An intelligent scheduling system for job shop scheduling of parallel batch processing machines

Abstract

This invention relates to the field of intelligent scheduling and control technology, specifically to an intelligent scheduling system for parallel batch processing machine job shop scheduling. The system first acquires the arrival time, preparation time, maximum in-process time limit, and oxidation rate constant of each batch, and sets multiple tentative start-up times. Then, for each start-up time, it filters processable and scrapped batches, and performs virtual loading sorting based on the maximum in-process time limit and batch capacity. Next, it linearly calculates the yield impact ratio of each loaded batch by combining waiting time and oxidation rate constant, and constructs a net output evaluation value for each start-up time by comprehensively considering loading volume, scrap volume, and yield loss, and calculates the delay gain index relative to the initial time. Finally, the scheduling module selects the optimal start-up time based on the delay gain index. This system effectively balances the conflict between "waiting for full load" and "avoiding overdue scrapping" under unsteady material conditions, improving the efficiency of equipment output and scheduling decisions.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling and control technology, and specifically to an intelligent scheduling system for job shop scheduling of parallel batch processing machines. Background Technology

[0002] In discrete manufacturing fields such as semiconductor manufacturing (Fab), parallel batch machines (such as diffusion furnaces, oxidation furnaces, and CVD equipment) are expensive bottleneck resources. The operating characteristics of such equipment are: multiple workpieces (batches) can be accommodated simultaneously in a single operation, but once the process is started, it cannot be interrupted or materials added in the middle, and the process cycle is relatively long.

[0003] In actual production, loading is carried out sequentially in batches. Upstream materials typically exhibit non-steady-state, pulsating flow characteristics, and different batches of materials have strict maximum in-process time (Q-Time) constraints. If the waiting time of materials in front of the furnace exceeds Q-Time, it will lead to scrapping. If the start-up timing is decided solely based on the current batches in stock, it is difficult to cope with the trade-off between the "waiting benefits" and "timeout risks" brought about by the non-steady-state material flow, which can easily lead to idle equipment or high-value batches being scrapped due to waiting. Summary of the Invention

[0004] To address the technical problem in related technologies of balancing the "waiting benefits" and "timeout risks" caused by unsteady incoming material flows, which can easily lead to equipment idleness or high-value batches being scrapped due to waiting, this invention provides an intelligent scheduling system for parallel batch processing machine job shop scheduling. The specific technical solution adopted is as follows: This invention proposes an intelligent scheduling system for job shop scheduling in parallel batch processing machines. The system includes: The acquisition module is used to acquire the initial time of the scheduling process and the arrival time, preparation time, maximum in-process time, and oxidation rate constant of different semiconductor batches; and to determine different tentative power-on times after the initial time. The loading analysis module is used to filter out the processable batches and scrapped batches at each startup time from all batches based on different startup times, the arrival time of each batch, the preparation time, and the maximum in-process time limit; for all processable batches, virtual processing sorting is performed according to the maximum in-process time limit and batch capacity to obtain the loading batches. The timing analysis module is used to perform linear analysis of the yield impact based on the arrival time and oxidation rate constant of the semiconductor materials in the loading batch, and to determine the yield impact ratio of the loading batch; based on the capacity of the loading batch and the scrapped batch, and the yield impact ratio of each loading batch, to determine the net output evaluation value at each startup time; and combined with the difference between the net output evaluation value at each startup time and the initial time, to determine the delay gain index. The scheduling module is used to schedule power-on by combining the delay gain metrics at the initial moment and at all power-on moments.

[0005] Furthermore, the step of selecting the processable batches and scrapped batches for each startup time from all batches based on different startup times combined with the arrival time, preparation time, and maximum in-process time limit of each batch includes: Calculate the sum of the arrival time and preparation time for each batch, and use it as the processing time. Batches whose processing time is earlier than the machine start-up time are considered as processable batches; The sum of the arrival time and the maximum in-use time limit is used as the scrapping time; Batches whose processing time is later than their scrapping time will be considered scrapped batches.

[0006] Furthermore, the virtual processing sorting of all processable batches based on the maximum work-in-process time and batch capacity to obtain loading batches includes: Based on the earliest deadline priority strategy, all processable batches are sorted according to the order of the maximum in-process time limit. Loading is carried out batch by batch according to the sorting result and the capacity of the processable batches until no processable batch can be fully loaded. The loaded batches are then regarded as the loading batches.

[0007] Furthermore, the step of performing a linear analysis of the yield impact based on the arrival time and oxidation rate constant of the loaded batch of semiconductor material to determine the proportion of the yield impact of the loaded batch includes: The time interval between the arrival time and the power-on time is taken as the physicochemical decay period. Calculate the product of the physicochemical decay time period and the oxidation rate constant, and normalize it to represent the degree of physicochemical decay. The percentage of yield impact is determined based on the degree of physicochemical degradation and the maximum percentage of the preset yield of the loaded batch of semiconductor materials.

[0008] Furthermore, determining the yield impact percentage based on the degree of physicochemical degradation and the maximum percentage of the preset yield impact of the loaded batch of semiconductor materials includes: Calculate the product of the degree of physicochemical degradation and the maximum percentage of the preset yield impact, and use it as the percentage of yield impact.

[0009] Furthermore, the determination of the net output evaluation value at each startup time based on the capacity of the loading batch and the scrapped batch, and the percentage impact of the yield on each loading batch, includes: The yield risk level is determined by combining the proportion of all loading batches and their impact on yield. Based on the capacity and yield risk of scrapped batches, determine the amount of capacity loss; The difference between the total quantity of all loaded batches and the amount of capacity loss is used as the net output evaluation value.

[0010] Furthermore, the determination of yield risk by combining all loaded batches with their respective proportions of yield impact includes: Calculate the product of the capacity and the percentage of yield impact for each loading batch to obtain the batch yield impact. The sum of the batch yield impact values ​​of all loaded batches is taken as the yield risk value.

[0011] Furthermore, determining the capacity loss based on the capacity and yield risk of scrapped batches includes: Calculate the product of the capacity of the scrapped batch and the preset scrapping penalty weight to obtain the scrapping penalty amount; The sum of the scrap penalty and the yield risk is taken as the production capacity loss.

[0012] Furthermore, the determination of the delay gain index by combining the difference in net output evaluation values ​​between each startup time and the initial time includes: The difference between the net output evaluation value at each startup time and the initial time is calculated and used as the delay gain index.

[0013] Furthermore, the power-on scheduling, which combines the delay gain metrics from the initial time and all power-on times, includes: If all delay gain indicators are less than or equal to 0, then the initial time is taken as the target time; Otherwise, the power-on time with the maximum delay gain index shall be taken as the target time; Each loading batch is scheduled and prepared based on the target time and the preparation time for each loading batch.

[0014] The present invention has the following beneficial effects: The embodiments of this invention can effectively solve the problem of balancing "waiting benefits" and "timeout risks" under non-steady-state incoming material flow in related technologies: the system performs virtual loading simulations for multiple tentative start-up times based on arrival time, preparation time, maximum in-process time limit, and oxidation rate constant, and combines oxidation decay model to quantify the yield impact ratio of each loading batch, and then calculates the net output evaluation value and delay gain index of each start-up time by comprehensively considering the capacity of scrapped batches and yield degradation loss; thus, the scheduling decision no longer relies solely on physical full load or simple deadline rules, but dynamically evaluates the real benefits of delayed waiting under a unified economic value scale, thereby avoiding the scrapping of high-value batches due to blind waiting, and preventing equipment idleness caused by premature start-up, which significantly improves the effective output and scheduling robustness of parallel batch processors under non-steady-state incoming material conditions. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a structural diagram of an intelligent scheduling system for parallel batch processing machine job shop scheduling provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an intelligent scheduling system for parallel batch processing machine job shop scheduling proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details a specific scheme for an intelligent scheduling system for parallel batch processing machine job shop scheduling provided by the present invention.

[0020] Please see Figure 1 This diagram illustrates an intelligent scheduling system architecture for parallel batch processing machine job shop scheduling according to an embodiment of the present invention. The system includes: The acquisition module 101 is used to acquire the initial time of the scheduling process and the arrival time, preparation time, maximum in-process time limit and oxidation rate constant of different semiconductor batches; and to determine different tentative power-on times after the initial time.

[0021] In discrete manufacturing fields such as semiconductor manufacturing (Fab), parallel batch machines (such as diffusion furnaces, oxidation furnaces, and CVD equipment) are expensive bottleneck resources. The operating characteristics of such equipment are: multiple workpieces (batches) can be accommodated simultaneously in a single operation, but once the process is started, it cannot be interrupted or materials added in the middle, and the process cycle is relatively long.

[0022] In actual production, upstream incoming materials typically exhibit non-steady-state, pulsating flow characteristics, and different batches of materials have strict maximum in-process time (Q-Time) constraints. If the waiting time of materials in front of the furnace exceeds Q-Time, it will lead to scrapping. If the start-up timing is decided solely based on the current in-stock batches, it is difficult to cope with the trade-off between the "waiting benefits" and "timeout risks" brought about by the non-steady-state incoming material flow, which can easily lead to equipment idleness or high-value batches being scrapped due to waiting.

[0023] To ensure the consistency of all time logic within this scheduling calculation cycle, the system first performs environment initialization. Therefore, this embodiment of the invention sets the initial time of the scheduling process. At the moment scheduling is triggered (e.g., when the previous round of work on the device is completed or periodic polling is triggered), the system reads the current system clock and locks that time as the initial time.

[0024] The initial moment can serve as the zero point for calculating the remaining time and extrapolating future arrival intervals. It is no longer updated with the passage of the physical clock, thus eliminating time drift errors caused by computation time.

[0025] It should be noted that the initial startup time can either be used immediately for the current round of operations, or a certain period of waiting can be allowed depending on the current batches to be processed. That is, different startup times can be determined after the initial startup time, and the current round of operations can be processed at each startup time. In this embodiment of the invention, the startup time can be a periodic time after the initial startup time, for example, every half hour.

[0026] The oxidation rate constant characterizes a key process parameter of a semiconductor material (such as a silicon wafer) surface or thin film under specific process environments (such as high temperature and oxygen content). It refers to the rate constant of oxidation reaction of a semiconductor material under specific environmental conditions, describing the rate of degradation of the material's physicochemical properties due to oxidation reaction per unit time in that scenario. Its value is determined by product type, thin film material, and historical yield statistics, and is expressed in hours, used to quantify the degree of physicochemical degradation. Optionally, the oxidation rate constant can be, for example, 0.1 / h.

[0027] The arrival time refers to the moment when the corresponding batch arrives at the processing equipment. This arrival can specifically mean arrival in the preparation library of the corresponding processing flow. For example, if a batch is still being processed in the previous flow and has not yet reached its arrival time, but is now in the preparation library of the current flow after the previous flow is completed, the moment it is located in that preparation library will be used as the arrival time, indicating that the batch is physically ready. Starting from the arrival time, semiconductor materials will gradually experience degradation of their physicochemical properties due to the influence of the storage environment.

[0028] It should be noted that each batch may not be able to be processed directly upon arrival and requires certain preparation work. Therefore, a preparation time is set to provide a buffer for each batch. This preparation work may be the time required for the overall start-up of the equipment or the pre-processing time of the batch materials, which can be adjusted according to the actual situation. For example, it can be 10 minutes.

[0029] Some process steps (such as annealing after ion implantation, etching after photoresist development, and cleaning before metal deposition) are extremely sensitive to the wafer surface condition. If the waiting time is too long, it can lead to surface contamination or thickening of the natural oxide layer, diffusion of doped atoms or accumulation of defects, and an increase in interface state density. These changes can directly cause device electrical parameters to drift, yield to decrease, or even scrap the entire batch.

[0030] Therefore, the maximum in-process time limit indicates the maximum time each batch can wait after its arrival time. For example, if it is 4 hours, then if the corresponding batch has not been processed within 4 hours after its arrival time, it can be determined that it has been severely affected by the environment and the entire batch needs to be scrapped.

[0031] The loading analysis module 102 is used to filter out the processable batches and scrapped batches for each startup time from all batches based on different startup times combined with the arrival time, preparation time and maximum in-process time limit of each batch; for all processable batches, virtual processing sorting is performed according to the maximum in-process time limit and batch capacity to obtain the loading batches.

[0032] In parallel batch processing machine scheduling, if physical loading decisions are made directly for all batches to be processed, it is difficult to predict which batches are still in the effective process window and which have exceeded the deadline at different start-up times. Furthermore, it is impossible to assess the conflict between the full load potential of the equipment and the urgency of the deadline, which can easily lead to high-risk batches being missed or low equipment utilization.

[0033] To address this issue, the present invention focuses on constructing a virtual processing feasibility simulation mechanism for multiple candidate start-up times. Based on each tentative start-up time, it analyzes batches that have arrived but not yet been processed before the current tentative start-up time. Combining the arrival time, preparation time, and maximum in-process time limit of each batch, it dynamically filters out processable batches that have completed preparation and have not exceeded the time limit at that time, as well as scrapped batches that have missed the process window.

[0034] Furthermore, in some embodiments of the present invention, based on different startup times combined with the arrival time, preparation time, and maximum in-process time limit of each batch, the processable batches and scrapped batches at each startup time are selected from all batches, including: calculating the sum of the arrival time and preparation time of each batch as the processing time; selecting batches whose processing time is earlier than the startup time as processable batches; selecting the sum of the arrival time and the maximum in-process time limit as scrapped times; and selecting batches whose processing time is later than the scrapped times as scrapped batches.

[0035] For example, if the arrival time is 0:00 and the preparation time is 30 minutes, then the processing time is 0:30, which means that during the normal loading and processing process, the processing preparation can be completed at 0:30.

[0036] Then, batches whose processing time is earlier than the machine start-up time are identified as processable batches, that is, batches that can be prepared before the machine start-up time and can be loaded for processing immediately. However, batches that cannot be effectively processed within the maximum work-in-process time limit due to their arrival time being too early and the machine start-up time being too late are identified as scrapped batches.

[0037] Furthermore, in some embodiments of the present invention, all processable batches are virtually processed and sorted according to the maximum in-process time limit and batch capacity to obtain loading batches, including: sorting all processable batches according to the order of the maximum in-process time limit based on the earliest deadline priority strategy, loading batches one by one according to the sorting result and the capacity of the processable batches, until any processable batch cannot be fully loaded, and the loaded batches are taken as loading batches.

[0038] The capacity of each batch can be specified as the number of wafers to be processed in each batch, and the batch capacity shall not exceed the maximum loading capacity of the parallel batch processor in a single operation. The "number of wafers that can be processed in each batch" is accumulated sequentially according to the order, until the total number exceeds the "maximum loading capacity of the equipment".

[0039] In this embodiment of the invention, an Earliest Deadline First (EDF) strategy is used to virtually sort and load processable batches within limited capacity. This strategy is driven by the rigid time constraints of Q-Time sensitive processes in semiconductor manufacturing: if high-risk batches nearing their maximum in-process deadline are not prioritized, even if the equipment is subsequently fully loaded, individual batches may be scrapped due to exceeding the deadline, leading to a loss of overall yield or wasted capacity. Therefore, simply pursuing full physical load (such as prioritizing large batches) or First-Come, First-Served (FCFS) cannot simultaneously guarantee urgency and resource utilization.

[0040] From a physical perspective, "maximum in-process time limit" represents the end of the process survival window for each batch. The order of these limits directly reflects the time vulnerability level of the batch. The earlier the time limit, the lower the tolerance for waiting, and the higher the scheduling priority should be. "Batch capacity," on the other hand, reflects the physical loading limit of the equipment in a single operation (such as the maximum number of wafers in a diffusion furnace).

[0041] This invention, through coupling the two, simulates the real loading process in a virtual environment: iterates through the processable batches in ascending order of their deadlines, attempting to load them sequentially. If the remaining capacity is insufficient to accommodate the next complete batch, it continues iterating until all complete batches cannot be loaded, at which point loading stops. This "prioritize urgency, stop when not full" mechanism avoids sacrificing critical batches to achieve full load and also prevents severe underloading of the equipment due to premature truncation.

[0042] The batches that can be loaded after traversal are taken as the loading batches. The loading batches represent the batches that can be fully loaded into the parallel batch processing machine for processing under the corresponding boot time.

[0043] The timing analysis module 103 is used to perform linear analysis of yield impact based on the arrival time and oxidation rate constant of the semiconductor materials in the loading batch, and to determine the yield impact ratio of the loading batch; based on the capacity of the loading batch and the scrapped batch, and the yield impact ratio of each loading batch, to determine the net output evaluation value at each startup time; and combined with the difference between the net output evaluation value at each startup time and the initial time, to determine the delay gain index.

[0044] In parallel batch processing machine scheduling decisions, if only physical load or the number of scrapped batches are used as evaluation indicators, the implicit yield loss caused by waiting cannot be reflected. That is, although the scrapping does not exceed the time limit, the effective output has been reduced due to physical and chemical degradation such as oxidation, thus overestimating the actual benefits of delayed start-up.

[0045] To address this issue, this invention constructs a unified benefit measurement mechanism that integrates explicit scrap and implicit defects. Based on the loading batch corresponding to each power-on moment, linear decay modeling is performed using the arrival time and oxidation rate constant of its semiconductor material to quantify the proportion of yield impact caused by the accumulation of waiting time for each batch. Furthermore, by combining the capacity and yield impact of the loading batch and the scrap batch, the net output evaluation value at the power-on moment is calculated, and the delay gain index is defined by the difference between the net output at the initial moment and the net output at the initial moment.

[0046] Furthermore, in some embodiments of the present invention, a linear analysis of the yield impact is performed based on the arrival time and oxidation rate constant of the loaded batch of semiconductor materials to determine the yield impact ratio of the loaded batch. This includes: taking the time interval between the arrival time and the power-on time as the physicochemical decay time period; calculating the product of the physicochemical decay time period and the oxidation rate constant, and normalizing it to obtain the physicochemical decay degree; and determining the yield impact ratio based on the physicochemical decay degree and the preset maximum yield impact ratio of the loaded batch of semiconductor materials.

[0047] It is understandable that if the physical and chemical degradation time period is greater than or equal to the maximum in-process time limit, the corresponding batch is a scrap batch. If the physical and chemical degradation time period is less than the maximum in-process time limit, the corresponding batch will still undergo oxidation, resulting in a certain yield impact. Therefore, this embodiment of the invention analyzes this part of the yield impact.

[0048] In this embodiment of the invention, the attenuation from the arrival time to the power-on time is analyzed. Therefore, the time interval between the arrival time and the power-on time is directly taken as the physicochemical attenuation period. Then, the product of the physicochemical attenuation period and the oxidation rate constant is calculated and normalized to represent the degree of physicochemical attenuation. Taking an oxidation rate constant of 0.1 / h as an example, if the physicochemical attenuation period is 3 hours, the corresponding physicochemical attenuation degree is 0.3. Then, the maximum and minimum values ​​are normalized to facilitate subsequent calculations. Specifically, the maximum and minimum values ​​can be determined through historical experience to achieve the normalization effect.

[0049] The yield impact percentage is determined based on the degree of physicochemical degradation and the maximum percentage of the preset yield impact of the loaded batch of semiconductor materials. This includes calculating the product of the degree of physicochemical degradation and the maximum percentage of the preset yield impact, which is used as the yield impact percentage.

[0050] The maximum percentage impact of preset yield is a key process parameter in this invention used to characterize the maximum yield loss ratio that a specific batch of semiconductor materials can tolerate under extreme waiting conditions.

[0051] The preset maximum percentage impact on yield is essentially an upper limit threshold determined by product type, process layer characteristics, and historical yield data. It is used to linearly map the dimensionless "physicochemical degradation degree" (value range [0,1]) to the actual percentage impact on yield. Its typical value range is 3% to 10%, for example, it can be set to 7% for high-k dielectric layers.

[0052] Based on the capacity of the loaded and scrapped batches, and the yield impact ratio of each loaded batch, determine the net output evaluation value at each startup time, including: determining the yield risk based on all loaded batches and their yield impact ratio; determining the capacity loss based on the capacity and yield risk of the scrapped batches; and using the difference between the total amount of all loaded batches and the capacity loss as the net output evaluation value.

[0053] Traditional scheduling methods, when evaluating the merits of a strategy, only count the number of wafers loaded into the equipment, ignoring the fact that some wafers have already experienced a decline in yield due to waiting for oxidation, and also failing to include the losses of batches that are scrapped beyond the allotted time in a reasonable cost. This one-sided evaluation will cause the system to tend to delay startup in pursuit of full load, thus sacrificing the effective output of high-value batches.

[0054] Among them, the characteristics of three dimensions—physical loading, implicit yield loss, and explicit scrap loss—are integrated into a quantifiable and comparable economic equivalent output value, namely the net output evaluation value, which is used for intelligent decision-making that balances yield baseline and capacity efficiency.

[0055] Furthermore, in some embodiments of the present invention, the yield impact is analyzed: the yield risk is determined by combining all loading batches and their yield impact ratio, including: calculating the product of the capacity of each loading batch and the yield impact ratio to obtain the batch yield impact; and using the sum of the batch yield impact of all loading batches as the yield risk.

[0056] The yield risk measure represents the capacity of yield risk caused by oxidation. This capacity is the number of wafers in each loading batch. For example, if there are 100 wafers in a loading batch, the corresponding yield impact ratio is 3%, and the calculated batch yield impact is 3 wafers. Since there are different loading batches in one loading, all loading batches are integrated and analyzed. The sum of the batch yield impact of all loading batches is used as the yield risk measure.

[0057] Furthermore, in some embodiments of the present invention, the capacity loss caused by scrapping is analyzed: based on the capacity and yield risk of the scrapped batch, the capacity loss is determined, including: calculating the product of the capacity of the scrapped batch and the preset scrapping penalty weight to obtain the scrapping penalty; and taking the sum of the scrapping penalty and the yield risk as the capacity loss.

[0058] The default scrap penalty weight is a constant greater than 1, which is used to reflect the higher economic loss of scrapped batches compared to batches with degraded yield. Its typical value range is 1.5 to 10, and specifically, it can be 1.5.

[0059] Since the severity of scrapping is greater than the impact on yield, wafers affected by yield can be used as raw materials for lower-end chips through post-processing. However, scrapped batches, having exceeded the maximum in-processing time limit, have suffered irreversible physical and chemical damage (such as severe oxidation and contamination), making the entire batch unusable for subsequent processes and requiring disposal or rework, resulting in 100% capacity loss and additional processing costs. Therefore, a penalty analysis is conducted by setting a preset scrapping penalty weight greater than 1.

[0060] When the scrap batch consists of 100 wafers, the product of the scrap batch capacity of 100 wafers and the preset scrap penalty weight of 1.5 is calculated to obtain a scrap penalty of 150 wafers. Then, the sum of the scrap penalty and the yield risk is taken as the capacity loss. This capacity loss is a loss value that combines the effects of scrap and yield. The larger the capacity loss value, the more capacity loss occurred at the corresponding start-up time.

[0061] Next, a net output analysis is performed, using the difference between the total quantity of all loaded batches and the capacity loss as the net output evaluation value. The total quantity of all loaded batches is the sum of the number of wafers in all loaded batches.

[0062] In this embodiment of the invention, the power-on time is the time after the initial time. Due to the delay effect, some batches in transit will be put into the warehouse so that they can be loaded in time. However, some batches that are already waiting in the warehouse will have a risk of yield loss and scrap. Therefore, the delay between different power-on times and the initial time will lead to changes in efficiency. The delay gain index is used to characterize this change in efficiency.

[0063] Furthermore, in some embodiments of the present invention, the delay gain index is determined by combining the difference between the net output evaluation value at each startup time and the initial time, including: calculating the difference between the net output evaluation value at each startup time and the initial time as the delay gain index.

[0064] In other words, the net output evaluation value at the initial moment is calculated first, and then the net output evaluation value at each startup moment is calculated. The difference between the net output evaluation value at each startup moment and the net output evaluation value at the initial moment is used as the delay gain index.

[0065] The scheduling module 104 is used to perform power-on scheduling by combining the delay gain index at the initial time and at all power-on times.

[0066] Since it can be powered on directly at the initial moment or at different times, the specific power-on time is determined based on the delay gain index.

[0067] Furthermore, in some embodiments of the present invention, power-on scheduling is performed by combining the delay gain index of the initial time and all power-on times, including: when all delay gain indices are less than or equal to 0, the initial time is taken as the target time; otherwise, the power-on time with the largest delay gain index is taken as the target time; and each loading batch is scheduled and prepared according to the target time and the preparation time of each loading batch.

[0068] If all delay gain metrics are less than or equal to 0, it indicates that delayed loading is not as good as loading immediately at the initial moment; therefore, the initial moment is taken as the target moment. If all delay gain metrics are greater than 0, the power-on moment with the highest delay gain metric is taken as the target moment.

[0069] If multiple power-on times have the same delay gain index and all are at their maximum values, then the earliest power-on time is selected as the target time.

[0070] Understandably, if the target time is 3:00, and the preparation time for a certain batch is half an hour, then the loading batch needs to be scheduled and prepared no later than 2:30 to avoid being unprepared just before startup.

[0071] The embodiments of this invention can effectively solve the problem of balancing "waiting benefits" and "timeout risks" under non-steady-state incoming material flow in related technologies: the system performs virtual loading simulations for multiple tentative start-up times based on arrival time, preparation time, maximum in-process time limit, and oxidation rate constant, and combines oxidation decay model to quantify the yield impact ratio of each loading batch, and then calculates the net output evaluation value and delay gain index of each start-up time by comprehensively considering the capacity of scrapped batches and yield degradation loss; thus, the scheduling decision no longer relies solely on physical full load or simple deadline rules, but dynamically evaluates the real benefits of delayed waiting under a unified economic value scale, thereby avoiding the scrapping of high-value batches due to blind waiting, and preventing equipment idleness caused by premature start-up, which significantly improves the effective output and scheduling robustness of parallel batch processors under non-steady-state incoming material conditions.

[0072] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0073] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An intelligent scheduling system for parallel batch machine job shop scheduling, characterized in that, The system comprises: an acquisition module, configured to acquire an initial time of a scheduling process, arrival times of different semiconductor batches, preparation durations, maximum in-process time limits, and oxidation rate constants of the semiconductor materials; and determine different tentative start-up times after the initial time; a loading analysis module, configured to screen workable batches and scrap batches for each start-up time from all batches according to the arrival times, the preparation durations, and the maximum in-process time limits of each batch in combination with the different start-up times; and perform virtual processing sequencing on all workable batches according to the maximum in-process time limits and batch capacities to obtain loaded batches; an aging analysis module, configured to perform a yield influence linear analysis according to the arrival times of the semiconductor materials of the loaded batches and the oxidation rate constants to determine a yield influence proportion of the loaded batches; determine a net output evaluation value of each start-up time according to the capacities of the loaded batches and the scrap batches and the yield influence proportion of each loaded batch; and determine a delay gain index by combining the difference between the net output evaluation value of each start-up time and the initial time; a scheduling module, configured to perform start-up scheduling in combination with the delay gain index of the initial time and all start-up times.

2. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 1, wherein, The screening of workable batches and scrap batches for each start-up time from all batches according to the arrival times, the preparation durations, and the maximum in-process time limits of each batch in combination with the different start-up times comprises: calculating the sum of the arrival time and the preparation duration of each batch as a processing time; regarding the batches with the processing time earlier than the start-up time as workable batches; regarding the sum of the arrival time and the maximum in-process time limit as a scrap time; regarding the batches with the processing time later than the scrap time as scrap batches.

3. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 1, wherein, The virtual processing sequencing on all workable batches according to the maximum in-process time limits and batch capacities to obtain loaded batches comprises: sequencing all workable batches according to the maximum in-process time limits in a priority order based on an earliest deadline first strategy; and loading the workable batches according to the sequencing result and the capacities of the workable batches one batch by one batch until any workable batch cannot be completely loaded, and regarding the loaded batches as loaded batches.

4. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 1, wherein, The yield influence linear analysis according to the arrival times of the semiconductor materials of the loaded batches and the oxidation rate constants to determine a yield influence proportion of the loaded batches comprises: regarding the time interval between the arrival time and the start-up time as a physical-chemical decay time period; calculating the product of the physical-chemical decay time period and the oxidation rate constant as a physical-chemical decay degree after normalization processing; determining the yield influence proportion according to the physical-chemical decay degree and a preset maximum yield influence proportion of the semiconductor materials of the loaded batches.

5. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 4, wherein, The determination of the yield influence proportion according to the physical-chemical decay degree and the preset maximum yield influence proportion of the semiconductor materials of the loaded batches comprises: calculating the product of the physical-chemical decay degree and the preset maximum yield influence proportion as the yield influence proportion.

6. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 1, wherein, The determination of a net output evaluation value of each start-up time according to the capacities of the loaded batches and the scrap batches and the yield influence proportion of each loaded batch comprises: determining a yield risk amount in combination with all loaded batches and the yield influence proportion thereof; determining a capacity loss amount based on the capacities of the scrap batches and the yield risk amount; and The difference between the total amount of all loaded batches and the amount of capacity loss is used as the net output evaluation value.

7. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 6, wherein, The yield risk is determined by combining the proportion of all loaded batches and their yield impact, including: Calculate the product of the capacity and the percentage of yield impact for each loading batch to obtain the batch yield impact. The sum of the batch yield impact values ​​of all loaded batches is taken as the yield risk value.

8. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 6, wherein, The determination of capacity loss based on the capacity and yield risk of scrapped batches includes: Calculate the product of the capacity of the scrapped batch and the preset scrapping penalty weight to obtain the scrapping penalty amount; The sum of the scrap penalty and the yield risk is taken as the production capacity loss.

9. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 1, wherein, The delay gain index is determined by combining the difference between the net output evaluation value at each startup time and the initial time, including: The difference between the net output evaluation value at each startup time and the initial time is calculated and used as the delay gain index.

10. An intelligent scheduling system for parallel batch machine job shop scheduling as recited in claim 1, wherein, The method of scheduling power-on by combining the delay gain metrics of the initial time and all power-on times includes: If all delay gain indicators are less than or equal to 0, then the initial time is taken as the target time; Otherwise, the power-on time with the maximum delay gain index shall be taken as the target time; Each loading batch is scheduled and prepared based on the target time and the preparation time for each loading batch.

Images