Dynamic feeding decision optimization system of wafer manufacturing system considering bottleneck drift

Abstract

The invention relates to a wafer manufacturing feeding decision-making system for bottleneck drift dynamic coping, which comprises a data acquisition module, a bottleneck identification module and a dynamic feeding control decision-making module, and is characterized in that the data acquisition module is in butt joint with a production management system to acquire equipment, process and order related data; the bottleneck identification module constructs a weighted directed graph model, formulates a bottleneck judgment index and a threshold value, and adopts a Transform model to predict a bottleneck trend; the dynamic feeding control decision module designs a strategy combining a fixed period and priority dynamic adjustment, and reasonable feeding is achieved. The problem that an existing wafer manufacturing feeding decision method cannot accurately recognize and predict bottleneck drifting and cannot reasonably and dynamically make a feeding decision is solved, accurate recognition and prediction of bottleneck drifting in the wafer manufacturing process are achieved, the feeding strategy is dynamically adjusted based on the bottleneck drifting, the equipment utilization rate is increased, the production period is shortened, and the production efficiency is improved. Dynamic changes of bottlenecks in a wafer manufacturing system can be effectively handled, and the overall productivity and order delivery efficiency of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of wafer manufacturing technology, specifically to a dynamic material feeding decision optimization system for wafer manufacturing systems that takes into account bottleneck drift, belonging to the fields of semiconductor manufacturing, intelligent manufacturing, and artificial intelligence technology. Background Technology

[0002] In the semiconductor industry, wafer manufacturing is a core process with a complex production process involving hundreds of steps, numerous pieces of equipment, and a large amount of work-in-process. Bottlenecks in the production system have a significant impact on overall production efficiency, and in actual production, bottlenecks can shift due to factors such as production progress, order changes, and equipment status.

[0003] Traditional wafer manufacturing material feeding decisions are mostly based on static models, assuming fixed bottlenecks and relying on experience or simple rules, such as fixed feeding intervals or quantities. Such static strategies have many drawbacks in the complex and ever-changing wafer manufacturing environment: First, the inability to accurately identify bottleneck drift in real time leads to an unreasonable quantity of work-in-process. If the original strategy is still followed after bottleneck drift, it may cause a backlog of work-in-process before the bottleneck process, occupying a large amount of inventory space and capital, prolonging product waiting time, and reducing production efficiency; or the quantity of work-in-process may be insufficient, resulting in idle equipment and the inability to fully utilize production capacity.

[0004] Secondly, traditional methods are ineffective at predicting future bottleneck trends. Production managers cannot make reasonable decisions in advance to cope with bottleneck changes. For example, when a bottleneck is about to shift, the lack of a prediction mechanism makes it impossible to adjust the material feeding strategy in time, which can easily lead to production plan chaos and affect the timely delivery of orders.

[0005] In summary, existing wafer manufacturing material feeding decision-making methods are severely inadequate in addressing bottleneck drift issues. There is an urgent need for a system that can accurately identify and predict bottleneck drift and optimize material feeding decisions based on this, in order to improve the production efficiency of wafer manufacturing systems and ensure order delivery efficiency. Summary of the Invention

[0006] To address the problem that existing wafer manufacturing feed decision-making methods cannot accurately identify and predict bottleneck drift and make reasonable dynamic feed decisions, a wafer manufacturing feed decision-making system for dynamic response to bottleneck drift is proposed. This system enables accurate identification and prediction of bottleneck drift and dynamic adjustment of feed strategies, providing decision support for the efficient operation of wafer manufacturing systems.

[0007] The technical solution of this invention is as follows: A dynamic material feeding decision optimization system for wafer manufacturing considering bottleneck drift includes: a data acquisition module, a bottleneck identification module, and a dynamic material feeding control decision module. The data acquisition module interfaces with the production management system to acquire equipment, process, and order-related data. The bottleneck identification module constructs a weighted directed graph model, formulates bottleneck judgment indicators and thresholds, and uses a Transformer model to predict bottleneck trends. The dynamic material feeding control decision module designs a strategy combining fixed cycles and dynamic priority adjustments to achieve rational material feeding. The data acquisition module connects to the production management system via an interface to obtain equipment operating status, process flow records, and work-in-process quantity data in real time and accurately, and transmits the acquired data to the bottleneck identification module. The bottleneck identification module includes: a workshop operation modeling submodule, a bottleneck trend prediction submodule, and a bottleneck determination submodule. The workshop operation modeling submodule constructs a weighted directed graph model with equipment groups as nodes and process connections as directed edges. Edge weights characterize the material flow efficiency between processes, and node weights integrate the processing capacity of equipment groups with the current equipment load to form a comprehensive load index. The bottleneck trend prediction submodule uses a Transformer model with a graph attention mechanism (GAT) to predict bottleneck trends. GAT captures the relationships between equipment groups and outputs node embedding features, while the Transformer model handles the temporal dependencies in the weighted directed graph model. The bottleneck determination submodule formulates dynamically adjusted bottleneck determination indicators and thresholds and outputs them to the dynamic material feeding control decision module to match actual production conditions. The dynamic material feeding control decision module includes an order priority adjustment submodule and a material feeding quantity allocation submodule. The order priority adjustment submodule determines the processing priority of different orders based on the urgency of the orders and the complexity of the product process. The material feeding quantity allocation submodule determines the material feeding quantity of different orders based on the bottleneck determination result of the bottleneck determination of the equipment group by the bottleneck determination submodule.

[0008] Furthermore, the data acquisition module specifically includes: connecting to the production management system through a specific interface; this interface follows a standard data transmission protocol; the acquired data covers equipment operating parameters, process flow information, and work-in-process related data; the data acquisition frequency is set according to actual production needs; the acquired data is transmitted in real time to the system's data storage module, stored using a distributed database, and used as input for the bottleneck identification module.

[0009] Furthermore, the bottleneck identification module specifically includes: a workshop operation modeling submodule, a bottleneck trend prediction submodule, and a bottleneck determination submodule; The workshop operation modeling submodule constructs a weighted directed graph model using equipment groups as nodes and process connections as directed edges; a unique identifier is assigned to each equipment group. ; Determine the direction of the directed edge according to the process flow; edge weight It is set according to the average time or rate of material flow between processes in historical data. Let the average material from process e to process f (e < f) be , that is ; Calculation of node weight, comprehensively considering the processing capacity of the equipment group and the current in-process load ; Processing capacity of the equipment group Obtained by calculating the rated production rate of the equipment and the daily available operating duration , that is ; Current in-process load Calculated according to the ratio of the number of in-process products currently being processed and waiting to be processed by this equipment group to its maximum processing capacity , that is ; The two are fused according to a certain weight to obtain the node weight of the node , that is , where ; Bottleneck trend prediction sub-module: The input of the bottleneck trend prediction model is the weighted directed graph structure model output by the workshop operation modeling sub-module; The graph attention network is constructed to process the association relationship between equipment groups. For each node i, by weighted aggregation with the feature information of adjacent nodes j, let the feature vector of node i be , the feature vector of adjacent node j be , and the weight coefficient be , then the propagated feature vector of node i , where is the activation function, is the set of adjacent nodes of node i. Through this process, the association features between nodes are obtained and the node embedding features are output; Then the node embedding features are input into the Transformer model. The self-attention mechanism of the Transformer processes the weighted directed graph structure model to capture the temporal dependence relationship between different nodes; Through the multi-head attention mechanism, let the number of heads be , for each head , calculate the attention coefficient , where and are the query matrix and the key matrix respectively, is the matrix dimension, T is the transpose symbol, and the attention distribution is obtained through the softmax function, and the final output value is , and then the results are concatenated to obtain the output of the bottleneck trend prediction sub-module and input it to the bottleneck determination sub-module; Bottleneck determination submodule: By setting the node weight threshold range ,when At that time, it was determined that the equipment group might become a bottleneck; secondly, every cycle time... ,in It is a time adjustment factor. It is the average processing time of the nth equipment group out of m equipment groups. It is the effective utilization rate coefficient of the equipment group. It refers to the production cycle time; during the evaluation process, feedback on bottlenecks in actual production is used to determine the adjustment range of the threshold. Finally, the results are output to the dynamic feeding control decision module.

[0010] Furthermore, the dynamic feeding control decision module specifically includes: The dynamic material feeding control decision module is divided into an order priority adjustment submodule and a material feeding quantity allocation submodule. Based on the complexity of the wafer manufacturing process, it divides the production process into multiple fixed cycles. At the beginning of each cycle, order priorities and quantities are adjusted. Order Priority Control Submodule: Regarding dynamic priority adjustment, it adjusts priorities based on order urgency. Product process complexity Determine product priorities The result is then output to the material quantity allocation submodule, where... , and These are the weighting coefficients; Material input quantity allocation submodule: Within each cycle, to reduce the impact of bottleneck drift, the material input quantity for each product type is allocated. When the node weight of a device group exceeds the bottleneck determination threshold, and the bottleneck trend prediction submodule indicates that it has a high probability of becoming a bottleneck in the future, the order quantity is adjusted. For device group s, for product u, its material input quantity... ,in This is the theoretical maximum number of work-in-process items in equipment group s. The bottleneck drift risk suppression coefficient is determined by adjusting the amount of product u fed into the equipment group to reduce the workload flowing into the equipment group and prevent it from exacerbating bottleneck drift due to overload.

[0011] A dynamic material feeding decision optimization method for a wafer manufacturing system considering bottleneck drift is implemented based on the aforementioned dynamic material feeding decision optimization system for a wafer manufacturing system considering bottleneck drift, comprising: Step 1) Data collection and preprocessing: Obtain operational data from the workshop operation management system and represent and store it uniformly; Step 2) Modeling the workshop operation process: Using equipment groups as nodes and process connection relationships as directed edges, construct a weighted directed graph model; Step 3) Scheduling status assessment: Based on the bottleneck drift prediction model, predict the bottleneck trend in future cycles, and identify potential drifting bottlenecks through preset thresholds to provide early warning of dynamic changes in bottlenecks. Step 4) Dynamic feeding decision: Dynamic feeding is carried out according to a fixed cycle.

[0012] The beneficial effects of this invention are as follows: 1. By constructing a workshop operation diagram structure model, production line topology information is integrated into bottleneck analysis, which more comprehensively represents the interaction between equipment.

[0013] 2. By combining the Transformer model with graph attention mechanism, a more accurate determination of drift bottlenecks is achieved.

[0014] 3. By combining a fixed-cycle material feeding strategy with a priority-based dynamic challenge strategy, the production rhythm is ensured, and the material feeding plan can be quickly adjusted according to the bottleneck status.

[0015] The system of this invention includes a data acquisition module, a bottleneck identification module, and a dynamic material feeding control decision module. Through the collaborative operation of this three-layer architecture, it achieves accurate identification and prediction of bottleneck drift during wafer manufacturing, and dynamically adjusts the material feeding strategy accordingly to improve equipment utilization and shorten production cycles. This invention can effectively address the dynamic changes in bottlenecks in wafer manufacturing systems, improving overall system capacity and order delivery efficiency. Attached Figure Description

[0016] Figure 1 This is a structural diagram of the dynamic material feeding decision optimization system for the wafer manufacturing system of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0018] A dynamic material feeding decision optimization system for wafer manufacturing considering bottleneck drift includes: a data acquisition module, a bottleneck identification module, and a dynamic material feeding control decision module. The data acquisition module interfaces with production management systems such as MES (Manufacturing Execution System) to acquire key data on equipment operating status, process flow records, and work-in-process quantity (related to equipment, process, and order). The bottleneck identification module constructs a weighted directed graph model, formulates bottleneck judgment indicators and thresholds, and uses a Transformer model to predict bottleneck trends. The dynamic material feeding control decision module designs a strategy combining fixed-cycle and priority-based dynamic adjustment to achieve rational material feeding.

[0019] The data acquisition module connects to production management systems such as MES through an interface to obtain equipment operating status, process flow records, and work-in-process quantity data in real time and accurately. The acquired data is then transmitted to the bottleneck identification module to provide reliable data support for subsequent analysis. Compared with traditional manual data collection, this improves the timeliness and accuracy of the data.

[0020] The bottleneck identification module includes: a workshop operation modeling submodule, a bottleneck trend prediction submodule, and a bottleneck determination submodule. The workshop operation modeling submodule constructs a weighted directed graph model using equipment groups as nodes and process connections as directed edges. Edge weights characterize the material flow efficiency between processes, and node weights integrate the processing capacity of equipment groups with the current equipment load to form a comprehensive load index, comprehensively and intuitively reflecting the node load of workshop equipment groups. This overcomes the limitations of traditional single-factor analysis and improves the accuracy of bottleneck identification. The bottleneck trend prediction submodule uses a Transformer model with a Graph Attention (GAT) mechanism to predict bottleneck trends. GAT captures the relationships between equipment groups and outputs node embedding features, while the Transformer model handles the temporal dependencies in the weighted directed graph model. This combination overcomes the shortcomings of traditional simple statistical or linear prediction methods in handling complex nonlinear relationships and temporal dependencies, significantly improving prediction accuracy. The bottleneck determination submodule formulates dynamically adjusted bottleneck determination indicators and thresholds and outputs them to the dynamic material feeding control decision module to adapt to actual situations such as production process improvements and product type changes, thereby improving identification reliability.

[0021] The dynamic material feeding control decision module includes an order priority adjustment submodule and a material feeding quantity allocation submodule. The order priority adjustment submodule determines the processing priority of different orders based on the urgency of the orders and the complexity of the product process. The material feeding quantity allocation submodule determines the material feeding quantity of different orders based on the bottleneck determination result of the bottleneck determination of the equipment group by the bottleneck determination submodule.

[0022] Example: The dynamic feeding decision optimization system of the wafer manufacturing system considering bottleneck drift in the present invention includes a data acquisition module, a bottleneck identification module, and a dynamic feeding control decision module. Each layer closely collaborates to achieve the optimized management of the wafer manufacturing system.

[0023] The data acquisition module is connected to production management systems such as MES through a specific interface. This interface follows standard data transmission protocols to ensure the efficient and accurate transmission of data. The collected data covers equipment operation parameters, process flow information, and in-process product-related data. The data acquisition frequency can be set according to actual production requirements. The collected data is transmitted to the data storage module of the system in real time and stored using a distributed database as the input for the bottleneck identification module.

[0024] The bottleneck identification module mainly includes a workshop operation modeling sub-module, a bottleneck trend prediction sub-module, and a bottleneck determination sub-module.

[0025] Workshop operation modeling sub-module: Taking equipment groups as nodes and process connection relationships as directed edges, a weighted directed graph model is constructed. A unique identifier is set for each equipment group. . Determine the direction of the directed edge according to the process flow. The edge weight is set according to the average time or rate of material flow between processes in historical data. Let the average material from process e to process f (e < f) be , that is . For the calculation of the node weight, the processing capacity of the equipment group and the current in-process product load are comprehensively considered. The processing capacity of the equipment group is calculated through the rated production rate of the equipment and the daily available operation duration , that is . The current in-process product load is calculated according to the ratio of the number of in-process products currently being processed and waiting to be processed in this equipment group to its maximum processing capacity , that is . The two are fused according to a certain weight to obtain the node weight of the node, that is <​​​​​​​​​​ Then the feature vector of node i after propagation ,in It is an activation function. This is the set of neighboring nodes of node i. This process obtains the association features between nodes and outputs the node embedding features. These node embedding features are then input into the Transformer model. The Transformer's self-attention mechanism processes the weighted directed graph structure model, capturing the temporal dependencies between different nodes. Specifically, a multi-head attention mechanism is used, with the number of heads set to 1. For each head Calculate the attention coefficient ,in and These are the query matrix and the key matrix, respectively. Let T be the matrix dimension and T be the transpose. The attention distribution is obtained after passing through the softmax function. The final output value is The results are then concatenated to obtain the output of the bottleneck trend prediction submodule, which is then input into the bottleneck determination submodule.

[0027] Bottleneck determination submodule: By setting the node weight threshold range ,when At that time, it was determined that the equipment group might become a bottleneck. Secondly, at each cycle time... ,in It is a time adjustment factor. It is the average processing time of the nth equipment group out of m equipment groups. It is the effective utilization rate coefficient of the equipment group. This refers to the production cycle time. During the evaluation process, feedback on bottlenecks encountered in actual production is used to determine the adjustment range of the threshold. Finally, the results are output to the dynamic feeding control decision module.

[0028] The dynamic material feeding control decision module is mainly divided into an order priority adjustment submodule and a material feeding quantity allocation submodule. Based on the complexity of the wafer manufacturing process, it divides the production process into multiple fixed cycles. At the beginning of each cycle, order priorities and quantities are adjusted.

[0029] Order Priority Control Submodule: Regarding dynamic priority adjustment, it adjusts priorities based on order urgency. Product process complexity Determine product priorities The result is then output to the material quantity allocation submodule, where... , and This is a weighting coefficient. Products with urgent orders or high process complexity are assigned higher priority. Priority adjustments are made in real time; for example, when a new urgent order is received, the priority of related products is immediately increased. In actual production, complex products generally have higher order sources, market competitiveness, and profit margins, and are therefore also given higher priority.

[0030] Material Feed Quantity Allocation Submodule: Within each cycle, to reduce the impact of bottleneck drift, the material feed quantity is allocated for each product type. When the node weight of a device group exceeds the bottleneck determination threshold, and the bottleneck trend prediction submodule indicates a high probability of it becoming a bottleneck in the future, the order quantity is adjusted. Specifically, for device group s, for product u, its material feed quantity... ,in This is the theoretical maximum number of work-in-process items in equipment group s. The bottleneck drift risk suppression coefficient is determined by adjusting the amount of product u fed into the equipment group to reduce the workload flowing into the equipment group and prevent it from exacerbating bottleneck drift due to overload.

[0031] The system's workflow is as follows: (1) Data collection and preprocessing: Obtain operational data from workshop operation management systems such as MES, and represent and store it in a unified manner.

[0032] (2) Modeling of workshop operation process: Using equipment groups as nodes and process connection relationships as directed edges, a weighted directed graph model is constructed.

[0033] (3) Scheduling status assessment: Based on the bottleneck drift prediction model, the bottleneck trend of future cycles is predicted, and potential drift bottlenecks are identified through preset thresholds to provide early warning of bottleneck dynamic changes.

[0034] (4) Dynamic feeding decision: Dynamic feeding is carried out according to a fixed cycle.

[0035] The above-described embodiments are merely one implementation of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention should be determined by the appended claims.

Claims

1. A dynamic material feeding decision optimization system for a wafer manufacturing system considering bottleneck drift, characterized in that, include: The system comprises a data acquisition module, a bottleneck identification module, and a dynamic material feeding control decision module. The data acquisition module interfaces with the production management system to acquire equipment, process, and order-related data. The bottleneck identification module constructs a weighted directed graph model, establishes bottleneck judgment indicators and thresholds, and uses a Transformer model to predict bottleneck trends. The dynamic material feeding control decision module designs a strategy combining fixed-cycle and priority-based dynamic adjustments to achieve rational material feeding. The data acquisition module interfaces with the production management system to acquire real-time and accurate data on equipment operating status, process flow records, and work-in-process quantities, and transmits this data to the bottleneck identification module. The bottleneck identification module includes a workshop operation modeling submodule, a bottleneck trend prediction submodule, and a bottleneck judgment submodule. The workshop operation modeling submodule constructs a weighted directed graph model with equipment groups as nodes and process connections as directed edges, where edge weights represent the material flow between processes. Material flow efficiency is calculated by integrating the processing capacity of the equipment group with the current equipment load to form a comprehensive load index. The bottleneck trend prediction submodule uses a Transformer model with a fusion graph attention mechanism (GAT) to predict bottleneck trends. GAT captures the correlation between equipment groups and outputs node embedding features, while the Transformer model handles the temporal dependencies in the weighted directed graph model. The bottleneck determination submodule formulates dynamically adjusted bottleneck determination indicators and thresholds and outputs them to the dynamic material feeding control decision module to match the actual production situation. The dynamic material feeding control decision module includes an order priority adjustment submodule and a material feeding quantity allocation submodule. The order priority adjustment submodule determines the processing priority of different orders based on the urgency of the orders and the complexity of the product process. The material feeding quantity allocation submodule determines the material feeding quantity of different orders based on the bottleneck determination results of the equipment group bottleneck by the bottleneck determination submodule.

2. The dynamic material feeding decision optimization system for wafer manufacturing system considering bottleneck drift according to claim 1, characterized in that, The data acquisition module specifically includes: connecting to the production management system through a specific interface; this interface follows a standard data transmission protocol; the acquired data covers equipment operating parameters, process flow information, and work-in-process related data; the data acquisition frequency is set according to actual production needs; the acquired data is transmitted in real time to the system's data storage module, stored using a distributed database, and used as input for the bottleneck identification module.

3. The dynamic material feeding decision optimization system for wafer manufacturing system considering bottleneck drift according to claim 1, characterized in that, The bottleneck identification module specifically includes: a workshop operation modeling submodule, a bottleneck trend prediction submodule, and a bottleneck determination submodule; Workshop operation modeling sub-module: Construct a weighted directed graph model with equipment groups as nodes and process connection relationships as directed edges; set a unique identifier for each equipment group ; Determine the direction of the directed edge according to the process flow; edge weight Set according to the average time or rate of material flow between processes in historical data. Let the average material from process e to process f be , that is , e < f; Calculation of node weight, comprehensively considering the processing capacity of the equipment group and the current in-process load ; Processing capacity of the equipment group Obtained by calculating the rated production rate of the equipment and the daily available operation duration , that is ; Current in-process load Calculated according to the ratio of the number of in-process products currently being processed and waiting to be processed in this equipment group to its maximum processing capacity , that is ; The two are fused according to a certain weight to obtain the node weight of the node , that is , where ; Bottleneck Trend Prediction Submodule: The input to the bottleneck trend prediction model is the weighted directed graph structure model output by the workshop operation modeling submodule. It processes the relationships between equipment groups by constructing a graph attention network. For each node i, it performs weighted aggregation with the feature information of its neighboring nodes j. Let the feature vector of node i be... The feature vector of the neighboring node j is The weighting coefficient is Then the feature vector of node i after propagation ,in It is an activation function. Let be the set of neighboring nodes of node i. This process obtains the association features between nodes and outputs the node embedding features. Then, these node embedding features are input into the Transformer model. The Transformer's self-attention mechanism processes the weighted directed graph structure model, capturing the temporal dependencies between different nodes. Finally, a multi-head attention mechanism is used, with the number of heads set to . For each head Calculate the attention coefficient ,in and These are the query matrix and the key matrix, respectively. Let T be the matrix dimension and T be the transpose. The attention distribution is obtained after passing through the softmax function. The final output value is The results are then concatenated to obtain the output of the bottleneck trend prediction submodule, and then input into the bottleneck determination submodule. Bottleneck determination submodule: By setting the node weight threshold range ,when At that time, it was determined that the equipment group might become a bottleneck; secondly, every cycle time... ,in It is a time adjustment factor. It is the average processing time of the nth equipment group out of m equipment groups. It is the effective utilization rate coefficient of the equipment group. It refers to the production cycle time; during the evaluation process, feedback on bottlenecks in actual production is used to determine the adjustment range of the threshold. Finally, the results are output to the dynamic feeding control decision module.

4. The dynamic material feeding decision optimization system for wafer manufacturing system considering bottleneck drift according to claim 1, characterized in that, The dynamic feeding control decision module specifically includes: The dynamic material feeding control decision module is divided into an order priority adjustment submodule and a material feeding quantity allocation submodule. Based on the complexity of the wafer manufacturing process, it divides the production process into multiple fixed cycles. At the beginning of each cycle, order priorities and quantities are adjusted; Order Priority Control Submodule: Regarding dynamic priority adjustment, it adjusts priorities based on order urgency. Product process complexity Determine product priorities The result is then output to the material quantity allocation submodule, where... , and These are the weighting coefficients; Material input quantity allocation submodule: Within each cycle, to reduce the impact of bottleneck drift, the material input quantity for each product type is allocated. When the node weight of a device group exceeds the bottleneck determination threshold, and the bottleneck trend prediction submodule indicates that it has a high probability of becoming a bottleneck in the future, the order quantity is adjusted. For device group s, for product u, its material input quantity... ,in This is the theoretical maximum number of work-in-process items in equipment group s. The bottleneck drift risk suppression coefficient is determined by adjusting the amount of product u fed into the equipment group to reduce the workload flowing into the equipment group and prevent it from exacerbating bottleneck drift due to overload.

5. A dynamic material feeding decision optimization method for a wafer manufacturing system considering bottleneck drift, implemented based on the dynamic material feeding decision optimization system for a wafer manufacturing system considering bottleneck drift as described in any one of claims 1-4, comprising: Step 1) Data collection and preprocessing: Obtain operational data from the workshop operation management system and represent and store it uniformly; Step 2) Modeling the workshop operation process: Using equipment groups as nodes and process connection relationships as directed edges, construct a weighted directed graph model; Step 3) Scheduling status assessment: Based on the bottleneck drift prediction model, predict the bottleneck trend in future cycles, and identify potential drifting bottlenecks through preset thresholds to provide early warning of dynamic changes in bottlenecks. Step 4) Dynamic feeding decision: Dynamic feeding is carried out according to a fixed cycle.

Images