Semiconductor wafer manufacturing system scheduling method based on dynamic adjustment A*heuristic function

Abstract

The invention particularly relates to a semiconductor wafer manufacturing system scheduling method based on a dynamic adjustment A * heuristic function, which comprises the following steps of: establishing a Petri net model according to a processing flow of combined equipment related to wafer manufacturing, and initializing system resources in the Petri net model; and searching an optimal scheduling path in the Petri net model by using an improved A * algorithm based on node depth dynamic adjustment heuristic function weight. A dynamic adjustment heuristic function mechanism is introduced into the A * algorithm, and the scheduling of the wafer manufacturing system is optimized in combination with the node depth weight, so that path selection at different stages can be more flexibly adapted in the search process. By constructing a cost value priority comparator based on combination of node depth and a heuristic function, nodes with higher priorities can be more effectively screened out, and search on invalid paths is reduced. And dynamically adjusted cost evaluation is introduced into the Petri network model, so that the use of resources such as single-cavity and double-cavity machines is relatively balanced, and the condition of resource limitation or overload is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor wafer manufacturing, and particularly to a scheduling method for a semiconductor wafer manufacturing system based on dynamically adjusting the A* heuristic function. Background Art

[0002] As a supporting industry for integrated circuit manufacturing, the semiconductor industry provides key materials and manufacturing equipment for the integrated circuit industry. The semiconductor material required for producing circuit chips is a silicon wafer, and the silicon crystal carrier of the circuit chip is called a wafer. Wafer manufacturing is a manufacturing industry with extremely high complexity today. Due to the very complex processing process of wafers, in order to ensure the quality and accuracy of wafers, more and more wafer manufacturers choose to use combined equipment to process wafers. The operation of the combined equipment refers to the transportation and processing process of a fixed number of wafers by a manipulator between different processing modules. The processing modules and the manipulator are all resources competed by each wafer. How to allocate resources more reasonably is the process of system scheduling.

[0003] Petri net is a modeling tool used for designing and analyzing systems, and can strongly describe some behavioral characteristics such as asynchrony, concurrency, and deadlock during the operation of combined equipment through digital and graphical expressions. Therefore, it is widely used in related fields such as semiconductor wafer manufacturing.

[0004] The A* algorithm is an algorithm that combines a heuristic function and is applied to path search. Path search is actually a breadth-first search, with breadth as the priority, gradually spreading outwards until the end point is found. The definition of the heuristic function directly affects the search performance of the A* algorithm. In order to improve the efficiency and accuracy of optimal path selection, an improved evaluation function is used for optimization based on the A* algorithm. By introducing a weighted process of node depth, a heuristic function is innovatively designed. The introduction of the A* algorithm can optimize the research on the operation control of the semiconductor wafer manufacturing system.

[0005] When scheduling the semiconductor wafer manufacturing system in combination with the A* algorithm, the design and calculation process of the algorithm often need to consider the influence of time factors, and avoid some unnecessary states according to the marking and time characteristics to improve the algorithm efficiency. Therefore, it is necessary to comprehensively consider the optimization strategy of optimal path search.

[0006] It should be noted that the information disclosed in the above background art section is only used to strengthen the understanding of the background of the present invention. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] To solve the problems of complex scheduling in the wafer manufacturing process and improve the performance and accuracy of the algorithm in the process of searching for the optimal path, the present invention provides a scheduling method for a semiconductor wafer manufacturing system based on dynamically adjusting the A* heuristic function.

[0008] Other features and advantages of the present invention will become apparent from the following detailed description or will be partially learned through the practice of the present invention.

[0009] According to a first aspect of the present invention, there is provided a scheduling method for a semiconductor wafer manufacturing system based on dynamically adjusting the A* heuristic function, the method comprising:

[0010] Establish a timed Petri net model according to the processing flow of the combined equipment involved in wafer manufacturing, and initialize the system resources in the timed Petri net model;

[0011] Use an improved A* algorithm based on dynamically adjusting the heuristic function weight according to the node depth to search for the optimal scheduling path in the timed Petri net model; including:

[0012] Put the initial state into the OPEN list, indicating the state to be expanded;

[0013] Take out the state with the minimum node cost from the OPEN list for expansion, where a weight related to the node depth is introduced before the heuristic function included in the node cost;

[0014] Generate all possible successor states according to the available transitions of the current state;

[0015] Calculate the node cost for each successor state that satisfies the transition firing rule, and add it to the OPEN list according to the cost value in the order of arrangement, and take out the state with the minimum node cost from the OPEN list for expansion until the target node;

[0016] Connect the start node, multiple expanded nodes and the target node to generate the optimal scheduling path.

[0017] In some exemplary embodiments, the combined equipment includes a vacuum lock module, a processing module and a manipulator, and the wafer processing flow is as follows:

[0018] The wafer is transferred from the loading and unloading position to the vacuum lock module by the manipulator, enters the processing module after completing the vacuum environment conversion, and returns to the loading and unloading position after processing.

[0019] In some exemplary embodiments, the timed Petri net model adopts a process-oriented modeling method, and the tokens correspond to different states in the movement process of the combined equipment, where:

[0020] Places are divided into processing places and resource places. In the initial state, the processing places represent the modules to be processed, and the tokens in them represent the wafers to be processed; the resource places represent the resource control places for controlling the capacity, and the tokens in them represent the resources in the equipment.

[0021] Transitions represent the operations of the manipulator, including picking, placing, and exchanging actions.

[0022] In some exemplary embodiments, the initialization process of the system resources in the timed Petri net model includes:

[0023] Configure the initial markings of the processing places and resource places. The number of tokens in the processing places is consistent with the number of wafers to be processed, and the number of tokens in the resource places is consistent with the available resources of the equipment.

[0024] In some exemplary embodiments, the calculation formula of the node cost is specifically expressed as:

[0025] f′(n) = g(n) + ω·h(n)

[0026]

[0027] Wherein, g(n) is the actual cost from the initial node to the current node, h(n) is the value of the improved heuristic function, α is the adjustment parameter, curDepth() is the depth of the node, and ω is the weight function.

[0028] In some exemplary embodiments, the value of the improved heuristic function is specifically expressed as:

[0029] h max (M) = max{h R (M), h RWT (M)}

[0030]

[0031] Wherein, represents the minimum processing time for using the resource r j in the marking M, RWT i represents the shortest time required for a token to move from the place p i to the target place, represents the i-th element in an n×1-dimensional vector, and this vector contains the remaining processing times of all places p i ; here, the marking M in the timed Petri net model is the node n in the A* algorithm.

[0032] In some exemplary embodiments, the transition firing rule includes determining the firing condition and firing timing of each transition in a timed Petri net;

[0033] The firing condition means that to fire a certain transition, all input places connected to the transition must satisfy certain token conditions. Specifically, there must be a sufficient number of tokens in the input places;

[0034] The firing timing refers to the time point at which a transition is fired, provided that given time regions are satisfied simultaneously under the condition that the transition is enabled.

[0035] In some exemplary embodiments, the method further includes performing a pruning operation on the OPEN table, and the pruning operation includes:

[0036] When the OPEN table reaches a preset capacity threshold, only the top K nodes with the smallest cost values are retained, and the remaining nodes are removed to reduce memory occupancy and computational overhead.

[0037] According to a second aspect of the present invention, there is provided a storage medium having stored thereon a computer program, which when executed by a processor implements the semiconductor wafer manufacturing system scheduling method based on dynamically adjusting the A* heuristic function described in the first aspect above.

[0038] According to a third aspect of the present invention, there is provided a computer program product having stored thereon a computer program, which when executed by a processor implements the semiconductor wafer manufacturing system scheduling method based on dynamically adjusting the A* heuristic function described in the first aspect above.

[0039] According to a fourth aspect of the present invention, there is provided an electronic device, including:

[0040] a processor; and

[0041] a memory for storing executable instructions of the processor;

[0042] wherein the processor is configured to implement the semiconductor wafer manufacturing system scheduling method based on dynamically adjusting the A* heuristic function described in the first aspect above when executing the executable instructions.

[0043] The scheduling method of the semiconductor wafer manufacturing system based on dynamically adjusting the A* heuristic function provided by the embodiments of the present invention introduces a mechanism for dynamically adjusting the heuristic function into the A* algorithm, and combines the node depth weight to optimize the scheduling of the wafer manufacturing system. The above algorithm can more flexibly adapt to path selection in different stages during the search process. For complex wafer manufacturing scenarios, this dynamics can more accurately reflect the actual production situation, thereby more precisely finding the optimal scheduling path. By constructing a cost value priority comparator based on the combination of node depth and heuristic function, nodes with higher priorities can be more effectively screened out, reducing the search on invalid paths. Introducing dynamically adjusted cost evaluation in the Petri net model relatively balances the use of resources such as single and double-chamber machines, reducing the situation of resource constraints or overload. Finally, pruning operations are performed on the size of the OPEN table, which limits the number of nodes that need to be retained during the execution of the algorithm, considering the memory overhead issue while optimizing the algorithm efficiency.

[0044] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0046] Figure 1 It is a flowchart of a scheduling algorithm for a semiconductor wafer manufacturing system optimized based on dynamically adjusting the A* heuristic function provided by an embodiment of the present invention;

[0047] Figure 2 It is a schematic diagram of a time Petri net model of a combined device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present invention will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0049] In addition, the attached drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0050] In view of the disadvantages and deficiencies of the prior art, in this exemplary embodiment, a scheduling method for a semiconductor wafer manufacturing system based on dynamically adjusting the A* heuristic function is provided. Referring to Figure 1 as shown, the specific steps may include the following:

[0051] Step S1: Establish a Petri net model according to the processing flow of the combined equipment involved in wafer manufacturing, and initialize the system resources in the Petri net model;

[0052] Step S2: Use an improved A* algorithm based on dynamically adjusting the heuristic function weight according to the node depth to search for the optimal scheduling path in the Petri net model, including:

[0053] Put the initial state into the OPEN list, indicating the state to be expanded;

[0054] Take out the state with the minimum node cost from the OPEN list for expansion, where a weight related to the node depth is introduced before the heuristic function included in the node cost;

[0055] Generate all possible successor states according to the available transitions of the current state;

[0056] Calculate the node cost for each successor state, and add them to the OPEN list in the order of the cost values. Take out the state with the minimum node cost from the OPEN list for expansion until the target node;

[0057] Connect the starting node, multiple expanded nodes and the target node to generate the optimal scheduling path.

[0058] Next, each step of the phased array radar design method in this exemplary embodiment will be described in more detail with reference to the accompanying drawings and embodiments.

[0059] In step S1, a Petri net model is established according to the processing flow of the combined equipment involved in wafer manufacturing, and the system resources in the Petri net model are initialized.

[0060] Exemplarily, in a wafer manufacturing scenario, a timed Petri net model is established for the processing flow of combined equipment such as single- and dual-chamber machines. The combined equipment module in the embodiments of the present invention includes: a TM (Transfer Module) single-arm robot, a TM (Transfer Module) dual-arm robot, an LP (Load Port) loading and unloading position, an AL (Aligner) wafer alignment module, a PM (Processing Module) processing module, and an LL (Load Lock) vacuum lock. Among them, the TM is a transfer module, and the robot can perform detailed operations such as wafer placement, retrieval, and exchange, and can operate on a single wafer or two wafers at a time. In the combined equipment in the embodiments of the present invention, the wafer processing process is generally as follows: After the wafer is taken out from the LP by the TM, it is first placed in the LL module (the wafer processing process needs to be carried out in a vacuum environment). After switching to the vacuum environment, the wafer flows to the PM module for processing. After the processing is completed in sequence according to the set procedures, the wafer is put into the LL module again. After the vacuum environment is released, the wafer will be put into the LP again.

[0061] Further, in the processing process of the combined equipment in the present invention, multiple wafers are input for processing each time it runs. Therefore, for establishing the Petri net model of the combined equipment, POPN (Process-oriented Petri Net), that is, a process-oriented modeling method, is selected. POPN is modeled based on the processing process of the manufacturing system, and the processing process of the workpiece is intuitively presented by constructing a model. This modeling method requires a clear understanding of the procedures for workpiece processing. On this basis, after modeling each procedure and the resources in the system, the established models are connected according to the processing sequence. In this model, the markings correspond to different states of the combined equipment during operation, the places are used to represent each module or resource in the equipment, and the actions of the robot are represented by transitions. In the initial state, the token in the processing place represents the wafer to be processed, and the token in the resource place represents the resources in the equipment, such as the occupancy of the robot.

[0062] When combining the Petri net with the manufacturing system, in order to better consider the time factor of the system, a Petri net with time factors is selected. The Petri net has two types of nodes, places and transitions, which are connected by directed arcs. Among them, each marking corresponds to each state of the system, the place represents the state of a certain module of the system, and the transition can change the state. Therefore, the process of modeling the combined equipment is to use places, transitions, and connection arcs to characterize the state changes of the system. Based on the high complexity of the processing process of the combined equipment, POPN (Process-oriented PetriNet, POPN), that is, a process-oriented modeling method, is used during operation.

[0063] Furthermore, in the timed Petri net constructed in the embodiment of the present invention, the markings correspond to different states during the movement of the combined equipment. The places are divided into processing places and resource places. In the initial state, the processing places represent the modules to be processed, and the tokens inside represent the wafers to be processed. The resource places represent the resource control places for controlling the capacity, and the tokens therein represent the resources in the equipment.

[0064] Please refer to Figure 2 , Figure 2 FIG. is a schematic diagram of a timed Petri net model of a combined equipment provided by an embodiment of the present invention. This equipment contains 3 double-arm manipulators, namely ATR, VTR1, and VTR2. ATR is responsible for picking and placing wafers in the non-vacuum section, while VTR1 and VTR2 are responsible for picking and placing wafers in the two combined equipments. LAB and LCD are two vacuum lock modules. The processing process of wafers requires a strict vacuum environment, and this module can realize the conversion between the vacuum end and the atmosphere end. The transitions represent the operations of the manipulators, represented by squares, including pick, place, and swap actions. LP1 represents the resource place that has not been processed yet, with an initial value of 13 pieces. The general equipment processing flow is as follows: After the wafers to be processed are taken away from LP1, they enter the vacuum lock LAB / LCD. After passing through the vacuum lock, they enter the processing chamber CHA for processing. After processing is completed, they enter the vacuum lock LAB / LCD again and enter the cooling module. After cooling, they are placed into LP1Comp.

[0065] More specifically, during the process of constructing the timed Petri net model in the embodiment of the present invention, it also includes determining the transition firing rules of the model. The firing rules define when and how to change from one marking to another. In a timed Petri net, it is necessary to determine the firing conditions and firing times of each transition. To fire a certain transition, all input places connected to this transition must meet certain token conditions, specifically, there must be a sufficient number of tokens in the input places. The firing time refers to the time point when the transition is fired, provided that it can be fired while meeting the given time region under the condition that the transition is enabled.

[0066] In step S2, an improved A* algorithm based on dynamically adjusting the heuristic function weight according to the node depth is used to search for the optimal scheduling path in the Petri net model.

[0067] The process of using the A* search algorithm based on cost estimation to search the Petri net model is as follows:

[0068] (1) Put the initial state into the OPEN list, indicating the state to be expanded, and the CLOSED list is used to store the states that have been expanded.

[0069] (2) Take out the state with the minimum node cost f′(n) from the OPEN list for expansion;

[0070] (3) Generate all possible successor states according to the available transitions of the current state;

[0071] (4) In the Petri net model, a transition needs to meet certain enabling conditions to be triggered and transfer to a new state;

[0072] (5) Calculate f′(n) for each successor state and add it to the OPEN list according to the cost value in ascending order;

[0073] (6) Move the expanded state to the CLOSED list to prevent repeated processing.

[0074] In an embodiment of the present invention, the Petri net model is scheduled using a dynamic weighted A* algorithm. As a traditional heuristic search algorithm, the core expression of the A* algorithm is f(n) = g(n) + h(n), where g(n) represents the actual cost from the initial node to the current node, h(n) represents the estimated cost from the current node to the target node, and f(n) represents an approximation of the cost of the entire path. There are two important sets in the process of the A* algorithm to find a path, named the OPEN list and the CLOSED list respectively. The OPEN list is a default ascending priority queue set that stores all unexpanded nodes according to the cost value, and each time a node with the smallest heuristic value is selected for expansion, while the CLOSED list is used to store the nodes that have been expanded.

[0075] Furthermore, the above node cost is described in detail.

[0076] Establish an initial heuristic function, calculate the node depth of each node expanded during the search process, introduce the weight of the node depth into the estimation function, and dynamically adjust the heuristic function with the node depth.

[0077] In the A* algorithm, the setting of the heuristic function is crucial. Its main role is to guide the search direction and estimate the cost from the current node to the target node. Its formula expression is: f(n) = g(n) + h(n). g(n) is the path cost from the starting node to the current node n, h(n) is the heuristic estimate value representing the estimated cost from the current node to the target node, and f(n) represents the evaluation function of node n.

[0078] Among them

[0079] marking.getgTime() is the global time of the current node. During the path search process, the node depths of all expanded nodes in the OPEN table are calculated, and a weight function ω related to the node depth curDepth() is defined. The function is to attach more importance to the heuristic function h(n) when the depth is small, and reduce the dependence on h(n) when the node depth is large. The specific expression formula for introducing the node depth weight function is:

[0080]

[0081] Among them, α is a tuning parameter used to control the rate of weight decay. A larger α value will accelerate the decrease of the weight, enabling the algorithm to rely on the known path cost faster. When calculating the total cost function f(n), the original heuristic function is replaced with the new weight function:

[0082] f′(n) = g(n) + ω·h(n)

[0083]

[0084] h max (n) = max{h R (n), h RWT (n)}

[0085] Furthermore, the total remaining time of the resources r j required for the marked places in the marking M is expressed as:

[0086]

[0087] Among them, H(r j ) represents the holder of the resource r j .

[0088] The minimum processing time of using the resource r j in the marking M is expressed as Among them, represents the least processing time of using the resource r i in the place p j .

[0089] Finally, an estimation function h R (M) can be obtained, that is, the cost required from the marking M to the final marking, specifically expressed as:

[0090] Therefore, in an embodiment of the present invention, an improved heuristic function is used, which combines two heuristic functions. One function is h R (M) as described above, and the other heuristic function has the calculation formula: where RWT i represents the shortest time required for a token to move from place p i to the target place, represents the i-th element in an n×1 dimensional vector that contains the remaining processing times of all places p i Finally, an improved heuristic function that combines the above two heuristic functions h RWT (M) and h R (M) can be obtained, specifically expressed as: h max (M) = max{h R (M), h RWT (M)}.

[0091] Furthermore, a priority comparator is constructed. After calculating the node cost values that combine the node depth and the heuristic function, they are input into the priority comparator for screening.

[0092] The role of the priority comparator is to sort the nodes so as to select the optimal nodes for expansion. The traditional A* algorithm simply calculates the sum of the actual cost of the current node and the estimated cost from the current node to the target node. The embodiments in the present invention simultaneously consider the weight adjustment of the node depth. The advantage of such a setting is that it can dynamically adjust the heuristic estimation according to the depth of the node, making the selection of nodes more in line with the actual needs of the problem, and enabling the heuristic function to have different influences at different depths. For example, when the depth of the expanded node is relatively small, more attention is paid to the heuristic function h(n), and when the depth is relatively large, the dependence on h(n) is reduced.

[0093] After using the priority comparator, the OPEN list will be automatically sorted according to the f′(n) values of each node. When entering the expanded node, the newly generated nodes are inserted into the OPEN list, and then the priority order of the nodes is automatically maintained according to the priority comparator. At this time, the nodes in the OPEN list have been sorted according to the priority, and the optimal node is always located at the head of the queue. When extracting nodes from the OPEN list during the A* search algorithm, the element at the head of the queue can be directly taken out as the optimal node for expansion.

[0094] Furthermore, pruning operations are performed in the OPEN list, and only some nodes with better function values are retained. The improved A* algorithm based on dynamically adjusting the heuristic function weight according to the node depth is used to find the optimal solution in the Petri net model.

[0095] Specifically, the specific process of using the A* search algorithm based on cost estimation to search the Petri net model is as follows:

[0096] (1) Put the initial state into the OPEN list, indicating the state to be expanded, and the CLOSED list is used to store the states that have been expanded;

[0097] (2) Take out the state with the minimum node cost f′(n) from the OPEN list for expansion;

[0098] (3) Generate all possible successor states according to the available transitions of the current state;

[0099] (4) In the Petri net model, a transition needs to meet certain enabling conditions to be triggered and transfer to a new state;

[0100] (5) Calculate f′(n) for each successor state and add it to the OPEN list according to the cost value in the sorted order;

[0101] (6) Move the expanded state to the CLOSED list to prevent repeated processing.

[0102] When the OPEN list reaches a certain size, by pruning the OPEN list and retaining the better states, it is possible to reduce computing resources and implement a more efficient scheduling algorithm. By evaluating and screening the states currently stored in the OPEN list, only the states with better cost values are retained, thereby reducing those redundant states that are less likely to be the optimal solution. This pruning process is based on the dynamic adjustment strategy of the A* heuristic function, effectively reducing unnecessary computational overhead, avoiding excessive expansion of inefficient paths, and reducing the occupation of memory and processing resources. At the same time, this pruning mechanism ensures the core performance of the algorithm, that is, while maintaining the ability to search for the optimal solution, it searches and optimizes the scheduling path in a more efficient manner. In this way, the system can accelerate the scheduling optimization process in complex semiconductor manufacturing scenarios, realize the rationalization of resource allocation, and improve the overall scheduling efficiency.

[0103] The nodes with better cost values are selected through a priority comparator and put into the OPEN list. To prevent the OPEN list from expanding excessively and consuming a large amount of memory, pruning operations are performed on the OPEN list. By restricting the size of the OPEN list, this method can preferentially retain some nodes with higher potential, discard some nodes that may not have optimal solutions, reduce the number of nodes to be processed, thereby reducing the computational cost, and enabling the algorithm to avoid searching relatively suboptimal paths and thus avoid wasting resources.

[0104] Based on a Petri net model of a semiconductor wafer manufacturing system optimized by dynamically adjusting the A* heuristic function designed according to the present invention, determine the best completion time, running time, and expansion identification number:

[0105] Calculate the extended identification number according to the current identification, and judge the current identification: if it is the target identification, output the extended identification number, and at the same time output the Makespan (optimal completion time) and running time corresponding to the initial identification to the target identification; if it is not the target identification, calculate the transition cost value in the transition sequence according to the given heuristic evaluation function above, arrange them in order in the OPEN table according to the calculated cost value, select the head element in the priority queue as the firing transition and expand the new current node, and return to the step of calculating the extended identification number with the changed current identification until the final target node is found.

[0106] The final output data results of the embodiments of the present invention include: Makespan, running time, and extended identification number. Among them, Makespan is the optimal completion time, that is, the global time to reach the target identification. The extended identification number is the number of nodes expanded in the A* search path. The running time is the time counted after the computer executes the entire A* algorithm.

[0107] Table 1 shows the comparison results of the output indicators of the ordinary heuristic function and the dynamically adjusted A* heuristic function optimization algorithm designed by the present invention

[0108] Makespan / s Running time / s Number of extended identifiers Ordinary heuristic function 397691 6.001 319182 Dynamically adjusted A* heuristic function 398385 0.618 9140

[0109] It can be seen from Table 1 that the dynamically adjusted A* heuristic function optimization algorithm designed by the present invention has great advantages in running time and extended identification number. The reduction of the extended identification number means that the searched state space is greatly reduced. In complex systems, the state space may grow exponentially, especially in concurrent and resource-competitive systems. Reducing the extended identification number helps to control the scale of the state space and avoid the problem of state explosion. At the same time, the reduction of the running time also shows that by reducing the extended identification number, the computational complexity can be reduced and the running speed of the algorithm can be improved. During the expansion of the state space, each identification needs to be stored in memory. If the identification system is expanded without control, the problem of out-of-memory may occur. Reducing the extended identification number can also significantly reduce the memory consumption, enabling the analysis process to run under limited memory resources, and is more suitable for large-scale Petri net models.

[0110] A semiconductor wafer manufacturing system scheduling method based on the optimization of the A* heuristic function with dynamic adjustment. By introducing dynamic adjustment of the node depth into the heuristic function, the algorithm can more flexibly adapt to path selection at different stages during the search process. This newly proposed estimation function pays more attention to the heuristic function h(n) when the depth is small through the calculation of the node depth, and reduces the dependence on h(n) when the depth is large. By dynamically adjusting the weight of the estimation function in the heuristic calculation, the algorithm can approach the optimal solution faster, saving search time and improving the convergence speed of the scheduling algorithm, which is particularly crucial in the wafer manufacturing process with high real-time scheduling requirements. Further, by constructing a priority comparator based on the combination of node depth and heuristic function, nodes with higher priority can be more effectively screened out, reducing the search for invalid paths. By pruning the OPEN table, the number of nodes to be retained during the execution of the algorithm is restricted. While optimizing the algorithm efficiency, it also effectively reduces the memory overhead, meeting the large-scale data processing requirements in complex scheduling scenarios.

[0111] It should be noted that, on the other hand, the present application also provides a storage medium, which may be included in an electronic device; or may exist separately without being assembled into the electronic device. When the above one or more programs are executed by an electronic device, the electronic device is caused to implement the method as described in the following embodiments. For example, the electronic device may implement each step of the method as Figure 1 shown.

[0112] In one embodiment, the present application provides a computer program product, including a computer program, which when executed by a processor implements the steps in the above method embodiments.

[0113] In addition, the above drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present invention, rather than for limiting purposes. It is easy to understand that the processes shown in the above drawings do not indicate or limit the time sequence of these processes. Additionally, it is also easy to understand that these processes may be executed synchronously or asynchronously in, for example, multiple modules.

[0114] Those skilled in the art will readily think of other embodiments of the present invention after considering the specification and practicing the invention herein. The present application aims to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not disclosed by the present invention. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present invention are pointed out by the claims.

[0115] It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is defined only by the appended claims.

Claims

1. A scheduling method for a semiconductor wafer manufacturing system based on dynamically adjusting the A* heuristic function, characterized in that The method includes: Establishing a timed Petri net model according to the processing flow of the combined equipment involved in wafer manufacturing, and initializing the system resources in the timed Petri net model; Searching for an optimal scheduling path in the timed Petri net model by using an improved A* algorithm with dynamically adjusted heuristic function weights based on node depth; including: Putting the initial state into the OPEN list, representing the states to be expanded; Taking out the state with the minimum node cost from the OPEN list for expansion, where a weight related to the node depth is introduced before the heuristic function included in the node cost; Generating all possible successor states according to the available transitions of the current state; Calculating the node cost for each successor state that satisfies the transition firing rule, and adding them to the OPEN list according to the cost values in the sorted order, taking out the state with the minimum node cost from the OPEN list for expansion until the target node; Connecting the start node, multiple expanded nodes and the target node to generate the optimal scheduling path.

2. The method according to claim 1, characterized in that, The combined equipment includes a vacuum lock module, a processing module and a manipulator, and the wafer processing flow is as follows: The wafer is transferred from the loading / unloading position to the vacuum lock module by the manipulator, enters the processing module after completing the vacuum environment conversion, and returns to the loading / unloading position after processing.

3. The method according to claim 2, wherein The timed Petri net model adopts a process-oriented modeling method, and the markings correspond to different states during the movement of the combined equipment, where: Places are divided into processing places and resource places. In the initial state, the processing place represents the processing module, and the token in it represents the wafer to be processed; the resource place represents the resource control place for controlling the capacity, and the token in it represents the resources in the equipment; Transitions represent manipulator operations, including picking, placing and exchanging actions.

4. The method according to claim 3, characterized in that, The initialization processing of the system resources in the timed Petri net model includes: Configuring the initial markings of the processing places and resource places. The number of tokens in the processing place is the same as the number of wafers to be processed, and the number of tokens in the resource place is the same as the number of available resources of the equipment.

5. The method according to claim 1, characterized in that, The specific formula for calculating the node cost is expressed as: f′(n)=g(n)+ω·h(n) where g(n) is the actual cost from the initial node to the current node, h(n) is the value of the improved heuristic function, α is the adjustment parameter, curDepth() is the node depth, and ω is the weight function.

6. The method according to claim 5, characterized in that The value of the improved heuristic function is specifically expressed as: h max (M) = max{h R (M), h RWT (M)} Among them, represents the minimum processing time for using resource r in the marking M, j RWT i represents the shortest time required for a token to move from the place p i to the target place, represents the i-th element in an n×1 dimensional vector, and this vector contains the remaining processing time in all places p i ; here, the marking M in the time Petri net model is the node n in the A* algorithm.

7. The method according to claim 1, characterized in that, The transition firing rule includes determining the firing condition and firing time of each transition in the timed Petri net; The firing condition means that to fire a certain transition, all input places connected to this transition must meet certain token conditions, specifically, there must be enough tokens in the input places; The firing time means the time point to fire the transition. On the premise that the transition is enabled, it can be fired as long as the given time region is satisfied.

8. The method according to claim 1, wherein The method also includes performing pruning operations in the OPEN list, and the pruning operations include: When the OPEN list reaches the preset capacity threshold, only the top K nodes with the smallest cost values are retained, and the remaining nodes are removed to reduce memory occupancy and computational overhead.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the semiconductor wafer manufacturing system scheduling method based on dynamically adjusting the A* heuristic function as described in any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the semiconductor wafer manufacturing system scheduling method based on dynamically adjusting the A* heuristic function as described in any one of claims 1 to 8.

Images