subsystem; K(m,n): the weight of the convolution kernel (filter) at position (m,n), used to extract local features (such as edges or abnormal regions) in the wafer data; M,N: the size of the convolution kernel (such as 3x3), representing the range of local regions considered when analyzing wafer data; b: bias term, used to adjust the output range of the feature map and optimize the sensitivity of anomaly detection.

[0060] In this data analysis system, the CNN algorithm processes multidimensional input data (temperature distribution, current density, material stress) through convolution operations to generate feature maps. The convolution kernel slides across the wafer data to extract local anomaly features (such as hot spots or cracks), providing a foundation for subsequent heat map generation and anomaly localization. Bias terms and activation functions (such as ReLU) further enhance the significance of the features.

[0061] Kriging interpolation algorithm

[0062] The specific formula is as follows:

[0063]

[0064] Wherein, weight λ i The variation function satisfies:

[0065]

[0066] Where Z(x0): the predicted value of the interpolation point x0, representing the temperature, current density, or stress value at a certain location on the wafer in the thermal map generation subsystem. i Given sampling point x: i The observed values ​​correspond to the actual data measured by the data acquisition subsystem on the wafer surface. λ i : The weight of the i-th sampling point, representing the degree of influence of that point on the predicted value, calculated from the spatial distance and the variogram. n: The number of sampling points involved in the interpolation, depending on the acquisition density of the wafer data. γ(h): The variogram value, representing the spatial variability of the wafer data at a distance h (e.g., non-uniformity of temperature distribution). h: The spatial distance between sampling points, in μm, related to the resolution of the wafer thermal map. N(h): The number of sampling point pairs within a distance h, used for statistical analysis of the variogram.

[0067] In the heatmap generation subsystem, the Kriging interpolation algorithm generates a continuous multidimensional heatmap based on discrete data points provided by the data analysis subsystem. The algorithm analyzes the spatial correlation of wafer data through a variogram function, calculates the weight of each sampling point, and ensures that the interpolation results reflect the true distribution of temperature, current density, and stress, supporting accurate division of circular, rectangular, or annular regions.

[0068] Time series analysis algorithm

[0069] The specific formula is as follows:

[0070] ARIMA(p, d, q) model:

[0071] Z t =φ1Z t-1 +φ2Z t-2 +...+φ p Z t-p+∈ t -θ1∈ t-1 -θ2∈ t-2 -...-θ q ∈ t-q

[0072] Z t The predicted value for time represents the predicted trend of wafer anomaly parameters (such as temperature uniformity or defect rate) in the process optimization subsystem; Z t-1 Z t-2 ,...,Z t-p : Historical values ​​for the first p time points, corresponding to the statistical data sequence generated by the data analysis subsystem; φ1, φ2, ..., φ p : Autoregressive (AR) coefficient, representing the weight of historical data on the current prediction; p: Autoregressive order, representing the number of historical time steps considered; ∈ t Random errors in time represent unpredictable fluctuations in wafer data; ∈ t-1 ,∈ t-2 ,...,∈ t-q Error values ​​at the first q time points, used for smoothing prediction; θ1, θ2, ..., θ q : Moving average (MA) coefficient, representing the adjustment effect of error on forecast; q: Moving average order, representing the number of error steps considered; d: Difference order, representing how many times the original data is differencing to eliminate the trend (e.g., d=1 represents one difference).

[0073] In the process optimization subsystem, the ARIMA algorithm analyzes the abnormal data sequences provided by the anomaly detection subsystem to predict the evolution trend of temperature uniformity or defect rate. The system fits the model with historical statistical data, calculates key parameters of the process window (such as the uniformity threshold ±2℃), generates optimization suggestions (such as adjusting the etching time), and feeds them back to the data acquisition subsystem to optimize the acquisition frequency. Specific Implementation Example 3:

[0075] like Figure 1-3 As shown, the following is a description of the specific application logic steps of the core algorithm in the intelligent wafer thermal mapping analysis and anomaly area localization system:

[0076] I. The specific application logic of the algorithm and its application steps in the invention

[0077] 1. Convolutional Neural Network (CNN) Algorithm

[0078] Application logic:

[0079] CNNs extract spatial features (such as edges or intensity variations of anomalous regions) from multidimensional wafer data through multiple layers of convolution and pooling operations, providing a foundation for subsequent heatmap generation and anomaly localization. The logic includes input data preprocessing, convolutional feature extraction, pooling dimensionality reduction, and output feature map generation.

[0080] Application steps in the invention:

[0081] Step 1: The data analysis subsystem receives the raw data (such as temperature distribution, current density, and material stress) from the data acquisition subsystem and organizes it in the form of a two-dimensional matrix (e.g., a 256x256 pixel temperature matrix).

[0082] Step 2: Use a pre-trained 3x3 convolution kernel (such as an edge detection kernel) to scan the input matrix, calculate the feature value at each position, and generate multiple feature maps (such as temperature gradient map, current density change map).

[0083] Step 3: Apply max pooling (e.g., 2x2 window) to the feature map to preserve significant features (e.g., abnormal high points), reduce data dimensionality, and improve computational efficiency.

[0084] Step 4: Pass the pooled feature map to the heatmap generation subsystem to generate a multi-dimensional heatmap, and at the same time pass it to the anomaly detection and localization subsystem for anomaly region identification.

[0085] 2. Kriging interpolation algorithm

[0086] Application logic:

[0087] Kriging interpolation, based on spatial correlation, interpolates discrete wafer data points into a continuous distribution map, ensuring the spatial smoothness and accuracy of the heatmap. The logic includes variogram calculation, weight allocation, and interpolation prediction.

[0088] Application steps in the invention:

[0089] Step 1: The heat map generation subsystem receives discrete data points (such as temperature values ​​and their coordinates [x,y]) from the data analysis subsystem.

[0090] Step 2: Calculate the spatial variability between data points, analyze the temperature difference between adjacent points (e.g., 1 μm apart), and construct a variogram model.

[0091] Step 3: Assign weights to each known data point based on the variation function. The weights decrease as the distance increases, ensuring that values ​​closer to the prediction point have a greater impact.

[0092] Step 4: Use a weighted average method to predict the temperature value of the unknown point and generate a continuous heat map that covers temperature, current density and stress.

[0093] Step 5: Based on the user-defined circular, rectangular, or annular area, overlay the boundary onto the heatmap and transmit it to the anomaly detection and localization subsystem and the output subsystem.

[0094] 3. Time series analysis algorithms

[0095] Application logic:

[0096] Time series analysis algorithms predict future trends in wafer anomaly parameters by analyzing historical data sequences, thereby optimizing process windows. The logic includes data stabilization, model fitting, and trend prediction.

[0097] Application steps in the invention:

[0098] Step 1: The process optimization subsystem receives the abnormal data sequence (such as the defect rate per minute) from the anomaly detection subsystem.

[0099] Step 2: Perform differential processing on the data to eliminate trends (e.g., calculate the defect rate change between adjacent time points).

[0100] Step 3: Based on historical data, determine the autoregression order p, the difference order d, and the moving average order q (e.g., ARIMA(1,1,1)), and fit the model parameters.

[0101] Step 4: Use the fitted model to predict the defect rate or temperature uniformity at future time points, and calculate the process window threshold (e.g., ±2℃).

[0102] Step 5: Based on the prediction results, generate process improvement suggestions (such as "adjust the etching time to 5 seconds") and pass them to the output subsystem. Specific Implementation Example 4:

[0104] like Figure 1-3 As shown, the following is a detailed hardware composition and hardware description of each subsystem in Embodiment 1:

[0105] Hardware composition and description of the data acquisition subsystem:

[0106] The data acquisition subsystem is responsible for acquiring multi-parameter data from the wafer and transmitting it to downstream modules. Its hardware includes an infrared sensor array for high-precision acquisition of wafer surface temperature distribution; optical imaging equipment (such as a Baslerac A1300-30gm camera, 1.3 megapixels, 30fps frame rate, with a 10μm / pixel lens) to record current density and material stress distribution through optical imaging; an environmental sensing module, including temperature and humidity sensors, barometric pressure sensors, and electromagnetic interference detectors, to monitor production environmental variables in real time to support data calibration; a data acquisition card (such as NIUSB-6210, 16-bit ADC, 250kS / s sampling rate, 16 analog input channels) to convert sensor signals into digital signals and add timestamps; and a microcontroller (such as an STM32F407, 168MHz main frequency, 1MB Flash, 192KB RAM) to coordinate sensor data acquisition, control sampling frequency, and perform preprocessing. The hardware uses infrared sensors and optical devices to acquire multi-dimensional data on wafer temperature, current density, and stress. An environmental sensing module supplements the environmental data. The data acquisition card and microcontroller digitize the analog signals and transmit them to the data analysis subsystem via a high-speed data bus (such as PCIe) to ensure the real-time performance and integrity of the data.

[0107] Hardware composition and description of the data analysis subsystem:

[0108] The data analysis subsystem utilizes high-performance hardware to run CNN algorithms to analyze wafer data. This hardware includes a high-performance GPU for executing convolutional neural network algorithms, processing multidimensional data, and extracting features; a central processing unit (CPU) to coordinate data processing tasks and calculate statistical data (such as mean and variance); 64GB DDR4 memory (3200MHz) to store temporary data and feature maps for high-speed analysis; a solid-state drive (SSD) (e.g., a Samsung 970 EVO Plus 1 TB SSD with read / write speeds of 3500 / 3300MB / s) to store CNN model parameters and historical data for environmental self-calibration and model training; and a data interface module (e.g., PCIe 4.0 x16) to receive raw data from the data acquisition subsystem and output feature maps and statistical data. The GPU-based CNN extracts wafer anomaly features, the CPU assists in calculating statistical data for user-defined regions (circular, rectangular, and annular), memory and SSD support large data processing, and the data interface module ensures efficient communication with upstream and downstream subsystems to meet real-time analysis requirements.

[0109] Hardware composition and description of the heatmap generation subsystem:

[0110] The heatmap generation subsystem is responsible for generating continuous multidimensional heatmaps. Its hardware includes a graphics processor (such as an NVIDIA Quadro RTX 4000, 8GB GDDR6 memory, 2304 CUDA cores, and 6.1 TFLOPS computing power) to run the Kriging interpolation algorithm to generate high-resolution heatmaps; a central processing unit (CPU) to handle region partitioning logic and coordinate heatmap generation tasks; 32GB DDR4 memory (3000MHz) to cache intermediate interpolation results and heatmap data; high-speed memory (such as a WD Black SN850 500GB, read / write speed 7000 / 5100MB / s) to store heatmap templates and historical data for fast loading; and a video output interface (such as DisplayPort 1.4) to transfer heatmap data to the output subsystem. The GPU executes the interpolation algorithm to transform discrete data into continuous heatmaps, the CPU processes user-defined region partitioning, memory and storage support efficient data processing, and the video output interface ensures real-time transmission of heatmaps to the output subsystem to meet visualization requirements.

[0111] Hardware composition and description of the anomaly detection and localization subsystem:

[0112] The anomaly detection and localization subsystem uses deep learning to locate wafer anomalies. Its hardware includes a deep learning accelerator that runs deep learning models (CNN+RPN) and K-means algorithms to identify and locate anomalies; a central processing unit (such as an Intel Xeon E-2278G, 8 cores, 16 threads, 5.0GHz) for 3D coordinate system calculations and fingerprint database updates; 48GB DDR4 ECC memory (2666MHz) for storing anomaly features and fingerprint database data; a solid-state drive (such as an Intel Optane 905P 480GB, read / write speed 2600 / 2200MB / s) for persistently storing the anomaly fingerprint database and localization results; and a data communication module (such as USB 3.2 Gen2) for receiving heatmap data and outputting localization results. The deep learning accelerator identifies the anomaly region and calculates its precise location (accuracy 0.1μm), the CPU assists in coordinate mapping and fingerprint database updates, memory and SSD support fast data access, and the communication module ensures efficient transfer of results to the process optimization and output subsystem.

[0113] Hardware composition and description of the process optimization subsystem:

[0114] The process optimization subsystem analyzes abnormal trends and generates recommendations. Its hardware includes a high-performance processor (e.g., AMD EPYC7313P, 16 cores, 32 threads, 3.7GHz) that runs the ARIMA algorithm to analyze abnormal trends and generate optimization suggestions; 64GB DDR4 ECC memory (2933MHz) to store time-series data and prediction results; a solid-state drive (e.g., Samsung 980 PRO 2TB, read / write speeds 7000 / 5100MB / s) to store historical data and a process parameter library; a data interface module (e.g., PCIe 4.0 x8) to receive abnormal data and output optimization suggestions; and a network module (e.g., Intel X550-T2, dual-port 10GbE) to communicate with production line equipment and provide feedback on optimization parameters. The high-performance processor predicts abnormal trends and calculates process window parameters, while the memory and SSD support big data analysis. The data interface and network modules transmit recommendations to the output subsystem and feed them back to the data acquisition subsystem, enabling process improvements.

[0115] Hardware composition and description of the output subsystem

[0116] The output subsystem presents the analysis results and supports interaction. Its hardware includes an augmented reality device (such as the Microsoft HoloLens 2, 52° field of view, 2048x1080 resolution, gesture recognition accuracy ±1mm) to project heatmaps and anomaly areas onto the real wafer and support gesture interaction; a high-resolution display (such as the Dell UltraSharp U2720Q, 27-inch, 4K resolution 3840x2160) to display virtual wafer models and statistical charts; a central processing unit (such as the Intel Core i7-11700, 8 cores, 16 threads, 4.9GHz) to process interactive control signals and data rendering; a graphics processing unit (such as the NVIDIA GeForce RTX 3060, 12GB GDDR6 memory, 3584 CUDA cores) to render heatmaps and virtual models; and an audio module (such as the Realtek ALC897) to output voice alerts in case of anomalies. The AR device and high-resolution display present the results intuitively, the CPU and GPU handle interaction and rendering, and the audio module provides anomaly notifications to ensure timely user response.

[0117] Hardware composition and description of the central control unit (core of the main system):

[0118] The central control unit, as the core of the main system, coordinates the operation of various subsystems. Its hardware includes an embedded controller (such as a Raspberry Pi 4 Model B, 8GB RAM, quad-core 1.5GHz) that executes control flow logic; a clock module (such as a DS3231RTC, ±2ppm accuracy) that provides high-precision timestamps to synchronize data acquisition and processing; a communication bus (such as a PCIe 4.0 switch) that connects the subsystems to achieve high-speed data transmission; and a power management module (such as a TITPS65988, multi-output, 95% efficiency) that provides stable power to the subsystems. The central control unit coordinates the operation of the subsystems sequentially (data acquisition → analysis → heatmap generation → anomaly location → process optimization → output). The clock module ensures time synchronization, and the communication bus and power module support efficient data flow and stable operation. Specific Implementation Example 5:

[0120] like Figure 1-3 As shown, the following use cases are provided based on the above solution:

[0121] Use Case 1: Temperature Anomaly Detection During Wafer Lithography

[0122] background:

[0123] During the 3nm process of photolithography, a semiconductor factory discovered that the yield of some finished wafers had decreased, and suspected that the uneven temperature distribution of the photolithography machine caused abnormal exposure of the photoresist.

[0124] Problem description:

[0125] When the lithography machine is running, local hot spots may appear on the wafer surface (exceeding the upper limit of the process window of 35°C), which will affect the uniformity of the photoresist and cause pattern defects.

[0126] System application process:

[0127] Data collection:

[0128] The data acquisition subsystem activates the infrared sensor array (FLIRLepton 3.5) and optical imaging equipment (Baslerac A1300-30gm) to acquire real-time temperature distribution and current density data on the wafer surface. The environmental sensing module (SensirionSHT31) records the workshop humidity (50% RH) and air pressure (1013 hPa). The data is timestamped and then transmitted to the data analysis subsystem via the PCIe bus.

[0129] Data Analysis:

[0130] The data analysis subsystem uses a high-performance GPU (NVIDIA RTX 3090) to run a CNN algorithm to extract temperature anomaly features. Users define a circular detection area (center coordinates [50, 50], radius 20 μm) through the output subsystem (HoloLens 2). The system calculates the mean temperature (36.2℃) and variance (1.5) of this area, and the results are passed to the heat map generation subsystem.

[0131] Heatmap generation:

[0132] The heatmap generation subsystem runs the Kriging interpolation algorithm on a GPU (NVIDIA Quadro RTX 4000) to generate a continuous temperature heatmap (resolution 1 μm / pixel). Hot spots are marked in circular areas (highlighted in red), and the data is sent to the anomaly detection and localization subsystem.

[0133] Anomaly detection and localization:

[0134] The anomaly detection and localization subsystem runs a CNN+RPN model using a deep learning accelerator (JetsonAGXXavier) to identify hotspot boundaries (coordinates [48,52] to [52,48]). Combined with the 3D coordinate system, the localization accuracy reaches 0.1μm. The fingerprint database (K-means update) classifies it as "hotspot", and the result is passed to the process optimization subsystem.

[0135] Process optimization:

[0136] The process optimization subsystem runs the ARIMA algorithm on the processor (AMD EPYC7313P), analyzes the hotspot trend, predicts that it will spread within 10 minutes if no adjustment is made, and suggests "reducing the lithography machine power to 90%". It also suggests feeding back to the lithography machine through the network module (Intel X550-T2).

[0137] Output presentation:

[0138] The output subsystem projects a heat map using an AR device, displaying hotspot locations and optimization suggestions. Users confirm adjustments via gestures, and the system updates the heat map in real time to verify the effect.

[0139] result:

[0140] After adjustment, the wafer temperature returned to the process window (33℃±2℃), the photoresist uniformity improved, and the yield increased from 85% to 92%. The system successfully located and resolved the temperature anomaly problem.

[0141] Use Case 2: Crack Defect Location in Wafer Deposition Process

[0142] background:

[0143] During the chemical vapor deposition (CVD) process, a wafer fab discovered that some wafers had microcracks, which affected subsequent processing. It was necessary to quickly locate the cracks and optimize the process.

[0144] Problem description:

[0145] During the deposition process, uneven stress distribution in the wafer material may lead to cracks (stress exceeding 200MPa). Traditional detection methods are time-consuming and have low accuracy.

[0146] System application process:

[0147] Data collection:

[0148] The data acquisition subsystem uses an optical imaging device to acquire material stress data (Baslerac A1300-30gm, 10μm / pixel resolution), an infrared sensor to record temperature distribution, an environmental sensing module to monitor electromagnetic interference (10μT), and the data is transmitted to the data analysis subsystem after being timestamped by a microcontroller (STM32F407).

[0149] Data Analysis:

[0150] The data analysis subsystem runs a CNN algorithm on a GPU (NVIDIA RTX 3090) to extract stress anomaly features. The user defines a rectangular region (coordinates [20,20] to [40,40]), calculates the average stress (210 MPa) and the number of anomalies (15) within the region, and sends the results to the heatmap generation subsystem.

[0151] Heatmap generation:

[0152] The heat map generation subsystem generates stress heat maps using the Kriging interpolation algorithm. High stress points within a rectangular area are marked in yellow, and the data is transmitted to the anomaly detection and location subsystem.

[0153] Anomaly detection and localization:

[0154] The anomaly detection and localization subsystem runs a deep learning model to locate the crack boundary (coordinates [25,25] to [30,28], accuracy 0.1μm), updates the fingerprint database to the "crack" category, and sends the results to the process optimization subsystem.

[0155] Process optimization:

[0156] The process optimization subsystem uses ARIMA to analyze stress trends and predicts that cracks may propagate. It recommends "reducing the deposition rate to 0.5 nm / s" and suggests feeding this information back to the data acquisition subsystem to adjust the acquisition frequency (2 Hz).

[0157] Output presentation:

[0158] The output subsystem displays the stress heat map and crack location on a high-resolution display (Dell U2720Q), the AR device projects the crack area, the user confirms the optimization plan, and the system updates the heat map to verify the effect.

[0159] result:

[0160] After adjusting the deposition rate, the stress dropped to 180 MPa, the crack did not propagate, the wafer yield increased to 95%, and the system quickly located the crack and optimized the process.

[0161] Use Case 3: Optimization of Current Density Anomalies in Wafer Etching Process

[0162] background:

[0163] During the plasma etching process, a factory discovered that the performance of some wafer circuits was unstable, suspecting that the uneven etching depth was caused by abnormal current density distribution.

[0164] Problem description:

[0165] During the etching process, the local current density on the wafer is too high (exceeding 50 μA / cm). 2 This could lead to over-etching, affecting circuit integrity.

[0166] System application process:

[0167] Data collection:

[0168] The data acquisition subsystem collects current density data through optical imaging equipment, records temperature through infrared sensors, and monitors air pressure (1010 hPa) through an environmental sensing module. The data is then digitized by the data acquisition card (NIUSB-6210) and transmitted to the data analysis subsystem.

[0169] Data Analysis:

[0170] The data analysis subsystem uses a CNN algorithm to extract current density features. A user-defined annular region (inner diameter 10 μm, outer diameter 20 μm) is used to calculate the mean (52 μA / cm²). 2 The results are calculated as follows: (1) and (2) variance (2.0), and then sent to the heatmap generation subsystem.

[0171] Heatmap generation:

[0172] The heat map generation subsystem generates a current density heat map, with high-density points within the annular area marked in red. The data is then transmitted to the anomaly detection and location subsystem.

[0173] Anomaly detection and localization:

[0174] The anomaly detection and localization subsystem identifies areas with excessively high current density (coordinates [15,15] to [18,18]), classifies them as "current anomalies" in the fingerprint database, and has a localization accuracy of 0.1 μm. The results are then sent to the process optimization subsystem.

[0175] Process optimization:

[0176] The process optimization subsystem analyzes the current density trend, predicts the risk of over-etching, and recommends "reducing the plasma power to 80%", suggesting that the feedback be sent to the etching equipment via the network module.

[0177] Output presentation:

[0178] The output subsystem projects heatmaps and abnormal areas using AR devices, displays statistical charts on the screen, allows users to confirm adjustments, and updates the heatmap to verify the effect.

[0179] result:

[0180] After adjustment, the current density decreased to 45 μA / cm. 2 The etching depth is uniform, the circuit performance is stable, and the yield has increased from 88% to 94%. The system has effectively optimized the process window.

[0181] The three case studies demonstrate the system's application in photolithography (temperature anomalies), deposition (crack defects), and etching (current density anomalies), showcasing the real-time data acquisition, the analytical capabilities of CNN and Kriging interpolation, the intelligent optimization using K-means and ARIMA, and the intuitiveness of AR output. Through hardware and algorithm collaboration, the system quickly locates anomalies and provides improvement suggestions, enhancing wafer manufacturing efficiency and quality.

[0182] 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.

[0183] 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 intelligent wafer thermal mapping analysis and anomaly region location system, characterized in that: The system includes a management system, which comprises a data acquisition subsystem, a data analysis subsystem, a heatmap generation subsystem, an anomaly detection and location subsystem, a process optimization subsystem, and an output subsystem. The specific design of each subsystem is as follows: The data acquisition subsystem is used to acquire raw data on temperature distribution, current density and material stress of the wafer through infrared sensors and optical imaging, and then transmits the raw data to the data analysis subsystem after being timestamped. The data analysis subsystem receives raw data from the data acquisition subsystem, automatically analyzes data features using a convolutional neural network algorithm, generates multidimensional heat maps and statistical data, and transmits the analysis results to the heat map generation subsystem and the anomaly detection and localization subsystem. The heatmap generation subsystem receives the analysis results from the data analysis subsystem, generates a multidimensional heatmap based on the interpolation algorithm, the heatmap supports custom division of circular, rectangular and annular regions, and transmits the generated heatmap data to the anomaly detection and localization subsystem and the output subsystem; The anomaly detection and localization subsystem receives statistical data from the data analysis subsystem and heat map data from the heat map generation subsystem. It uses a deep learning model to identify abnormal areas, accurately locates anomaly boundaries, and transmits the anomaly localization results and statistical characteristics to the process optimization subsystem and the output subsystem. The process optimization subsystem receives the anomaly location results from the anomaly detection and location subsystem and the statistical data from the data analysis subsystem. Based on the time series analysis algorithm, it calculates key parameters such as temperature uniformity and defect rate of the process window, generates process improvement suggestions, and transmits the suggested data to the output subsystem. The output subsystem receives thermal map data from the thermal map generation subsystem, anomaly location results from the anomaly detection and location subsystem, and improvement suggestions from the process optimization subsystem. It presents the analysis results intuitively through a virtual wafer model and supports user interactive control.

2. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The management system is equipped with a central control unit, which coordinates the operation of each subsystem. The central control unit triggers the following sequence: data acquisition → data analysis → heat map generation → anomaly detection and location → process optimization → output. After anomaly detection is completed, a control signal is fed back to the data acquisition subsystem to adjust the acquisition frequency. The data analysis subsystem supports custom data statistics functions for circular, rectangular, and annular regions. Users input region parameters through the output subsystem. After receiving the region parameters, the data analysis subsystem calculates the temperature distribution, average current density, and number of anomalies within the specified region based on the region division algorithm, and transmits the statistical results to the heat map generation subsystem to update the heat map.

3. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The heat map generation subsystem uses the Kriging interpolation algorithm to generate multidimensional heat maps, ensuring the spatial continuity of temperature and electrical performance data, and dynamically adjusts the display priority of circular, rectangular, or annular areas in the heat map based on feedback from the anomaly detection and location subsystem.

4. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The anomaly detection and localization subsystem identifies abnormal regions using a deep learning model, calculates the precise location of the anomaly boundary using the wafer 3D coordinate system, and transmits the localization data to the process optimization subsystem and the output subsystem in JSON format.

5. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The process optimization subsystem uses time series analysis algorithms to analyze the evolution trend of abnormal regions, combines statistical data to calculate the temperature uniformity threshold and defect rate tolerance of the process window, generates optimization suggestions, and feeds them back to the data acquisition subsystem through control flow to optimize the next round of data acquisition.

6. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The output subsystem includes an augmented reality interactive interface. After receiving heat map data from the heat map generation subsystem, it projects the heat map onto the real wafer using an AR device. Users can input control signals through gestures to trigger the data analysis subsystem to recalculate the statistical data of a specified area.

7. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The data acquisition subsystem integrates an environmental sensing module, which collects humidity, air pressure, and electromagnetic interference data in real time through built-in sensors and transmits the environmental data to the data analysis subsystem. The data analysis subsystem uses an environmental variable self-calibration analysis algorithm to ensure the accuracy of statistical data.

8. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The anomaly detection and localization subsystem includes an adaptive anomaly "fingerprint" database. After receiving heatmap data from the heatmap generation subsystem, it updates the anomaly features in the fingerprint database through a clustering algorithm. When a new anomaly is detected, the classification result is transmitted to the output subsystem to generate a statistical report.

9. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The process optimization subsystem includes an anomaly repair suggestion generator. After receiving the anomaly location results from the anomaly detection and location subsystem, it generates repair suggestions based on expert rules and simulation algorithms, and transmits the predicted heatmap of the suggestion effect to the output subsystem.

10. The intelligent wafer thermal mapping analysis and anomaly region location system according to claim 1, characterized in that: The control flow design supports anomaly priority management. When the anomaly detection and location subsystem identifies a serious anomaly, the central control unit interrupts the normal process, prioritizes triggering the process optimization subsystem to generate emergency repair suggestions, and notifies the user in real time through the output subsystem.

Images