A semiconductor production line production scheduling method, medium, device and product

Abstract

The application discloses a semiconductor production line scheduling method, medium, equipment and product, relates to the technical field of production line scheduling optimization, and comprises the following steps: acquiring a semiconductor production line order set, an equipment set, a process set and scheduling attributes; generating a candidate scheduling set; generating a scheduling comprehensive score function according to the scheduling attributes and the candidate scheduling set; maximizing the sum of scheduling comprehensive scores of the candidate scheduling set as a scheduling objective function, and solving a scheduling result under the constraint of a set process order and processing time; when an abnormal situation occurs, constructing an affected scheduling set and an unaffected scheduling set based on the scheduling result, constructing a rearrangement benefit of the affected scheduling set and a scheduling disturbance of the unaffected scheduling set; constructing a local rearrangement optimization model based on the rearrangement benefit of the affected set and the scheduling disturbance of the unaffected set, and obtaining a final scheduling result. The problems of high rearrangement cost of the semiconductor production line, lack of a global perspective and failure of a static plan are solved.

Description

Technical Field

[0001] This invention relates to the field of production line scheduling optimization technology, and in particular to a semiconductor production line scheduling method, medium, equipment and product. Background Technology

[0002] Semiconductor production line scheduling refers to the comprehensive determination of when, on which equipment, and which process step for each order is performed, based on a combination of order demand, process routes, and the real-time capacity and status of the production line equipment. Orders include order quantity, due date, priority, product model, and rush / interim order flags. Processes are the technological routes corresponding to each product, defining the process sequence, process constraints, and typical man-hours. Equipment includes each piece of equipment's type, capacity matrix, chamber structure, status, historical efficiency, calendar, and maintenance schedule.

[0003] Existing semiconductor production line scheduling methods can be categorized into traditional experience and simple rule-based methods, heuristic scheduling methods, discrete event simulation methods, mathematical optimization methods, classic intelligent algorithms, and emerging AI / reinforcement learning methods.

[0004] Traditional experience and simple rule-based scheduling methods include manual / Excel scheduling, First-In-First-Out (FIFO), Delivery-Oriented (DDO) scheduling, and fixed priority scheduling. These methods only consider local, static factors and cannot account for multi-dimensional conditions such as equipment status, work-in-process (WIP) distribution, and process constraints, resulting in low scheduling quality and significant capacity waste. They also respond poorly to dynamic disturbances, making it easy for plans to deviate from actual production. Ultimately, they are ill-suited to support the demands of modern wafer fabs with their high-variety, low-volume, and complex reentrant processes.

[0005] Heuristic scheduling methods, based on complex dispatch rules designed with factors such as region, product, and equipment configuration, form the foundation of most fabrication production line (Fab) RTS / dispatch systems. However, heuristic scheduling can only provide a reasonably feasible solution, not a mathematically optimal one. Potential capacity and cycle time benefits are not maximized, and the rules require continuous manual fine-tuning based on actual production line conditions, resulting in high maintenance costs and insufficient sensitivity to production line changes. When faced with extremely complex constraints, rule combinations explode, leading to poor interpretability and unstable performance.

[0006] Discrete event simulation scheduling generates schedules by constructing plant-wide or regional equipment / Work-in-Progress (WIP) models and simulating future batch arrivals and equipment states. It is commonly used for overall forecasting and rule verification. However, each simulation calculation is very time-consuming, making it difficult to support high-frequency real-time adjustments. It has extremely high requirements for input data quality; data delays or errors can lead to prediction biases that accumulate. Simulation itself is merely an evaluation tool, not a direct optimization method, and typically can only provide suboptimal schedules in conjunction with rules; embedding optimization algorithms drastically increases the computational load. Model maintenance is cumbersome, requiring extensive reconfiguration when equipment changes, resulting in insufficient flexibility.

[0007] Mathematical optimization methods, such as mixed-integer programming and constrained programming, model batch allocation and equipment assignment, using a weighted objective function to find the optimal schedule. However, model performance deteriorates non-linearly with problem size, making real-time performance difficult to guarantee. It is extremely sensitive to data quality and the completeness of constraint modeling; any deviation from reality will result in an infeasible schedule. Furthermore, local optimization without global information can easily lead to local optimization but global suboptimal results.

[0008] Classical intelligent optimization algorithms, such as evolutionary algorithms and neighborhood search, are used for approximate optimization and are often applied in academic research or some advanced APS modules. However, these algorithms cannot guarantee optimality, their parameters rely on empirical tuning, and they lack stability. They also have limited adaptability to dynamic events, often requiring re-running after sudden disturbances, resulting in a delayed response. Furthermore, they still face computational efficiency bottlenecks in large-scale fabs, making it difficult to achieve real-time reordering within minutes.

[0009] Emerging AI / deep reinforcement learning methods utilize historical data to train DRLs, GNNs, and other algorithms for real-time work assignment decisions. Training relies on large amounts of high-quality historical data; while semiconductor plants have abundant data, it often contains noise and distribution drift. Models frequently require retraining after changes in production line configuration, product mix, or process technology. Robustness from the laboratory to real-world dynamic fabs remains insufficient.

[0010] In summary, existing scheduling technologies generally suffer from the following problems when orders change frequently and equipment status changes dynamically in semiconductor production lines: the cost of rescheduling is too high or there is a lack of a global perspective when orders change frequently; sudden equipment failures or project managers can quickly invalidate static plans; in addition, the inability to unify data sensitivity, real-time performance and optimality ultimately leads to a serious disconnect between the planning and execution layers, making it difficult to simultaneously meet the comprehensive requirements of delivery, capacity and cycle time. Summary of the Invention

[0011] The purpose of this invention is to address the problems of high rescheduling costs or lack of a global perspective caused by frequent order changes and dynamic changes in equipment status in semiconductor production lines, as well as the failure of static plans. This invention proposes a semiconductor production line scheduling method, comprising the following steps: S1. Obtain the semiconductor production line order set, equipment set, and process set, as well as the production scheduling attributes of the orders, equipment, and processes; S2. Generate a candidate production schedule set for assigning orders to equipment in the process; S3. Generate a comprehensive production scheduling score function based on production scheduling attributes and candidate production scheduling sets; S4. Maximize the sum of the comprehensive production scores of the candidate production schedule set as the production schedule objective function, and solve the production schedule result under the constraints of the set process sequence and processing time. S5. When an abnormal situation occurs, construct an affected production schedule set and an unaffected production schedule set based on the production scheduling results, and construct the rearrangement benefit of the affected production schedule set and the production schedule disturbance of the unaffected production schedule set. S6. Based on the rearrangement benefits of the affected set and the scheduling disturbance of the unaffected set, a local rearrangement optimization model is constructed to obtain the final scheduling result.

[0012] Furthermore, the comprehensive production scheduling scoring function is as follows:

[0013] Where S(i,j,p,t) represents the comprehensive production scheduling score of the i-th order assigned to the j-th equipment for process p at time t; G(i,j,p,t) represents the basic scoring item of the i-th order assigned to the j-th equipment for process p at time t; and L(i,j,p,t) represents the penalty item of the i-th order assigned to the j-th equipment for process p at time t. The basic scoring items are:

[0014] in, This represents the dynamic weight of the k-th production scheduling attribute at time t, which is adjusted in real time based on the production scheduling attribute. This represents the score of the k-th scheduling attribute when the i-th order is assigned to the j-th device at time t; The penalty item is:

[0015] in, and These are the weighting coefficients for the device replacement penalty and the anomaly penalty, respectively. This represents the machine change penalty term at time t for the i-th order being assigned to the j-th equipment to perform process p. This represents the equipment exception penalty item at time t where the i-th order is assigned to the j-th equipment to perform process p.

[0016] Furthermore, the method for adjusting the dynamic weights is as follows: Define the on-site status vector based on production scheduling attributes; Generate a weighted response function based on the on-site state vector:

[0017] in, This represents the weighted response of the k-th production scheduling attribute at time t. This represents the basic weight parameter of the k-th production scheduling attribute. This vector represents the response coefficients of the k-th production scheduling attribute to the field state vector. This represents the state vector at time t. The dynamic weights are obtained by normalizing the weight response.

[0018] Furthermore, the production scheduling objective function is:

[0019] in, Let S(i,j,p,t) be the candidate production scheduling set at time t, and let S(i,j,p,t) represent the comprehensive production scheduling score of the i-th order at time t being assigned to the j-th equipment to perform process p. Let x(i,j,p) be a binary decision variable representing the candidate production scheduling set. When x(i,j,p)=1, it means that at time t, the process p of the i-th order is assigned to the j-th device; when x(i,j,p)=1, it means that at time t, the process p of the i-th order is assigned to the j-th device. When )=0, it means that at time t, the process p of the i-th order has not been assigned to the j-th device; The constraints include: Process sequence constraints are expressed as follows:

[0020] in, This represents the start time of the next process after process p in the i-th order. This indicates the next process after process p. This represents the end time of process p in the i-th order. This represents the preparation or transfer time between process p of the i-th order and the next process after process p; The equipment processing time constraint is expressed as follows:

[0021]

[0022] Where End(i,p) represents the end time of process p in the i-th order, and Start(i,p) represents the start time of process p in the i-th order. This represents the processing time at time t when the i-th order is assigned to the j-th device to perform operation p. This represents the production quantity of the i-th order. This represents the historical corrected production efficiency of the j-th device in process p. To prevent the denominator from being 0; The constraint that equipment processing times do not overlap is expressed as follows:

[0023]

[0024] in, This represents the start time of process p in the r-th order. This represents the end time of process p in the r-th order. Represent a positive number. For sorting variables, The i-th order was assigned to the j-th device to perform process p before the r-th order. The i-th order is assigned to the j-th device to perform process p after the r-th order.

[0025] Furthermore, the local rearrangement optimization model is as follows:

[0026]

[0027]

[0028] in, Let represent the rearranged product set at time t. This represents the rearrangement payoff of the production set affected at time t. This represents the scheduling disturbance of the unaffected scheduling set at time t, where η is the stability constraint coefficient. This represents the set of production schedules affected at time t. Let S(i,j,p,t) represent the candidate production scheduling set at time t, and let S(i,j,p,t) represent the comprehensive production scheduling score of the i-th order at time t being assigned to the j-th equipment to execute process p. Let x(i,j,p) be the decision variable for assigning process p of order i to equipment j at time t. When x(i,j,p)=1, it means that process p of the i-th order is assigned to the j-th device; when x(i,j,p)=1, it means that process p of the i-th order is assigned to the j-th device. When )=0, it means that at time t, the process p of the i-th order has not been assigned to the j-th device. Let represent the decision variable where process p of order i is assigned to equipment j at time t-1.

[0029] Furthermore, the production scheduling parameters are optimized according to the following steps: Production scheduling parameters should include at least: ,in, This represents the set of production scheduling parameters at time t. This represents the basic weight parameter of the k-th production scheduling attribute at time t. Let represent the vector of response coefficients of the k-th production scheduling attribute at time t to the field state vector. and These are the weight coefficients for the machine switching penalty term and the anomaly penalty term at time t, respectively. Let be the stability constraint coefficient of the local rearrangement model at time t; Production losses were determined based on actual production feedback. The scheduling parameters are iteratively updated based on the scheduling loss.

[0030] The present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described semiconductor production line scheduling method.

[0031] The present invention also proposes an electronic device, including a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to execute the above-described semiconductor production line scheduling method.

[0032] The present invention also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-described semiconductor production line scheduling method.

[0033] The beneficial effects of the technical solution provided by this invention are: This invention generates a candidate production schedule set based on orders, equipment, and processes. It then generates a comprehensive production schedule scoring function based on scheduling attributes and candidate schedules, including basic scoring items and penalty items. The dynamic weights of each scheduling attribute in the basic scoring items are adjusted in real-time according to the on-site status of the scheduling attributes. The objective function is to maximize the sum of the comprehensive production schedule scores of the candidate production schedule sets, resulting in a real-time optimal production schedule reflecting the on-site status. In case of anomalies, scores are recalculated only within the affected set, while maintaining the stability of the scheduling results in unaffected areas. Through this local rescheduling mechanism, the system can effectively reduce the scope and cost of rescheduling in abnormal scenarios while ensuring the overall rationality of the production schedule, thereby improving the real-time performance, stability, and interpretability of the semiconductor six-process production scheduling process. This addresses the problems of excessively high rescheduling costs or lack of a global perspective caused by frequent order changes and dynamic changes in equipment status in semiconductor production lines, as well as the failure of static plans. Attached Figure Description

[0034] Figure 1 This is a flowchart of a semiconductor production line scheduling method according to an example of the present invention; Figure 2 This is a block diagram of an electronic device according to an exemplary embodiment of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0036] A flowchart of a semiconductor production line scheduling method according to an example of the present invention is shown below. Figure 1Specifically, it includes: S1. Obtain the semiconductor production line order set, equipment set, and process set, as well as the production scheduling attributes of the orders, equipment, and processes.

[0037] This invention takes a semiconductor packaging and testing production line as an example, involving six major processes: die bonding, wire bonding, molding, electroplating, lead trimming, and testing. Let the order set be: ,in, Let n be the i-th order and n be the total number of orders; the device set is: ,in, Let j be the j-th device, and s be the total number of devices; the process set is: In this context, DB stands for die bonding, WB for wire bonding, MD for encapsulation, ELP for electroplating, TF for wire cutting, and TE for testing.

[0038] The scheduling attributes of orders, equipment, and processes include: order importance, order urgency, delivery timeliness, equipment time efficiency, historical UPH, current equipment status, product model matching degree, packaging form matching degree, fixed machine rules, and upstream and downstream process capacity matching relationship.

[0039] S2. Generate a candidate production schedule set for assigning orders to equipment in the process.

[0040] A candidate production schedule set is generated based on order requirements, process routes, and the real-time capabilities and status of production line equipment. , Let t represent the candidate production scheduling set at time t. This indicates that at time t, the i-th order is assigned to the j-th device to perform process p.

[0041] Define decision variables for candidate production scheduling ,when When, it means that at time t, the process p of order o(i) is assigned to equipment m(j); when At time t, it means that the process p of order o(i) at time t has not been assigned to equipment m(j).

[0042] Let M(p) represent the set of equipment capable of processing operation p. Then, each order satisfies a unique assignment constraint on the corresponding operation:

[0043] If equipment m(j) does not have the processing capability of process p, or if the equipment violates the fixed machine-specific product rule, then let x(i,j,p,t)=0.

[0044] S3. Generate a comprehensive production scheduling score function based on production scheduling attributes and candidate production scheduling sets.

[0045] For candidate scheduling Define the production scheduling comprehensive scoring function as follows, and decompose the comprehensive score into basic scoring items and penalty items:

[0046] Where S(i,j,p,t) represents the comprehensive production scheduling score of the i-th order assigned to the j-th equipment for process p at time t, G(i,j,p,t) represents the basic scoring item of the i-th order assigned to the j-th equipment for process p at time t, and L(i,j,p,t) represents the penalty item of the i-th order assigned to the j-th equipment for process p at time t.

[0047] The basic scoring items are:

[0048] in, This represents the dynamic weight of the k-th production scheduling attribute at time t, which is adjusted in real time based on the production scheduling attribute. This represents the score value of the k-th scheduling attribute when the i-th order is assigned to the j-th device at time t.

[0049] This invention defines a score for seven production scheduling attributes, and the score for each production scheduling attribute is defined as follows:

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056] Wherein, F(i,t) is the importance score of the i-th order at time t; U(i,t) is the urgency score of the i-th order at time t; D(i,t) is the timeliness score of the delivery of the i-th order at time t; E(j,p,t) is the time efficiency score of the j-th equipment executing process p at time t; A(i,j,t) is the product model matching degree of the i-th order assigned to the j-th equipment at time t; B(i,j,t) is the packaging form matching degree of the i-th order assigned to the j-th equipment at time t; and C(p,t) is the capacity synergy score of the upstream and downstream processes of process p at time t.

[0057] The order urgency score is calculated using the Sigmoid function:

[0058] Where α is the urgency growth coefficient, Let d(i) be the planned completion time of order o(i), t be the current time, and τ be the urgency warning time window. The closer the order is to the delivery date, the smaller d(i)-t is, the larger the value of U(i,t) is, and the higher the corresponding production scheduling priority.

[0059] Time efficiency scoring is based on a normalized calculation of historical UPH for the equipment. First, the maximum historical corrected UPH for process p is defined:

[0060] in, This represents the maximum historical corrected production efficiency of process p at time t. This represents the historical corrected production efficiency of equipment m(l) under process p. Let M(p) represent the l-th device, and M(p) represent the set of devices capable of processing step p.

[0061] Time efficiency rating:

[0062] in, This represents the historical corrected production efficiency of the j-th device executing process p at time t. To prevent extremely small constants with a denominator of zero.

[0063] The range of values ​​for product model matching degree A(i,j,t) is defined as follows:

[0064] Specifically, A(i,j,t) = 1 when the product model of the equipment's historical processing matches the product model of the order; A(i,j,t) = a1 when the product series of the equipment's historical processing matches the product series of the order; and A(i,j,t) = 0 when they do not match. Where 0... <a1<1。

[0065] The range of values ​​for the encapsulation matching degree B(i,j,t) is defined as follows:

[0066] Specifically, when the historical packaging form of the equipment matches the order packaging form, B(i,j,t)=1; when the packaging category is the same but the sub-models are different, B(i,j,t)=b1; when the two do not match, B(i,j,t)=0. Where 0... <b1<1。

[0067] Upstream and downstream process capacity synergy scoring is used to constrain capacity balance between die bonding and wire bonding, and between molding and subsequent processes. First, the capacity load ratio of process p is defined:

[0068] In the formula, Let Q(p,t) represent the capacity load ratio of process p at time t, Q(p,t) represent the work-in-process or pending-process quantity of process p at time t, and Cap(p,t) represent the current available capacity of process p at time t.

[0069] Let next(p) represent the next process after process p, then the capacity deviation between upstream and downstream processes is defined as:

[0070] in, This represents the capacity deviation between process p and the next process at time t. This represents the capacity load ratio of the next process after process p at time t.

[0071] The corresponding process collaboration scoring expression is:

[0072] In the formula, β is the sensitivity coefficient for capacity imbalance. The higher the matching degree of upstream and downstream capacity, the smaller Δ(p,next(p),t) is, and the closer C(p,t) is to 1; the greater the capacity difference, the smaller the value of C(p,t) is, and the system will automatically increase the capacity coordination weight or trigger a local rearrangement.

[0073] The penalty items are:

[0074] in, and These are the weighting coefficients for the device replacement penalty and the anomaly penalty, respectively. This represents the machine change penalty term at time t for the i-th order being assigned to the j-th equipment to perform process p. This represents the equipment exception penalty item at time t where the i-th order is assigned to the j-th equipment to perform process p.

[0075] The dynamic weight w(k,t) is a variable parameter that is adjusted in real time based on order delivery dates, equipment load, process backlog, and abnormal events. The field state vector is defined as follows:

[0076] In the formula, These are: order delivery urgency, equipment load imbalance, upstream and downstream capacity imbalance, and intensity of abnormal events.

[0077] For ease of calculation, the weighted response function for the k-th criterion is defined as follows:

[0078] in, This represents the weighted response of the k-th production scheduling attribute at time t. This represents the basic weight parameter of the k-th production scheduling attribute. This vector represents the response coefficients of the k-th production scheduling attribute to the field state vector. Let t represent the state vector at time t.

[0079] The dynamic weights are calculated using the Softmax normalization method.

[0080] in, This represents the weighted response of the r-th production scheduling attribute at time t, where K represents the number of production scheduling attributes in the field state vector, corresponding to... K=4. When an order is nearing its delivery date, equipment load is unbalanced, process capacity is mismatched, or abnormal events increase, the corresponding criterion weight will automatically increase.

[0081] S4. Maximize the sum of the comprehensive production scores of the candidate production schedule set as the production schedule objective function, and solve the production schedule result under the constraints of the set process sequence and processing time.

[0082] The production scheduling objective function is:

[0083] in, Let S(i,j,p,t) be the candidate production scheduling set at time t, and let S(i,j,p,t) represent the comprehensive production scheduling score of the i-th order at time t being assigned to the j-th equipment to perform process p. Let x(i,j,p) be a binary decision variable representing the candidate production scheduling set. When x(i,j,p)=1, it means that at time t, the process p of the i-th order is assigned to the j-th device; when x(i,j,p)=1, it means that at time t, the process p of the i-th order is assigned to the j-th device. When )=0, it means that at time t, the process p of the i-th order has not been assigned to the j-th device; The following constraints must be met simultaneously: Process sequence constraints are expressed as follows:

[0084] in, This represents the start time of the next process after process p in the i-th order. This indicates the next process after process p. This represents the end time of process p in the i-th order. This represents the preparation or transfer time between process p of the i-th order and the next process after process p; The equipment processing time constraint is expressed as follows:

[0085]

[0086] Where End(i,p) represents the end time of process p in the i-th order, and Start(i,p) represents the start time of process p in the i-th order. This represents the processing time at time t when the i-th order is assigned to the j-th device to perform operation p. This represents the production quantity of the i-th order. This represents the historical corrected production efficiency of the j-th device in process p. This is a number that prevents the denominator from being zero.

[0087] The constraint that equipment processing times do not overlap is expressed as follows:

[0088]

[0089] in, This represents the start time of process p in the r-th order. This represents the end time of process p in the r-th order. Represents a sufficiently large positive number. For sorting variables, The i-th order was assigned to the j-th device to perform process p before the r-th order. The i-th order is assigned to the j-th machine to perform operation p after the r-th order. This constraint ensures that only one order can be processed on the same machine at any given time, thus avoiding scheduling conflicts.

[0090] S5. When an abnormal situation occurs, construct an affected production set and an unaffected production set based on the production scheduling results, and construct the rearrangement benefit of the affected production set and the production scheduling disturbance of the unaffected production set.

[0091] When events such as manual order insertion, order cancellation, equipment malfunction, or work order timeout occur, the system does not perform a global rescheduling; instead, it constructs a set of affected production schedules. , This is the set of orders affected by the abnormal event. The set of devices affected by the abnormal event. This represents the set of processes affected by the abnormal event. The set of unaffected production schedules is then obtained. - .

[0092] The system recalculates scores within the affected production scheduling set Ω(t), while striving to maintain stable production scheduling results in unaffected regions. For ease of representation, the rearrangement benefit for the affected regions is defined as:

[0093] The production scheduling disturbance in the unaffected area is defined as:

[0094] in, This represents the rearrangement payoff of the production set affected at time t. This represents the production scheduling disturbance of the production scheduling set that is unaffected at time t. This represents the set of production schedules affected at time t. Let S(i,j,p,t) represent the candidate production scheduling set at time t, and let S(i,j,p,t) represent the comprehensive production scheduling score of the i-th order at time t being assigned to the j-th equipment to execute process p. Let x(i,j,p) be the decision variable for assigning process p of order i to equipment j at time t. When x(i,j,p)=1, it means that process p of the i-th order is assigned to the j-th device; when x(i,j,p)=1, it means that process p of the i-th order is assigned to the j-th device. When )=0, it means that at time t, the process p of the i-th order has not been assigned to the j-th device. Let represent the decision variable where process p of order i is assigned to equipment j at time t-1.

[0095] S6. Based on the rearrangement benefits of the affected set and the scheduling disturbance of the unaffected set, a local rearrangement optimization model is constructed to obtain the final scheduling result.

[0096] The local rearrangement optimization model is then:

[0097] in, Let represent the rearranged product set at time t. This is a stability constraint coefficient used to reduce production scheduling disturbances in unaffected areas. This represents the set of production schedules affected at time t. Through this local rescheduling mechanism, the system can effectively reduce the scope and cost of rescheduling in abnormal scenarios while ensuring the overall rationality of production scheduling, thereby improving the real-time performance, stability, and interpretability of the semiconductor six-process production scheduling process.

[0098] Optimize production scheduling parameters according to the following steps: Production scheduling parameters should include at least: ,in, This represents the set of production scheduling parameters at time t. This represents the basic weight parameter of the k-th production scheduling attribute at time t. Let represent the vector of response coefficients of the k-th production scheduling attribute at time t to the field state vector. and These are the weight coefficients for the machine replacement penalty term and the anomaly penalty term at time t, respectively. Let be the stability constraint coefficient of the local rearrangement model at time t. This represents the urgency growth coefficient at time t in the urgency score. This represents the sensitivity coefficient to capacity imbalance at time t in the capacity synergy score of upstream and downstream processes.

[0099] Based on actual production feedback, the production scheduling loss is determined, and the loss function is as follows:

[0100] in, To lose weight, This indicates the degree of order delay at time t. This represents the device idle rate at time t. This indicates the degree of capacity imbalance between upstream and downstream processes at time t. This indicates the timeout level of the work order at time t. This indicates the degree of disturbance caused by production schedule changes at time t.

[0101] Iterative updates to scheduling parameters based on scheduling losses can be achieved using gradient calculation:

[0102] in, For learning rate, This represents the feasible region projection operation, used to ensure that each weight, penalty coefficient, and threshold remains within a preset range. and Let be the set of production scheduling parameters at time t and time t+1. Indicates about The gradient operator.

[0103] When gradient calculation is not used, an incremental update method based on feedback error can also be used:

[0104] in, Represents the time t. The feedback error corresponds to each production scheduling attribute, where K is the number of production scheduling attributes, which is set to 4 in this invention. For example, when the order delay error is large, the weights of on-time delivery and order urgency are increased; when the equipment idle rate is high, the weights of equipment utilization and load balancing are increased; when there is a large imbalance between upstream and downstream capacity, the weight of process coordination is increased. This indicates the step size for weight updates; a larger step size indicates a more drastic weight adjustment. and This represents the dynamic weight of the k-th production scheduling attribute at time t and time t+1. This represents the weight of the r-th production scheduling attribute at time t.

[0105] In one exemplary embodiment, a computer-readable storage medium is included, which stores a computer program that, when executed by a processor, implements the semiconductor production line scheduling method described above.

[0106] Please see Figure 2 In one exemplary embodiment, the device further includes an electronic device including at least one processor, at least one memory, and at least one communication bus.

[0107] The memory stores a computer program, which includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the semiconductor production line scheduling method described above.

[0108] In one exemplary embodiment, a computer program product is proposed, including a computer program / instructions that, when executed by a processor, implement the steps of the semiconductor production line scheduling method described above.

[0109] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A semiconductor production line scheduling method, characterized in that, Includes the following steps: S1. Obtain the semiconductor production line order set, equipment set, and process set, as well as the production scheduling attributes of the orders, equipment, and processes; S2. Generate a candidate production schedule set for assigning orders to equipment in the process; S3. Generate a comprehensive production scheduling score function based on production scheduling attributes and candidate production scheduling sets; S4. Maximize the sum of the comprehensive production scores of the candidate production schedule set as the production schedule objective function, and solve the production schedule result under the constraints of the set process sequence and processing time. S5. When an abnormal situation occurs, construct an affected production schedule set and an unaffected production schedule set based on the production scheduling results, and construct the rearrangement benefit of the affected production schedule set and the production schedule disturbance of the unaffected production schedule set. S6. Based on the rearrangement benefits of the affected set and the scheduling disturbance of the unaffected set, a local rearrangement optimization model is constructed to obtain the final scheduling result.

2. The semiconductor production line scheduling method according to claim 1, characterized in that, The comprehensive scoring function for production scheduling is: Where S(i,j,p,t) represents the comprehensive production scheduling score of the i-th order assigned to the j-th equipment for process p at time t; G(i,j,p,t) represents the basic scoring item of the i-th order assigned to the j-th equipment for process p at time t; and L(i,j,p,t) represents the penalty item of the i-th order assigned to the j-th equipment for process p at time t. The basic scoring items are: in, This represents the dynamic weight of the k-th production scheduling attribute at time t, which is adjusted in real time based on the production scheduling attribute. This represents the score of the k-th scheduling attribute when the i-th order is assigned to the j-th device at time t; The penalty item is: in, and These are the weighting coefficients for the device replacement penalty and the anomaly penalty, respectively. This represents the machine change penalty term at time t for the i-th order being assigned to the j-th equipment to perform process p. This represents the equipment exception penalty item at time t where the i-th order is assigned to the j-th equipment to perform process p.

3. The semiconductor production line scheduling method according to claim 2, characterized in that, The method for adjusting dynamic weights is as follows: Define the on-site status vector based on production scheduling attributes; Generate a weighted response function based on the on-site state vector: in, This represents the weighted response of the k-th production scheduling attribute at time t. This represents the basic weight parameter of the k-th production scheduling attribute. This vector represents the response coefficients of the k-th production scheduling attribute to the field state vector. This represents the state vector at time t. The dynamic weights are obtained by normalizing the weight response.

4. The semiconductor production line scheduling method according to claim 1, characterized in that, The production scheduling objective function is: in, Let S(i,j,p,t) be the candidate production scheduling set at time t, and let S(i,j,p,t) represent the comprehensive production scheduling score of the i-th order at time t being assigned to the j-th equipment to perform process p. Let x(i,j,p) be a binary decision variable representing the candidate production scheduling set. When x(i,j,p)=1, it means that at time t, the process p of the i-th order is assigned to the j-th device; when x(i,j,p)=1, it means that at time t, the process p of the i-th order is assigned to the j-th device. When )=0, it means that at time t, the process p of the i-th order has not been assigned to the j-th device; The constraints include: Process sequence constraints are expressed as follows: in, This represents the start time of the next process after process p in the i-th order. This indicates the next process after process p. This represents the end time of process p in the i-th order. This represents the preparation or transfer time between process p of the i-th order and the next process after process p; The equipment processing time constraint is expressed as follows: Where End(i,p) represents the end time of process p in the i-th order, and Start(i,p) represents the start time of process p in the i-th order. This represents the processing time at time t when the i-th order is assigned to the j-th device to perform operation p. This represents the production quantity of the i-th order. This represents the historical corrected production efficiency of the j-th device in process p. To prevent the denominator from being 0; The constraint that equipment processing times do not overlap is expressed as follows: in, This represents the start time of process p in the r-th order. This represents the end time of process p in the r-th order. Represent a positive number. For sorting variables, The i-th order was assigned to the j-th device to perform process p before the r-th order. The i-th order is assigned to the j-th device to perform process p after the r-th order.

5. A semiconductor production line scheduling method according to claim 3, characterized in that, The local rearrangement optimization model is as follows: in, Let represent the rearranged product set at time t. This represents the rearrangement payoff of the production set affected at time t. This represents the scheduling disturbance of the unaffected scheduling set at time t, where η is the stability constraint coefficient. This represents the set of production schedules affected at time t. Let S(i,j,p,t) represent the candidate production scheduling set at time t, and let S(i,j,p,t) represent the comprehensive production scheduling score of the i-th order at time t being assigned to the j-th equipment to execute process p. Let x(i,j,p) be the decision variable for assigning process p of order i to equipment j at time t. When x(i,j,p)=1, it means that process p of the i-th order is assigned to the j-th device; when x(i,j,p)=1, it means that process p of the i-th order is assigned to the j-th device. When )=0, it means that at time t, the process p of the i-th order has not been assigned to the j-th device. Let represent the decision variable where process p of order i is assigned to equipment j at time t-1.

6. A semiconductor production line scheduling method according to claim 5, characterized in that, Optimize production scheduling parameters according to the following steps: Production scheduling parameters should include at least: ,in, This represents the set of production scheduling parameters at time t. This represents the basic weight parameter of the k-th production scheduling attribute at time t. Let represent the vector of response coefficients of the k-th production scheduling attribute at time t to the field state vector. and These are the weight coefficients for the machine switching penalty term and the anomaly penalty term at time t, respectively. Let be the stability constraint coefficient of the local rearrangement model at time t; Production losses were determined based on actual production feedback. The scheduling parameters are iteratively updated based on the scheduling loss.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

8. An electronic device, characterized in that, The device includes a processor and a memory, the processor being interconnected with the memory, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to perform the method as described in any one of claims 1 to 6.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the method described in any one of claims 1 to 6.

Images