Semiconductor furnace tube area batch processing equipment scheduling method and equipment based on rule decoding

Abstract

The invention belongs to the related technical field of semiconductor manufacturing scheduling and intelligent optimization, and discloses a semiconductor furnace tube area batch processing equipment scheduling method and equipment based on rule decoding. The invention provides an integrated solving method of coding initialization, rule active decoding, fitness evaluation and elite retention, genetic evolution and hierarchical neighborhood enhancement. According to the method, a process sequence decision, a machine allocation decision, a batch formation decision and a time window control decision are incorporated into a unified solution chain, the problem of conflict accumulation caused by staged isolated decisions in an existing method is avoided, and process sorting, machine allocation, batch formation and time window control are cooperatively processed in the unified solution chain. Comprehensive optimization of the maximum completion time, the batch rate and the time window violation rate is realized, and an efficient, stable and executable scheduling scheme can be provided for batch processing equipment in a semiconductor furnace tube area.

Description

Technical Field

[0001] This invention belongs to the technical field of semiconductor manufacturing scheduling and intelligent optimization, and more specifically, relates to a method and device for scheduling batch processing equipment in semiconductor furnace tube area based on rule decoding. Background Technology

[0002] The Fourth Industrial Revolution is driving the global manufacturing industry to accelerate its transformation towards intelligence and digitalization. As the core of the electronic information industry, the semiconductor industry is experiencing exponential growth in market demand for semiconductor products. Enterprises urgently need to break through traditional efficiency bottlenecks through flexible production models with multiple varieties and small batches in order to adapt to the core demands of industrial upgrading.

[0003] Wafer manufacturing is a fundamental link in the semiconductor industry, and the furnace area, which undertakes key processes such as oxidation and thin film deposition, has become a core bottleneck for capacity improvement. This area faces three significant challenges: First, high-precision processes result in long processing cycles; for example, advanced thin film deposition can take several hours per cycle and requires strict quality control. Second, in batch processing, batch division and processing sequence planning are complex, directly impacting production efficiency and quality. Third, time window constraints mean that wafers processed beyond their designated timeframe can easily lead to performance degradation; for example, failure to passivate wafers promptly after etching can cause surface oxidation, affecting chip yield and posing a significant challenge to production management.

[0004] Currently, production scheduling in the furnace tube area mainly relies on manual experience. Under the challenges of long processing cycles, complex batch processing, time window constraints, heterogeneous equipment, and reentrant processes, the efficiency of the traditional scheduling mode fluctuates greatly and it is difficult to meet the needs of semiconductor manufacturing. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a method and device for scheduling semiconductor furnace tube area batch processing equipment based on rule decoding, which aims to solve the problem of low batching efficiency in the scheduling process of existing semiconductor furnace tube area batch processing equipment.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for scheduling batch processing equipment in a semiconductor furnace tube area based on rule decoding is provided, comprising the following steps: S1, based on workshop workpieces, processes, selectable machines, batch processing constraints and time window constraints, generates an initial population by combining random initialization with maximum batch efficiency initialization, and uses a two-layer chromosome composed of process sequence encoding and machine selection encoding to represent individuals in the initial population; S2, performs rule-based active decoding on individuals in the initial population, converting the encoded information corresponding to the individuals into an executable scheduling scheme that satisfies the sequence of processes, batch processing constraints, and time window constraints; S3. The fitness of the obtained executable scheduling scheme is evaluated, and an elite retention strategy is implemented on individuals based on the fitness evaluation results to obtain the dominant individuals to participate in the next round of evolution, forming a new population. S4, perform genetic operations on individuals in the new population to obtain candidate individuals; S5. Perform hierarchical neighborhood search on the scheduling scheme corresponding to the candidate individual. Based on the critical path and critical block information, take the critical batch as the local search object and perform cross-machine batch movement and same-machine batch movement in sequence to obtain the optimal scheduling scheme. Among them, the critical path, critical block and critical batch are identified based on the scheduling scheme corresponding to the candidate individual.

[0007] Furthermore, a comprehensive fitness function is constructed based on the maximum completion time, batching rate, and time window violation rate, and the fitness of the obtained executable scheduling scheme is evaluated based on the comprehensive fitness function.

[0008] Furthermore, the comprehensive fitness function is:

[0009] in, Represents an individual The overall fitness value, and These represent the normalized results for total processing time, batch rate, and time window violation rate, respectively. For the corresponding weight coefficients, and satisfying .

[0010] Furthermore, the initial population consists of two parts: the first part is randomly initialized individuals, which are obtained by randomly arranging the process and randomly selecting feasible machines; the second part is initialized individuals with maximum batching efficiency, which are obtained by prioritizing the selection of machines that can match the already formed batches from the set of available machines. If there are no machines that can be batched, the earliest machine that can start operation is selected.

[0011] Furthermore, the current process to be scheduled is extracted sequentially according to the process sequence code, and the target processing machine for the current process to be scheduled is determined by combining the machine selection code.

[0012] Furthermore, the current scheduled process is type-identified. When the current scheduled process is an intermediate process within a time window, a feasible insertion position is first searched based on the active insertion rule. Then, the completion time of the process is compared with the latest start time of its corresponding subsequent process. If the completion time of the process is not later than the latest start time of the subsequent process, the current decoding result is maintained. If the completion time of the process is later than the latest start time of the subsequent process, the set of available machines for the process is traversed, and the earliest startable machine is reselected for processing. When the current process is not an intermediate process within a time window, the current... If a compatible batch exists at the insertion position, the process is merged into the compatible batch; if not, a batch that can be grouped is searched forward and backward along the current insertion position, and the batch movement direction is determined according to the principle of minimizing additional waiting time; if no batch that can be grouped is found, a new batch is created at the current insertion position; when the current process is the starting process of a time window, the latest start time of each related process on the time window chain is calculated in reverse based on the ending process of the corresponding time window, and the latest start time is used as the constraint basis for subsequent process decoding, machine reselection, and neighborhood movement.

[0013] Further, in step S4, a roulette wheel selection strategy is first used to select parent individuals from the current population. Then, crossover and mutation are performed on the process sequence encoding and machine selection encoding respectively. For process sequence encoding, the POX operator is used to recombine the parent process sequence, forming a new process arrangement while retaining the relative order of some priority processes. An insertion mutation operator is used to adjust the position of local processes. For machine selection encoding, a uniform crossover operator is used to exchange the machine allocation genes of different parent individuals, and a single-point mutation operator is used to randomly change the equipment selection result of the target process.

[0014] Furthermore, the critical paths and critical blocks in the scheduling scheme are identified, and the critical batches are divided into first-batch batches, intra-batch batches, and last-batch batches; target batches on the critical paths are selected. It iterates through the set of available machines, searching for idle or batch processing periods on the target machine in chronological order; for candidate insertion positions, it uses the previous batch at the insertion position. Completion time and the next batch after the insertion position Start time Determine the candidate insertion interval and combine it with the target batch. Batch structure time window interval Perform a feasibility assessment; when the overlap length between the candidate insertion interval and the time window interval is greater than the target batch processing time. At that time, the target batch is moved to the candidate position, thereby realizing the redistribution of load among devices; After completing the cross-machine batch move, further perform the same-machine batch move. While keeping the processing machine where the target batch is located unchanged, when the target batch is the first batch of the block, search for its subsequent insertable time period; when the target batch is the last batch of the block, search for its preceding insertable time period; when the target batch is a batch within the block, search for the preceding and following insertable time periods on its machine in chronological order.

[0015] The present invention also provides a semiconductor furnace tube area batch processing equipment scheduling system based on rule decoding. The system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it performs the semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described above.

[0016] The present invention also provides a computer-readable storage medium storing machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the rule-based decoding semiconductor furnace tube area batch processing device scheduling method as described above.

[0017] In summary, compared with the prior art, the rule-based decoding-based semiconductor furnace tube area batch processing equipment scheduling method and equipment provided by the present invention have the following beneficial effects: (1) This invention incorporates process sequence decision, machine allocation decision, batch formation decision and time window control decision into a unified solution chain, avoiding the problem of conflict accumulation caused by isolated decision-making in stages in the existing methods. Furthermore, based on the critical path and critical block identification results, this invention limits the local search object to critical batches and adopts a hierarchical neighborhood enhancement mechanism that combines cross-machine batch movement and same-machine batch movement to improve the scheduling scheme from the two levels of inter-device resource allocation and intra-device processing sorting, thereby reducing invalid searches and improving batching efficiency.

[0018] (2) In the active decoding stage, the present invention introduces a reverse transmission mechanism based on the latest start time of the time window chain, which transforms the time window constraint from post-verification to pre-constraint in the decoding process, thereby improving the feasibility of the scheduling scheme under complex time window conditions.

[0019] (3) The present invention takes the maximum completion time, batching rate and time window violation rate as comprehensive optimization objectives, and can take into account the production cycle, equipment utilization efficiency and time window compliance, providing an efficient, stable and executable scheduling scheme for semiconductor furnace tube area batch processing equipment.

[0020] (4) A process-machine dual-layer coding structure is adopted, and a hybrid heuristic method of "random generation + maximum batch efficiency" is used to generate the initial solution of workshop scheduling. The coding sequence is included as an individual in the population, thereby providing sequence information and equipment allocation information for subsequent decoding.

[0021] (5) Active decoding and elite strategy: Perform rule-based active decoding on individuals in the population to generate specific scheduling schemes; at the same time, evaluate the fitness of the decoding schemes through the fitness function, select the current best solution to join the elite population, eliminate the last second-best solution to achieve dynamic updates, and retain the historical best solution.

[0022] (6) Population genetic manipulation: Genetic manipulation is performed on individuals in the population to promote evolution: at the process level, POX crossover (recombining process order) and insertion mutation (adjusting process arrangement) are used, and at the machine level, uniform crossover (randomly recombining machine allocation) and single-point mutation (changing single machine allocation) are used.

[0023] (7) Second-level neighborhood local optimization: After the population completes the genetic evolution operation, a second-level neighborhood search algorithm is introduced for deep optimization: the first-level neighborhood performs cross-machine batch movement to achieve overall processing resource optimization; the second-level neighborhood implements same-machine batch movement to achieve processing sequence optimization within the equipment; the two-level search work together to improve the local optimality of the solution. Attached Figure Description

[0024] Figure 1 This is a flowchart of a semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the rule-based active decoding process in an embodiment of the present invention; Figure 3 This is a schematic diagram of crossover and mutation operations in an embodiment of the present invention, wherein (a), (b), (c) and (d) correspond to the POX crossover operator, uniform crossover operator, insertion mutation operator and single-point mutation operator, respectively; Figure 4 This is a schematic diagram of cross-machine neighborhood operation in an embodiment of the present invention; Figure 5 This is a schematic diagram of machine-neighborhood operations in an embodiment of the present invention; Figure 6 This is a scheduling Gantt chart obtained from an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0026] This invention provides a rule-based decoding-based scheduling method for batch processing equipment in semiconductor furnace tube areas. It is an integrated solution method encompassing "encoding initialization—rule-based active decoding—fitness evaluation and elite retention—genetic evolution—hierarchical neighborhood reinforcement." The method employs a two-layer encoding structure combining process sequence encoding and machine selection encoding, and generates an initial population through random initialization and maximum batching efficiency initialization. Subsequently, active decoding based on time window constraints and batching rules is performed on individuals, generating executable scheduling schemes while simultaneously forming batch information, machine processing sequences, and the latest start information of the time window chain. Based on this, superior individuals are screened through fitness evaluation and elite retention mechanisms, and candidate individuals are generated through genetic operations at the process and machine levels. Furthermore, a hierarchical neighborhood search is performed based on the critical path and critical block identification results. First, cross-machine batch movement is performed to optimize resource allocation between equipment, and then intra-machine batch movement is performed to optimize the batch processing order within the equipment, thereby achieving targeted reinforcement of critical batches. If the termination condition is not met, the method returns to the decoding and evaluation stage to continue iteration until the optimal scheduling scheme is output.

[0027] This invention achieves comprehensive optimization of maximum completion time, batching rate, and time window violation rate by coordinating process sequencing, machine allocation, batch formation, and time window control in a unified solution chain. It can provide an efficient, stable, and executable scheduling scheme for batch processing equipment in semiconductor furnace tube areas.

[0028] The scheduling method mainly includes the following steps: Step S1: Based on workshop workpieces, processes, selectable machines, batch processing constraints, and time window constraints, an initial population is generated by combining random initialization with maximum batch efficiency initialization. Individuals in the initial population are represented by a two-layer chromosome composed of process sequence encoding and machine selection encoding.

[0029] Step S1 yields the encoded individuals available for decoding. The dual-layer encoding uses a fixed-length encoding, with the encoding length equal to the total number of all workpiece processes. The process sequence encoding is used to characterize the scheduling order of each workpiece process, while the machine selection encoding is used to characterize the target processing machine corresponding to each process, thus providing sequence information and equipment allocation information for subsequent decoding.

[0030] The initial population consists of two parts: the first part is randomly initialized individuals, which are obtained by randomly arranging the process and randomly selecting feasible machines to ensure population diversity; the second part is individuals initialized with maximum batching efficiency, which are obtained by prioritizing the selection of machines that can match the batches that have been formed from the set of available machines. If there are no machines that can be batched, the earliest machine that can start is selected to improve the batching quality of the initial individuals.

[0031] Step S2 involves performing rule-based active decoding on individuals in the initial population, converting the encoded information corresponding to the individuals into an executable scheduling scheme that satisfies the sequence of processes, batch processing constraints, and time window constraints, while simultaneously generating batch information, machine processing sequence, and the latest start information of the time window chain.

[0032] Perform rule-based active decoding. For the double-layer coded individuals generated in step S1, extract the scheduling processes in sequence according to the process sequence codes, and determine the corresponding equipment by combining the machine selection codes.

[0033] The current scheduling process is extracted sequentially according to the process sequence code, and the target processing machine for the current scheduling process is determined by combining the machine selection code. The current scheduling process is then type-identified, and decoding is performed according to the following rules: When the current scheduling process is in the middle of a time window, a feasible insertion position is first searched based on the active insertion rule, and then the completion time of the process is compared with the latest start time of its corresponding subsequent process. If the completion time of the process is not later than the latest start time of the subsequent process, the current decoding result is maintained; if the completion time of the process is later than the latest start time of the subsequent process, the set of available machines for the process is traversed, and the earliest startable machine is reselected for processing. To reduce the risk of time window violations, when the current process is not an intermediate process within a time window, it first determines whether a compatible batch exists at the current insertion position. If so, the process is merged into the compatible batch; if not, a search is conducted forward and backward along the current insertion position for batches that can be grouped together, and the batch movement direction is determined based on the principle of minimizing additional waiting time. If no batches that can be grouped together are found, a new batch is created at the current insertion position. When the current process is the starting process of a time window, the latest start time of each related process in the time window chain is calculated in reverse based on the ending process of the corresponding time window, and the latest start time is used as the constraint basis for subsequent process decoding, machine reselection, and neighborhood movement. Through the above rule decoding method, batch information, machine processing sequence, and latest start time information of the time window chain are simultaneously generated while generating an executable scheduling scheme, thereby providing an input basis for fitness evaluation in step S3 and hierarchical neighborhood search in step S5.

[0034] Step S3: The fitness of the obtained executable scheduling scheme is evaluated, and an elite retention strategy is implemented on the individuals according to the fitness evaluation results to obtain the dominant individuals to participate in the next round of evolution, forming a new population.

[0035] A comprehensive fitness function is constructed based on the maximum completion time, batching rate, and time window violation rate. The comprehensive fitness function is as follows:

[0036] in, Represents an individual The overall fitness value, and These represent the normalized results for total processing time, batch rate, and time window violation rate, respectively. For the corresponding weight coefficients, and satisfying Among them, the batching rate is a positive indicator, obtained through items The evaluation direction is converted to minimize the total processing time and time window violation rate; the individual's quality is determined based on the comprehensive fitness function, and the best individual in the current generation is directly retained to the next generation through the elite retention strategy.

[0037] Specifically, for the executable scheduling scheme obtained after active decoding in step S2, the maximum completion time, batching rate, and time window violation rate are calculated as evaluation indicators. Since the dimensions, numerical ranges, and optimization directions of each evaluation indicator are different, they are first normalized before constructing the comprehensive fitness function. Specifically, the normalized results of the total processing time, batching rate, and time window violation rate are expressed as follows:

[0038] in, , and Each represents an individual The maximum completion time, batching rate, and time window violation rate of the corresponding scheduling scheme; , , , , and These represent the minimum and maximum values ​​of the corresponding evaluation index in the current population, respectively.

[0039] Construct a comprehensive fitness function based on the normalization results:

[0040] in, For individuals The overall fitness value, , and Let be the weight coefficient, and satisfy... Since a higher batch rate is a better indicator, therefore, the following approach is adopted. This is transformed into a minimization evaluation direction consistent with the maximum completion time and time window violation rate. Individuals in the population are ranked according to their overall fitness value, and the best individuals of the current generation are directly retained to the next generation using an elite retention strategy.

[0041] Step S4: Perform genetic operations on individuals in the new population to obtain candidate individuals.

[0042] Specifically, a roulette wheel selection strategy is first used to select parent individuals from the current population. Then, crossover and mutation are performed on both the process sequence encoding and machine selection encoding. For process sequence encoding, the POX operator is used to recombine the parent process sequences, forming a new process arrangement while retaining the relative order of some priority processes. An insertion mutation operator is used to adjust the positions of local processes to enhance the ability to perturb the processing order. For machine selection encoding, a uniform crossover operator is used to exchange the machine allocation genes of different parent individuals, and a single-point mutation operator is used to randomly change the equipment selection results of the target process to enhance the search ability for process-machine matching relationships. After the genetic operations are completed, the newly generated individuals are re-decoded using the method in step S2 to update the batch formation results, equipment processing order, and time window constraints after the encoding changes.

[0043] Step S5: Perform hierarchical neighborhood search on the scheduling scheme corresponding to the candidate individual. Based on the critical path and critical block information, take the critical batch as the local search object and perform cross-machine batch movement and intra-machine batch movement in sequence. Among them, the critical path, critical block and critical batch are identified based on the scheduling scheme corresponding to the candidate individual.

[0044] To perform a hierarchical neighborhood search, the critical path and critical block in the scheduling scheme are first identified, and the critical batches are divided into first batch of the block, intra-block batches, and last batch of the block. The critical path and critical block information is used to limit the local search objects and reduce invalid searches. Among them, the critical batch is the batch located on the critical path that has a significant impact on the maximum completion time, batching rate, or time window feasibility.

[0045] Specifically, select target batches on the critical path. It iterates through the set of available machines, searching for idle or batch processing periods on the target machine in chronological order; for candidate insertion positions, it uses the previous batch at the insertion position. Completion time and the next batch after the insertion position Start time Determine the candidate insertion interval and combine it with the target batch. Batch structure time window interval Perform a feasibility assessment; when the overlap length between the candidate insertion interval and the time window interval is greater than the target batch processing time. This means simultaneously satisfying both the batch processing time constraint and the time window constraint consisting of the earliest and latest start times of the target batch:

[0046] When the target batch is moved to the candidate position, the load between devices is redistributed and the resource allocation between devices is improved.

[0047] After completing cross-machine batch movement, further intra-machine batch movement is performed. While keeping the processing machine of the target batch unchanged, when the target batch is the first batch of a block, its subsequent insertable time intervals are searched; when the target batch is the last batch of a block, its preceding insertable time intervals are searched; when the target batch is a batch within a block, its preceding and following insertable time intervals on its machine are searched in chronological order. When the candidate insertion interval and the time window interval... The existence of overlapping intervals means that both the target batch time window constraint and the processing time constraint are satisfied simultaneously:

[0048] When the target batch is moved to a candidate position, the batch processing order on the same device is optimized and the waiting time between batches is reduced.

[0049] in, For batch The start time of the batch structure time window, which is the maximum value of the earliest completion time of all preceding processes in the batch; For batch The batch structure time window end time is the minimum value of the latest start time of all subsequent processes within that batch. Through the above two-level neighborhood search, the processing position of critical batches is first improved from the perspective of inter-equipment redistribution, and then waiting and blocking are compressed from the perspective of internal equipment rearrangement, thereby outputting locally enhanced improved individuals.

[0050] The hierarchical neighborhood search adopts an iterative mechanism of "traversing key batches - executing corresponding neighborhoods - re-identifying key paths if improvements are made"; when the objective function is improved after a key batch is moved, the current individual is updated and key path identification and key block division are re-executed; when there is no further improvement in all key batches, the local search ends.

[0051] Step S6: Determine whether the maximum number of iterations or fitness convergence condition is met; if not, proceed to step S2; if met, output the optimal scheduling scheme.

[0052] Specifically, determine whether the maximum number of iterations or fitness convergence condition is met; if not, return to step S2 to continue the closed-loop optimization of "decoding-evaluation-evolution-reinforcement"; if met, output the optimal scheduling scheme.

[0053] The present invention will be further described in detail below with reference to specific embodiments.

[0054] Please see Figure 1 This invention provides a method for solving the scheduling problem of semiconductor furnace tube area batch processing equipment based on rule decoding, the steps of which are as follows: Step 1. Data Preparation. Collect data such as the set of workpieces to be scheduled, the set of processes for each workpiece, the set of machines available for each process, batch processing capacity constraints, process compatibility relationships, processing time, time window start and end process relationships, and equipment availability periods, and use this data as input for Step 2.

[0055] Step 2. Two-layer encoding and initial population generation. For each individual, construct an equal-length two-layer chromosome: the first layer records the process scheduling order, and the second layer records the machine selection result for each process. To balance diversity and initial quality, a portion of the individuals in the population are generated completely randomly, while the other portion is generated using the maximum batching efficiency rule. That is, machines that can form batches with similar workpieces are prioritized; if no suitable batching machine exists, the earliest machine that can start is selected. The individuals output in this step directly enter the decoding stage.

[0056] Step 3. Rule-based active decoding. For example... Figure 2As shown, the scheduling processes are read one by one according to the process sequence code, and the target processing equipment is determined by combining the machine selection code. Specifically, the type of the current process is first identified. If the current process is an intermediate process within a time window, the earliest feasible insertion position is searched according to the active insertion rule, and the completion time of the process on the current machine is calculated. Then, the completion time is compared with the latest start time of the corresponding subsequent process. If the completion time of the process is not later than the latest start time of the subsequent process, the current decoding result is maintained. If the completion time of the process is later than the latest start time of the subsequent process, it indicates that there is a risk of time window violation. At this time, the set of available machines for the process is traversed, and the earliest startable machine is reselected for processing to reduce the probability of time window exceeding the limit. If the current process is not an intermediate process within a time window, first determine if a compatible batch already exists at the current insertion position. If it does, the current process is directly merged into that batch. If not, search for compatible batches forward and backward from the current insertion position, calculate the additional waiting time caused by batch movement in different directions, and select the scheme with the smaller additional waiting time to determine the batch movement direction. If no compatible batch still exists, create a new batch at the current insertion position. When the current process is the starting process of a time window, calculate the latest start time of each process in the chain backward along the time window chain based on the corresponding ending process position of the time window, and store the result for feasibility determination of subsequent process insertion, machine reselection, and neighborhood movement. Through the above decoding process, the encoded individual can be converted into an executable scheduling scheme that satisfies the process sequence relationship, batch processing constraints, and time window constraints.

[0057] Step 4. Fitness Evaluation and Elite Retention. For the executable scheduling scheme obtained from the rule decoding in the previous step, calculate the maximum completion time, batching rate, and time window violation rate, and construct a comprehensive fitness function based on these. Considering the different dimensions, value ranges, and optimization directions of each evaluation indicator, normalize each indicator to obtain the normalized value of the maximum completion time. Normalized value of batch rate Normalized value of violation rate within time window Based on this, a comprehensive fitness function is constructed. ,in, Represents an individual The overall fitness value, , and This represents the weight coefficient of the corresponding evaluation indicator, and satisfies... Since the batch rate is a positive indicator, it is adopted. This is transformed into a minimization evaluation direction consistent with the total processing time and time window violation rate. Individuals in the population are ranked according to their comprehensive fitness values, and an elite retention strategy is adopted to directly retain the best individuals of the current generation to the next generation.

[0058] Step 5. Genetic manipulation. For example... Figure 3 As shown, firstly, based on the individual fitness values ​​obtained in step 4, a roulette wheel selection strategy is used to select parent individuals from the current population. Then, crossover and mutation are performed on the process sequence encoding and machine selection encoding respectively. For process sequence encoding, the POX operator is used: firstly, a subset of workpiece-corresponding processes are selected from the process set as a priority retention set. The processes in this set retain their original relative order in the offspring, and the remaining processes are filled sequentially according to their appearance order in another parent generation, thus generating a new process sequence. Based on this, an insertion mutation operator is further used to randomly select a process and insert it into another position in the same sequence to achieve local processing order adjustment. For machine selection encoding, a uniform crossover operator is used to select machine allocation genes from different parents position by position with a preset probability, thus forming a new machine selection sequence. Then, a single-point mutation operator is used to randomly select the machine selection position corresponding to a process and reassign it within its selectable machine set to change the matching relationship between the process and the processing equipment. Since crossover and mutation operations simultaneously affect the process sequence, batch formation results, and temporal relationships on the time window chain, the new individuals generated after genetic operations are re-decoded using the method in step 3 to reconstruct an executable scheduling scheme.

[0059] Step 6. Hierarchical Neighborhood Reinforcement. This step takes the executable scheduling scheme obtained from the active decoding in Step 5 as input and performs a hierarchical neighborhood search to locally reinforce and optimize critical batches. Specifically, first, critical paths and critical blocks are identified according to the current scheduling scheme, and critical batches located in critical blocks are used as local search objects; then, two levels of neighborhood search are performed sequentially in the order of "first cross-machine batch movement, then same-machine batch movement". Cross-machine batch movement is used to improve load configuration between devices, while same-machine batch movement is used to reduce waiting and congestion on the same device, thereby obtaining improved individuals after local reinforcement.

[0060] (1) Cross-machine batch movement like Figure 4 As shown, for the target batch on the critical path It iterates through its set of available machines and searches for idle or batchable processing periods on the target machine in chronological order. For any candidate insertion position, it considers the previous batch at that insertion position. Completion time and the next batch after the insertion position Start time Determine the candidate insertion interval At the same time, in conjunction with the target batch Batch structure time window interval Determine whether the candidate position meets the processing time constraint and the time window constraint; the overlap length between the candidate insertion interval and the time window interval is not less than the target batch processing time. When a candidate position is deemed feasible, it satisfies the following conditions:

[0061] When a candidate location meets the above constraints, the target batch is moved to that location to redistribute the load among the devices; otherwise, the process continues to traverse the next candidate location.

[0062] (2) Moving within the same machine batch After completing the cross-machine batch move, further perform the same-machine batch move, such as... Figure 5 As shown. While keeping the processing machine where the target batch is located unchanged, when the target batch is the first batch of a block, search for its subsequent insertable time periods; when the target batch is the last batch of a block, search for its preceding insertable time periods; when the target batch is a batch within a block, search for its preceding and following insertable time periods on its machine in chronological order, and determine whether it satisfies the time window constraint and processing time constraint; when there is an overlap between the candidate insertion interval and the time window interval, the candidate position is considered feasible, i.e., it satisfies:

[0063] When a candidate position meets the above constraints, the target batch is moved to the candidate position to optimize the batch processing order on the same device and reduce the waiting time between batches; if it does not meet the constraints, the search continues for the next candidate position.

[0064] (3) Neighborhood solution update After each cross-machine batch move or intra-machine batch move, the updated scheduling scheme is decoded and re-evaluated. If the moved individual is better than the original individual, the neighborhood solution is accepted, and the search continues with the updated individual. If the moved individual is not better than the original individual, the neighborhood solution is discarded, and the original state is restored. After all critical batches have completed the above two-level neighborhood search, the locally enhanced improved individual is output, and the subsequent iteration process begins.

[0065] Step 7. Termination and Output. When the number of iterations reaches the preset upper limit, or when the optimal fitness no longer improves after several consecutive generations, the process ends, and the scheduling scheme corresponding to the optimal individual is output. This scheduling scheme includes batch composition, the machine corresponding to each batch, the start and end processing times of each batch, and time window violation situations.

[0066] Through the above steps, this embodiment forms a closed-loop solution process from encoded representation to executable scheduling output. Step 2 provides the encoding carrier; step 3 transforms the encoding into a scheduling result that satisfies the constraints; step 4 quantitatively evaluates the decoding result and retains elites; step 5 expands the search range; step 6 implements targeted reinforcement for critical batches; and step 7 controls the termination of the iteration. Figure 6 This is a scheduling Gantt chart of a specific example obtained using the method of the present invention.

[0067] The present invention also provides a semiconductor furnace tube area batch processing equipment scheduling system based on rule decoding. The system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it performs the semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described above.

[0068] The present invention also provides a computer-readable storage medium storing machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the rule-based decoding semiconductor furnace tube area batch processing device scheduling method as described above.

[0069] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for scheduling batch processing equipment in a semiconductor furnace tube area based on rule decoding, characterized in that, The steps are as follows: S1, based on workshop workpieces, processes, selectable machines, batch processing constraints and time window constraints, generates an initial population by combining random initialization with maximum batch efficiency initialization, and uses a two-layer chromosome composed of process sequence encoding and machine selection encoding to represent individuals in the initial population; S2, performs rule-based active decoding on individuals in the initial population, converting the encoded information corresponding to the individuals into an executable scheduling scheme that satisfies the sequence of processes, batch processing constraints, and time window constraints; S3. The fitness of the obtained executable scheduling scheme is evaluated, and an elite retention strategy is implemented on individuals based on the fitness evaluation results to obtain the dominant individuals to participate in the next round of evolution, forming a new population. S4, perform genetic operations on individuals in the new population to obtain candidate individuals; S5. Perform hierarchical neighborhood search on the scheduling scheme corresponding to the candidate individual. Based on the critical path and critical block information, take the critical batch as the local search object and perform cross-machine batch movement and same-machine batch movement in sequence to obtain the optimal scheduling scheme. Among them, the critical path, critical block and critical batch are identified based on the scheduling scheme corresponding to the candidate individual.

2. The semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described in claim 1, characterized in that: A comprehensive fitness function is constructed based on the maximum completion time, batching rate, and time window violation rate. The fitness of the resulting executable scheduling scheme is evaluated based on the comprehensive fitness function.

3. The semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described in claim 2, characterized in that: The comprehensive fitness function is: in, Represents an individual The overall fitness value, and These represent the normalized results for total processing time, batch rate, and time window violation rate, respectively. For the corresponding weight coefficients, and satisfying .

4. The semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described in claim 1, characterized in that: The initial population consists of two parts: the first part is randomly initialized individuals, which are obtained by randomly arranging the process and randomly selecting feasible machines; the second part is initialized individuals with the maximum batching efficiency, which are obtained by prioritizing the selection of machines that can match the already formed batches from the set of available machines. If there are no machines that can be batched, the earliest machine that can start operation is selected.

5. The semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described in any one of claims 1-4, characterized in that: Extract the current process to be scheduled in sequence according to the process sequence code, and determine the target processing machine for the current process to be scheduled by combining the machine selection code.

6. The semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described in claim 5, characterized in that: The current scheduled process is type-identified. When the current scheduled process is in the middle of a time window, a feasible insertion position is first searched based on the active insertion rule. Then, the completion time of the process is compared with the latest start time of its corresponding subsequent process. If the completion time of the process is not later than the latest start time of the subsequent process, the current decoding result is maintained. If the completion time of the process is later than the latest start time of the subsequent process, the set of available machines for the process is traversed, and the earliest startable machine is reselected for processing. When the current process is not in the middle of a time window, it is first determined whether there is a compatible batch at the current insertion position. If there is, the process is merged into the compatible batch. If not, a batch that can be grouped is searched forward and backward along the current insertion position, and the batch movement direction is determined according to the principle of minimizing additional waiting time. If no batch that can be grouped is found, a new batch is created at the current insertion position. When the current process is the starting process of a time window, the latest start time of each related process on the time window chain is calculated in reverse based on the ending process of the corresponding time window, and the latest start time is used as the constraint basis for subsequent process decoding, machine reselection, and neighborhood movement.

7. The semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described in any one of claims 1-4, characterized in that: In step S4, a roulette wheel selection strategy is first used to select parent individuals from the current population. Then, crossover and mutation are performed on the process sequence encoding and machine selection encoding respectively. For process sequence encoding, the POX operator is used to recombine the parent process sequence, forming a new process arrangement while retaining the relative order of some priority processes. An insertion mutation operator is used to adjust the position of local processes. For machine selection encoding, a uniform crossover operator is used to exchange the machine allocation genes of different parent individuals, and a single-point mutation operator is used to randomly change the equipment selection result of the target process.

8. The semiconductor furnace tube area batch processing equipment scheduling method based on rule decoding as described in any one of claims 1-4, characterized in that: Identify the critical path and critical blocks in the scheduling scheme, and divide the critical batches into first-batch batches, intra-batch batches, and last-batch batches; select target batches on the critical path. It iterates through the set of available machines, searching for idle or batch processing periods on the target machine in chronological order; for candidate insertion positions, it uses the previous batch at the insertion position. Completion time and the next batch after the insertion position Start time Determine the candidate insertion interval and combine it with the target batch. Batch structure time window interval Perform a feasibility assessment; when the overlap length between the candidate insertion interval and the time window interval is greater than the target batch processing time. At that time, the target batch is moved to the candidate position, thereby realizing the redistribution of load among devices; After completing the cross-machine batch move, further perform the same-machine batch move. While keeping the processing machine where the target batch is located unchanged, when the target batch is the first batch of the block, search for its subsequent insertable time period; when the target batch is the last batch of the block, search for its preceding insertable time period; when the target batch is a batch within the block, search for the preceding and following insertable time periods on its machine in chronological order.

9. A semiconductor furnace tube area batch processing equipment scheduling system based on rule decoding, characterized in that: The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it performs the rule-based decoding semiconductor furnace tube area batch processing equipment scheduling method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the rule-based decoding semiconductor furnace tube area batch processing equipment scheduling method according to any one of claims 1-8.

Images