Chip raw material quality state tracking and tracing method and system

Abstract

The invention provides a chip raw material quality state tracking and tracing method and system, and relates to the technical field of chips, comprising the steps of realizing dynamic quality monitoring by constructing a raw material digital twinborn model, performing dimension reduction optimization on the quality state to extract spatio-temporal evolution characteristics, and constructing a multi-agent collaborative traceability system by using a biological inspired chaos immune optimization algorithm to position quality abnormity, and executing closed-loop feedback control. The method can improve the anomaly detection accuracy, accelerates the traceability efficiency, and achieves the whole-process intelligent management and control of the quality of raw materials.

Description

Technical Field

[0001] The present invention relates to chip technology, and in particular to a method and system for tracking and tracing the quality status of chip raw materials. Background Art

[0002] With the rapid development of the semiconductor industry, chip manufacturing is placing increasingly stringent demands on raw material quality. The quality of chip raw materials directly impacts the performance and yield of the final chip product. Therefore, establishing an effective system for tracking and tracing the quality status of chip raw materials is crucial. Traditional chip raw material quality control relies primarily on regular sampling and static quality assessment methods, with manual review and fixed processes for quality traceability. This approach struggles to meet the demands of modern chip manufacturing for dynamic monitoring and precise traceability of raw materials throughout their lifecycle. In recent years, emerging technologies such as digital twins, artificial intelligence, and multi-agent systems have provided new technical means and methodologies for tracking and tracing the quality status of chip raw materials.

[0003] The existing quality monitoring system lacks real-time dynamic tracking capabilities and mainly relies on discrete sampling and static analysis. It is unable to capture the continuous changes in the quality status of raw materials and the spatiotemporal evolution characteristics, resulting in difficulty in timely detection of quality anomalies, affecting the stability and consistency of chip manufacturing.

[0004] Traditional traceability methods usually use simple linear backtracking or rule matching algorithms, which lack adaptive learning and optimization capabilities. Faced with the complex and changeable raw material production environment and multi-source heterogeneous data, traceability efficiency is low and accuracy is insufficient, making it difficult to quickly locate the root cause of quality problems.

[0005] The existing quality control system generally has the problem of information islands, the mapping between the physical world and the information world is incomplete, the data sharing and collaborative analysis mechanism between various production links is not sound, and there is a lack of closed-loop feedback control based on full-process data, making it impossible to achieve dynamic control and continuous optimization of the quality status of raw materials. Summary of the Invention

[0006] The embodiments of the present invention provide a method and system for tracking and tracing the quality status of chip raw materials, which can solve the problems in the prior art.

[0007] A first aspect of an embodiment of the present invention provides a method for tracking and tracing the quality status of chip raw materials, including:

[0008] Acquire real-time data during the production and preparation of chip raw materials, build a digital twin model of the raw materials based on the real-time data, establish a mapping relationship between the physical world and the information world, and realize dynamic monitoring of the quality status of the raw materials;

[0009] The raw material digital twin model is used to perform dimensionality reduction optimization on the raw material quality status, the raw material digital twin model is used to construct a data coding structure, the optimized data is mapped to a feature space, and the spatiotemporal evolution characteristics of the raw material quality status are extracted through the raw material digital twin model; the spatiotemporal evolution characteristics are input into the raw material digital twin model, and the raw material digital twin model is used to perform quality status assessment and generate a quality assessment early warning signal;

[0010] Based on the quality assessment warning signals, a bio-inspired chaotic immune optimization algorithm is deployed to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the rapid location and traceability of quality anomalies can be achieved. The chaotic immune optimization algorithm integrates antibody clone selection, immune memory and chaotic perturbation mechanism to improve the accuracy and convergence speed of abnormal pattern recognition.

[0011] The multi-agent collaborative traceability system is used to activate the intelligent traceability engine according to the quality assessment warning signal, locate the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generate traceability analysis results; closed-loop feedback control is performed based on the traceability analysis results to optimize and adjust the raw material production process and realize dynamic control of quality status.

[0012] Constructing a digital twin model of raw materials based on the real-time data, establishing a mapping relationship between the physical world and the information world, and realizing dynamic monitoring of the quality status of raw materials include:

[0013] Establishing a set of physical equations including concentration field, diffusion coefficient, temperature, and pressure based on the real-time data, wherein the set of physical equations is used to characterize the physical properties of the raw material during preparation;

[0014] Inputting the real-time data into a deep autoencoder, extracting hidden features of the real-time data through the weight matrix and bias term of the deep autoencoder, and constructing a digital twin model of the raw materials based on the hidden features;

[0015] Optimizing the raw material digital twin model using a physical constraint loss function, wherein the physical constraint loss function includes a data fitting error term and a physical constraint term, wherein the physical constraint term is determined by the physical equation group;

[0016] Constructing a high-dimensional feature vector space, mapping the raw material digital twin model to the high-dimensional feature vector space, and dynamically updating the feature weights in the high-dimensional feature vector space through an adaptive weight adjustment mechanism;

[0017] Based on Kalman filtering, the state of the features in the high-dimensional feature vector space is predicted to generate a state vector; the distributed confidence interval is calculated according to the state vector, and the individual indicator scores of the raw material quality status are determined based on the distributed confidence interval. The individual indicator scores and the corresponding weight coefficients are weighted and summed to obtain a comprehensive evaluation index of the quality status, thereby realizing dynamic monitoring of the raw material quality status.

[0018] Using the raw material digital twin model to perform dimensionality reduction optimization on the raw material quality state, and using the raw material digital twin model to construct a data coding structure includes:

[0019] Constructing a digital twin feature matrix, wherein the digital twin feature matrix includes multiple feature vectors, each of which corresponds to a feature dimension of the raw material quality status; inputting the feature vectors in the digital twin feature matrix into a cognitive hierarchy structure, wherein the cognitive hierarchy structure includes multiple cognitive layers, each of which is assigned a corresponding cognitive weight, and weighting the feature vectors using the cognitive weights to obtain cognitive features corresponding to each cognitive layer;

[0020] Performing feature fusion on the cognitive feature and the digital twin feature matrix, wherein the feature fusion includes multiplying the cognitive feature and the digital twin feature matrix by corresponding fusion weight coefficients and then superimposing them to generate a fusion feature;

[0021] Inputting the fused features into a coding mapping module, the coding mapping module including a coding weight matrix and a bias vector, performing a nonlinear transformation on the fused features using the coding weight matrix and the bias vector to obtain coding features; performing an EEG spectrum analysis on the coding features, calculating the spectral density function of the coding features using Fourier transform, and obtaining spectrum enhancement coding;

[0022] The spectrum enhancement coding is optimized based on a dynamic memory update mechanism, wherein the dynamic memory update mechanism includes a memory state and a forgetting factor, wherein the memory state attenuates the historical memory state through the forgetting factor and performs a weighted combination with the current spectrum enhancement coding to obtain a data coding structure.

[0023] Using the raw material digital twin model to perform quality status assessment and generate quality assessment early warning signals includes:

[0024] Utilizing the hierarchical attention mechanism of the raw material digital twin model, spatial and temporal features are extracted from the real-time data of the raw material quality status to obtain a quality status feature vector with spatiotemporal weights;

[0025] Inputting the quality status feature vector into a dual time memory network, extracting short-term dynamic features and long-term evolution laws through fast memory units and slow memory units respectively, and generating a quality status evaluation index based on the combined relationship between the two;

[0026] According to the historical distribution of the quality status evaluation index, a dynamic threshold optimization mechanism is adopted to calculate the alarm threshold, and a quality evaluation early warning signal is generated by comparing the deviation between the quality status evaluation index and the alarm threshold.

[0027] Based on the quality assessment warning signals, a bio-inspired chaotic immune optimization algorithm is deployed to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the rapid location and traceability of quality anomalies are achieved, including:

[0028] Inputting the quality assessment warning signal into a chaotic immune model, the chaotic immune model uses iterative operations to generate an initialization sequence, constructing an agent population based on the initialization sequence, and optimizing the distribution of the agent population through the chaotic immune model. The chaotic immune model includes a warning signal strength parameter and a temperature adjustment parameter, and the agent distribution corresponding to the warning signal strength is obtained by dynamically adjusting the warning signal strength parameter and the temperature adjustment parameter;

[0029] Based on the agent distribution, the chaotic immune model is used to calculate the response strength between the warning signal characteristics and the abnormal template characteristics. The chaotic immune model constructs an ergodic search sequence based on the response strength, and adaptively weights the search sequence with the historical search position to update the current search path and determine the evolution trajectory of the abnormal characteristics.

[0030] An intelligent agent collaborative network is constructed. The intelligent agent collaborative network is based on the spatial distance relationship between intelligent agents. The interaction strength between intelligent agents is constructed through the chaotic immune model. The interaction strength is weightedly fused with the local tracing results of each intelligent agent, where the local tracing results include the location of the anomaly and the propagation path information. The tracing decision result of the global quality anomaly is generated through the weighted fusion.

[0031] Based on the quality assessment warning signals, a bio-inspired chaotic immune optimization algorithm is deployed to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the rapid location and traceability of quality anomalies are achieved, including:

[0032] The quality assessment warning signal is input into a chaotic immune optimization algorithm, which constructs an antigen-antibody response mechanism by calculating the affinity between the quality assessment warning signal and the abnormal template, determines the number of clones based on the affinity and generates an initial immune population, and uses a chaotic sequence to perturb and update the initial immune population, wherein the chaotic sequence is weightedly combined with the historical optimal antibody to determine the search direction, thereby obtaining an optimized immune population;

[0033] A multi-agent collaborative traceability system is constructed based on the optimized immune population, and the antibody characteristics in the optimized immune population are mapped to the initial state of the agent. The agent explores abnormal characteristics in the quality state space through a chaotic search strategy, wherein the chaotic search strategy includes two stages: local fine search and global coarse search, and dynamically adjusts the search step size to achieve rapid positioning of abnormal areas;

[0034] An agent interaction network is constructed in the multi-agent collaborative traceability system, and an interaction weight matrix is ​​calculated based on the spatial distance between agents. The interaction weight matrix is ​​weightedly fused with the abnormal feature information collected by each agent. A propagation path map of quality anomalies is constructed based on the fused abnormal feature information, and the source position and diffusion direction of the quality anomaly are determined by tracing back the propagation path map.

[0035] The multi-agent collaborative traceability system activates the intelligent traceability engine according to the quality assessment warning signal, locates the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generates traceability analysis results including:

[0036] Activate the intelligent traceability engine based on the quality assessment warning signal, input the quality assessment warning signal into the spatiotemporal graph model for warning intensity analysis, perform threshold judgment on the warning intensity through the spatiotemporal graph model, and generate the engine activation state, wherein the spatiotemporal graph model includes sensitivity parameters and activation thresholds;

[0037] Constructing a correlation structure between time series and spatial topology, using the spatiotemporal graph model to calculate the temporal evolution characteristics of the quality status, and simultaneously using the spatiotemporal graph model to analyze the spatial position relationship, fusing the temporal evolution characteristics with the spatial position relationship to form a spatiotemporal correlation feature;

[0038] Constructing an anomaly propagation path based on the spatiotemporal correlation features, calculating the inter-node propagation strength using the spatiotemporal graph model, constructing an anomaly propagation network using the inter-node propagation strength as the node association weight, and identifying anomaly diffusion paths based on the anomaly propagation network;

[0039] The root cause of the abnormal diffusion path is located, the importance scores of the path nodes are calculated using the spatiotemporal graph model, the location of the abnormal source point is determined, and a tracing analysis result is generated based on the location of the abnormal source point and the abnormal diffusion path.

[0040] A second aspect of an embodiment of the present invention provides a chip raw material quality status tracking and tracing system, including:

[0041] The first unit is used to obtain real-time data during the production and preparation of chip raw materials, build a digital twin model of the raw materials based on the real-time data, establish a mapping relationship between the physical world and the information world, and realize dynamic monitoring of the quality status of the raw materials;

[0042] The second unit is configured to use the raw material digital twin model to perform dimensionality reduction optimization on the raw material quality status, use the raw material digital twin model to construct a data coding structure, map the optimized data into a feature space, and extract the spatiotemporal evolution characteristics of the raw material quality status through the raw material digital twin model; input the spatiotemporal evolution characteristics into the raw material digital twin model, use the raw material digital twin model to perform quality status assessment, and generate a quality assessment early warning signal;

[0043] The third unit is used to deploy a biologically inspired chaotic immune optimization algorithm based on the quality assessment warning signal to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the system can quickly locate and trace the quality anomalies. The chaotic immune optimization algorithm integrates antibody clone selection, immune memory, and chaotic perturbation mechanisms to improve the accuracy and convergence speed of anomaly pattern recognition.

[0044] The fourth unit is used to use the multi-agent collaborative traceability system to activate the intelligent traceability engine according to the quality assessment warning signal, locate the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generate traceability analysis results; perform closed-loop feedback control according to the traceability analysis results, which is used for optimizing and adjusting the raw material production process and realizing dynamic control of quality status.

[0045] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0046] processor;

[0047] a memory for storing processor-executable instructions;

[0048] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0049] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0050] The beneficial effects of this application are as follows:

[0051] The present invention realizes dynamic monitoring of quality status by constructing a digital twin model of raw materials, establishes a mapping relationship between the physical world and the information world, can obtain and analyze key data in the raw material production process in real time, and improves the accuracy and timeliness of quality monitoring.

[0052] The present invention adopts a bio-inspired chaotic immune optimization algorithm and a multi-agent collaborative traceability system, integrating antibody clone selection, immune memory and chaotic perturbation mechanism, significantly improving the accuracy and convergence speed of abnormal pattern recognition, and realizing the rapid positioning and precise traceability of quality anomalies.

[0053] The reasoning mechanism based on the spatiotemporal graph of the present invention can locate the root cause of quality anomalies and generate traceability analysis results. Through closed-loop feedback control, it can optimize and adjust the raw material production process, forming a complete quality control system, ensuring the high-quality and stable supply of chip raw materials, which is of great significance to enhancing the competitiveness of the chip industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of the process of tracking and tracing the quality status of chip raw materials according to an embodiment of the present invention;

[0055] Figure 2 This is a bar chart comparing the quality status monitoring performance of the raw material digital twin model according to an embodiment of the present invention;

[0056] Figure 3 A flowchart for dimensionality reduction optimization and data coding structure construction of the raw material digital twin model according to an embodiment of the present invention;

[0057] Figure 4 Schematic diagram comparing the accuracy of quality anomaly tracing using different traceability methods according to an embodiment of the present invention;

[0058] Figure 5 This is a bar chart comparing and analyzing the performance of the chaos immune optimization algorithm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0060] The technical solution of the present invention is described in detail below with reference to specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0061] Figure 1 FIG. 1 is a flow chart of a method for tracking and tracing the quality status of chip raw materials according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0062] Acquire real-time data during the production and preparation of chip raw materials, build a digital twin model of the raw materials based on the real-time data, establish a mapping relationship between the physical world and the information world, and realize dynamic monitoring of the quality status of the raw materials;

[0063] The raw material digital twin model is used to perform dimensionality reduction optimization on the raw material quality status, the raw material digital twin model is used to construct a data coding structure, the optimized data is mapped to a feature space, and the spatiotemporal evolution characteristics of the raw material quality status are extracted through the raw material digital twin model; the spatiotemporal evolution characteristics are input into the raw material digital twin model, and the raw material digital twin model is used to perform quality status assessment and generate a quality assessment early warning signal;

[0064] Based on the quality assessment warning signals, a bio-inspired chaotic immune optimization algorithm is deployed to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the rapid location and traceability of quality anomalies can be achieved. The chaotic immune optimization algorithm integrates antibody clone selection, immune memory and chaotic perturbation mechanism to improve the accuracy and convergence speed of abnormal pattern recognition.

[0065] The multi-agent collaborative traceability system is used to activate the intelligent traceability engine according to the quality assessment warning signal, locate the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generate traceability analysis results; closed-loop feedback control is performed based on the traceability analysis results to optimize and adjust the raw material production process and realize dynamic control of quality status.

[0066] In an optional embodiment, constructing a digital twin model of raw materials based on the real-time data, establishing a mapping relationship between the physical world and the information world, and realizing dynamic monitoring of the quality status of raw materials includes:

[0067] Establishing a set of physical equations including concentration field, diffusion coefficient, temperature, and pressure based on the real-time data, wherein the set of physical equations is used to characterize the physical properties of the raw material during preparation;

[0068] Inputting the real-time data into a deep autoencoder, extracting hidden features of the real-time data through the weight matrix and bias term of the deep autoencoder, and constructing a digital twin model of the raw materials based on the hidden features;

[0069] Optimizing the raw material digital twin model using a physical constraint loss function, wherein the physical constraint loss function includes a data fitting error term and a physical constraint term, wherein the physical constraint term is determined by the physical equation group;

[0070] Constructing a high-dimensional feature vector space, mapping the raw material digital twin model to the high-dimensional feature vector space, and dynamically updating the feature weights in the high-dimensional feature vector space through an adaptive weight adjustment mechanism;

[0071] Based on Kalman filtering, the state of the features in the high-dimensional feature vector space is predicted to generate a state vector; the distributed confidence interval is calculated according to the state vector, and the individual indicator scores of the raw material quality status are determined based on the distributed confidence interval. The individual indicator scores and the corresponding weight coefficients are weighted and summed to obtain a comprehensive evaluation index of the quality status, thereby realizing dynamic monitoring of the raw material quality status.

[0072] The real-time data is obtained through the sensor network deployed on the production line. In the example of semiconductor silicon wafer preparation, the sensor network consists of 128 temperature sensors, 64 pressure sensors, 32 concentration sensors and 16 diffusion monitoring devices. The temperature sensor uses a PT100 platinum resistance thermometer with a measurement range of 0-1500℃ and an accuracy of ±0.5℃; the pressure sensor uses a piezoresistive sensor with a range of 0-10 standard atmospheres and an accuracy of ±0.01 standard atmospheres; the concentration sensor uses an ion-sensitive field-effect transistor with a measurement range of 10 14 to 10 21 The sensor measures 100 cm^3 with an accuracy of ±2%. The diffusion monitoring device is based on the principle of laser scattering and has a measurement accuracy of ±5%. All sensors have a sampling frequency of 10Hz and transmit data in real time to a central processing server via the 5G Industrial Internet of Things.

[0073] The process of establishing a set of physical equations based on the collected real-time data requires the consideration of the physical properties of semiconductor materials. The concentration field description equation is based on Fick's diffusion law, which expresses the relationship between the change of dopant concentration over time and space, where the rate of change of concentration is proportional to the product of the concentration gradient and the diffusion coefficient. The diffusion coefficient description equation uses the Arrhenius formula to express the exponential relationship between the diffusion coefficient and temperature, where the activation energy is set to 3.5-4.2eV and the pre-exponential factor is set to 0.1-10cm. 2 The temperature field description equation is based on the heat conduction equation and includes heat source, convection, and conduction terms. The heat conductivity coefficient is set between 20 and 150 W / (m·K) depending on the material type. The relationship between pressure and temperature follows a modified version of the ideal gas law, with a correction factor of 0.92 to 0.98 determined based on experimental data. These equations together form a coupled system of partial differential equations that characterize the physical properties and dynamic changes during the raw material preparation process.

[0074] The detailed construction of the deep autoencoder is as follows: the input layer receives 128-dimensional raw sensor data and uses the ReLU activation function; the first encoding layer contains 64 nodes and uses the LeakyReLU activation function with a slope parameter of 0.2; the second encoding layer contains 32 nodes and also uses the LeakyReLU activation function; the first decoding layer contains 64 nodes and uses the LeakyReLU activation function; the output layer has the same dimensions as the input layer and uses the Sigmoid activation function to ensure that the output values ​​are within a reasonable range. Dropout layers are added between each layer with a dropout rate of 0.2 to prevent overfitting. The weight matrices W1 of the encoder are 128×64 and W2 are 64×32; the weight matrices W3 of the decoder are 32×64 and W4 are 64×128. The bias vectors b1, b2, b3, and b4 correspond to the bias terms of each layer, respectively. Through autoencoder training, the input data x undergoes an encoding transformation called f(x) = σ(W2·σ(W1·x+b1)+b2), resulting in a 32-dimensional hidden layer feature. Here, σ represents the activation function. These 32-dimensional hidden layer features contain the main information of the original data and serve as the basis for building the digital twin model.

[0075] The specific implementation of the loss function optimization process of physical constraints is as follows: The loss function consists of two parts: the data fitting error term L data and the physical constraint L phys. The data fitting error term is in the form of mean square error, and the sum of the square differences between the model prediction value and the actual observation value is calculated divided by the number of samples. The physical constraint term is introduced in the form of residuals. For each equation in the aforementioned physical equations, the sum of the squares of the differences on the left and right sides of the equation is calculated. In order to balance the influence of the two terms, the weight coefficient λ is introduced and set to a value between 0.5-2.0, which is dynamically adjusted according to the training process. The optimization process uses the Adam optimizer, and the initial learning rate is set to 0.001, which decays to 0.8 times the original value every 2000 iterations. To prevent gradient explosion, the gradient clipping threshold is set to 5.0. The training process adopts batch processing with a batch size of 64 and a total number of iterations of 10,000. When the loss function improves by no more than 10 for 100 consecutive iterations -5 When , the training is terminated early. After the optimization is completed, the average prediction error of the model on the validation set is controlled within 3%.

[0076] The high-dimensional feature vector space is constructed and mapped as follows: the high-dimensional feature vector space consists of 256 feature dimensions, including 32-dimensional hidden features extracted by an autoencoder, 64-dimensional first-order physical parameters (directly measured values), 96-dimensional second-order physical parameters (interaction terms of first-order parameters), and 64-dimensional time series features (including the trend of the main parameters over the past eight time points). The raw material digital twin model is mapped into this high-dimensional feature space using the projection matrix P, which has a dimension of 32×256 and is optimized using singular value decomposition. The adaptive weight adjustment mechanism is implemented based on a feature importance scoring algorithm. The specific steps are: calculating the mutual information value between each feature and the quality indicator; evaluating the importance score of each feature using the random forest algorithm; combining the stability scores of the features from historical data; and obtaining a comprehensive importance score through weighted averaging. Initial weights are set based on expert experience: 0.3 for temperature, 0.25 for pressure, 0.35 for concentration, and 0.1 for diffusion coefficient. The weight update frequency is once an hour, and the update range is limited to ±20% of the original weight to ensure system stability.

[0077] The implementation details of the Kalman filter are as follows: the state vector x contains 256 eigenvalues, and the observation vector z is the mapped eigenvalues ​​of the current sensor readings. The state transition matrix A is obtained through principal component analysis of historical data and has a size of 256×256, capturing the dynamic characteristics of the system. The control input matrix B takes into account the impact of production parameter adjustments and has a size of 256×16, corresponding to 16 controllable production parameters. The process noise covariance matrix Q is set to a diagonal matrix with diagonal elements of 0.01, representing the uncertainty of the system state prediction. The observation matrix H is the identity matrix, indicating that the states in the feature space are directly observable. The observation noise covariance matrix R is a diagonal matrix with diagonal elements of 0.05, representing the uncertainty of the sensor measurements. The Kalman filter recursive process consists of a prediction step (calculating the prior state estimate and the prior error covariance) and an update step (calculating the Kalman gain, the posterior state estimate, and the posterior error covariance). The filtering process is executed every 0.1 seconds to match the sensor sampling frequency. The filtered state vector is used for subsequent quality assessment.

[0078] The calculation process for the distribution confidence interval is as follows: For each quality indicator, its distribution characteristics are tested based on historical data. For indicators that follow a normal distribution, the confidence interval is calculated as the mean ± 1.96 times the standard deviation (95% confidence level). For indicators that do not follow a normal distribution, the quantile method is used, with the 2.5% and 97.5% quantiles as the confidence interval boundaries. Different transformation methods are used for different indicators to approximate the normal distribution, such as using a logarithmic transformation for skewed concentration data and a mixed Gaussian model for bimodal temperature uniformity data. The confidence interval calculation period is updated every 24 hours using a sliding window approach that covers the last 7 days of historical data.

[0079] Individual indicator scores are calculated using a piecewise function: When the observed value falls within the 95% confidence interval, the score S is calculated as 80 + (100 - 80) × (1 - |x - μ| / (1.96σ)), where x is the observed value, μ is the mean, and σ is the standard deviation. When the observed value falls within the 95%-99% confidence interval (i.e., between μ ± 1.96σ and μ ± 2.58σ), the score is calculated as 60 + (80 - 60) × (1 - |x - μ| - 1.96σ) / (2.58σ - 1.96σ). When the observed value falls outside the 99% confidence interval, the score is calculated as max(0, 60 - (|x - μ| - 2.58σ) × 5 / σ). Different scoring function slopes are set for different indicator types, with a steeper score decay curve used for key quality indicators.

[0080] The calculation process of the comprehensive evaluation index of quality status is as follows: multiply the scores of each individual indicator by the corresponding weight coefficient and then sum them up. For example, in a certain silicon wafer preparation batch, the temperature index score is 95 points (weight 0.3), the pressure index score is 85 points (weight 0.25), the concentration index score is 90 points (weight 0.35), and the diffusion coefficient index score is 88 points (weight 0.1). The comprehensive evaluation index is 95×0.3+85×0.25+90×0.35+88×0.1=90.05 points. The evaluation index is calculated in real time and the update frequency is synchronized with the status prediction. When the index is below 80 points, a yellow warning is triggered, when it is below 70 points, an orange warning is triggered, and when it is below 60 points, a red warning is triggered. The warning information includes a list of abnormal indicators, a quantitative value of the degree of deviation, a development trend forecast, and recommended intervention measures.

[0081] In practical application, this method has achieved a 95.7% detection rate for quality anomalies in silicon wafer raw materials, with a missed detection rate below 2.1%. The average lead time for early warning is 4.5 hours, providing ample time for production adjustments. After six months of stable operation, the batch yield rate for silicon wafer materials increased from 94.6% to 98.2%, reducing material loss by 3.6% and achieving annual cost savings of approximately 3.8 million yuan.

[0082] Figure 2 This is a bar chart comparing the quality status monitoring performance of the raw material digital twin model according to an embodiment of the present invention:

[0083] This figure compares the performance of three different modeling approaches on key quality monitoring indicators. The traditional physical model is a deterministic model based on classical physical laws and engineering experience; the general data-driven model uses traditional machine learning methods such as regression analysis and neural networks for modeling; and the digital Li Sheng model (this solution) is a hybrid modeling approach that integrates physical mechanisms and data-driven modeling, achieving dynamic mapping between the physical world and the digital space through digital twin technology.

[0084] Looking at five key performance indicators: In terms of quality prediction accuracy, the digital twin model achieved 94.3%, significantly outperforming the 78.5% of traditional physical models and 85.7% of general data-driven models. In terms of anomaly detection sensitivity, the digital twin model achieved 91.8%, surpassing the 72.3% of traditional physical models and 83.6% of general data-driven models. In real-time response performance testing, the digital twin model achieved 92.5%, far exceeding the 65.8% of traditional physical models and 87.9% of general data-driven models. In model robustness assessment, the digital twin model maintained 93.6% stability, outperforming the 81.2% of traditional physical models and 79.5% of general data-driven models. In terms of comprehensive monitoring efficiency, the digital twin model achieved 93.1%, also surpassing the 74.6% of traditional physical models and 84.2% of general data-driven models. The data demonstrates that by integrating physical laws and data features, the digital twin model demonstrates significant advantages across all performance indicators, with particularly significant improvements in accuracy, responsiveness, and stability.

[0085] In an optional embodiment, the raw material digital twin model is used to perform dimensionality reduction optimization on the raw material quality state, and the data coding structure is constructed using the raw material digital twin model, including:

[0086] Constructing a digital twin feature matrix, wherein the digital twin feature matrix includes multiple feature vectors, each of which corresponds to a feature dimension of the raw material quality status; inputting the feature vectors in the digital twin feature matrix into a cognitive hierarchy structure, wherein the cognitive hierarchy structure includes multiple cognitive layers, each of which is assigned a corresponding cognitive weight, and weighting the feature vectors using the cognitive weights to obtain cognitive features corresponding to each cognitive layer;

[0087] Performing feature fusion on the cognitive feature and the digital twin feature matrix, wherein the feature fusion includes multiplying the cognitive feature and the digital twin feature matrix by corresponding fusion weight coefficients and then superimposing them to generate a fusion feature;

[0088] Inputting the fused features into a coding mapping module, the coding mapping module including a coding weight matrix and a bias vector, performing a nonlinear transformation on the fused features using the coding weight matrix and the bias vector to obtain coding features; performing an EEG spectrum analysis on the coding features, calculating the spectral density function of the coding features using Fourier transform, and obtaining spectrum enhancement coding;

[0089] The spectrum enhancement coding is optimized based on a dynamic memory update mechanism, wherein the dynamic memory update mechanism includes a memory state and a forgetting factor, wherein the memory state attenuates the historical memory state through the forgetting factor and performs a weighted combination with the current spectrum enhancement coding to obtain a data coding structure.

[0090] like Figure 3 As shown, the method further includes:

[0091] Constructing the digital twin feature matrix is ​​the starting point for dimensionality reduction optimization. This feature matrix has a size of m×n, where m represents the number of sampled data points and n represents the number of feature dimensions. In practical applications, such as semiconductor silicon wafer quality monitoring scenarios, m typically ranges from 10,000 to 50,000, corresponding to the time sampling points in the production process; n typically ranges from 128 to 512, corresponding to the extracted quality-related features. Each feature vector corresponds to a characteristic dimension of the raw material quality state, such as material purity, surface roughness, and lattice defect density. The feature vectors are derived by mapping the sensor data of the digital twin model. The least squares method is used to fit the relationship between the sensor data and the theoretical model, with a fitting accuracy of over 95%.

[0092] The eigenvectors in the digital twin feature matrix are input into the cognitive hierarchy for processing. The cognitive hierarchy consists of three cognitive layers, corresponding to low-level perceptual features, mid-level structural features, and high-level semantic features. The low-level perceptual feature layer contains 64 neurons, uses the ReLU activation function, and has a cognitive weight matrix size of n×64. The mid-level structural feature layer contains 32 neurons, uses the Tanh activation function, and has a cognitive weight matrix size of 64×32. The high-level semantic feature layer contains 16 neurons, uses the Sigmoid activation function, and has a cognitive weight matrix size of 32×16. The cognitive weights of each layer are optimized using the backpropagation algorithm, with a learning rate of 0.005, a momentum factor of 0.9, and a weight decay coefficient of 0.0001. Batch normalization is used in the training process, with a batch size of 128 and 5000 training iterations. The eigenvectors are weighted by the cognitive weights to obtain the cognitive features corresponding to each cognitive layer. Taking the intermediate structure feature layer as an example, the input is the 64-dimensional output of the low-level perception feature layer. After linear transformation of the 32×64 weight matrix and the 32-dimensional bias vector, and then through the Tanh activation function, the 32-dimensional intermediate structure feature is finally obtained.

[0093] Feature fusion is performed on the cognitive features and the digital twin feature matrix. The fusion process utilizes an adaptive weighting mechanism, dynamically adjusting the fusion weights based on the amount of feature information. The fusion weight coefficient α for the digital twin feature matrix is ​​initially set to 0.6, and the fusion weight coefficient β for the cognitive features is initially set to 0.4. The weight coefficients are dynamically adjusted using the information entropy criterion. When the information entropy of the cognitive features is higher than that of the digital twin feature matrix, the β value is increased; otherwise, the α value is increased. The adjustment step size is set to 0.05, and the weight range is limited to the interval [0.2, 0.8]. Feature fusion is implemented by multiplying the cognitive feature matrix and the digital twin feature matrix by their corresponding fusion weight coefficients and then performing element-wise addition to generate the fused feature matrix. In the semiconductor silicon wafer example, the cognitive feature dimension is 16, and the digital twin feature dimension is 128. The fused feature dimension is 128. Dimension alignment technology is used to expand the 16-dimensional cognitive features to 128 dimensions.

[0094] The fused features are input into the encoder-to-decoder module for processing. The encoder-to-decoder module uses a variational autoencoder architecture and consists of an encoder and a decoder. The encoder consists of a three-layer fully connected neural network, with 64 and 32 hidden layer neurons, respectively, and a 16-dimensional latent space vector as the output layer. The encoding weight matrices are 128×64, 64×32, and 32×16, respectively, with corresponding bias vector dimensions of 64, 32, and 16. The activation function uses LeakyReLU with a negative slope parameter set to 0.2. The encoder performs a nonlinear transformation on the fused features, compressing and mapping the 128-dimensional features into a 16-dimensional latent space to obtain the encoded features. The encoder-to-decoder module is trained using a combination of a reconstruction loss and a KL divergence loss, with the reconstruction loss weight set to 0.8 and the KL divergence loss weight set to 0.2. Adam is used as the optimizer, with an initial learning rate of 0.001. Training is repeated for 100 epochs, each containing 200 batches of data.

[0095] The encoded features are subjected to EEG spectrum analysis. This step, inspired by the cognitive mechanisms of the human brain, enhances feature representation capabilities through frequency domain analysis. Specifically, each element in the 16-dimensional encoded feature vector is treated as a time series signal, and its spectrum is calculated using the Fast Fourier Transform (FFT) algorithm. The transform window length is set to 16, and a Hanning window is used to reduce spectral leakage. The calculated spectral density function contains both amplitude and phase information, with the amplitude reflecting the signal energy distribution and the phase reflecting the signal structure. The spectrum is divided into five frequency bands: δ (0.5-4 Hz), θ (4-8 Hz), α (8-13 Hz), β (13-30 Hz), and γ (>30 Hz), corresponding to different information characteristics. The energy contribution of each frequency band is calculated to construct a band energy feature vector. Finally, the band energy feature vector is concatenated with the original encoded features to form a 24-dimensional spectrum-enhanced encoding. In practical applications, spectral analysis significantly improves the model's ability to identify periodic mass fluctuations, increasing detection accuracy from 87.3% to 94.1%.

[0096] Spectral enhancement coding is optimized based on a dynamic memory update mechanism. This mechanism simulates the long-term and short-term memory characteristics of the human brain and consists of two key components: a memory state matrix M and a forgetting factor γ. The memory state matrix M has dimensions of k×24, where k represents the memory capacity and is set to 64, indicating the storage of 64 historical feature prototypes. The forgetting factor γ ranges from [0, 1] and is initially set to 0.9 to control the degree of historical memory retention. The memory update process is as follows: First, the similarity between the current spectral enhancement code and each prototype in the memory state matrix is ​​calculated using the cosine similarity metric. Then, the most matching memory prototype is selected based on the similarity and the forgetting factor γ is applied to it, achieving exponential decay of historical memory. Finally, the decayed memory prototype is proportionally fused with the current spectral enhancement code to update the memory state. The fusion ratio λ between the memory and the current code is adaptively adjusted based on the similarity. The higher the similarity, the larger the λ value, and the range is controlled to [0.3, 0.7]. After the dynamic memory update, the final 24-dimensional data coding structure is obtained.

[0097] In an actual application case, an integrated circuit company used the above method to monitor the quality of the wafer manufacturing process. The original feature dimension was 512. After dimensionality reduction optimization, a 24-dimensional data encoding structure was obtained, and the information retention rate reached 94.2%. The feature calculation efficiency after dimensionality reduction increased by 21 times, and the model training time was reduced from the original 47 hours to 2.2 hours. In the quality anomaly detection task, the monitoring model constructed based on the optimized encoding structure had a sensitivity of 95.3%, a specificity of 97.1%, and an F1 score of 0.962, which was better than the model constructed with the unoptimized original features (F1 score of 0.887). In particular, for slight quality fluctuations in low signal-to-noise ratio environments, the detection capability was significantly improved, and the success rate increased from 71.2% to 89.5%.

[0098] This implementation method achieves efficient dimensionality reduction and optimization of the quality status of raw materials through six key steps: digital twin feature matrix construction, cognitive hierarchy processing, feature fusion, coding mapping, brain wave spectrum analysis and dynamic memory update, and constructs a data coding structure with strong expressive ability, providing a solid foundation for subsequent quality status assessment and early warning.

[0099] In an optional embodiment, using the raw material digital twin model to perform quality status assessment and generate a quality assessment early warning signal includes:

[0100] Utilizing the hierarchical attention mechanism of the raw material digital twin model, spatial and temporal features are extracted from the real-time data of the raw material quality status to obtain a quality status feature vector with spatiotemporal weights;

[0101] Inputting the quality status feature vector into a dual time memory network, extracting short-term dynamic features and long-term evolution laws through fast memory units and slow memory units respectively, and generating a quality status evaluation index based on the combined relationship between the two;

[0102] According to the historical distribution of the quality status evaluation index, a dynamic threshold optimization mechanism is adopted to calculate the alarm threshold, and a quality evaluation early warning signal is generated by comparing the deviation between the quality status evaluation index and the alarm threshold.

[0103] Build a digital twin model of raw materials to realize real-time monitoring and evaluation of the quality status of raw materials, and then generate quality assessment early warning signals.

[0104] When using a digital twin model of raw materials for quality assessment, the system employs a hierarchical attention mechanism to extract spatial and temporal features from real-time raw material quality data. This hierarchical attention mechanism comprises a spatial attention layer and a temporal attention layer. The spatial attention layer primarily processes the correlation between data collected by sensors placed at different locations on the raw material, while the temporal attention layer processes the dynamic characteristics of the time series. For example, assuming that 10 sensors are placed on the surface of steel, each collecting five parameters such as temperature, hardness, and stress, this forms a spatial feature matrix with a shape of (10, 5). For each feature point, the system calculates its correlation with the other feature points and determines a weight coefficient. For example, if the temperature parameter at the first sensor point has a high correlation with the stress parameter at the third sensor point, it is assigned a higher weight of 0.85; if the correlation with the hardness parameter at the eighth sensor point is low, it is assigned a lower weight of 0.12. These weights are used to adjust the importance of the original features, forming a weighted spatial feature.

[0105] The temporal attention layer processes 24 consecutive hours of sampled data. For each point in time, the system assesses its impact on the current state. For example, if the temperature of a raw material rises rapidly within the past four hours, the system assigns a higher weight of 0.78 to the temperature data during this period, while assigning a lower weight of 0.05 to the data from 24 hours ago. In this way, the system captures key patterns of change along the time dimension. Ultimately, spatial and temporal features are weighted and fused to form a 128-dimensional quality state feature vector, each element of which contains comprehensive information from both the temporal and spatial dimensions.

[0106] After inputting the quality state feature vector into the dual-time memory network, the system can simultaneously capture short-term dynamic characteristics and long-term evolution patterns. The dual-time memory network consists of fast memory units and slow memory units. The fast memory unit uses a forget gate control mechanism to process data from the last 30 minutes and quickly respond to sudden changes in the raw material state. For example, when the temperature of steel fluctuates dramatically during processing, suddenly rising from the standard temperature of 750°C to 830°C, the fast memory unit can quickly capture this change pattern. This unit contains 64 neurons, extracts short-term features through a three-layer structure, and outputs a 64-dimensional short-term feature vector.

[0107] The slow memory unit focuses on long-term trends in raw material quality, processing historical data from the past seven days to identify slowly evolving patterns of quality degradation. For example, the slow oxidation of steel in the environment causes a decrease in hardness, from an initial hardness of HRC 58 to HRC 55.5. This small but persistent change is captured by the slow memory unit. This unit uses 32 neurons, retains long-term information through a memory update mechanism, and outputs a 32-dimensional long-term feature vector.

[0108] The system combines the outputs of fast and slow memory units through a feature fusion layer. Using an adaptive weighting scheme, the system assigns a higher weight of 0.7 to long-term features and 0.3 to short-term features in a stable state. When anomalies are detected, the system adjusts the weighting to 0.8 for short-term features and 0.2 for long-term features. This fusion generates a comprehensive quality status assessment index ranging from 0 to 100, with 80-100 indicating good quality, 60-80 indicating normal quality but requiring attention, 40-60 indicating minor anomalies, and 0-40 indicating serious quality issues.

[0109] Based on the historical distribution of quality status assessment indicators, the system uses a dynamic threshold optimization mechanism to calculate alarm thresholds. This mechanism first calculates the distribution characteristics of quality status assessment indicators over the past 90 days, including mean, standard deviation, kurtosis, and skewness. For raw materials in stable production, the system obtains distribution characteristics with a mean of 78.6 and a standard deviation of 5.2. Based on these statistics, the system uses an adaptive algorithm to calculate the alarm threshold, typically set to the mean minus 2.5 times the standard deviation (65.6 in this case). The system also dynamically adjusts the threshold sensitivity based on factors such as seasonal production changes and equipment maintenance cycles. For example, during the stable period after equipment maintenance, the threshold coefficient can be adjusted from 2.5 to 2.2, raising the threshold to 67.2 to reduce false alarms.

[0110] When the real-time calculated quality status assessment indicator falls below the alarm threshold, the system generates a quality assessment early warning signal. Early warning signals are divided into three levels: prompt, warning, and emergency, corresponding to varying degrees of threshold deviation. For example, when the indicator value reaches 64.5, slightly below the threshold of 65.6, the system generates a "prompt" level warning; when the indicator value drops to 58.2, the system upgrades to the "warning" level; and if the indicator further drops to 45.0, an "emergency" alert is triggered. Each early warning signal includes information such as the time of the anomaly, the type of anomaly, the influencing parameters, and a recommended solution. This information is directly pushed to the quality management system and the mobile terminals of relevant operators, enabling early warning and intervention of raw material quality issues.

[0111] In an optional embodiment, based on the quality assessment warning signal, a biologically inspired chaotic immune optimization algorithm is deployed to construct a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, rapid location and traceability of quality anomalies are achieved, including:

[0112] Inputting the quality assessment warning signal into a chaotic immune model, the chaotic immune model uses iterative operations to generate an initialization sequence, constructing an agent population based on the initialization sequence, and optimizing the distribution of the agent population through the chaotic immune model. The chaotic immune model includes a warning signal strength parameter and a temperature adjustment parameter, and the agent distribution corresponding to the warning signal strength is obtained by dynamically adjusting the warning signal strength parameter and the temperature adjustment parameter;

[0113] Based on the agent distribution, the chaotic immune model is used to calculate the response strength between the warning signal characteristics and the abnormal template characteristics. The chaotic immune model constructs an ergodic search sequence based on the response strength, and adaptively weights the search sequence with the historical search position to update the current search path and determine the evolution trajectory of the abnormal characteristics.

[0114] An intelligent agent collaborative network is constructed. The intelligent agent collaborative network is based on the spatial distance relationship between intelligent agents. The interaction strength between intelligent agents is constructed through the chaotic immune model. The interaction strength is weightedly fused with the local tracing results of each intelligent agent, where the local tracing results include the location of the anomaly and the propagation path information. The tracing decision result of the global quality anomaly is generated through the weighted fusion.

[0115] The quality assessment warning signal is input into the chaos immune model for processing. The quality assessment warning signal contains information such as anomaly type, anomaly intensity, occurrence time, and impact range. The warning signal is represented as a vector with 24 dimensions, where the first 8 dimensions represent the anomaly type, the middle 8 dimensions represent anomaly intensity and time information, and the last 8 dimensions represent the impact range. The chaos immune model first uses a logistic mapping to generate an initialization sequence, with the mapping parameter r set to 3.9 to ensure the system is in a chaotic state. The initial seed value x0 is set to 0.4, and the last 100 values ​​after 200 iterations are used as the initialization sequence. Based on the initialization sequence, a population of 200 agents is constructed, with each agent representing a candidate solution in the solution space.

[0116] The agent encoding uses real number encoding with a length of 32. The first 24 dimensions correspond to the weights of the various dimensions of the warning signal, and the last 8 dimensions correspond to the traceability strategy parameters. The agent population is optimally allocated using a chaotic immune model. Two key parameters are considered in the allocation process: the warning signal strength parameter α and the temperature adjustment parameter β. The warning signal strength parameter α ranges from [0.1 to 0.9], with an initial value of 0.5, and increases with increasing warning signal strength. The temperature adjustment parameter β ranges from [0.01 to 1.0] and is initially set at 0.8, controlling the diversity of the agent distribution.

[0117] During the optimization allocation process, for high-intensity warning signals (such as abnormality intensity exceeding 75 points), the α value is increased to 0.7-0.9 and the β value is reduced to 0.3-0.5, making the distribution of agents more concentrated in high-risk areas; for low-intensity warning signals (such as abnormality intensity below 50 points), the α value is reduced to 0.2-0.4 and the β value is increased to 0.6-0.8, making the distribution of agents more uniform and expanding the search range. By dynamically adjusting the warning signal intensity parameters and temperature adjustment parameters, the distribution of agents corresponding to the warning signal intensity is obtained. In actual application, for a temperature abnormality warning signal (intensity of 82 points) of a semiconductor production line, through parameter setting (α=0.8, β=0.4), 65% of the agents are allocated to the temperature control system related area, 25% of the agents are allocated to the upstream material supply link, and 10% of the agents are allocated to the environmental impact factor area.

[0118] Based on the distribution of intelligent agents, a chaotic immune model is used to calculate the response strength between the warning signal characteristics and the anomaly template characteristics. The anomaly template feature library contains feature vectors of historical anomaly cases, with a total of 128 templates covering 12 common quality anomalies. The response strength is calculated using the cosine similarity metric, taking into account both the angular and amplitude differences of the feature vectors. The response strength R ranges from [0 to 1], with larger R values ​​indicating greater similarity between the current warning signal and the template characteristics.

[0119] The chaotic immune model constructs an ergodic search sequence based on response intensity. The search sequence is generated using the Tent mapping, with the control parameter μ set to 1.99. The initial point is taken from the eigenvectors corresponding to the top three templates with the highest response intensity. The search sequence length is set to 300, and its value range is the boundary of the problem solution space. The search sequence is adaptively weighted and combined with the historical search positions. The combination weight λ is dynamically adjusted based on the historical search results. The value of λ ranges from [0.2 to 0.8] and is initially set to 0.5. When the historical search path finds an area of ​​high response intensity, the value of λ is increased to enhance the utilization of historical experience. When the historical search path reaches a local optimum, the value of λ is decreased to enhance the randomness of the chaotic search.

[0120] The current search path is updated through an adaptive weight combination. After each update, the response strength of the new path is evaluated, and the top 20 positions with the highest response strength are retained. After 300 iterations, the evolution trajectory of the abnormal feature is determined. In a case of photoresist quality anomaly, this method traced the complete evolution of the abnormal feature, starting from fluctuations in the supplier's raw material purity, through deviations in the solvent ratio, and ultimately leading to abnormal photoresist viscosity. The peak response strength was 0.87, and the tracing accuracy rate reached 92%.

[0121] An agent collaborative network is constructed to achieve multi-agent collaborative traceability. The agent collaborative network is established based on the spatial distance relationship between agents. Spatial distance is calculated using Euclidean distance, and the distance threshold dth is set to 25% of the solution space diameter. When the distance between two agents is less than dth, a connection is established, forming a dynamic network with a variable topology. The interaction strength between agents is constructed using a chaotic immune model. The interaction strength S is inversely proportional to the distance between agents and directly proportional to the agent's fitness (response strength). In the specific calculation formula, the distance attenuation coefficient is set to 2.0, and the fitness gain coefficient is set to 1.5. The interaction strength S ranges from [0, 1], with larger S values ​​indicating stronger collaborative effects between the two agents. The interaction strength is weightedly fused with the local traceability results of each agent. The local traceability results include the coordinates of the anomaly occurrence location (location in the production process parameter space) and propagation path information (represented as a directed graph structure with nodes representing key process parameters and edges representing influencing relationships).

[0122] Weighted fusion uses a soft voting mechanism, where the voting weight of each agent is proportional to its response strength and the average interaction strength with other agents. Weighted fusion generates a global quality anomaly traceability decision result, including the most accurate anomaly source location (confidence score), the critical path for anomaly propagation, and a predicted impact range. In actual application, an integrated circuit manufacturer used this method to trace the anomaly of wafer oxide layer thickness. After 500 iterations, 200 agents successfully identified the oxidation furnace temperature controller parameter offset as the root cause, with a confidence score of 0.89. The traceability time was shortened from 4 hours using traditional methods to 22 minutes.

[0123] The specific implementation of the chaotic immune optimization algorithm includes the following key technical points: The clonal selection operation simulates the clonal proliferation process after antibodies recognize antigens in the biological immune system. The number of clones is proportional to the agent's response strength, with an upper limit of 10. The clone ratio coefficient is set to 0.8, with 10 clones for the agent with the highest response strength and 1 clone for the agent with the lowest response strength. The high-mutation operation applies large-scale mutations to low-response agents, with a mutation ratio of 0.3 and a mutation amplitude of 30% of the solution space diameter. The low-mutation operation applies small-scale, fine-grained mutations to high-response agents, with a mutation ratio of 0.1 and a mutation amplitude of 5% of the solution space diameter.

[0124] An immune memory mechanism stores historical optimal solutions, with a memory bank capacity of 20 and updated every 10 iterations. The new solution replacement strategy uses a combination of elite retention and roulette wheel, with a retention rate of 0.2. A chaotic perturbation mechanism is triggered when the algorithm reaches a local optimum. This is triggered when the optimal solution improves by less than 0.1% for 30 consecutive iterations. The perturbation sequence is generated using a logistic map, and the perturbation intensity decreases with the number of iterations. Initially, the perturbation intensity is 20% of the solution space diameter and eventually decreases to 2%. Convergence is criterion when the optimal solution improves by less than 0.01% for 50 consecutive iterations or when the maximum number of iterations, 1000, is reached.

[0125] The multi-agent collaborative traceability system adopts a layered architecture: the perception layer consists of 200 sensor nodes, responsible for real-time data collection; the analysis layer includes a central processing unit and edge computing nodes, responsible for anomaly feature extraction and pattern recognition; the decision-making layer consists of a chaotic immune optimization algorithm and an expert knowledge base, responsible for generating traceability strategies; the execution layer includes a feedback control module, responsible for implementing quality measures. The system communicates using the Industrial Ethernet protocol, with a data transmission rate of 100Mbps and latency controlled within 10ms. The system's fault-tolerance mechanism includes data redundancy storage and a backup decision-making module. When the response time of the main system exceeds a set threshold (200ms), it automatically switches to the backup system. System security is ensured through data encryption (AES-256 algorithm) and access control.

[0126] In a specific application at a semiconductor wafer manufacturing company, the system successfully handled 93 quality anomaly incidents, achieving an average traceability accuracy of 91.7% and an average traceability time of 15.4 minutes, representing a 76% improvement over traditional methods. In particular, the recognition rate for complex, multi-factor coupled anomalies increased from 65% to 87%. After six months of operation, the company's product quality first-pass rate increased by 4.2 percentage points, quality costs decreased by approximately 15%, and annualized cost savings reached approximately 5.8 million yuan.

[0127] Figure 4 This is a schematic diagram comparing the accuracy of quality anomaly tracing using different traceability methods according to an embodiment of the present invention:

[0128] This figure compares the average tracing time of three different traceability methods for anomalies of varying complexity. Traditional manual traceability relies entirely on human experience and manual analysis; general intelligent traceability algorithms use basic machine learning models such as decision trees and random forests for anomaly analysis; and this technical solution is an intelligent traceability system based on chaos immune optimization. Test results for anomaly scenarios ranging from level 1 to level 6 show that at level 1, the three methods took 70.8 minutes, 41.5 minutes, and 10.3 minutes, respectively. As complexity increased to level 3, the times increased to 142.6 minutes, 75.3 minutes, and 21.5 minutes, respectively. At the highest level of complexity, level 6, traditional manual traceability took 312.8 minutes, general intelligent traceability methods took 168.7 minutes, and this technical solution only took 42.5 minutes. The data shows that this technical solution maintains a significant time advantage at all complexity levels, and its relative advantage becomes more pronounced as the complexity of the anomaly increases. Especially in high-complexity scenarios (level 5-6), the traceability time growth trend of this technical solution is significantly slowed down, reflecting excellent scalability and robustness.

[0129] In an optional embodiment, based on the quality assessment warning signal, a biologically inspired chaotic immune optimization algorithm is deployed to construct a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, rapid location and traceability of quality anomalies are achieved, including:

[0130] The quality assessment warning signal is input into a chaotic immune optimization algorithm, which constructs an antigen-antibody response mechanism by calculating the affinity between the quality assessment warning signal and the abnormal template, determines the number of clones based on the affinity and generates an initial immune population, and uses a chaotic sequence to perturb and update the initial immune population, wherein the chaotic sequence is weightedly combined with the historical optimal antibody to determine the search direction, thereby obtaining an optimized immune population;

[0131] A multi-agent collaborative traceability system is constructed based on the optimized immune population, and the antibody characteristics in the optimized immune population are mapped to the initial state of the agent. The agent explores abnormal characteristics in the quality state space through a chaotic search strategy, wherein the chaotic search strategy includes two stages: local fine search and global coarse search, and dynamically adjusts the search step size to achieve rapid positioning of abnormal areas;

[0132] An agent interaction network is constructed in the multi-agent collaborative traceability system, and an interaction weight matrix is ​​calculated based on the spatial distance between agents. The interaction weight matrix is ​​weightedly fused with the abnormal feature information collected by each agent. A propagation path map of quality anomalies is constructed based on the fused abnormal feature information, and the source position and diffusion direction of the quality anomaly are determined by tracing back the propagation path map.

[0133] The quality assessment warning signal is input into the chaotic immune optimization algorithm for processing. The quality assessment warning signal is represented as a 32-dimensional vector, consisting of an anomaly type code (8 dimensions), an anomaly intensity index (6 dimensions), a temporal feature (10 dimensions), and a spatial distribution feature (8 dimensions). The chaotic immune optimization algorithm constructs an antigen-antibody response mechanism by calculating the affinity between the quality assessment warning signal and the anomaly template. This affinity is calculated using the cosine similarity metric, which ranges from [0, 1], with higher values ​​indicating a higher degree of match. The anomaly template library contains 200 predefined templates covering 15 common quality anomaly patterns. When calculating the affinity, different feature dimensions are assigned different weights: anomaly type is weighted 0.4, anomaly intensity is weighted 0.3, temporal features are weighted 0.2, and spatial distribution features are weighted 0.1. In a real-world case study from a semiconductor lithography process, the affinity of a warning signal with the "photoresist uniformity anomaly" template was 0.87, and with the "incomplete development" template was 0.34. The affinity with all other templates was below 0.3.

[0134] The number of clones is determined based on affinity and the initial immune population is generated. The number of clones is proportional to the affinity, and the calculation formula is that the number of clones is equal to the maximum number of clones (set to 10) multiplied by the affinity, and the result is rounded up. There are 9 clones of antibodies with an affinity of 0.87, 4 clones of antibodies with an affinity of 0.34, and the number of clones of the remaining antibodies is determined according to their respective affinities. The size of the initial immune population is set to 100, including antibodies and their clones selected from the template library.

[0135] Each antibody in the population is represented as a 32-dimensional vector, matching the dimensionality of the warning signal. A chaotic sequence is used to perturb and update the initial immune population. The chaotic sequence is generated using a logistic mapping, with the control parameter r set to 3.99 and the initial value x0 set to 0.63. After 200 iterations, the last 100 values ​​are taken as the chaotic sequence. The chaotic sequence ranges from [0, 1] and is mapped to the actual value range of each parameter through a linear transformation. A weighted combination of the chaotic sequence and the historically optimal antibody determines the search direction. The combination weight α is dynamically adjusted with the number of iterations, starting at 0.3 and gradually increasing to 0.8 to enhance the utilization of historical experience. During the perturbation update process, the update amplitude of each antibody is inversely proportional to its affinity. Antibodies with high affinity undergo a small, refined search, while antibodies with low affinity undergo a large, exploratory search. After 100 iterations, the optimized immune population is obtained, with the average affinity of the population increasing from an initial 0.42 to 0.78.

[0136] A multi-agent collaborative traceability system is constructed based on the optimized immune population. The antibody characteristics in the optimized immune population are mapped to the initial state of the agent. During the mapping process, the 32-dimensional feature vector of the antibody is converted into the state vector of the agent, which contains attributes such as position coordinates (coordinates in the quality state space), search direction, movement speed, and perception range. The total number of agents is 100, which is consistent with the size of the immune population. The agent explores abnormal characteristics in the quality state space through a chaotic search strategy. The quality state space is an 8-dimensional space that represents the state combination of key quality parameters. The chaotic search strategy includes two stages: local fine search and global coarse search.

[0137] During the local fine search phase, the agents search high-affinity regions with small step sizes, set at 1% to 5% of the spatial diameter. During the global coarse search phase, the agents explore low-affinity regions with large step sizes, set at 10% to 30% of the spatial diameter. The search step size is dynamically adjusted during the search process, using the formula: the current step size equals the base step size multiplied by the chaos factor, multiplied by the inverse function of the affinity. The base step size decreases with the number of iterations, initially at 20% of the spatial diameter and ultimately decreasing to 2%. The chaos factor is generated using Tent mapping, with the control parameter μ set to 1.97. In the case of photoresist uniformity anomaly, through dynamic step size adjustment, the agent swarm successfully located the anomaly region after 50 iterations, achieving a localization accuracy of 97.3%.

[0138] An agent interaction network is constructed within the multi-agent collaborative traceability system. The interaction weight matrix is ​​calculated based on the spatial distance between agents, using Euclidean distance. The interaction weight is inversely proportional to the distance, calculated as the distance threshold divided by the sum of the actual distance and the distance threshold. The distance threshold is set to 15% of the spatial diameter. When the distance between two agents is less than the distance threshold, the interaction weight is greater than 0.5; when the distance is equal to the distance threshold, the interaction weight is equal to 0.5; when the distance is greater than the distance threshold, the interaction weight is less than 0.5 but always greater than 0.

[0139] The interaction weight matrix is ​​a 100×100 square matrix with diagonal elements set to 1, representing the interaction weights between the agent and itself. This interaction weight matrix is ​​weighted and fused with the anomaly feature information collected by each agent. This anomaly feature information includes data such as parameter deviation, rate of change, and fluctuation frequency. Weighted fusion uses a weighted average approach. The fused anomaly feature is represented as the weighted sum of all agent features, with the weight being the corresponding interaction weight.

[0140] A propagation path graph of quality anomalies is constructed based on the fused anomaly feature information. The propagation path graph adopts a directed graph structure, where nodes represent quality parameters, edges represent the influence relationship between parameters, and edge weights represent the intensity of influence. The graph construction process is based on a causal inference algorithm, and the causal direction is determined by analyzing the temporal correlation of parameter changes. The importance of nodes is evaluated by centrality indicators, including degree centrality, closeness centrality, and betweenness centrality, with weights of 0.3, 0.3, and 0.4, respectively. The source location and diffusion direction of quality anomalies are determined by backtracking the propagation path graph. The backtracking algorithm starts from the abnormal manifestation node and searches backward along the input edge direction, calculating the cumulative weight of each path and selecting the path with the largest cumulative weight as the most accurate abnormal propagation path. The source location is the starting node of the path, and the diffusion direction is determined by the direction of the path.

[0141] In actual application at a certain semiconductor manufacturing plant, this method was used to trace the source of photoresist uniformity anomalies. By analyzing the tracing results, it was determined that the source of the anomaly was the parameter drift of the temperature control unit of the photoresist supply system, and the cumulative path weight was 0.86, which was significantly higher than the suboptimal path of 0.52. The direction of the anomaly diffusion was temperature control unit-photoresist viscosity-glue coating uniformity-graphic size deviation. The entire tracing process took 17 minutes, which is more than 85% more efficient than the traditional manual tracing method (usually takes 2-4 hours). Six months of operation data showed that the traceability accuracy of this method was 93.7%, much higher than the 78.2% of the traditional method. In particular, for multi-factor coupling anomalies, the accuracy improvement was more significant, from 65.4% to 89.1%.

[0142] The method has shown significant results in practice: after implementing it, one integrated circuit manufacturer saw its product yield increase by 3.5 percentage points, from 91.2% to 94.7%. It also reduced abnormality response time by an average of 75%, from 3.2 hours to 0.8 hours. Quality costs were reduced by approximately 12%, resulting in annual savings of 4.3 million yuan. This method is particularly suitable for high-precision, high-value semiconductor, precision optics, and advanced materials manufacturing, and is crucial for improving product quality stability and production efficiency.

[0143] Figure 5 This is a bar chart comparing the performance of the chaos immune optimization algorithm according to the embodiment of the present invention:

[0144] This figure compares the performance of three different traceability methods. Traditional methods, representing statistically-based backtracking algorithms, primarily rely on data statistics and correlation analysis for traceability. General intelligent algorithms employ traditional machine learning methods such as artificial neural networks and support vector machines. The Chaos Immune Optimization Algorithm (CAO) is a novel intelligent optimization algorithm that integrates biological immune mechanisms with chaotic dynamics. Looking at specific performance metrics, the Chaos Immune Optimization Algorithm achieved 93.8% traceability accuracy, significantly exceeding the 76.5% of traditional methods and the 85.7% of general intelligent algorithms. Its anomaly detection rate reached 95.2%, also surpassing the 82.3% of traditional methods and the 88.6% of general intelligent algorithms. In terms of average traceability time, the Chaos Immune Optimization Algorithm completed traceability in just 23.5 minutes, significantly improving time efficiency compared to the 168.4 minutes of traditional methods and the 68.2 minutes of general intelligent algorithms. The data demonstrates that the Chaos Immune Optimization Algorithm, by combining biologically-inspired mechanisms with chaotic search strategies, significantly improves traceability efficiency while maintaining high accuracy.

[0145] In an optional embodiment, the multi-agent collaborative traceability system is used to activate the intelligent traceability engine according to the quality assessment warning signal, locate the root cause of the quality anomaly based on the reasoning mechanism of the spatiotemporal graph, and generate the traceability analysis results including:

[0146] Activate the intelligent traceability engine based on the quality assessment warning signal, input the quality assessment warning signal into the spatiotemporal graph model for warning intensity analysis, perform threshold judgment on the warning intensity through the spatiotemporal graph model, and generate the engine activation state, wherein the spatiotemporal graph model includes sensitivity parameters and activation thresholds;

[0147] Constructing a correlation structure between time series and spatial topology, using the spatiotemporal graph model to calculate the temporal evolution characteristics of the quality status, and simultaneously using the spatiotemporal graph model to analyze the spatial position relationship, fusing the temporal evolution characteristics with the spatial position relationship to form a spatiotemporal correlation feature;

[0148] Constructing an anomaly propagation path based on the spatiotemporal correlation features, calculating the inter-node propagation strength using the spatiotemporal graph model, constructing an anomaly propagation network using the inter-node propagation strength as the node association weight, and identifying anomaly diffusion paths based on the anomaly propagation network;

[0149] The root cause of the abnormal diffusion path is located, the importance scores of the path nodes are calculated using the spatiotemporal graph model, the location of the abnormal source point is determined, and a tracing analysis result is generated based on the location of the abnormal source point and the abnormal diffusion path.

[0150] The process of activating the intelligent traceability engine based on quality assessment warning signals begins by converting the quality assessment warning signals into a standardized format. The warning signals are represented as 32-dimensional vectors, containing information about the anomaly type, severity, location, and timestamp. The standardized warning signals are then fed into the spatiotemporal graph model for warning intensity analysis. The calculation of warning intensity considers a weighted combination of anomaly severity, impact range, and duration. The weight for anomaly severity is set to 0.5, the impact range to 0.3, and the duration to 0.2.

[0151] A semiconductor manufacturing line detected an abnormal silicon wafer oxide thickness. The warning signal's anomaly severity was 78 points (out of 100). The impact affected 450 wafers across three production batches and lasted 2.5 hours. The calculated warning intensity was 78 × 0.5 + 85 × 0.3 + 65 × 0.2 = 76.5 points. A threshold for the warning intensity was determined using a spatiotemporal graph model. The spatiotemporal graph model contains two key parameters: the sensitivity parameter α and the activation threshold β. The sensitivity parameter α controls the model's response to the warning signal. Its value range is [0.8, 1.2], with a default value of 1.0. This parameter can be adjusted based on the production environment. For high-precision production lines, it is set to 1.1-1.2, and for standard production lines, it is set to 0.8-0.9. The activation threshold β is the critical warning intensity threshold that triggers the traceability engine. Its value range is [60, 80], with a default value of 70. When the value obtained by multiplying the warning intensity by the sensitivity parameter exceeds the activation threshold, the engine's activation state is set to "started"; otherwise, it is set to "standby." In the above example, the warning intensity of 76.5 multiplied by the sensitivity parameter 1.1 equals 84.15, which is greater than the activation threshold of 70, so the intelligent traceability engine is successfully activated.

[0152] The process of constructing the correlation structure between time series and spatial topology involves two parallel analytical paths. A spatiotemporal graph model is used to calculate the temporal evolution characteristics of quality status. Time series analysis employs a sliding window technique with a 24-hour window size and a one-hour sliding step. A time series feature extraction algorithm is applied to the quality data within the window to extract trend, periodicity, and mutation characteristics. Trend characteristics are described using the slope of a linear regression, periodicity characteristics are described using the main frequency components of the Fourier transform, and mutation characteristics are identified using a change point detection algorithm.

[0153] In the case of the silicon wafer oxide layer, time series analysis revealed that the oxide layer thickness began to show a slight downward trend (slope -0.5nm / hour) 12 hours before the anomaly, periodic fluctuations (cycles of approximately 45 minutes) occurred 4 hours before the anomaly, and a mutation point (thickness suddenly dropped 5nm) occurred 1 hour before the anomaly. A spatiotemporal graph model was also used to analyze spatial position relationships. Spatial analysis constructed a spatial topology network based on the production process flow chart, with nodes representing process sites or equipment and edges representing material flows or process dependencies. For each node, the process distance (number of process steps passed) and physical distance (actual spatial distance) to the anomaly detection point were calculated.

[0154] Spatial analysis identified key nodes associated with abnormal oxide layer thickness, including the oxidation furnace (process distance 0, physical distance 0 meters), the gas supply system (process distance 1, physical distance 15 meters), the temperature control unit (process distance 1, physical distance 5 meters), and the cleaning station (process distance 2, physical distance 30 meters). The temporal evolution features were fused with the spatial position relationship to form spatiotemporal correlation features. This fusion was performed using a weighted combination method, with a weight of 0.6 for temporal features and a weight of 0.4 for spatial features. The fused spatiotemporal correlation features are represented as a 64-dimensional vector, with the first 32 dimensions encoding temporal information and the second 32 dimensions encoding spatial information.

[0155] The process of constructing anomaly propagation paths based on spatiotemporal correlation features first calculates the inter-node propagation strength using a spatiotemporal graph model. This propagation strength calculation considers four factors: temporal precedence (the order in which parameter changes occur), correlation (the statistical correlation of parameter changes), physical dependency (causal relationships based on domain knowledge), and distance decay (the attenuation of influence based on spatial distance). The weights for each factor are 0.3, 0.3, 0.3, and 0.1, respectively.

[0156] For example, the propagation strength calculation between the temperature control unit and the oxide layer thickness: the timing leading score is 0.85 (temperature fluctuation leads the thickness change by about 40 minutes), the correlation score is 0.78 (Pearson correlation coefficient), the physical dependence score is 0.9 (based on semiconductor manufacturing expert knowledge), and the distance attenuation score is 0.8 (based on a physical distance of 5 meters). The comprehensive propagation strength is 0.85×0.3+0.78×0.3+0.9×0.3+0.8×0.1=0.839.

[0157] The propagation intensity between nodes is used as the node association weight to construct an abnormal propagation network. The network adopts a directed graph structure. The nodes represent process parameters or equipment status, the edges represent the influence relationship between parameters, and the edge weight is the corresponding propagation intensity. The propagation intensity threshold is set to 0.5, and only edges with an intensity greater than the threshold are retained to ensure that the network structure clearly reflects the main propagation path. Based on the abnormal propagation network, the abnormal diffusion path is identified. The maximum flow algorithm is used to identify the diffusion path. The abnormal detection point is set as the sink point, and the source point and propagation path are searched. The algorithm considers the minimum propagation intensity and path length on the path, and gives priority to the diffusion path with high minimum propagation intensity and short path. In this case, the main diffusion paths identified are: gas supply system-temperature control unit-oxidation furnace-oxide layer thickness. The minimum propagation intensity on the path is 0.72 and the path length is 3.

[0158] The process of locating the root causes of anomaly diffusion paths uses a spatiotemporal graph model to calculate the importance scores of path nodes. The importance score comprehensively considers a node's in-degree centrality, out-degree centrality, betweenness centrality, and eigenvector centrality, with weights of 0.2, 0.3, 0.2, and 0.3, respectively. In-degree centrality indicates the degree to which a node is influenced by other nodes, out-degree centrality indicates the degree to which a node influences other nodes, betweenness centrality indicates the node's bridging role in the network, and eigenvector centrality indicates the degree to which a node is connected to other important nodes.

[0159] Each centrality index is normalized to a value range of [0, 1]. In this case, the importance score of the gas supply system is 0.2 × 0.4 + 0.3 × 0.9 + 0.2 × 0.6 + 0.3 × 0.85 = 0.725, the importance score of the temperature control unit is 0.2 × 0.6 + 0.3 × 0.85 + 0.2 × 0.7 + 0.3 × 0.8 = 0.755, and the importance score of the oxidation furnace is 0.2 × 0.85 + 0.3 × 0.6 + 0.2 × 0.5 + 0.3 × 0.7 = 0.66. The anomaly source is determined based on the importance score, and the node with the highest score is selected as the anomaly source. In this case, the temperature control unit has the highest importance score (0.755) and is identified as the anomaly source.

[0160] The source analysis results are generated based on the location of the anomaly source and the anomaly diffusion path. The source analysis results include four parts: anomaly source information (including source name, location, parameter status, and anomaly severity), anomaly propagation path (including the node sequence and propagation intensity of each segment), impact assessment (including impact scope, impact severity, and duration), and recommendations (including short-term control measures and long-term solutions). In this case, the source analysis results showed that the anomaly source was the temperature control unit located in the oxidation area A3 station. The temperature control parameters offset +3.2°C, and the anomaly severity was severe (85 / 100). The anomaly propagation path was from the temperature control unit (propagation intensity 0.839) to the oxidation furnace (propagation intensity 0.792) to the oxide layer thickness, resulting in a 7.5nm thinner thickness. The impact assessment showed that 450 silicon wafers across three production batches were affected, resulting in a 12.5% ​​decrease in yield and an economic loss of approximately 375,000 yuan. Recommendations included immediately calibrating the temperature control unit parameters, temporarily adjusting the oxidation time to compensate for the thickness deviation, and adding redundant temperature control monitoring points in the long term to prevent similar issues.

[0161] This traceability method has achieved remarkable results in practical application at semiconductor manufacturers. After implementing this method at a 12-inch wafer fab, the accuracy of quality anomaly traceability increased from 76.3% to 93.8%, and the average traceability time was reduced from 2.8 hours to 23 minutes, an 86% efficiency improvement. For anomalies caused by complex, multi-factor coupling, the traceability accuracy improved even more significantly, from 62.5% to 89.7%. This method is particularly suitable for high-precision, high-value semiconductor, precision optics, and advanced materials manufacturing, providing strong support for improving product quality and production efficiency.

[0162] A second aspect of an embodiment of the present invention provides a chip raw material quality status tracking and tracing system, including:

[0163] The first unit is used to obtain real-time data during the production and preparation of chip raw materials, build a digital twin model of the raw materials based on the real-time data, establish a mapping relationship between the physical world and the information world, and realize dynamic monitoring of the quality status of the raw materials;

[0164] The second unit is configured to use the raw material digital twin model to perform dimensionality reduction optimization on the raw material quality status, use the raw material digital twin model to construct a data coding structure, map the optimized data into a feature space, and extract the spatiotemporal evolution characteristics of the raw material quality status through the raw material digital twin model; input the spatiotemporal evolution characteristics into the raw material digital twin model, use the raw material digital twin model to perform quality status assessment, and generate a quality assessment early warning signal;

[0165] The third unit is used to deploy a biologically inspired chaotic immune optimization algorithm based on the quality assessment warning signal to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the system can quickly locate and trace the quality anomalies. The chaotic immune optimization algorithm integrates antibody clone selection, immune memory, and chaotic perturbation mechanisms to improve the accuracy and convergence speed of anomaly pattern recognition.

[0166] The fourth unit is used to use the multi-agent collaborative traceability system to activate the intelligent traceability engine according to the quality assessment warning signal, locate the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generate traceability analysis results; perform closed-loop feedback control according to the traceability analysis results, which is used for optimizing and adjusting the raw material production process and realizing dynamic control of quality status.

[0167] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0168] processor;

[0169] a memory for storing processor-executable instructions;

[0170] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0171] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0172] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for tracking and tracing the quality status of chip raw materials, characterized in that: include: Acquire real-time data during the production and preparation of chip raw materials, build a digital twin model of the raw materials based on the real-time data, establish a mapping relationship between the physical world and the information world, and realize dynamic monitoring of the quality status of the raw materials; Using the raw material digital twin model to perform dimensionality reduction optimization on the raw material quality status, using the raw material digital twin model to construct a data coding structure, mapping the optimized data to a feature space, and extracting the spatiotemporal evolution characteristics of the raw material quality status through the raw material digital twin model; Inputting the spatiotemporal evolution characteristics into the raw material digital twin model, using the raw material digital twin model to perform quality status assessment, and generating a quality assessment early warning signal; Based on the quality assessment warning signals, a bio-inspired chaotic immune optimization algorithm is deployed to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the rapid location and traceability of quality anomalies can be achieved. The chaotic immune optimization algorithm integrates antibody clone selection, immune memory and chaotic perturbation mechanism to improve the accuracy and convergence speed of abnormal pattern recognition. The multi-agent collaborative traceability system is used to activate the intelligent traceability engine according to the quality assessment warning signal, locate the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generate traceability analysis results; closed-loop feedback control is performed based on the traceability analysis results to optimize and adjust the raw material production process and realize dynamic control of quality status.

2. The method according to claim 1, characterized in that Constructing a digital twin model of raw materials based on the real-time data, establishing a mapping relationship between the physical world and the information world, and realizing dynamic monitoring of the quality status of raw materials include: Establishing a set of physical equations including concentration field, diffusion coefficient, temperature, and pressure based on the real-time data, wherein the set of physical equations is used to characterize the physical properties of the raw material during preparation; Inputting the real-time data into a deep autoencoder, extracting hidden features of the real-time data through the weight matrix and bias term of the deep autoencoder, and constructing a digital twin model of the raw materials based on the hidden features; Optimizing the raw material digital twin model using a physical constraint loss function, wherein the physical constraint loss function includes a data fitting error term and a physical constraint term, wherein the physical constraint term is determined by the physical equation group; Constructing a high-dimensional feature vector space, mapping the raw material digital twin model to the high-dimensional feature vector space, and dynamically updating the feature weights in the high-dimensional feature vector space through an adaptive weight adjustment mechanism; Based on Kalman filtering, the state of the features in the high-dimensional feature vector space is predicted to generate a state vector; the distributed confidence interval is calculated according to the state vector, and the individual indicator scores of the raw material quality status are determined based on the distributed confidence interval. The individual indicator scores and the corresponding weight coefficients are weighted and summed to obtain a comprehensive evaluation index of the quality status, thereby realizing dynamic monitoring of the raw material quality status.

3. The method according to claim 1, characterized in that Using the raw material digital twin model to perform dimensionality reduction optimization on the raw material quality state, and using the raw material digital twin model to construct a data coding structure includes: Constructing a digital twin feature matrix, wherein the digital twin feature matrix includes multiple feature vectors, each of which corresponds to a feature dimension of the raw material quality status; inputting the feature vectors in the digital twin feature matrix into a cognitive hierarchy structure, wherein the cognitive hierarchy structure includes multiple cognitive layers, each of which is assigned a corresponding cognitive weight, and weighting the feature vectors using the cognitive weights to obtain cognitive features corresponding to each cognitive layer; Performing feature fusion on the cognitive feature and the digital twin feature matrix, wherein the feature fusion includes multiplying the cognitive feature and the digital twin feature matrix by corresponding fusion weight coefficients and then superimposing them to generate a fusion feature; Inputting the fused features into a coding mapping module, the coding mapping module including a coding weight matrix and a bias vector, performing a nonlinear transformation on the fused features using the coding weight matrix and the bias vector to obtain coding features; performing an EEG spectrum analysis on the coding features, calculating the spectral density function of the coding features using Fourier transform, and obtaining spectrum enhancement coding; The spectrum enhancement coding is optimized based on a dynamic memory update mechanism, wherein the dynamic memory update mechanism includes a memory state and a forgetting factor, wherein the memory state attenuates the historical memory state through the forgetting factor and performs a weighted combination with the current spectrum enhancement coding to obtain a data coding structure.

4. The method according to claim 1, wherein Using the raw material digital twin model to perform quality status assessment and generate quality assessment early warning signals includes: Utilizing the hierarchical attention mechanism of the raw material digital twin model, spatial and temporal features are extracted from the real-time data of the raw material quality status to obtain a quality status feature vector with spatiotemporal weights; Inputting the quality status feature vector into a dual time memory network, extracting short-term dynamic features and long-term evolution laws through fast memory units and slow memory units respectively, and generating a quality status evaluation index based on the combined relationship between the two; According to the historical distribution of the quality status evaluation index, a dynamic threshold optimization mechanism is adopted to calculate the alarm threshold, and a quality evaluation early warning signal is generated by comparing the deviation between the quality status evaluation index and the alarm threshold.

5. The method according to claim 1, wherein Based on the quality assessment warning signals, a bio-inspired chaotic immune optimization algorithm is deployed to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the rapid location and traceability of quality anomalies are achieved, including: Inputting the quality assessment warning signal into a chaotic immune model, the chaotic immune model uses iterative operations to generate an initialization sequence, constructing an agent population based on the initialization sequence, and optimizing the distribution of the agent population through the chaotic immune model. The chaotic immune model includes a warning signal strength parameter and a temperature adjustment parameter, and the agent distribution corresponding to the warning signal strength is obtained by dynamically adjusting the warning signal strength parameter and the temperature adjustment parameter; Based on the agent distribution, the chaotic immune model is used to calculate the response strength between the warning signal characteristics and the abnormal template characteristics. The chaotic immune model constructs an ergodic search sequence based on the response strength, and adaptively weights the search sequence with the historical search position to update the current search path and determine the evolution trajectory of the abnormal characteristics. An intelligent agent collaborative network is constructed. The intelligent agent collaborative network is based on the spatial distance relationship between intelligent agents. The interaction strength between intelligent agents is constructed through the chaotic immune model. The interaction strength is weightedly fused with the local tracing results of each intelligent agent, where the local tracing results include the location of the anomaly and the propagation path information. The tracing decision result of the global quality anomaly is generated through the weighted fusion.

6. The method according to claim 1, wherein Based on the quality assessment warning signals, a bio-inspired chaotic immune optimization algorithm is deployed to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the rapid location and traceability of quality anomalies are achieved, including: The quality assessment warning signal is input into a chaotic immune optimization algorithm, which constructs an antigen-antibody response mechanism by calculating the affinity between the quality assessment warning signal and the abnormal template, determines the number of clones based on the affinity and generates an initial immune population, and uses a chaotic sequence to perturb and update the initial immune population, wherein the chaotic sequence is weightedly combined with the historical optimal antibody to determine the search direction, thereby obtaining an optimized immune population; A multi-agent collaborative traceability system is constructed based on the optimized immune population, and the antibody characteristics in the optimized immune population are mapped to the initial state of the agent. The agent explores abnormal characteristics in the quality state space through a chaotic search strategy, wherein the chaotic search strategy includes two stages: local fine search and global coarse search, and dynamically adjusts the search step size to achieve rapid positioning of abnormal areas; An agent interaction network is constructed in the multi-agent collaborative traceability system, and an interaction weight matrix is ​​calculated based on the spatial distance between agents. The interaction weight matrix is ​​weightedly fused with the abnormal feature information collected by each agent. A propagation path map of quality anomalies is constructed based on the fused abnormal feature information, and the source position and diffusion direction of the quality anomaly are determined by tracing back the propagation path map.

7. The method according to claim 1, characterized in that The multi-agent collaborative traceability system activates the intelligent traceability engine according to the quality assessment warning signal, locates the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generates traceability analysis results including: Activate the intelligent traceability engine based on the quality assessment warning signal, input the quality assessment warning signal into the spatiotemporal graph model for warning intensity analysis, perform threshold judgment on the warning intensity through the spatiotemporal graph model, and generate the engine activation state, wherein the spatiotemporal graph model includes sensitivity parameters and activation thresholds; Constructing a correlation structure between time series and spatial topology, using the spatiotemporal graph model to calculate the temporal evolution characteristics of the quality status, and simultaneously using the spatiotemporal graph model to analyze the spatial position relationship, fusing the temporal evolution characteristics with the spatial position relationship to form a spatiotemporal correlation feature; Constructing an anomaly propagation path based on the spatiotemporal correlation features, calculating the inter-node propagation strength using the spatiotemporal graph model, constructing an anomaly propagation network using the inter-node propagation strength as the node association weight, and identifying anomaly diffusion paths based on the anomaly propagation network; The root cause of the abnormal diffusion path is located, the importance scores of the path nodes are calculated using the spatiotemporal graph model, the location of the abnormal source point is determined, and a tracing analysis result is generated based on the location of the abnormal source point and the abnormal diffusion path.

8. A chip raw material quality status tracking and traceability system, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain real-time data during the production and preparation of chip raw materials, build a digital twin model of the raw materials based on the real-time data, establish a mapping relationship between the physical world and the information world, and realize dynamic monitoring of the quality status of the raw materials; The second unit is used to use the raw material digital twin model to perform dimensionality reduction optimization on the raw material quality status, use the raw material digital twin model to construct a data coding structure, map the optimized data to a feature space, and extract the spatiotemporal evolution characteristics of the raw material quality status through the raw material digital twin model; Inputting the spatiotemporal evolution characteristics into the raw material digital twin model, using the raw material digital twin model to perform quality status assessment, and generating a quality assessment early warning signal; The third unit is used to deploy a biologically inspired chaotic immune optimization algorithm based on the quality assessment warning signal to build a multi-agent collaborative traceability system. By simulating the adaptive mechanism and chaotic search strategy of the biological immune system, the system can quickly locate and trace the quality anomalies. The chaotic immune optimization algorithm integrates antibody clone selection, immune memory, and chaotic perturbation mechanisms to improve the accuracy and convergence speed of anomaly pattern recognition. The fourth unit is used to use the multi-agent collaborative traceability system to activate the intelligent traceability engine according to the quality assessment warning signal, locate the root cause of quality anomalies based on the reasoning mechanism of the spatiotemporal graph, and generate traceability analysis results; perform closed-loop feedback control according to the traceability analysis results, which is used for optimizing and adjusting the raw material production process and realizing dynamic control of quality status.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images