Wafer scheduling method and semiconductor process device

By optimizing wafer scheduling through a cluster search algorithm, the problems of complex coding calculations and estimation errors in existing technologies are solved, enabling rapid wafer processing within the wafer cassette and efficient utilization of the chamber.

WO2026238084A1PCT designated stage Publication Date: 2026-11-19BEIJING NAURA MICROELECTRONICS EQUIP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2026/097123
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-16
Filing Date
2026-05-15
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Existing wafer scheduling methods require complex coding calculations, have long development cycles, high maintenance costs, and may have deviations in completion time estimation when equipment structure changes or process paths become complex, resulting in low chamber utilization.

Method used

By employing a cluster search algorithm, the action sequence corresponding to the new wafer that arrives at the process chamber fastest is determined, and the first action sequence that satisfies the pre-constraint is selected. The wafer is then retrieved from the loading/unloading position first, enabling concurrent and alternating wafer ejection and avoiding idle waiting in the chamber.

Benefits of technology

It improves the utilization rate of process chambers, reduces development cycle and maintenance costs, has simple logic, and produces more accurate and reliable results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2026097123_19112026_PF_FP_ABST
    Figure CN2026097123_19112026_PF_FP_ABST
Patent Text Reader

Abstract

The present application discloses a wafer scheduling method and a semiconductor process device. The method comprises: determining a first candidate action sequence on the basis of objective function values of action sequences obtained by means of beam search; from among all action sequences corresponding to a same newly-dispatched wafer, determining a second candidate action sequence that enables the wafer to arrive at a process chamber in the shortest time; in response to the number of actions in the candidate action sequences reaching an action-count threshold, determining a target action sequence on the basis of the objective function values; in response to the first action of the target action sequence satisfying a precondition, executing the first action of the target action sequence in the simulation, wherein the precondition is that newly-dispatched wafers executing different tasks are retrieved from load ports in the order of their arrival at the process chamber; and in response to completion of simulated scheduling for all wafers, outputting a scheduling sequence comprising actions sequentially executed in the simulation. The present application enables wafers to be processed as quickly as possible. In addition, the development cycle is short, the maintenance costs are low, and the results are more accurate and reliable.
Need to check novelty before this filing date? Find Prior Art

Description

Wafer scheduling methods and semiconductor process equipment Technical Field

[0001] This application belongs to the field of semiconductor technology, and in particular relates to a wafer scheduling method and semiconductor process equipment. Background Technology

[0002] Clustering equipment is one of the most widely used equipment structures in semiconductor manufacturing processes, and it is widely used in processes such as resist stripping, etching, and deposition. Semiconductor manufacturing plants have strict requirements on the time that equipment can process wafers; that is, the time from when the wafer cassette is placed into the equipment loading / unloading position to when all wafers in the wafer cassette have completed their processing and returned to the wafer cassette should be as short as possible. The processing time for wafers with different process requirements is generally different. If the order in which wafers leave the wafer cassette and enter the process chamber is not appropriate, it may cause some process chambers to be idle and waiting, making it impossible to complete the processing of the wafers in the wafer cassette as quickly as possible.

[0003] In related technologies, the completion time of different wafer arrangements is estimated during wafer entry, and the first wafer in the wafer exit sequence corresponding to the minimum completion time is selected for processing, ultimately obtaining the optimal wafer entry sequence. However, because the completion time of different wafer arrangements needs to be estimated according to a specific formula, it not only requires complex coding calculations, resulting in a long development cycle and high maintenance costs, but also the completion time estimation may be biased when the equipment structure changes or the process path becomes complex. For example, when there is a process chamber idle cleaning setting, the time it takes for the wafer to arrive at the process chamber is exactly when the cleaning threshold is reached, and the process chamber needs to be cleaned, which may lead to a deviation in the completion time estimation. Summary of the Invention

[0004] The purpose of this application is to provide a wafer scheduling method and semiconductor process equipment to solve the problems in related technologies, such as the need for complex coding calculations, long development cycles, high maintenance costs, and potential deviations in completion time estimation when equipment structure changes or process paths become complex.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0006] In a first aspect, embodiments of this application provide a wafer scheduling method, the method comprising: determining a first candidate action sequence from the current action sequences based on the objective function values ​​corresponding to each current action sequence obtained by cluster search, the action sequence including at least two actions traversed sequentially by a robotic arm; determining a second candidate action sequence from all action sequences corresponding to the same newly produced wafer, the first action in the action sequence corresponding to the newly produced wafer being retrieved from the loading / unloading position; responding to the number of actions in the current candidate action sequences reaching a preset action number threshold, determining a target action sequence from the current candidate action sequences based on the objective function values ​​corresponding to each current candidate action sequence; responding to the first action of the target action sequence satisfying a preset precondition, simulating the execution of the first action of the target action sequence, the precondition being that the newly produced wafers performing different tasks are retrieved from the loading / unloading position in the order of arrival at the process chamber; and responding to the completion of simulation scheduling of all wafers, outputting a scheduling sequence, the scheduling sequence including actions simulated sequentially.

[0007] In a second aspect, embodiments of this application provide a semiconductor process apparatus, including: a controller, a robot, a process chamber, and a loading / unloading position. The controller includes at least one processor and at least one memory, the memory storing a computer program. When the computer program is executed by the processor, it implements the steps of the method described in the first aspect.

[0008] This application embodiment retains the second candidate action sequence corresponding to the new wafer arriving at the process chamber fastest, and selects the action sequence that satisfies the first action as the target action sequence. The pre-constraint is that the new wafers performing different tasks are taken out from the loading / unloading position in the order of arrival at the process chamber. This achieves priority in selecting the wafer that arrives at the process chamber earliest for unloading from the loading / unloading position, thereby realizing the sequential decision of concurrent and alternating unloading of wafers performing different tasks. This avoids chamber idle waiting due to unreasonable unloading order, improves the utilization rate of the process chamber, and allows wafers in the wafer cassette to be processed as quickly as possible. Moreover, the logic is relatively simple, requiring no complex calculations, with a short development cycle and low maintenance cost. The actual arrival time of the wafer in the process chamber is determined by the cluster search method, which is a definite value and the result is more accurate and reliable than the estimated completion time in related technologies. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0010] Figure 1 is a schematic diagram of a typical cluster-type device;

[0011] Figure 2 is a schematic flowchart of a wafer scheduling method provided in an embodiment of this application;

[0012] Figure 3 is a schematic diagram of the principle of a tree search algorithm provided in an embodiment of this application;

[0013] Figure 4 is a schematic diagram of the principle of a bundle search algorithm provided in an embodiment of this application;

[0014] Figure 5 is a schematic diagram illustrating the principle of a tree search algorithm provided in an embodiment of this application when applied to the scheduling of semiconductor process equipment.

[0015] Figure 6 is a schematic diagram of the overall process of a wafer scheduling method provided in another embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, "and / or" in this application indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. It should be noted that all data involved in this application was obtained with the user's authorization.

[0018] Cluster tools are a widely used equipment structure in semiconductor manufacturing processes, extensively applied in processes such as resist stripping, etching, and deposition. As shown in Figure 1, a cluster tool typically includes multiple process modules (PM) 11 (corresponding to PM1 to PM4 in Figure 1), an alignment module 12, a robotic arm 13, a load lock 14, and load ports (LP) 15 (corresponding to LoadPort1 to LoadPort2 in Figure 1). The robotic arm 13 includes an atmospheric transfer module (ATM) and a vacuum transfer module (VTM). In semiconductor manufacturing plants, wafers are generally loaded into a wafer cassette (Foup). An overhead crane transports the wafer cassette to the loading / unloading position of the semiconductor process equipment. The robotic arm within the equipment then transports the wafers one by one to the corresponding process modules to complete the process. Finally, the robotic arm within the equipment returns the wafers to the wafer cassette at the loading / unloading position. After all the wafers in the wafer cassette have completed their process in the process chamber, the overhead crane will transport the wafer cassette to the next step for further process execution.

[0019] Semiconductor manufacturing plants have strict requirements on wafer processing time; the time from when the wafer cassette is placed into the equipment loading / unloading position to when all wafers in the cassette complete their processing and return to the cassette should be as short as possible. A single clustering device typically has multiple process chambers, which can be configured as identical or different process chambers. As wafer sizes increase, the need for a single clustering device to process multiple wafers with different process requirements in parallel also increases. For example, if a wafer cassette contains three wafers with different process requirements, and the clustering device also has three different process chambers, then the three types of wafers in the cassette can simultaneously enter their respective corresponding process chambers to complete their processing. The processing times for different types of wafers are generally different. If the order in which the wafers leave the cassette and enter the process chambers is inappropriate, some process chambers may become idle and wait, preventing the wafers in the cassette from being processed quickly enough.

[0020] Therefore, the core challenge in multi-task concurrent scenarios lies in the fact that the order in which wafers are ejected from the loading / unloading station directly affects the time it takes for each wafer to reach its target process chamber and the utilization rate of the chamber. In order to minimize the total completion time, the scheduling algorithm must be able to accurately predict the actual arrival time of each wafer in the process chamber under different ejection sequences and make the optimal ejection sequence decision accordingly.

[0021] The structure and function of each module in the cluster-type device shown in Figure 1 can be described as follows:

[0022] 1) LoadPort 15: Each loadport 15 can hold one wafer cassette, and each wafer cassette can hold 25 wafers. The 25 wafers can be of multiple different types.

[0023] 2) Aligner 12: It has one slot for placing a wafer for calibration.

[0024] 3) Process Chamber (PM) 11: Each process chamber 11 has a slot that can hold one wafer for processing at a time.

[0025] 4) LoadLock 14: Each loadlock 14 has two slots (represented as left and right slots in Figure 1 for convenience). Each slot can hold one wafer. The loadlock 14 can switch between atmospheric and vacuum states. When the loadlock 14 is in atmospheric state, the wafer located on the atmospheric manipulator (ATM) side can be fed into the loadlock 14. When the loadlock 14 is in vacuum state, the wafer located on the vacuum manipulator (VTM) side can be fed into the loadlock 14.

[0026] 5) Vacuum robot (VTM): It is a dual-arm robot with two arms at 180° and fixed. Each arm has a slot, and each slot can hold one wafer. The two arms cannot perform wafer picking and placing operations at the same time. The vacuum robot is responsible for moving wafers between the vacuum lock 14 and each process chamber 11 (e.g. PM1, PM2, PM3, PM4).

[0027] In the cluster-type equipment shown in Figure 1, the path and processing time of the wafer to be processed are generally specified by the process engineer. Typically, the wafer is ejected from the LoadPort, calibrated in the Aligner module, processed in the Process Chamber (PM), cooled in the Cooler, and then returned to the LoadPort. Each step consumes time. For example, the path of the wafer to be processed can be as follows: from the LoadPort, it is transported by an atmospheric robot (ATM) to the Aligner for calibration, then by a vacuum robot (VTM) to the Process Chamber (PM1 / 2 / 3) for processing. After processing, the VTM transports the wafer to the atmospheric robot (ATM) via a vacuum lock, and finally returns to the LoadPort. PM1 / 2 / 3 indicates that the wafer can be processed in any of these three process chambers.

[0028] When multiple types of wafers need to be processed in the cluster-type equipment shown in Figure 1: one wafer cassette can hold multiple wafers of different types, and different wafer cassettes can also hold wafers of different types. When multiple types of wafers are processed by the same equipment, the algorithm program within the equipment needs to coordinate the order in which the robotic arm handles different types of wafers so that the given wafer process task can be completed as quickly as possible. For example, type A wafers are in LoadPort1 and need to be processed in PM1 / 2, with a processing time of 120 seconds. Type B wafers are in LoadPort2 and need to be processed in PM3 / 4, with a processing time of 98 seconds. The goal is to process one type of wafer as quickly as possible, based on the PM utilization rate, and release the PM so that wafers for other tasks can be processed on this equipment, thereby improving overall capacity.

[0029] In addition, to ensure process quality, there are usually additional cleaning requirements for the process chamber. For example, after processing several wafers consecutively or after the process chamber has been idle for a certain period of time, it needs to be cleaned. For instance, a process chamber can be set to clean for 60 seconds after processing every two wafers. In other words, this means that after processing two wafers, the process chamber needs to be automatically cleaned for 60 seconds after the second wafer is removed before the third wafer can be placed in to continue the process.

[0030] In related technologies, the completion time of different wafer arrangements is estimated during wafer entry, and the first wafer in the wafer exit sequence corresponding to the minimum completion time is selected for processing, ultimately obtaining the optimal wafer entry sequence. However, because the completion time of different wafer arrangements needs to be estimated according to a specific formula, it not only requires complex coding calculations, resulting in a long development cycle and high maintenance costs, but also the completion time estimation may be biased when the equipment structure changes or the process path becomes complex. For example, when there is an idle cleaning setting in the process chamber, the time it takes for the wafer to arrive at the process chamber just reaches the cleaning threshold, and the process chamber needs to be cleaned, which may lead to a deviation in the completion time estimation. Thus, the accuracy requirement for wafer exit sequence decision-making in the multi-task scheduling scenario of related technologies constitutes a significant technical contradiction with the difficulty of providing accurate decisions due to estimation errors in estimation-based methods. To address this, this application proposes a wafer scheduling method and semiconductor process equipment with a more reasonable scheduling sequence.

[0031] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0032] Figure 2 is a flowchart illustrating a wafer scheduling method according to an embodiment of this application. As shown in Figure 2, the wafer scheduling method of this embodiment may specifically include the following steps:

[0033] S201, based on the objective function values ​​corresponding to each current action sequence obtained from the beam search, determine the first candidate action sequence among the current action sequences. The action sequence includes at least two actions that the robot arm traverses sequentially. In other words, the action sequence refers to an ordered arrangement of a series of actions executed sequentially by the robot arm, with each action representing a single film pick-up or placement operation completed by the robot arm.

[0034] In this embodiment of the application, the execution subject of the wafer scheduling method is the controller in the semiconductor process equipment (e.g., the clustering equipment), which can be set in the host computer and / or slave computer of the semiconductor process equipment.

[0035] Tree search algorithms are widely used in the scheduling of semiconductor process equipment. Different search algorithms are generally classified as breadth-first search, depth-first search, and combinations of different strategies to optimize computation time and results.

[0036] The breadth-first search strategy works as follows: iterates through the data in a first-order fashion, searching one level at a time. Each move selects the optimal move for the current situation (but not necessarily the global optimum), until a path of length n is found, at which point the optimal path is selected as the result. This approach is a greedy strategy, choosing the locally optimal branch at each node, but it is prone to getting trapped in local optima and rarely reaches the global optimum.

[0037] The depth-first search strategy involves traversing a single branch in advance until the target node is reached (i.e., traversing the entire tree; the black branches in Figure 3 represent the paths taken in each depth-first search). Then, all feasible paths are compared, and the path with the shortest execution time is selected as the result. This method requires traversing all branches and can obtain the theoretically optimal solution (global optimum), but it is computationally very large and time-consuming, failing to meet the real-time scheduling requirements of semiconductor process equipment (each computation task must take less than 5 seconds).

[0038] The beam search algorithm works as follows: A pre-defined number of branches, `num_beams`, is used. Each search, according to a set objective function, retains `num_beams` of the better results and continues the search, as shown in Figure 4. Each time step has five possible outputs, A to E. In Figure 4, `num_beams` = 2, meaning that each time step retains the two best sequences up to the current step. After the search is complete, the first `m` steps of the optimal path are executed. When `num_beams` = 1, the beam search algorithm degenerates into a greedy search. Compared to greedy search, beam search retains other branches, effectively avoiding getting trapped in local optima. Compared to depth-first search, beam search filters out suboptimal branches at each step through the objective function, significantly reducing the search range and computation time. Furthermore, selecting the first `m` steps of the optimal path for each step is more cautious and effectively avoids getting trapped in local optima.

[0039] When the tree search algorithm is applied to the scheduling of semiconductor process equipment, as shown in Figure 5, the equipment state is used as a node in the search process, and the robot's handling motion is used as a branch. The goal is to complete the processing of a given task (i.e., multiple wafers to be processed) in the shortest possible time. For example, in the initial state, all wafers to be processed are waiting for the robot to handle them in the LoadPort. At this time, the equipment state is equivalent to the initial state (i.e., ClusterToolState). When multiple wafers with different paths are ejected from the LoadPort, the algorithm derives the robot handling action for each type of wafer as a branch (as shown in Figure 5, t1 represents the wafer of the first task ejected from LoadPort1, and t2 represents the wafer of the second task ejected from LoadPort2). After selecting an action to execute, the equipment state changes from the initial state (ClusterToolState) to state 1 (State1). In this state, the robot has several optional actions to execute, thus generating states 1.1 to 1.h.

[0040] The number of branches (num_beams) retained, the search depth, and the number of steps executed after the search are completed in each beam search algorithm can be set according to the application scenario. The key factor affecting the algorithm's performance is the design of the objective function for retaining branches.

[0041] Since the goal of scheduling algorithms is to minimize the completion time of all wafers completing processing and returning to the LoadPort for a given task, using completion time as the objective function is a common practice. Furthermore, because semiconductor process equipment scheduling has strict requirements for solution time (controlled on the order of seconds), the search depth is usually limited to a set range. For example, if the search depth is set to 10, then after searching 10 branches (corresponding to 10 complete actions of the robotic arm), the search stops. The completion time cannot be determined using this limited search depth, but the overall completion time can be predicted by the time consumed by the current action, and this predicted completion time can be used as the objective function of the algorithm. Common methods for calculating predicted completion time are as follows:

[0042] Predicted completion time = Total current time / Current completion percentage

[0043] The calculation example is as follows: Assume the wafer path is from the LoadPort, through the LoadLock, to the process chamber PM1 for processing, and then back through the vacuum lock to the LoadPort. The time taken for transporting the wafer to the adjacent module is 10 seconds, the processing time in PM1 is 100 seconds, and there is no pressure conversion in the LoadLock (time taken is 0 seconds). Therefore, when there is only one wafer, the total completion time is 10 seconds (transport from LoadPort to Vacuum Lock) + 10 seconds (transport from Vacuum Lock to PM1) + 100 seconds (processing in PM1) + 10 seconds (transport from PM1 to Vacuum Lock) + 10 seconds (transport from Vacuum Lock to LoadPort) = 140 seconds. During the search process, when the wafer first arrives at the LoadLock, the time taken is 10 seconds, so the predicted completion time is 10 / (10 / 140) = 140 seconds. If the robotic arm handles other wafers in between, causing the wafer to take 20 seconds to reach LoadLock for the first time, then the predicted completion time is 20 / (10 / 140) = 280 seconds.

[0044] Because the commonly used calculations for predicted completion time only consider the current wafer completion rate and the current total time, without taking into account the idle time of the target process chambers for different types of wafers, they are difficult to apply to multi-job concurrent scenarios. In particular, since the time taken to move different types of wafers from the LoadPort to the LoadLock is the same, the type of wafer moved is almost equivalent to the objective function. However, once different types of wafers leave the LoadPort, it may block other types of wafers from entering the LoadLock, leading to suboptimal results. Furthermore, due to the greater complexity of the actual equipment structure, wafer paths (e.g., different numbers of process chambers required for different tasks) and cleaning configurations vary (e.g., processing two wafers in a process chamber for task 1 and cleaning once, processing three wafers in a process chamber for task 2 and cleaning once), it is difficult to use a single objective function to accommodate various scenarios. Therefore, it is also difficult to optimize multi-job concurrency by modifying the predicted completion time calculation method.

[0045] Therefore, this application embodiment improves upon the conventional cluster search algorithm by retaining the action sequence corresponding to the new wafer that arrives at the process chamber fastest during the branch search process. It determines the earliest time when wafers of different tasks arrive at their respective process chambers and selects the task wafer that can arrive at the process chamber earliest to be ejected from the LoadPort first. This enables the sequential decision of wafers of different tasks to be ejected concurrently and alternately, avoiding chamber idle waiting due to unreasonable ejection order, improving the utilization rate of the process chamber, and enabling wafers in the wafer cassette to complete processing as quickly as possible.

[0046] Specifically, for each action sequence obtained during the cluster search process, the action sequence includes at least two actions that the robot arm traverses in sequence, i.e., the action sequence obtained after one step of breadth search and at least one step of depth search.

[0047] The breadth-first search step involves iterating through the feasible next actions of the robotic arm based on the current simulation state of the semiconductor process equipment, and recording the corresponding changes in the current simulation state of the semiconductor process equipment. For example, when each module in the equipment has wafers and the robotic arm is empty, the next feasible action of the robotic arm is to pick up wafers from each module, which can be designated as Move1, Move2, Move3, Move4, Move5, etc.

[0048] The depth search step involves iterating through the feasible next actions of the robot arm based on the current simulation state changes and the current action sequences, recording the corresponding changes in the current simulation state. When this is the first depth search, a second search is performed based on all actions obtained in the breadth search step. For example, Move1 is the ATM retrieving a wafer from LoadPort1, so the next step is to place the retrieved wafer onto LoadLock. Let's call this action Move1.1, and the resulting action sequence is [Move1, Move1.1]. Move2 is the ATM retrieving a wafer from LoadPort2, so the next step is to place the retrieved wafer onto LoadLock. Let's call this action Move2.1, and the resulting action sequence is [Move2, Move2.1]. Assuming Move3 is when the VTM retrieves the wafer from the LoadLock, the next step is to place it in PM1 or PM2. Let the action of placing it in PM1 be Move3.1 and the action of placing it in PM2 be Move3.2. This will generate two action sequences: [Move3, Move3.1] and [Move3, Move3.2]. Other actions can generate corresponding action sequences in a similar way.

[0049] As shown in Figure 5, the state of a semiconductor process equipment refers to the distribution of wafers within the equipment and their earliest possible handling time. For example, during the initial calculation, the wafers have not yet left the LoadPort. At this time, the equipment state (State) indicates that all wafers are in the LoadPort, and the earliest time a wafer can be handled by the atmospheric robot is the current time t0. When the atmospheric robot removes a wafer from the LoadPort, the equipment state changes to State1 (i.e., one wafer is on the atmospheric robot, and the other wafers are still in the LoadPort). At this point, the earliest time a wafer on the atmospheric robot can be handled is the end time of the previous wafer-removing action, while the earliest time a wafer still in the LoadPort can be handled is still t0. Similarly, each time the robot completes an action, the distribution of wafers within the equipment and the earliest time each wafer can be handled changes, meaning the equipment state changes.

[0050] The objective function value of each action sequence is determined based on the recorded changes in the current simulation state. The objective function can be the predicted completion time mentioned above, and the corresponding objective function value is the calculated value of the predicted completion time. The calculation of the objective function value can refer to the aforementioned method for calculating the predicted completion time. When the predicted completion time is used as the objective function, it is based on the time consumed by the current action and the estimated time consumed by all subsequent actions, serving as an indicator of the quality of the current action. In the following embodiments of this application, the objective function is illustrated using the predicted completion time as an example. Alternatively, the objective function can also be the idle waiting time of the robotic arm performing the current action. For example, when the vacuum robotic arm can proceed to process chambers PM1, PM2, and PM3 to perform wafer retrieval, the wafers in the process chambers may still be undergoing processing. In this case, the robotic arm needs to wait for the process chambers to finish processing before it can retrieve the wafers. The objective function, by evaluating the idle time of the robotic arm retrieving wafers in the three PMs, can achieve the effect of prioritizing the wafer whose process has finished earliest. The above two objective functions can be used individually or in combination according to the needs of the scheduling scenario. The objective function can be adjusted and optimized according to different product scheduling needs. For example, the calculation method of predicted completion time can be adjusted according to the algorithm effect, or other evaluation indicators can be added (such as dwell time, different weights for different modules, etc.). This application does not impose too many restrictions on this. However, regardless of the objective function used, its evaluation direction is guided by shortening the total completion time. The following improvement mechanism of this application can be applied to different objective function designs.

[0051] The current action sequences are sorted according to their objective function values, either from smallest to largest or from largest to smallest. The first candidate action sequence is then determined based on this sorting. In practical applications, when the objective function is predicted completion time, the sorting direction is usually from smallest to largest, meaning the action sequence with the shortest predicted completion time is ranked higher. Taking predicted completion time as an example, the current action sequences are sorted according to their calculated predicted completion time values, from smallest to largest. The action sequences with the shortest predicted completion time have the highest priority, and the top num_beams action sequences are determined as the first candidate action sequences. It should be noted that the actual objective function value can be adjusted and supplemented based on whether the results are optimal. For example, the predicted completion time calculation formula can be optimized to make the predicted completion time calculation more accurate. However, the shortest completion time is always the primary evaluation metric.

[0052] The number of branches, `num_beams`, is an empirical value. Setting it too high results in more branches being retained, potentially leading to better results, but also increases computation time. Conversely, setting it too low means fewer branches are retained per search, which might lead to the accidental deletion of better results, resulting in suboptimal outcomes, but reduces computation time. In short, a large `num_beams` setting increases computation time, while a small setting results in insufficient information retention. A trade-off between search quality and computation time must be struck based on the specific scheduling scenario.

[0053] S202, among all the action sequences corresponding to the same newly produced wafer, the second candidate action sequence that arrives at the process chamber fastest is determined. The first action in the action sequence corresponding to the newly produced wafer is to take the wafer out of the loading / unloading position.

[0054] In the embodiments of this application, the newly produced wafer is the wafer corresponding to the action sequence in which the first action is to remove the wafer from the loading and unloading position, that is, the first action in the action sequence corresponding to the newly produced wafer is to remove the wafer from the loading and unloading position.

[0055] For each action sequence obtained from the cluster search, all action sequences in which the first action is to retrieve a wafer from the loading / unloading position are collected, i.e., all action sequences corresponding to each newly retrieved wafer are collected. That is, the order of action sequences obtained by sorting according to the objective function value in step S201 may be relatively low priority, and subsequent deep searches may not continue based on this. For example, the first action Move1 in action sequences [Move1, Move3] and [Move1, Move1.1] is ATM retrieving a wafer from LoadPort1, which may have a low priority. This step collects all action sequences in which the first action is to retrieve a wafer from LoadPort1 and LoadPort2.

[0056] For all action sequences corresponding to the same newly produced wafer, i.e., the first action is all action sequences of the same newly produced wafer, the action sequence that reaches the process chamber the fastest is determined as the second candidate action sequence.

[0057] Specifically, for all action sequences corresponding to the same newly produced wafer, they can be sorted in sequence according to at least one of the following indicators: whether each action in the action sequence operates on the same wafer, whether each action in the action sequence operates on wafers of the same path, whether each action in the action sequence empties the vacuum robot, and the predicted completion time after all actions in the action sequence are completed. Based on the sorting results, the second candidate action sequence that arrives at the process chamber fastest is determined.

[0058] It should be noted that the order of the four indicators mentioned above is fixed. This order is determined through industrial engineering analysis and design. These four indicators are used to select the action sequence that will allow the wafer from the LoadPort to reach the PM as quickly as possible, thus obtaining the earliest time the wafer arrives at the PM. The action sequence with the highest priority (e.g., 1) is selected as the second candidate action sequence for the newly delivered wafer.

[0059] It is understood that the sorting method of the above four indicators is a preferred embodiment of this application. In other possible implementations, at least one of the above four indicators may be selected for sorting. For example, when the scheduling scenario is relatively simple and the differences between the action sequences are obvious, sorting may be performed using only the indicator "whether each action in the action sequence is operating on this wafer"; or, two or three of the indicators may be selected for combined sorting, which is not limited in this application.

[0060] The above sorting process is equivalent to multidimensional sorting. First, sorting is performed based on the first indicator. If the first indicator of all action sequences is different, the optimal one is selected based on the sorting result of the first indicator and the sorting ends. If the first indicator of some action sequences is the same, sorting continues based on the second indicator while maintaining the order of the first indicator, and so on.

[0061] In particular, it is important to determine whether each action in the action sequence is an operation on the current wafer. For example, in the action sequences [Move1, Move3] and [Move1, Move1.1], where Move1 is the ATM taking wafer A from LoadPort1 and Move3 is the VTM taking another wafer B from LoadLock, then Move3 is not an operation on the current wafer A.

[0062] The following is an example of sorting based on four criteria: Suppose the action sequence for retrieving a wafer from LoadPort1 includes [Move1, Move1.1, Move2, Move2.1] and [Move1, Move3, Move3.1, Move1.1], etc. These action sequences are sorted according to the four criteria in sequence. First, they are sorted by whether the action operates on the current wafer. Therefore, [Move1, Move1.1, Move2, Move2.1] ranks higher than [Move1, Move3, Move3.1, Move1.1] because Move1.1 still operates on the current wafer, while Move3 does not. Similarly, they are sorted according to whether the action operates on wafers along the same path. A similar path means that the wafer passes through the same modules in the same order. Then, sort them according to whether the action is to clear the vacuum robot (VTM), and finally sort them according to the predicted completion time after all actions are completed. Select the first action sequence as the second candidate action sequence for the wafer to reach the process chamber the fastest.

[0063] S203, in response to the number of actions in each of the current candidate action sequences reaching a preset action number threshold, the target action sequence is determined from each of the current candidate action sequences based on the objective function value corresponding to each of the current candidate action sequences.

[0064] In this embodiment, the num_beams first candidate action sequences determined in step S201 and the second candidate action sequence corresponding to each newly produced wafer determined in step S202 are added to the set for further depth searching. For example, if num_beams = 10, then the 10 first candidate action sequences determined in step S201 are added to the set for further depth searching.

[0065] According to the traditional cluster search process, the action sequence for retrieving wafers from LoadPort may have a lower priority, and those not in the top 10 may be deleted, potentially leading to a local optimum and a global suboptimal result. In this case, the action sequence selected in step S202 can be added to the set. Assuming that step S202 retains one action sequence for retrieving wafers from LoadPort1 and one for retrieving wafers from LoadPort2, the total number of action sequences is 12.

[0066] A pre-set search depth `num_depth`, which is the threshold for the number of actions, is used. For example, `num_depth = 10`. When the number of actions in the candidate action sequence reaches 10, the set search depth is reached, and the search stops to reduce computational load and time. Based on the objective function values ​​of each of the current candidate action sequences (e.g., the 12 candidate action sequences mentioned above), the target action sequence is determined from among them. For example, for all candidate action sequences that have reached the search depth, they are sorted in ascending order of objective function (e.g., predicted completion time) values, and the first candidate action sequence is selected as the optimal action sequence, denoted as the target action sequence.

[0067] The search depth num_depth is an empirical value. A larger value results in a deeper search depth and better optimization, but also higher computation time; a smaller value results in a shallower search depth, lower computation time, but poorer optimization. This value must be greater than the number of steps from the wafer to the PM.

[0068] Furthermore, the wafer scheduling method in this application embodiment may also include the following steps: in response to the fact that the number of actions in each current candidate action sequence has not reached the action number threshold, return to the step "deep search one step", that is, return to the step "based on the current simulation state change, on the basis of each current action sequence, traverse the next feasible action of the robot arm, and record the corresponding current simulation state change".

[0069] For example, if num_depth = 10, when the number of actions in the candidate action sequence is less than 10, that is, the set search depth has not been reached, the search continues to the next step, that is, it returns to the step of the depth search one step, and the depth gradually increases until the set search depth is reached and the loop is exited.

[0070] S204, in response to the first action of the target action sequence satisfying the preset pre-constraint, the simulation executes the first action of the target action sequence. The pre-constraint is that the new wafers performing different tasks are taken out from the loading and unloading position in the order they arrive at the process chamber.

[0071] In the embodiments of this application, the pre-constraint refers to the constraint on the relationship between the order of robot arm actions. For example, assuming that the action of ATM picking up a piece from LoadPort1 is Move1 and the action of ATM picking up a piece from LoadPort2 is Move2, if Move1 is set as the pre-constraint of Move2, then Move2 must wait for Move1 to be executed before it can be executed. The pre-constraint stores the relationship between these different actions.

[0072] Before step S204, pre-set constraints are set according to the current candidate action sequence corresponding to each newly produced wafer. Specifically, the pre-set constraints can be that newly produced wafers performing different tasks are taken out from the loading and unloading position in the order of arriving at the process chamber.

[0073] Specifically, the pre-constraints can be set through the following steps: traverse each current candidate action sequence to obtain the earliest time when each newly produced wafer for different tasks arrives at the first process chamber; determine the order in which each newly produced wafer for different tasks arrives at the process chamber based on the earliest time when each newly produced wafer for different tasks arrives at the first process chamber; and set the pre-constraints based on the order in which each newly produced wafer for different tasks arrives at the process chamber.

[0074] Specifically, it iterates through each candidate action sequence and checks whether each newly ejected wafer from the LoadPort has reached the first process chamber. If it has, the arrival time in the first process chamber is recorded. For example, in the action sequence [Move1, Move1.1, Move3, Move3.1], where Move3.1 places the wafer from Move1 into the process chamber, the end time of the Move3.1 action is recorded as the time when the wafer from Move1 arrives in the first process chamber.

[0075] If multiple action sequences for the same newly ejected wafer from LoadPort reach the first process chamber, the earliest arrival time is selected to update the record of when the newly ejected wafer arrives at the first process chamber. An action sequence may involve multiple wafers. For example, [Move1, Move3, Move1.1] and [Move1, Move1.1] both complete the Move1.1 action, and the first action in both is ejection from LoadPort1. However, the Move3 action is interspersed in the first action sequence, which may delay the final completion time of Move1.1. This step aims to select the action sequence that completes Move1.1 earliest.

[0076] It's important to note that after the "breadth-first search" step, all breadth-first search steps can be identified as actions to retrieve wafers from the LoadPort. The time it takes for these newly retrieved wafers to reach the first process chamber is initialized to infinity. For example, in the feasible next actions of the robot arm in the "breadth-first search" step, suppose Move1 is the ATM retrieving a wafer from LoadPort1, Move2 is the ATM retrieving a wafer from LoadPort2, Move3 is the VTM retrieving a wafer from PM1, and Move4 is the VTM retrieving a wafer from PM2. Move1 and Move2 are the robot arm's next actions to retrieve wafers from the LoadPort. At this point, these wafers haven't reached the PM yet. We can initially record the time for these actions to reach the first PM as infinity. Later in the search process, when a wafer is found to have actually reached the first PM, the time is updated to the actual time.

[0077] Based on the earliest recorded arrival time of each newly ejected wafer in the first process chamber, the order in which they arrive is determined. The earlier the earliest arrival time of a new wafer in the first process chamber, the earlier it is in the order of arrival. Preconditions are then set based on the order of arrival: the first action in the action sequence corresponding to the earliest arrival time of each new wafer in the first process chamber is ordered.

[0078] For example, wafer A retrieved from LoadPort1 can quickly reach the target chamber PM1 (PM) in 100 seconds because PM1 is empty. Wafer B retrieved from LoadPort2, however, cannot reach PM3 until cleaning is complete, taking 150 seconds. Therefore, wafer A retrieved from LoadPort1 has a higher priority than wafer B retrieved from LoadPort2. Wafer A from LoadPort1 should be retrieved before wafer B from LoadPort2. A prerequisite should be added: the Move2 action for wafer B leaving LoadPort2 should precede the Move1 action for wafer A leaving LoadPort1. That is, Move2 must be executed after Move1; otherwise, the results will be suboptimal. If wafer B from LoadPort2 leaves first, it cannot be placed in PM3 in time, blocking the LoadLock and preventing wafer A from reaching PM1 for processing, causing PM1 to idle. Conversely, if wafer A from LoadPort1 exits first, it will arrive at PM1 first, and wafer B from LoadPort2 exiting at that time will not affect the arrival time of wafer B at PM3. Overall, wafer A exiting from LoadPort1 first has a shorter chamber idle time and higher chamber utilization than wafer B exiting from LoadPort2 first.

[0079] If the first action of the target action sequence determined in step S203 satisfies the preset preconditions, then the simulation executes the first action of the target action sequence. If the first action of the target action sequence determined in step S203 does not satisfy the preset preconditions, then the process returns to the breadth-first search step, that is, returns to the step "Based on the current simulation state of the semiconductor process equipment, traverse the next feasible actions of the robot arm and record the corresponding changes in the current simulation state of the semiconductor process equipment", and searches again. For example, if the first action of the target action sequence determined in step S203 is Move2, and the preset preconditions include Move1 as a prerequisite for Move2, then the process returns to the breadth-first search step and searches again; otherwise, the simulation executes the first action of the target action sequence and proceeds to the subsequent step S205.

[0080] Correspondingly, before step S201, the following steps may also be included: deleting action sequences that do not meet the preconditions in the current action sequences obtained by the bundle search.

[0081] Specifically, before determining the first candidate action sequence, it is also possible to check whether each action sequence obtained from the bundle search satisfies the preconditions. Branches that do not meet the preconditions are pruned, i.e., deleted. Action sequences that do not meet the preconditions are then removed and will not be further searched. For example, if the precondition is that the wafer from LoadPort1 is ejected before the wafer from LoadPort2, for the action sequences [Move1, Move1.1] and [Move2, Move2.1], Move1 is the ATM retrieving the wafer from LoadPort1, Move1.1 is placing the wafer just retrieved from LoadPort1 onto the LoadLock, and Move2 is the ATM retrieving the wafer from LoadPort2, Move2.1 is placing the wafer just retrieved from LoadPort2 onto the LoadLock. Move1 should precede Move2. However, since there is no Move1 before Move2, the action sequence [Move2, Move2.1] should be deleted and will not be further searched after it. This step is used to retain only the action sequence that meets the pre-constraints after determining the order in which new wafers are ejected from the LoadPort for different tasks.

[0082] S205, in response to the completion of simulation scheduling for all wafers, outputs a scheduling sequence, which includes the actions to be simulated sequentially.

[0083] In this embodiment of the application, it is determined whether the simulation scheduling of all wafers has been completed. If the simulation scheduling of all wafers has been completed, a scheduling sequence is output. The scheduling sequence includes actions to be simulated and executed in sequence. In subsequent actual scheduling, this scheduling sequence is used to schedule wafers with different tasks.

[0084] If the simulation scheduling of all wafers has not been completed, the next round of search will begin, the preconditions will be initialized to empty, and the step of the breadth-first search will be returned, that is, the step "Based on the current simulation state of the semiconductor process equipment, traverse the next feasible action of the robot and record the corresponding change step of the current simulation state of the semiconductor process equipment".

[0085] It should be noted here that in the above steps of pruning based on preconditions, if it is the first depth search of each round of search, that is, after the first step of the depth search, the preconditions are empty at this time, and the preconditions for subsequent depth searches are the preconditions updated in the previous depth search.

[0086] Furthermore, compared to methods that significantly increase computation time by increasing search depth or breadth, the embodiments of this application only add the path from the LoadPort to the PM, resulting in almost no increase in computation time. Specifically, in order to obtain the earliest time that a wafer ejected from the LoadPort arrives at the PM, related technologies often need to intersperse multiple handling operations unrelated to the current wafer during the search process, resulting in the need for a large number of deep search steps before the arrival time can be obtained. For example, to obtain the earliest time when a wafer in LP arrives at PM, the relevant search branches are as follows: LP1.5→LA, LB→Cooler, VTM→PM2, PM3→LB, ​​LA→PM1. That is, the first step is to move the wafer in slot 5 of LP1 to LA; the second step is to move the wafer in LB to Cooler; the third step is to move the wafer in VTM to PM2; the fourth step is to move the wafer in PM3 to LB; and the fifth step is to move the LP1.5 wafer in LA to PM1. A total of five steps of depth-first search are needed to obtain the earliest time when the LP1.5 wafer arrives at the target PM. In practice, due to the design of the objective function weights, the search depth may be even deeper. However, in this embodiment, because the LP wafer arrival is designed with a higher priority, the search branches are: move the LP1.5 wafer from LP to LA, and then move the LP1.5 wafer in LA to PM1. Only two steps are needed to obtain the earliest time when the LP1.5 wafer arrives at the target PM. Therefore, the search depth is smaller, and the computation time is relatively shorter. As can be seen, by forcibly retaining the action sequence of the new wafer that arrives at the PM the fastest and assigning it a higher priority, this application effectively shortens the search depth required to obtain the earliest time of wafer arrival at the PM, thereby improving the optimization effect of the scheduling result without increasing the computation time.

[0087] To clearly illustrate the wafer scheduling method of this application embodiment, the overall flow of the wafer scheduling method of this application embodiment will be described in detail below with reference to FIG6. As shown in FIG6, the wafer scheduling method of this application embodiment may specifically include the following steps:

[0088] S601, initialize the pre-constraints to be empty.

[0089] S602, Breadth-first search step: Based on the current simulation state of the semiconductor process equipment, traverse the next feasible actions of the robot arm and record the corresponding changes in the current simulation state of the semiconductor process equipment.

[0090] S603, initialize the time for the newly produced wafer to reach the first PM for all feasible actions of the next robotic arm to be infinite.

[0091] S604, Depth Search Step 1: Based on the current simulation state changes, traverse the next feasible actions of the robot arm based on the current action sequences, record the corresponding current simulation state changes, and calculate the objective function value based on the recorded current simulation state changes.

[0092] S605: Delete action sequences that do not meet the preconditions in the current action sequences, and stop the deep search.

[0093] S606, based on the objective function value corresponding to each current action sequence, determine the top num_beams action sequences as the first candidate action sequences.

[0094] S607, for all action sequences corresponding to the same newly produced wafer, sort them in order according to the following indicators: whether each action in the action sequence is operating on this wafer, whether each action in the action sequence is operating on wafers of the same path, whether each action in the action sequence is clearing the vacuum robot, and the predicted completion time after all actions in the action sequence are completed, and determine the action sequence with the highest ranking as the second candidate action sequence.

[0095] S608, determine whether the set search depth num_depth has been reached, that is, whether the number of actions in each candidate action sequence has reached the action count threshold num_depth. If yes, proceed to step S609. If no, return to step S604.

[0096] S609, traverse the current candidate action sequence to obtain the earliest time when each newly produced wafer, performing different tasks, arrives at the first process chamber.

[0097] S610 adds pre-constraints based on the earliest time that each newly produced wafer, performing different tasks, arrives at the first process chamber.

[0098] S611, Based on the objective function value corresponding to each of the current candidate action sequences, determine the optimal target action sequence among the current candidate action sequences.

[0099] S612, determine whether the first action of the target action sequence satisfies the preconditions. If yes, proceed to step S613. If no, return to step S602.

[0100] S613, simulates the execution of the first action in the target action sequence.

[0101] S614, determine whether simulation scheduling for all wafers has been completed. If yes, proceed to step S615. If no, return to step S601.

[0102] S615, outputs the scheduling sequence, which includes actions to be simulated and executed sequentially.

[0103] In summary, the wafer scheduling method of this application retains the second candidate action sequence corresponding to the new wafer arriving at the process chamber the fastest, and selects the action sequence that satisfies the pre-constraint as the target action sequence. The pre-constraint is that new wafers performing different tasks are taken out from the loading and unloading position in the order of arrival at the process chamber. This achieves the priority of selecting the wafer that arrives at the process chamber earliest to be unloaded from the loading and unloading position, thereby realizing the order decision of concurrent and alternating wafer unloading for wafers performing different tasks. This avoids chamber idle waiting due to unreasonable unloading order, improves the utilization rate of the process chamber, and allows the wafers in the wafer cassette to be processed as quickly as possible. This application directly determines the actual arrival time of each wafer in the process chamber based on simulation, and uses this determined value as the basis for unloading order decision. Compared with related technologies that require designing complex codes to estimate the completion time of different types of wafers, this application does not require designing, coding, and maintaining complex completion time estimation formulas. Therefore, the logic is relatively simple, no complex calculations are required, the development cycle is short, and the maintenance cost is low. Determining the actual arrival time of the wafer in the process chamber using a cluster search method yields a definitive value, which is more accurate and reliable than the estimated completion time in related technologies. After one step of depth-first search, branches that do not meet the preconditions are pruned, optimizing the search results. By sequentially sorting all action sequences corresponding to the same newly emerged wafer according to four indicators, the action sequence that gets the newly emerged wafer to the process chamber as quickly as possible can be selected, thus determining the earliest arrival time of the newly emerged wafer in the process chamber.

[0104] This application also provides a semiconductor process apparatus. As shown in FIG1, the semiconductor process apparatus includes: a controller (not shown in FIG1), a robot arm 13, a process chamber 11, and a loading / unloading position 15. The controller is disposed in a host computer and / or a slave computer. The controller includes at least one processor and at least one memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the steps of any of the wafer scheduling method embodiments described above.

[0105] The semiconductor process equipment of this application retains the second candidate action sequence corresponding to the new wafer that arrives at the process chamber the fastest, and selects the action sequence that satisfies the first action as the target action sequence. The pre-constraint is that new wafers performing different tasks are taken out from the loading and unloading position in the order of arrival at the process chamber. This achieves the priority of selecting the wafer that arrives at the process chamber earliest to be unloaded from the loading and unloading position, thereby realizing the order decision of concurrent and alternating wafer unloading for wafers performing different tasks. This avoids the chamber idle waiting caused by unreasonable unloading order, improves the utilization rate of the process chamber, and allows the wafers in the wafer cassette to be processed as quickly as possible. This application directly determines the actual arrival time of each wafer in the process chamber based on simulation, and uses this determined value as the basis for the unloading order decision. Compared with related technologies that require designing complex codes to estimate the completion time of different types of wafers, this application does not require designing, coding, and maintaining complex completion time estimation formulas. Therefore, the logic is relatively simple, no complex calculations are required, the development cycle is short, and the maintenance cost is low. Determining the actual arrival time of the wafer in the process chamber using a cluster search method yields a definitive value, which is more accurate and reliable than the estimated completion time in related technologies. After one step of depth-first search, branches that do not meet the preconditions are pruned, optimizing the search results. By sequentially sorting all action sequences corresponding to the same newly emerged wafer according to four indicators, the action sequence that gets the newly emerged wafer to the process chamber as quickly as possible can be selected, thus determining the earliest arrival time of the newly emerged wafer in the process chamber.

[0106] This application also proposes a readable storage medium storing one or more computer programs, the one or more computer programs including instructions. When the program or instructions are executed by a processor in a semiconductor process apparatus including multiple applications, the processor in the semiconductor process apparatus is able to execute the various processes of the above-described wafer scheduling method embodiments, and is specifically used to execute the steps of any of the above-described wafer scheduling method embodiments.

[0107] The readable storage medium of this application retains the second candidate action sequence corresponding to the new wafer arriving at the process chamber fastest, and selects the action sequence that satisfies the pre-constraint as the target action sequence. The pre-constraint is that new wafers performing different tasks are taken out from the loading / unloading position in the order of arrival at the process chamber. This achieves priority loading of wafers arriving at the process chamber earliest from the loading / unloading position, thereby realizing the sequential decision of concurrent and alternating wafer loading for wafers performing different tasks. This avoids chamber idle waiting due to unreasonable loading order, improves the utilization rate of the process chamber, and allows wafers in the wafer cassette to be processed as quickly as possible. Moreover, the logic is relatively simple, requiring no complex calculations, with a short development cycle and low maintenance cost. The actual arrival time of the wafer in the process chamber is determined by the cluster search method, which is a fixed value and more accurate and reliable than the estimated completion time in related technologies. After one step of depth search, branches that do not meet the pre-constraint are pruned, thereby optimizing the search results. By sorting all action sequences corresponding to the same newly produced wafer according to the four indicators, the action sequence that will get the newly produced wafer to the process chamber as quickly as possible can be selected, so as to determine the earliest time when the newly produced wafer arrives at the process chamber.

[0108] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0109] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0110] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0111] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of action steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0114] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0115] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0116] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0117] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0118] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0119] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0120] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A wafer scheduling method, characterized by, The method comprises the following steps: determining a first candidate action sequence from current action sequences according to target function values corresponding to the current action sequences obtained by bundle search, wherein the action sequence comprises at least two actions of sequentially traversing a robot; determining a second candidate action sequence which reaches a process chamber fastest from all action sequences corresponding to a same new wafer, wherein a first action in the action sequence corresponding to the new wafer is taking out the wafer from a loading and unloading site; in response to a number of actions in each current candidate action sequence reaching a preset action number threshold, determining a target action sequence from the current candidate action sequences according to the target function values corresponding to the current candidate action sequences; in response to a first action of the target action sequence satisfying a preset precondition, simulating execution of the first action of the target action sequence, wherein the precondition is that the new wafers performing different tasks are taken out from the loading and unloading site according to the order of reaching the process chamber; in response to completing simulation scheduling of all wafers, outputting a scheduling sequence, wherein the scheduling sequence comprises sequentially simulated actions.

2. The method of claim 1, wherein, Before the step of determining the first candidate action sequence from the current action sequences according to the target function values corresponding to the current action sequences obtained by bundle search, the method further comprises the following steps: traversing a next possible action of the robot according to a current simulation state of a semiconductor process equipment and recording a corresponding current simulation state change of the semiconductor process equipment; traversing the next possible action of the robot based on the current simulation state change and recording the corresponding current simulation state change on the basis of the current action sequence; determining the target function value according to the recorded current simulation state change.

3. The method of claim 2, wherein, The method further comprises the following steps: in response to the number of actions in each current candidate action sequence not reaching the action number threshold, returning to the step of traversing the next possible action of the robot based on the current simulation state change and recording the corresponding current simulation state change on the basis of the current action sequence.

4. The method of claim 1, wherein, The step of determining the second candidate action sequence which reaches the process chamber fastest from all action sequences corresponding to the same new wafer comprises the following steps: sequentially sorting all action sequences corresponding to the same new wafer according to at least one index, wherein the index comprises whether each action in the action sequence is operating the wafer, whether each action in the action sequence is operating the same path wafer, whether each action in the action sequence is emptying the vacuum robot, and predicted completion time after all actions in the action sequence are completed; determining the second candidate action sequence which reaches the process chamber fastest according to a sorting result.

5. The method of claim 2, wherein, Before the step of simulating execution of the first action of the target action sequence in response to the first action of the target action sequence satisfying the preset precondition, the method further comprises the following steps: traversing each current candidate action sequence to obtain an earliest time of reaching a first process chamber of each new wafer performing different tasks; determining an order of arrival of the new-out wafers at the process chamber according to earliest time of arrival of each of the new-out wafers at the process chamber for performing different tasks; setting the precedence constraint according to the order of arrival of the new-out wafers at the process chamber for performing different tasks.

6. The method of claim 5, wherein, Further comprising: in response to the first action of the target action sequence not satisfying the precedence constraint, returning to the step of traversing the next feasible action of the robot according to the current simulation state of the semiconductor process equipment and recording the corresponding change in the current simulation state of the semiconductor process equipment; before determining the first candidate action sequence from the current action sequences obtained by the beam search, further comprising: deleting the action sequences not satisfying the precedence constraint from the current action sequences obtained by the beam search.

7. The method of claim 2, wherein, Further comprising: in response to the simulation scheduling of all wafers not being completed, initializing the precedence constraint as empty and returning to the step of traversing the next feasible action of the robot according to the current simulation state of the semiconductor process equipment and recording the corresponding change in the current simulation state of the semiconductor process equipment.

8. The method of claim 1, wherein, The target function value is a calculated value of the predicted completion time.

9. The method of claim 8, wherein, The predicted completion time is a ratio of the current total time consumption to the current completion ratio.

10. A semiconductor process apparatus, characterized by, Comprising: a controller, a robot, a process chamber and a load / unload station, the controller comprising at least one processor and at least one memory, the memory storing a computer program, the computer program being executed by the processor to implement the steps of the method according to any one of claims 1-9.