Data processing methods, apparatus, devices, and storage media for semiconductor devices

Abstract

This invention relates to the field of data processing technology, and more particularly to a data processing method, apparatus, device, and storage medium for semiconductor devices. The core of the method lies in achieving end-to-end time synchronization through a logical global clock, and combining data processing path selection with precise control instruction generation to construct a closed-loop processing architecture that features unified time, collaborative computation, and precise control. By unifying the time base through a logical global clock, the problem of timing disorder caused by the fragmentation of Event Time and Processing Time in traditional frameworks is effectively solved. Furthermore, by selecting data processing paths based on task attributes, the collaboration between low-latency processing of simple tasks and high-precision analysis of complex tasks is achieved, overcoming the defects of fragmented streaming and batch computing resources. The precise control instruction generation mechanism ensures a strong correlation between control actions and physical timing, meeting the microsecond-level control precision requirements of semiconductor devices.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data processing method, apparatus, device, and storage medium for semiconductor devices. Background Technology

[0002] In the semiconductor manufacturing field, high-precision control of equipment and real-time data analysis are the core links to ensure process quality. As semiconductor devices develop towards the micro-nano scale, core equipment such as lithography machines and etching machines have put forward stringent requirements on the real-time performance, timing accuracy and batch processing computing collaboration capabilities of data processing.

[0003] Although current mainstream streaming and batch processing frameworks have achieved the integration of streaming computing and batch processing, they still reveal fundamental shortcomings that cannot be ignored in the specific scenario of semiconductor device control:

[0004] First, there is the problem of inconsistent time bases. Streaming and batch processing frameworks typically rely on two types of bases: the event time of the data itself and the processing time of the processing nodes. The former has millisecond or even microsecond-level precision differences due to the diverse sources, while the latter, even when synchronized through the NTP protocol, will still experience millisecond-level clock drift between distributed nodes. This fragmentation of time directly leads to disordered data calculation timing across nodes. For example, data generated earlier may be processed later, which seriously affects the accuracy of the timing of control command generation.

[0005] Secondly, the water level triggering mechanism used to deal with out-of-order data requires manually setting delay parameters, which inevitably causes the calculation results to lag behind the actual physical time, failing to meet the stringent requirements of semiconductor equipment for hard real-time control, such as scenarios where the exposure triggering error needs to be controlled within 1ms during wafer scanning.

[0006] More importantly, the generation time of control instructions in the general framework lacks a strong correlation with the physical execution time of the device. For example, an instruction may be generated at 8:00:00:000 local time on the node, but the actual execution time may deviate by 1-2ms due to the clock asynchrony between the device controller and the processing node. Such a small deviation is enough to cause serious process defects such as wafer exposure position shift and uneven thin film deposition thickness.

[0007] It is evident that existing technologies still need improvement and enhancement. Summary of the Invention

[0008] In order to overcome the shortcomings of the prior art, the present invention aims to provide a data processing method for semiconductor devices, which solves the problems of inconsistent time base, insufficient hard real-time performance, and high latency in batch processing in the prior art.

[0009] The first aspect of this invention provides a data processing method for a semiconductor device, comprising: when receiving raw sensor data, invoking a logical global clock service to configure a current input timestamp for the raw sensor data; when any operator is executed, querying the logical global clock service to obtain the current execution time, completing the operator execution process based on the sensor data configured with the current input timestamp and the obtained current execution time, and obtaining the intermediate result output after the operator execution; when receiving a task to be processed, obtaining the task attribute corresponding to the task to be processed, confirming the data processing path based on the task attribute and the intermediate result; processing the task to be processed based on the confirmed data processing path, and obtaining the processing result; querying the logical global clock service to obtain the current control time, and generating a control instruction corresponding to the task to be processed based on the processing result, combined with the current input timestamp configured for the raw sensor data and the current control time.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the step of calling the logical global clock service to configure the current input timestamp for the original sensor data when the original sensor data is received includes: when the original sensor data is received, the original sensor data includes multiple sensor information; calling the logical global clock service to obtain the current input timestamp at the microsecond level; and writing the obtained current input timestamp into the data header of each sensor information respectively, so as to use the current input timestamp as the unique time attribute of each sensor information.

[0011] Optionally, in a second implementation of the first aspect of the present invention, the step of querying the logical global clock service to obtain the current execution time when any operator is executed, completing the operator execution process based on sensor data configured with the current input timestamp and the obtained current execution time, and obtaining the intermediate result output after the operator execution includes: when any operator is executed, obtaining the preset window boundary corresponding to the operator; confirming the actual window boundary according to the current input timestamp and the preset window boundary; querying the logical global clock service to obtain the current execution time; if the current execution time is ≥ the actual window boundary, then the operator execution process is completed, and the intermediate result output after the operator execution is obtained.

[0012] Optionally, in a third implementation of the first aspect of the present invention, the step of obtaining the task attribute corresponding to the task to be processed when a task to be processed is received, and confirming the data processing path based on the task attribute and intermediate results, includes: obtaining the task attribute corresponding to the task to be processed when a task to be processed is received, wherein the task attribute includes simple logic tasks and complex computation tasks; when the task attribute is a simple logic task, a real-time streaming path is used to process the task to be processed; when the task attribute is a complex computation task, a batch path is used to process the task to be processed in combination with intermediate results.

[0013] Optionally, in the fourth implementation of the first aspect of the present invention, the step of processing the task to be processed based on the confirmed data processing path and obtaining the processing result includes: obtaining the real-time operating parameters of the device, and combining the real-time operating parameters of the device, the raw sensor data configured with the current input timestamp, and the intermediate results to generate context state information for realizing the interaction between the real-time streaming path and the batch path; when the data processing path is a real-time streaming path, performing streaming processing on the task to be processed, obtaining streaming processing information corresponding to the task to be processed, and writing the streaming processing information into the context state information to obtain the streaming processing result; when the data processing path is a batch path, querying the logical global clock service to obtain the current processing time, and updating the context state information based on the current processing time; processing the task to be processed based on the updated context state information to obtain batch processing information corresponding to the task to be processed; writing the batch processing information into the updated context state information to obtain the batch processing result, and returning the batch processing result to the real-time streaming path.

[0014] Optionally, in a fifth implementation of the first aspect of the present invention, the step of querying the global clock service of the logic to obtain the current control time, and generating a control instruction corresponding to the task to be processed based on the processing result, combined with the current input timestamp configured by the original sensor data and the current control time, includes: querying the global clock service of the logic to obtain the current control time, and updating the processing result based on the current control time; calculating the motion delay time based on the updated processing result, and confirming the control target time based on the current control time and the motion delay time; and generating a control instruction corresponding to the task to be processed based on the processing result and the confirmed control target time.

[0015] Optionally, in a sixth implementation of the first aspect of the present invention, the generation of control instructions corresponding to the task to be processed further includes: acquiring a real-time status signal of the device; when the real-time status signal indicates that the device is in a processing state, adjusting the resource allocation between the real-time flow path and the batch path according to a preset processing resource ratio to ensure that the resource ratio of the real-time flow path is higher than that of the batch path; when the real-time status signal indicates that the device is in a standby state, adjusting the resource allocation between the real-time flow path and the batch path according to a preset standby resource ratio to ensure that the resource ratio of the batch path is higher than that of the real-time flow path.

[0016] A second aspect of the present invention provides a data processing apparatus for a semiconductor device, comprising: a configuration module, configured to, upon receiving raw sensor data, invoke a logical global clock service to configure a current input timestamp for the raw sensor data; an execution module, configured to, when any operator is executed, query the logical global clock service to obtain the current execution time, complete the operator execution process based on the sensor data configured with the current input timestamp and the obtained current execution time, and obtain intermediate results output after operator execution; a confirmation module, configured to, upon receiving a task to be processed, obtain the task attributes corresponding to the task to be processed, and confirm the data processing path based on the task attributes and intermediate results; a processing module, configured to process the task to be processed based on the confirmed data processing path and obtain the processing result; and a control module, configured to, query the logical global clock service to obtain the current control time, and generate a control instruction corresponding to the task to be processed based on the processing result, combined with the current input timestamp configured on the raw sensor data and the current control time.

[0017] A third aspect of the present invention provides a data processing apparatus for a semiconductor device, the data processing apparatus for a semiconductor device comprising: a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to cause the data processing apparatus for a semiconductor device to perform the steps of the data processing method for a semiconductor device described in any of the preceding claims.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the data processing method for a semiconductor device described in any of the preceding claims.

[0019] The core of the technical solution of this invention lies in achieving end-to-end time synchronization through a logical global clock, and combining data processing path selection and precise control command generation to construct a closed-loop processing architecture with unified time, collaborative computing, and precise control. By unifying the time base through a logical global clock, the timing disorder caused by the fragmentation of Event Time and Processing Time in the traditional framework is effectively solved. Furthermore, by selecting data processing paths based on task attributes, the collaboration between low-latency processing of simple tasks and high-precision analysis of complex tasks is achieved, overcoming the defects of fragmented streaming and batch computing resources. The precise control command generation mechanism ensures a strong correlation between control actions and physical timing, meeting the microsecond-level control precision requirements of semiconductor equipment, and is suitable for semiconductor equipment such as lithography machines and etching machines that have stringent requirements for timing accuracy and real-time performance. Attached Figure Description

[0020] Figure 1A logic flowchart of a data processing method for a semiconductor device provided in an embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of the structure of a data processing apparatus for a semiconductor device provided in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of the structure of a data processing device for semiconductor devices provided in an embodiment of the present invention. Detailed Implementation

[0023] This invention provides a data processing method, apparatus, device, and storage medium for semiconductor devices. In this invention, the terms "first," "second," "third," "fourth," etc. (if present)," in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0024] This application discloses a data processing method for semiconductor devices. For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 In the context of wafer exposure control in lithography machines, one embodiment of the data processing method for semiconductor equipment in this invention includes:

[0025] 101. When raw sensor data is received, the logical global clock service is invoked to configure the current input timestamp for the raw sensor data;

[0026] In this embodiment, when the sensors of the lithography machine, such as wafer image sensors, cavity temperature sensors, and robotic arm position sensors, generate raw sensor data, the system calls the Logical Global Clock (LGC) service to configure a unique microsecond-level current input timestamp (Timestamp_In) for each data, replacing the device's local time or the timestamp inherent in the raw sensor data, to ensure a unified time base throughout the entire chain.

[0027] 102. When any operator is executed, query the logical global clock service to obtain the current execution time. Based on the sensor data configured with the current input timestamp and the obtained current execution time, complete the operator execution process and obtain the intermediate results output after the operator execution.

[0028] In this embodiment, each computational operator, such as the data cleaning operator, feature extraction operator, or rule judgment operator, needs to obtain the current execution time (Current_Exec) from LGC during execution. Based on the current input timestamp and the preset window logic, when Current_Exec meets the triggering condition, the calculation is immediately executed and intermediate results are output. The intermediate results can be "coordinates of abnormal image brightness areas" or "temperature fluctuation values", etc.

[0029] 103. When a task to be processed is received, obtain the task attributes corresponding to the task to be processed, and determine the data processing path based on the task attributes and intermediate results;

[0030] In this embodiment, after receiving a task to be processed, such as "wafer defect detection and exposure control", the system first parses the task to be processed to distinguish whether it is a simple logic task or a complex calculation task. Then, it determines the path based on the intermediate results: for simple logic tasks, a real-time streaming path is selected for processing; while for complex calculation tasks, a batch path is triggered for processing.

[0031] 104. Process the task to be processed based on the confirmed data processing path and obtain the processing results;

[0032] In this embodiment, the real-time flow path efficiently processes simple tasks, such as the simple logical task of temperature exceeding the threshold alarm, while the batch path calls the pre-loaded model to deal with complex calculation tasks, such as the complex calculation task of defect type identification. The real-time flow path and the batch path interact through context state information. The batch processing result is accompanied by the original timestamp, that is, the current input timestamp of the original sensor data, and flows back to the real-time flow path.

[0033] 105. Query the logical global clock service to obtain the current control time. Based on the processing results, combined with the current input timestamp and current control time configured by the original sensor data, generate control instructions corresponding to the task to be processed.

[0034] In this embodiment, based on the processing results, combined with the current input timestamp and the current control time (Current_Control) obtained from the LGC, the target time for instruction execution (T_target = Current_Control + motion delay) is calculated; then, a control instruction containing T_target is generated; for example, the control instruction is: adjust the exposure power by 3% at time T_target to ensure that the timing of the control instruction and the physical action of the device are strictly matched.

[0035] This application discloses a data processing method for semiconductor devices. Its core lies in achieving end-to-end time synchronization through a logical global clock, and combining data processing path selection with precise control instruction generation to construct a closed-loop processing architecture characterized by unified time, collaborative computation, and precise control. By unifying the time base through a logical global clock, the method effectively solves the timing disorder problem caused by the fragmentation of Event Time and Processing Time in traditional frameworks. Furthermore, by selecting data processing paths based on task attributes, it achieves collaboration between low-latency processing of simple tasks and high-precision analysis of complex tasks, overcoming the shortcomings of fragmented streaming and batch computing resources. The precise control instruction generation mechanism ensures a strong correlation between control actions and physical timing, meeting the microsecond-level control precision requirements of semiconductor devices. This method is suitable for semiconductor devices such as lithography machines and etching machines, which have stringent requirements for timing accuracy and real-time performance.

[0036] Furthermore, in this embodiment of the invention, the step of calling the logical global clock service to configure the current input timestamp for the original sensor data when the raw sensor data is received includes:

[0037] 201. When raw sensor data is received, the raw sensor data includes multiple sensor information entries;

[0038] In this embodiment, the raw sensor data of the lithography machine contains multiple parallel pieces of information, specifically including: the image sensor outputs a wafer surface image every 30 microseconds, with a resolution of 1024×1024 pixels; the temperature sensor collects the cavity temperature every 100 microseconds, with a measurement range of 20-50℃ and an accuracy of ±0.1℃; and the position sensor provides real-time feedback on the XYZ coordinates of the robotic arm with an accuracy of ±0.1 micrometers.

[0039] 202. Call the logical global clock service to obtain the current input timestamp in microseconds;

[0040] In this embodiment, the global logical clock service is invoked to obtain the current input timestamp (Timestamp_In) at the microsecond level. The format is a 16-bit integer, such as 1620000000000000μs, which corresponds to the physical time 8:00:00.000000.

[0041] 203. Write the acquired current input timestamp into the data header of each sensor information entry, so that the current input timestamp is used as the unique time attribute of each sensor information entry;

[0042] In this embodiment, Timestamp_In is written into the header of each sensor information as metadata. By binding a unified LGC timestamp to all sensor data, the problem of subsequent calculation timing disorder caused by the difference in time accuracy of different sensors is eliminated. For example, the original timestamps of the image sensor and the temperature sensor may differ by 50μs, but after LGC processing, their timestamps will be unified to the same microsecond-level time point, thereby ensuring the consistency and accuracy of data processing. By binding the timestamp to the data permanently, it is ensured that the results can be aligned based on the same time reference during batch processing, laying the foundation for cross-path data fusion.

[0043] Further, in this embodiment of the invention, when any operator is executed, querying the logical global clock service to obtain the current execution time, and based on the sensor data configured with the current input timestamp and the obtained current execution time, completing the operator execution process, and obtaining the intermediate results output after the operator execution, includes:

[0044] 301. When any operator is executed, obtain the preset window boundary corresponding to the operator;

[0045] In this embodiment, window parameters are set for different operators. For example, the preset window for the temperature aggregation operator is a "sliding window with a window size of 10ms and a sliding step of 5ms", and the preset window boundary is "[t_start, t_end) = [Timestamp_In, Timestamp_In + 10ms)"; the preset window for the image feature extraction operator is a "rolling window with a window size of 30μs", which matches the sampling period of the image sensor, and the preset window boundary is "[t_start, t_end) = [Timestamp_In, Timestamp_In + 30μs)".

[0046] 302. Based on the current input timestamp and the preset window boundaries, confirm the actual window boundaries;

[0047] In this embodiment, based on the current input timestamp (Timestamp_In) of the original data, the preset window boundary is converted into the actual logical time boundary, i.e., the actual window boundary. For example, if a certain image data Timestamp_In = 1620000000000000μs, then the actual window boundary is [16200000000000000, 1620000000030000)μs.

[0048] 303. Query the logical global clock service to obtain the current execution time. If the current execution time is greater than or equal to the actual window boundary, the operator execution process is completed, and the intermediate results output after the operator execution are obtained.

[0049] In this embodiment, when the operator is executed, the current execution time (Current_Exec) is obtained from the LGC. When Current_Exec reaches or exceeds the actual window boundary t_end, that is, when Current_Exec ≥ t_end, the calculation is triggered immediately. For example, if the actual window t_end of the temperature aggregation operator is 1620000000010000μs, then when Current_Exec equals 1620000000010000μs, the maximum, minimum and average temperature values ​​within the window are calculated immediately. This mechanism eliminates the artificial delay mechanism based on water level lines in the traditional framework. The calculation is directly triggered by the current execution time fed back by the LGC, ensuring that the window calculation is executed accurately on the logical timeline. The end-to-end delay can be controlled within 100μs, which can meet the hard real-time requirements of semiconductor devices.

[0050] Furthermore, in this embodiment of the invention, the step of obtaining the task attributes corresponding to the task when a task to be processed is received, and confirming the data processing path based on the task attributes and intermediate results, includes:

[0051] 401. When a task to be processed is received, the task attributes corresponding to the task to be processed are obtained, and the task attributes include simple logic tasks and complex calculation tasks.

[0052] 402. When the task attribute is a simple logical task, the task to be processed is processed using a real-time flow path;

[0053] In this embodiment, simple logic tasks must meet the following conditions: the calculation logic should be concise and clear, strictly follow fixed rules or preset thresholds, the latency requirement should be less than 100 microseconds, and the resource consumption should be as low as possible; for example, when the cavity temperature exceeds 25°C and lasts for 2 window periods, a cooling alarm is triggered; or when the position deviation of the robotic arm is less than 5 micrometers, the current motion parameters are maintained.

[0054] 403. When the task is a complex computational task, batch processing should be used to process the task based on intermediate results.

[0055] In this embodiment, complex computational tasks need to meet the following conditions: complex computational logic, such as model inference or multi-dimensional analysis; latency tolerance of 5-10 milliseconds; and high resource consumption; for example, identifying defect types in wafer images, or predicting exposure focus deviation based on multi-sensor data, including temperature, pressure, and image data.

[0056] In this embodiment, by refining the task attributes and intermediate result judgment, the on-demand allocation of computing resources is realized, which effectively avoids the problem of simple tasks occupying batch processing resources or complex tasks blocking the real-time stream, while taking into account both the requirements of hard real-time performance and complex analysis.

[0057] Furthermore, in this embodiment of the invention, the step of processing the task to be processed based on the confirmed data processing path and obtaining the processing result includes:

[0058] 501. Obtain the real-time operating parameters of the device, and combine the real-time operating parameters of the device, the raw sensor data configured with the current input timestamp, and the intermediate results to generate context state information for realizing the interaction between the real-time flow path and the batch path.

[0059] In this embodiment, the operating parameters of the semiconductor device are collected in real time, such as the robotic arm speed v=50mm / s and the cavity pressure P=101325Pa. Combined with the current input timestamp (Timestamp_In) of the original sensor data and intermediate results (such as the coordinates of abnormal areas), a comprehensive contextual state information is constructed.

[0060] 502. When the data processing path is a real-time streaming path, the task to be processed is stream-processed, the streaming processing information corresponding to the task to be processed is obtained, and the streaming processing information is written into the context state information to obtain the streaming processing result.

[0061] In this embodiment, for the streaming task to be processed, such as rapid marking of abnormal regions, the real-time streaming path performs simple logical calculations based on the context, first calculating the area of ​​the abnormal region as:

[0062] (150-100)×(250-200)=2500 ;

[0063] Because of 1 =1000000 0.0025 =2500 And 0.0025 <0.1 Therefore, it is marked as "further detection required", and then the streaming processing information is written into the stream_result field in the context state information, and finally the streaming processing result is output.

[0064] 503. When the data processing path is a batch path, query the logical global clock service to obtain the current processing time, and update the context state information based on the current processing time;

[0065] 504. Process the task to be processed based on the updated context state information to obtain batch processing information corresponding to the task to be processed.

[0066] 505. Write the batch processing information into the updated context state information, obtain the batch processing result, and return the batch processing result to the real-time stream path;

[0067] In this embodiment, the batch processing path obtains context state information from the real-time streaming path and obtains the current processing time (Current_Process) from the LGC to update the current_process_time field in the context state information, for example, updating it to 1620000000005000μs. When the task to be processed is the identification of defect types in an image, the CNN model is called to infer the abnormal region image, identify the defect type as "micro-dust" with a confidence level of 98%, and write this batch processing information into the batch_result field in the context state information to obtain the batch processing result. The batch processing result carries the current input timestamp of the original sensor data, Timestamp_In=16200000000000000μs, and flows back to the real-time streaming path to ensure alignment with the original data time sequence.

[0068] In this embodiment, context state information, through an efficient data processing mechanism, effectively achieves real-time data interaction between stream and batch paths. This real-time interaction mechanism not only improves the speed of data processing but also ensures the immediacy and accuracy of the data. During the batch processing, the batch processing results carry the current input timestamp of the original sensor data for backflow operation, effectively resolving the time sequence misalignment problem of stream and batch results that is common in traditional data processing frameworks. Through precise timestamp alignment, it is ensured that every detail of the complex calculation results can be accurately associated with its corresponding original data scene, thereby avoiding calculation errors caused by time sequence inconsistencies. This association mechanism provides solid and reliable data support for subsequent instruction generation, ensuring the accuracy and execution efficiency of control instructions, and further enhancing the stability and reliability of the overall system.

[0069] Furthermore, in this embodiment of the invention, the query logic global clock service obtains the current control time, and based on the processing result, combines the current input timestamp configured by the original sensor data and the current control time to generate a control command corresponding to the task to be processed, including:

[0070] 601. Query the logical global clock service to obtain the current control time, and update the processing result based on the current control time;

[0071] For example, the current control time is obtained from LGC, such as Current_Control=1620000000010000μs. The batch processing results, such as "defect = dust, coordinates (125, 225)", are integrated with the streaming processing results, such as "exposure needs adjustment", and the processing results are updated.

[0072] 602. Calculate the motion delay time based on the updated processing results, and confirm the control target time based on the current control time and the motion delay time;

[0073] For example, motion delay (Δ) can be the physical time it takes for the robotic arm to move from its current position to the defect coordinates. The formula is Δ = physical distance / robotic arm speed + mechanical response delay + temperature compensation value. Assuming the robotic arm's current position is (0, 0) and the defect coordinates are (125μm, 225μm), the physical distance is the straight-line distance between the two points, calculated to be 258.2μm. Since the robotic arm speed v = 50mm / s, the movement time is 0.257mm / 50mm / s = 0.00514s = 5.14ms. The mechanical response delay is 0.5ms. Because the actual ambient temperature is 24 degrees Celsius, slightly higher than the standard temperature, the temperature compensation value is 0.3ms. Therefore, the total motion delay Δ = 5.14 + 0.5 + 0.3 = 5.94ms, approximately 6ms.

[0074] 603. Generate control instructions corresponding to the task to be processed based on the processing results and the confirmed control target time;

[0075] For example, the control target time T_target is set to Current_Control plus Δ, i.e., T_target = 1620000000010000 + 6000 = 1620000000016000μs; based on the calculated control target time, a control command is generated: "At time T_target = 1620000000016000μs, adjust the exposure power at defect coordinates (125, 225) to +3%." This control command carries the T_target information and is sent to the controller of the semiconductor device.

[0076] In this embodiment, by accurately calculating the delay time during the motion process and the control target time, a close correlation is established between the generation time of the control command and the physical execution timing of the device. This effectively solves the process deviation problem caused by the disconnect between command generation and actual execution in the traditional control framework. Based on the optimized control mechanism, the exposure control error can be strictly controlled within 100 microseconds, thereby significantly improving the yield in the wafer processing process and providing a strong guarantee for efficient production in the semiconductor manufacturing industry.

[0077] Furthermore, in this embodiment of the invention, the generation of control instructions corresponding to the task to be processed further includes:

[0078] 701. Obtain the real-time status signals of the equipment;

[0079] In this embodiment, the status of the semiconductor equipment is monitored in real time by sensors and a real-time status signal is output. For example, in the processing state, when the lithography machine is in the wafer scanning exposure stage (lasting about 20 seconds), real-time control should be prioritized. In the standby state, such as the wafer replacement or cavity cleaning stage (lasting about 30 seconds), there are relatively few real-time tasks, and the resource ratio of the real-time flow path can be appropriately reduced.

[0080] 702. When the real-time status signal indicates that the equipment is in the processing state, adjust the resource allocation between the real-time flow path and the batch path according to the preset processing resource ratio to ensure that the resource ratio of the real-time flow path is higher than that of the batch path.

[0081] In this embodiment, when the real-time status signal display device is in processing mode, the CPU and memory resources are adjusted accordingly based on the preset processing resource allocation ratio, i.e., the streaming path accounts for 80% and the batch path accounts for 20%. Specifically, when the real-time status signal display device is in processing mode, the real-time streaming path is allocated 8 CPU cores and 2GB of memory to ensure that there is no delay in operator calculation and instruction generation; while the batch path only retains 1 CPU core and 1GB of memory to maintain the loading of the CNN model, but does not start new computing tasks.

[0082] 703. When the real-time status signal indicates that the device is in standby mode, adjust the resource allocation between the real-time flow path and the batch path according to the preset standby resource ratio to ensure that the resource ratio of the batch path is higher than that of the real-time flow path.

[0083] In this embodiment, when the real-time status signal display device is in standby mode, resource adjustments are made according to a preset standby resource allocation ratio, with the flow path accounting for 30% and the batch path accounting for 70%. Specifically, when the real-time status signal display device is in standby mode, the real-time flow path reserves 3 CPU cores and 1GB of memory to handle a small number of status monitoring tasks; the batch path is allocated 7 CPU cores and 5GB of memory to start offline tasks, such as historical defect data clustering tasks and model parameter optimization tasks.

[0084] In this embodiment, during processing, priority is given to ensuring the supply of real-time resources to guarantee the immediate response of critical tasks and avoid interference with the execution of hard real-time tasks due to complex calculations, thereby ensuring the stability and reliability of the entire system. In standby mode, idle resources are released and allocated to batch processing tasks. This not only effectively improves the efficiency of offline analysis but also maximizes the utilization of existing resources. Compared with traditional fixed resource allocation schemes, the dynamic adjustment strategy significantly improves resource utilization and also reduces hardware costs to a certain extent, saving enterprise operating expenses.

[0085] The data processing method for a semiconductor device in the embodiments of the present invention has been described above. The data processing apparatus for a semiconductor device in the embodiments of the present invention will be described below. Please refer to [link / reference]. Figure 2 One embodiment of the data processing apparatus for a semiconductor device in this invention includes:

[0086] The configuration module 801 is used to call the logical global clock service to configure the current input timestamp for the raw sensor data when the raw sensor data is received;

[0087] The execution module 802 is used to query the logical global clock service to obtain the current execution time when any operator is executed, and complete the operator execution process based on the sensor data configured with the current input timestamp and the obtained current execution time, and obtain the intermediate results output after the operator execution.

[0088] The confirmation module 803 is used to obtain the task attributes corresponding to the task when a task to be processed is received, and to confirm the data processing path based on the task attributes and intermediate results.

[0089] The processing module 804 is used to process the task to be processed based on the confirmed data processing path and obtain the processing results;

[0090] The control module 805 is used to query the logical global clock service to obtain the current control time. Based on the processing results, combined with the current input timestamp configured by the original sensor data and the current control time, it generates control instructions corresponding to the task to be processed.

[0091] Based on the same ideas as the methods in the above embodiments, the apparatus provided in this application can implement the methods in the above embodiments.

[0092] above Figure 2 The data processing apparatus for semiconductor devices in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The data processing apparatus for semiconductor devices in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0093] Figure 3This is a schematic diagram of a data processing device 900 for a semiconductor device according to an embodiment of the present invention. The data processing device 900 for a semiconductor device can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 910 and memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) for storing application programs 933 or data 932. The memory 920 and storage media 930 may be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the data processing device 900 for the semiconductor device. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the data processing device 900 for the semiconductor device to implement the steps of the data processing method for a semiconductor device provided in the above-described method embodiments.

[0094] The data processing device 900 for semiconductor devices may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The data processing device structure shown for a semiconductor device does not constitute a limitation on the data processing device for a semiconductor device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0095] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform steps of a data processing method for a semiconductor device.

[0096] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0097] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data processing method for semiconductor devices, characterized in that, include: When raw sensor data is received, the logical global clock service is invoked to configure the current input timestamp for the raw sensor data. The current input timestamp is in the microsecond range. When any operator is executed, the logical global clock service is queried to obtain the current execution time. Based on the sensor data configured with the current input timestamp and the obtained current execution time, the operator execution process is completed, and the intermediate results output after the operator execution are obtained. When a task to be processed is received, the task attributes corresponding to the task are obtained, and the data processing path is determined based on the task attributes and intermediate results. Specifically, when a task to be processed is received, the task attributes corresponding to the task are obtained, and the task attributes include simple logical tasks and complex computational tasks. When the task attribute is a simple logical task, a real-time streaming path is used to process the task. When the task attribute is a complex computational task, a batch processing path is used to process the task, taking into account the intermediate results. Based on the confirmed data processing path, the task to be processed is processed, and the processing result is obtained. Specifically, the real-time operating parameters of the device are obtained, and combined with the real-time operating parameters of the device, the raw sensor data configured with the current input timestamp, and the intermediate results, context state information for realizing the interaction between the real-time streaming path and the batch path is generated. When the data processing path is a real-time streaming path, the task to be processed is stream-processed, the streaming processing information corresponding to the task to be processed is obtained, and the streaming processing information is written into the context state information to obtain the streaming processing result. When the data processing path is a batch path, the logical global clock service is queried to obtain the current processing time, and the context state information is updated based on the current processing time. The task to be processed is processed based on the updated context state information to obtain the batch processing information corresponding to the task to be processed. The batch processing information is written into the updated context state information to obtain the batch processing result, and the batch processing result is returned to the real-time streaming path. The system queries the global clock service to obtain the current control time. Based on the processing results, and combined with the current input timestamp and current control time configured from the original sensor data, it generates control instructions corresponding to the task to be processed.

2. The data processing method for semiconductor devices according to claim 1, characterized in that, When raw sensor data is received, the process of calling the logical global clock service to configure the current input timestamp for the raw sensor data includes: When raw sensor data is received, the raw sensor data includes multiple sensor information entries; Call the logical global clock service to obtain the current input timestamp in microseconds; The current input timestamp is written into the data header of each sensor message to make the current input timestamp the unique time attribute of each sensor message.

3. The data processing method for semiconductor devices according to claim 1, characterized in that, When any operator is executed, the logical global clock service is queried to obtain the current execution time. Based on the sensor data configured with the current input timestamp and the obtained current execution time, the operator execution process is completed, and the intermediate results output after the operator execution are obtained, including: When any operator is executed, obtain the preset window boundary corresponding to the operator; Confirm the actual window boundaries based on the current input timestamp and the preset window boundaries; Query the global clock service to obtain the current execution time. If the current execution time is greater than or equal to the actual window boundary, the operator execution process is completed, and the intermediate results output after the operator execution are obtained.

4. The data processing method for semiconductor devices according to claim 1, characterized in that, The query logic uses a global clock service to obtain the current control time. Based on the processing result, and combined with the current input timestamp configured from the original sensor data and the current control time, it generates control instructions corresponding to the task to be processed, including: Query the logical global clock service to obtain the current control time, and update the processing results based on the current control time; The motion delay time is calculated based on the updated processing results, and the control target time is confirmed based on the current control time and the motion delay time. Based on the processing results and the confirmed control target time, control instructions corresponding to the task to be processed are generated.

5. The data processing method for semiconductor devices according to claim 1, characterized in that, The generation of control instructions corresponding to the task to be processed further includes: Acquire real-time status signals from the device; When the real-time status signal indicates that the equipment is in the processing state, the resource allocation between the real-time flow path and the batch path is adjusted according to the preset processing resource ratio to ensure that the resource ratio of the real-time flow path is higher than that of the batch path. When the real-time status signal indicates that the device is in standby mode, the resource allocation between the real-time flow path and the batch path is adjusted according to the preset standby resource ratio to ensure that the resource ratio of the batch path is higher than that of the real-time flow path.

6. A data processing apparatus for a semiconductor device, characterized in that, include: The configuration module is used to call the logical global clock service when raw sensor data is received to configure the current input timestamp for the raw sensor data, wherein the current input timestamp is in the microsecond range; The execution module is used to query the logical global clock service to obtain the current execution time when any operator is executed, and complete the operator execution process based on the sensor data configured with the current input timestamp and the obtained current execution time, and obtain the intermediate results output after the operator execution. The confirmation module is used to obtain the task attributes corresponding to the task when a task to be processed is received, and to confirm the data processing path based on the task attributes and intermediate results. Specifically, when a task to be processed is received, the task attributes corresponding to the task to be processed are obtained. The task attributes include simple logical tasks and complex computational tasks. When the task attribute is a simple logical task, a real-time streaming path is used to process the task. When the task attribute is a complex computational task, a batch processing path is used to process the task, taking into account the intermediate results. The processing module is used to process the task to be processed based on the confirmed data processing path and obtain the processing result. Specifically, it obtains the real-time operating parameters of the device, combines the real-time operating parameters of the device, the raw sensor data configured with the current input timestamp, and intermediate results to generate context state information for realizing the interaction between the real-time streaming path and the batch path. When the data processing path is a real-time streaming path, it performs streaming processing on the task to be processed, obtains the streaming processing information corresponding to the task to be processed, and writes the streaming processing information into the context state information to obtain the streaming processing result. When the data processing path is a batch path, it queries the logical global clock service to obtain the current processing time, updates the context state information based on the current processing time, and processes the task to be processed based on the updated context state information to obtain the batch processing information corresponding to the task to be processed. Write the batch processing information into the updated context state information, obtain the batch processing result, and return the batch processing result to the real-time stream path; The control module queries the global clock service to obtain the current control time. Based on the processing results, it combines the current input timestamp configured by the original sensor data and the current control time to generate control instructions corresponding to the task to be processed.

7. A data processing device for semiconductor equipment, characterized in that, The data processing device for semiconductor devices includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the data processing apparatus for a semiconductor device to perform the steps of the data processing method for a semiconductor device as claimed in any one of claims 1-5.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the data processing method for a semiconductor device as described in any one of claims 1-5.

Images