IMP-based intelligent beam current adjustment method, device and equipment and storage medium

Abstract

The present application relates to the technical field of information processing, and more particularly to an intelligent beam current adjustment method and device based on IMP, equipment and storage medium, the operation interface image of each IMP machine is obtained in real time through the KVM switch, the current beam current parameter and real-time state identifier in the operation interface image are recognized and extracted by using the OCR visual algorithm, the process requirement data of the current wafer batch number is obtained, the parameter and state data and the process requirement data are integrated, the integrated data is cleaned and normalized to obtain a preprocessed data set, the AI decision model is used to analyze the difference information between the current machine beam current state and the target process requirement based on the preprocessed data set, and the optimal beam current adjustment parameter is calculated according to the difference information, so as to improve the adjustment accuracy, adapt to complex process requirements, generate a control instruction according to the optimal beam current adjustment parameter, and IMP machine adjusts the related hardware module parameters according to the control instruction, thereby improving the intelligent level.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, and in particular to an intelligent beam current adjustment method and device based on IMP, equipment and a storage medium. BACKGROUND

[0002] In a semiconductor ion implantation process, beam current adjustment of an IMP machine (ion implantation machine) is a key link, directly affecting the accuracy of ion implantation parameters such as dose and energy, and relating to chip manufacturing yield. The traditional beam current adjustment process relies on manual operation or simple program control, and has problems such as low efficiency, large influence of precision by human factors, and poor adaptability to complex working conditions. With the continuous advancement of semiconductor technology to smaller nodes, the requirements for intelligentization, automation and high precision of beam current adjustment are increasingly urgent, and there is an urgent need for a beam current adjustment scheme that integrates multiple systems and uses AI algorithms for optimization. SUMMARY

[0003] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide an intelligent beam current adjustment method, device, equipment and storage medium based on IMP, which improves the level of intelligentization and automation, reduces human error, improves adjustment accuracy, adapts to complex process requirements, improves machine operation efficiency, and shortens process waiting time.

[0004] The first aspect of the present application provides an intelligent beam current adjustment method based on IMP, comprising: acquiring the operation interface image of each IMP machine in real time through a KVM switch, recognizing and extracting the current beam current parameter and real-time state identifier in the operation interface image by using an OCR visual algorithm; converting the current beam current parameter and the real-time state identifier into structured parameter and state data, obtaining process requirement data of the current wafer batch number from an EAP third-party system; integrating the parameter and state data and the process requirement data to obtain integrated data, performing data cleaning and normalization processing on the integrated data to obtain a preprocessed data set; inputting the preprocessed data set into a preset AI decision model, using the AI decision model to analyze the difference information between the current machine beam current state and the target process requirement based on the preprocessed data set, and calculating the optimal beam current adjustment parameter according to the difference information, performing compliance verification on the optimal beam current adjustment parameter, the compliance verification including safety threshold verification and process specification verification; when the verification is passed, generating a control instruction according to the optimal beam current adjustment parameter, and sending the control instruction to the IMP machine through the KVM switch, so that the IMP machine adjusts the related hardware module parameters according to the control instruction.

[0005] Optionally, in the first implementation manner of the first aspect, the acquiring, by the KVM switch, the operation interface image of each IMP machine in real time, recognizing and extracting, by an OCR visual algorithm, the current beam parameter and the real-time state identifier in the operation interface image comprises: automatically switching a video channel by the KVM switch according to a preset timing sequence to sequentially acquire real-time pictures of the operation interface of each IMP machine; locating a beam parameter display area and a state lamp area in the real-time picture of the operation interface based on a predefined machine interface template; pre-processing the beam parameter display area and the state lamp area, the pre-processing comprising grayscale processing, binaryzation processing and denoising filter processing; recognizing a number and a unit in the beam parameter display area using a pre-trained CRNN convolutional recurrent neural network model, and outputting the current beam parameter according to the number and the unit; and detecting a color and an icon type of the state lamp area using a pre-trained YOLOv5 target detection model, and outputting the real-time state identifier according to the color and the icon type.

[0006] Optionally, in the second implementation manner of the first aspect, the converting the current beam parameter and the real-time state identifier into structured parameter and state data, and acquiring process requirement data of a current wafer batch number from an EAP third-party system comprises: generating JSON data conforming to the SEMI E125 standard according to the current beam parameter and the real-time state identifier to obtain parameter and state data; identifying a wafer batch number currently loaded by the IMP machine; sending a query request to the EAP third-party system according to the wafer batch number; receiving a process recipe file returned by the EAP third-party system, and parsing the process recipe file to obtain process requirement data.

[0007] Optionally, in the third implementation manner of the first aspect, the integrating the parameter and state data and the process requirement data to obtain integrated data, and performing data cleaning and normalization processing on the integrated data to obtain a pre-processed data set comprises: aligning the parameter and state data and the process requirement data by taking an equipment ID of the IMP machine and a nanosecond-level timestamp as a correlation key to obtain integrated data; removing duplicate data in the integrated data, and filling missing values in the integrated data to obtain data-cleaning integrated data; and performing normalization processing on the data-cleaning integrated data to obtain a pre-processed data set.

[0008] Optionally, in a fourth implementation form of the first aspect of the present application, the step of inputting the preprocessed data set into the preset AI decision model, using the AI decision model to analyze the difference information between the current beam current state of the IMP machine and the target process requirement based on the preprocessed data set, and calculating the optimal beam current adjustment parameter according to the difference information, and the step of performing compliance verification on the optimal beam current adjustment parameter, the compliance verification including safety threshold verification and process specification verification, comprise: constructing an AI decision model using a neural network algorithm, collecting historical beam current adjustment success records of the IMP machine as a training set, the historical beam current adjustment success records including pre-adjustment parameters, execution actions, post-adjustment parameters and adjustment effects; training the AI decision model using the training set to enable the AI decision model to learn the mapping relationship between parameters and adjustment effects; converting the preprocessed data set into a feature vector conforming to the model input; inputting the preprocessed data set into the preset AI decision model, using the AI decision model to analyze the difference information between the current beam current state of the IMP machine and the target process requirement based on the preprocessed data set, and calculating the optimal beam current adjustment parameter according to the difference information; and performing compliance verification on the optimal beam current adjustment parameter, the compliance verification including safety threshold verification and process specification verification.

[0009] Optionally, in a fifth implementation form of the first aspect of the present application, the step of generating a control instruction according to the optimal beam current adjustment parameter when the verification passes, and sending the control instruction to the IMP machine through the KVM switch to enable the IMP machine to adjust the relevant hardware module parameters according to the control instruction, comprises: when the verification passes, generating JSON data conforming to the SEMI E84 standard according to the optimal beam current adjustment parameter to obtain a control instruction; sending the control instruction to the KVM switch through a KVM API to enable the KVM switch to convert the control instruction into a USB HID signal simulating a keyboard and mouse operation; and transmitting the USB HID signal to the IMP machine through a physically isolated channel to enable the IMP machine to adjust the relevant hardware module parameters according to the USB HID signal.

[0010] Optionally, in the sixth implementation form of the first aspect of the present application, when the verification passes, the control instruction is generated according to the optimal beam tuning parameter, and the control instruction is sent to the IMP machine by the KVM switch, so that the IMP machine adjusts the relevant hardware module parameters according to the control instruction, and the method further comprises: monitoring and recording parameter changes, execution time and effect data in the tuning process to obtain monitoring record information, and synchronously feeding back the monitoring record information to the EAP third-party system; when a fault alarm exception or a parameter out-of-limit exception occurs in the monitoring record information, a preset fault program is started and an alarm information is generated; the alarm information is sent to a management terminal; the KVM switch is used to continuously collect the latest operation interface image of the IMP machine after adjusting the relevant hardware module parameters, and an OCR visual algorithm is used to identify and extract the latest beam parameters and the latest state identifier in the latest operation interface image; whether the tuning effect meets the standard is determined according to the latest beam parameters and the latest state identifier; if not, the AI decision model is used to recalculate the optimal beam tuning parameter; and if yes, a current beam tuning success record is generated as a new training set, and the AI decision model is iteratively optimized by using the new training set.

[0011] The second aspect of the present application provides an intelligent beam tuning device based on IMP, comprising: an acquisition and identification extraction module, which is used to acquire the operation interface image of each IMP machine in real time through a KVM switch, and identify and extract the current beam parameters and real-time state identifier in the operation interface image by using an OCR visual algorithm; a conversion acquisition module, which is used to convert the current beam parameters and the real-time state identifier into structured parameter and state data, and acquire the process requirement data of the current wafer batch number from an EAP third-party system; an integration, cleaning and normalization module, which is used to integrate the parameter and state data and the process requirement data to obtain integrated data, and perform data cleaning and normalization processing on the integrated data to obtain a preprocessed data set; an analysis and calculation verification module, which is used to input the preprocessed data set into a preset AI decision model, analyze the difference information between the current machine beam state and the target process requirement based on the preprocessed data set by using the AI decision model, calculate the optimal beam tuning parameter according to the difference information, and perform compliance verification on the optimal beam tuning parameter, wherein the compliance verification comprises safety threshold verification and process specification verification; and a generation and sending module, which is used to generate a control instruction according to the optimal beam tuning parameter when the verification passes, and send the control instruction to the IMP machine by the KVM switch, so that the IMP machine adjusts the relevant hardware module parameters according to the control instruction.

[0012] Optionally, in the first implementation manner of the second aspect of the present application, the acquisition and recognition extraction module comprises: an acquisition unit configured to automatically switch video channels by the KVM switch according to a preset timing sequence, and sequentially acquire real-time pictures of operation interfaces of each IMP machine; a positioning unit configured to position a beam parameter display area and a state lamp area in the real-time picture of the operation interface based on a predefined machine interface template; a preprocessing unit configured to preprocess the beam parameter display area and the state lamp area, the preprocessing comprising grayscale processing, binarization processing and denoising filtering processing; a recognition output unit configured to recognize a number and a unit in the beam parameter display area using a pre-trained CRNN convolutional recurrent neural network model, and output a current beam parameter according to the number and the unit; and a detection output unit configured to detect a color and an icon type of the state lamp area using a pre-trained YOLOv5 target detection model, and output a real-time state identifier according to the color and the icon type.

[0013] Optionally, in the second implementation manner of the second aspect of the present application, the conversion and acquisition module comprises: a first generation unit configured to generate JSON data conforming to the SEMI E125 standard according to the current beam parameter and the real-time state identifier, and obtain parameter and state data; an identification unit configured to identify a wafer batch number currently loaded by the IMP machine; a first sending unit configured to send a query request to an EAP third-party system according to the wafer batch number; and a receiving and parsing unit configured to receive a process recipe file returned by the EAP third-party system, parse the process recipe file, and obtain process requirement data.

[0014] Optionally, in the third implementation manner of the second aspect of the present application, the integration, cleaning and normalization module comprises: an alignment unit configured to align the parameter and state data and the process requirement data by taking the equipment ID of the IMP machine and a nanosecond-level timestamp as an association key, and obtain integrated data; a removal and filling unit configured to remove repeated data in the integrated data, fill in missing values in the integrated data, and obtain integrated data after data cleaning; and a normalization unit configured to perform normalization processing on the integrated data after data cleaning, and obtain a preprocessed data set.

[0015] Optionally, in the fourth implementation manner of the second aspect of the present application, the analysis and calculation verification module comprises: a construction collecting unit configured to construct an AI decision model by using a neural network algorithm, collect historical beam current adjustment success records of the IMP machine as a training set, and the historical beam current adjustment success records comprise pre-adjustment parameters, execution actions, post-adjustment parameters and adjustment effects; a training unit configured to train the AI decision model by using the training set, so that the AI decision model learns a mapping relationship between parameters and adjustment effects; a conversion unit configured to convert the pre-processed data set into a feature vector conforming to the model input; an analysis and calculation unit configured to input the pre-processed data set into a preset AI decision model, analyze difference information between a current machine beam current state and target process requirements based on the pre-processed data set by using the AI decision model, and calculate optimal beam current adjustment parameters according to the difference information; and a verification unit configured to perform compliance verification on the optimal beam current adjustment parameters, and the compliance verification comprises safety threshold verification and process specification verification.

[0016] Optionally, in the fifth implementation manner of the second aspect of the present application, the generation and transmission module comprises: a second generation unit configured to generate JSON data conforming to the SEMI E84 standard according to the optimal beam current adjustment parameters to obtain a control instruction when the verification passes; a second transmission unit configured to transmit the control instruction to the KVM switch through a KVM API, so that the KVM switch converts the control instruction into a USB HID signal of an analog keyboard and mouse operation; and a transmission unit configured to transmit the USB HID signal to the IMP machine through a physically isolated channel, so that the IMP machine adjusts related hardware module parameters according to the USB HID signal.

[0017] Optionally, in the sixth implementation form of the second aspect of the present application, further comprising: a monitoring and recording feedback module, configured to monitor and record parameter changes, execution time and effect data in the adjustment process to obtain monitoring and recording information, and feed back the monitoring and recording information to the EAP third-party system; a start generation module, configured to start a preset fault program and generate an alarm information when an abnormal fault alarm or a parameter out-of-limit abnormality occurs in the monitoring and recording information; a sending module, configured to send the alarm information to a management terminal; a collection and recognition extraction module, configured to continuously collect a latest operation interface image of the IMP machine adjustment related hardware module parameter through the KVM switch, and recognize and extract a latest beam parameter and a latest state identifier in the latest operation interface image by using an OCR visual algorithm; a judgment module, configured to judge whether the adjustment effect meets the standard according to the latest beam parameter and the latest state identifier; a calculation module, configured to recalculate the optimal beam adjustment parameter by using the AI decision model if the adjustment effect does not meet the standard; and a generation optimization module, configured to generate a current beam adjustment success record as a new training set if the adjustment effect meets the standard, and iteratively optimize the AI decision model by using the new training set.

[0018] The third aspect of the present application provides an IMP-based intelligent beam adjustment device, comprising a memory and at least one processor, wherein the memory stores instructions; and the at least one processor invokes the instructions in the memory, so that the IMP-based intelligent beam adjustment device performs each step of the IMP-based intelligent beam adjustment method.

[0019] The fourth aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores instructions, and the instructions are executed by a processor to implement each step of the IMP-based intelligent beam adjustment method.

[0020] In the technical solution of the application, the operation interface image of each IMP machine is obtained in real time through the KVM switch, the current beam parameter and real-time state identifier in the operation interface image are recognized and extracted by using an OCR visual algorithm, the current beam parameter and real-time state identifier are converted into structured parameter and state data, the process requirement data of the current wafer batch number is obtained from an EAP third-party system, the parameter and state data and the process requirement data are integrated, the integrated data is subjected to data cleaning and normalization processing to obtain a preprocessed data set, the preprocessed data set is input into a preset AI decision model, the AI decision model is used to analyze the difference information between the current machine beam state and the target process requirement based on the preprocessed data set, and the optimal beam tuning parameter is calculated according to the difference information, so that the tuning accuracy is improved, the complex process requirement is adapted, the optimal beam tuning parameter is subjected to compliance verification, when the verification is passed, the control instruction is generated according to the optimal beam tuning parameter, the control instruction is sent to the IMP machine through the KVM switch, so that the IMP machine adjusts the related hardware module parameter according to the control instruction, the intelligentization and automation level are improved, the human error is reduced, the machine operation efficiency is improved, and the process waiting time is shortened. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A first flowchart of an intelligent beam tuning method based on IMP provided for an embodiment of the application;

[0022] Figure 2 A second flowchart of an intelligent beam tuning method based on IMP provided for an embodiment of the application;

[0023] Figure 3 A third flowchart of an intelligent beam tuning method based on IMP provided for an embodiment of the application;

[0024] Figure 4 A fourth flowchart of an intelligent beam tuning method based on IMP provided for an embodiment of the application;

[0025] Figure 5 A structure schematic diagram of an intelligent beam tuning device based on IMP provided for an embodiment of the application;

[0026] Figure 6 Another structure schematic diagram of an intelligent beam tuning device based on IMP provided for an embodiment of the application;

[0027] Figure 7 A structure schematic diagram of an intelligent beam tuning equipment based on IMP provided for an embodiment of the application. DETAILED DESCRIPTION

[0028] The application provides an IMP-based intelligent beam current adjustment method, device, equipment and storage medium, improves intelligent and automation level, reduces human error, improves adjustment accuracy, adapts to complex process requirements, improves machine operation efficiency, and shortens process waiting time.

[0029] The terms "first", "second", "third", "fourth" and the like in the description, claims, and drawings of the application, and those above and below (if any) are used for distinguishing between similar objects and not necessarily for describing a specific sequential or chronological order. It is to be understood that the use of the terms so termed herein is to be interpreted to allow that combinations of the multiples so called can be used in conjunction with each other. It is also to be understood that the use of the terms so termed herein allowing for the use of the terms in any order with the exception of the order illustrated or described herein. Furthermore, the terms "comprise" or "comprising" and any of their derivatives, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of steps or units uses "comprising" is not necessarily limited to those steps or units which are expressly listed, but can include other not expressly listed steps or units.

[0030] For the sake of understanding, the specific flow of the embodiments of the application is described below. Please refer to Figure 1 One embodiment of the IMP-based intelligent beam current adjustment method in the embodiments of the application includes the following steps.

[0031] 101, real-time acquisition of the operation interface image of each IMP machine through the KVM switch, identification and extraction of the current beam current parameter and real-time state identifier in the operation interface image by using the OCR visual algorithm;

[0032] In this embodiment, the KVM switch automatically switches to connect to different IMP machine video output ports according to a preset polling timing, captures the screen image of the operation interface (such as a touch screen or a display) in real time, uses the deployed OCR (optical character recognition) visual processing module, and combines the pre-defined interface area template to accurately identify and extract the key beam current parameters (such as beam current value and energy value) and state indicator lights and icons (such as running, alarm, standby state) displayed in the image.

[0033] 102, converting the current beam current parameter and real-time state identifier into structured parameter and state data, and obtaining process requirement data of the current wafer batch number from an EAP third-party system;

[0034] In this embodiment, the unstructured text information extracted by OCR (beam parameter values ​​and units) and the visually recognized status identifiers (color, icon type) are organized into structured JSON-formatted data (parameter and status data) in accordance with the SEMI E125 standard. Simultaneously, by reading the wafer batch number currently being processed by the IMP tool (e.g., from the tool log or load signal), the standard interface of the EAP (Equipment Automation Program) system (e.g., SECS / GEM, HSMS) is called to send a query request containing the batch number. The corresponding process recipe file returned by the EAP is retrieved and parsed to extract process requirement data such as the target beam setting value and the allowable tolerance range.

[0035] 103. Integrate the parameter and status data with the process requirement data to obtain integrated data, and perform data cleaning and normalization on the integrated data to obtain a preprocessed data set;

[0036] In this embodiment, the unique IMP device ID and high-precision (nanosecond-level) timestamp are used as association keys to time-align and integrate the real-time collected structural parameter and status data with the process requirement data obtained from the EAP. The integrated data is cleaned to remove duplicate records, and possible missing values ​​(such as EAP response delay) are filled with neighboring values ​​or preset default values. Subsequently, numerical parameters (such as beam current) are normalized (such as Min-Max or Z-Score), and data of different dimensions and ranges are converted to the [0,1] or standard normal distribution space to form a preprocessed dataset suitable for AI model input.

[0037] 104. Input the preprocessed data set into the preset AI decision model. Utilize the AI ​​decision model to analyze the difference between the current machine beam state and the target process requirements based on the preprocessed data set, calculate the optimal beam adjustment parameters based on the difference information, and perform compliance verification on the optimal beam adjustment parameters. The compliance verification includes safety threshold verification and process specification verification.

[0038] In this embodiment, the preprocessed data set is input into a preset AI decision model (e.g., one based on a deep neural network (DNN), reinforcement learning (RL), or gradient boosting tree (GBDT)). The model analyzes the difference between the actual beam state parameters (input features) of the current machine and the target process requirement data, calculates and outputs a set of optimized beam adjustment parameters (e.g., magnet current, focusing voltage adjustment amount). Before application, the system performs a double compliance check on the set of optimal parameters: a safety threshold check to ensure that parameter adjustment does not cause key equipment indicators (e.g., vacuum degree, cooling water temperature) to exceed hardware safety limits; and a process specification check to ensure that the adjusted parameters still meet the basic constraints and allowable range defined by the wafer process recipe.

[0039] 105. When the verification passes, a control instruction is generated according to the optimal beam current adjustment parameters, and the control instruction is sent to the IMP machine through the KVM switch, so that the IMP machine adjusts the relevant hardware module parameters according to the control instruction;

[0040] In this embodiment, after verification, a SEMI E84 control instruction is generated. The KVM switch converts the control instruction into a USB HID signal, which is transmitted to the IMP machine via the optical fiber isolation channel to drive the magnetic lens power supply of the IMP machine to adjust the parameters.

[0041] In an embodiment of the present invention, after the optimal beam adjustment parameters pass the compliance check, they are encapsulated into JSON format control instructions that comply with the SEMIE84 equipment control standard. By calling the API provided by the KVM switch, the control instructions are sent to the KVM switch. The built-in instruction parser of the KVM switch converts the JSON instructions into USB HID (human interface device) signals that simulate keyboard strokes (such as arrow keys and Enter key) and mouse clicks (such as button positions). These USB HID signals are transmitted to the target IMP machine through a physically isolated channel (such as a dedicated USB cable to ensure isolation from the control system network). The machine receives and parses these simulated operation signals, and automatically adjusts the parameter setting values ​​of its internal related hardware modules (such as ion source power supply and analysis magnet power supply) as if the operator were operating locally.

[0042] See also Figure 2 The second embodiment of the intelligent beam adjustment method based on IMP in the embodiment of the present invention includes:

[0043] 201. Automatically switch video channels according to preset timing through the KVM switch to obtain the real-time image of the operation interface of each IMP machine in turn;

[0044] In this embodiment, the KVM switch automatically performs polling switching among the video channels of the multiple IMP machines connected to it according to a preset time interval (such as every 5 seconds) or event trigger, ensuring that the real-time video stream or snapshot image of each machine operation interface is captured in sequence.

[0045] 202. Locate the beam parameter display area and status light area in the real-time screen of the operation interface based on the predefined machine interface template;

[0046] In this embodiment, based on the interface template pre-defined for each model of IMP machine (including the coordinate position and size information of the beam parameter display area and the status light area), image registration or feature matching technology is used to accurately locate the beam parameter digital display area and the status indicator and icon area that need to be identified in the acquired real-time image.

[0047] 203、Preprocess the beam parameter display area and the status light area, including grayscale processing, binarization processing, and denoising filtering processing;

[0048] In this embodiment, the located beam parameter display area image is preprocessed, first converted to a grayscale image to reduce color interference, then binarized (such as Otsu threshold method) to separate the text from the background, and finally denoising filtering (such as median filtering, Gaussian filtering) is applied to eliminate image noise and burrs, improve the OCR recognition accuracy, and the status light area is also grayscale and binarized, and may need morphological operations (such as opening operation, closing operation) to smooth the edges and fill small holes.

[0049] 204、Use a pre-trained CRNN convolutional recurrent neural network model to recognize the numbers and units in the beam parameter display area, and output the current beam parameter according to the numbers and units;

[0050] In this embodiment, the pre-trained CRNN convolutional recurrent neural network model is used to process the preprocessed beam parameter area image, and CRNN combines CNN to extract image features and RNN (such as LSTM) to process sequence characteristics, which can effectively recognize the number string (such as "150.5") and its corresponding unit text (such as "mA", "keV") in the area, and combine them to output the current beam parameter.

[0051] 205、Use a pre-trained YOLOv5 object detection model to detect the color and icon type of the status light area, and output the real-time status identifier according to the color and icon type;

[0052] In this embodiment, the pre-trained YOLOv5 object detection model is used to process the preprocessed status light area image, and YOLOv5 quickly detects the status light icon existing in the area, identifies its shape (such as circular, square) and accurate color (such as red, green, yellow), and according to the predefined rules (such as red light = alarm, green light = running, yellow light = standby / warning), the corresponding real-time status identifier is mapped and output.

[0053] 206、According to the current beam parameter and the real-time status identifier, generate JSON data conforming to the SEMI E125 standard, and obtain the parameter and status data;

[0054] In this embodiment, the beam parameter value and unit recognized by CRNN, and the status identifier (color, icon type, and mapped status semantics) recognized by YOLOv5 are organized into a clear JSON format string in key-value pairs according to the data structure requirements of the SEMI E125 communication standard, forming a standardized parameter and status data packet.

[0055] 207、Identify the wafer batch number currently loaded in the IMP machine;

[0056] In this embodiment, by analyzing the text displayed in a specific area on the IMP machine operation interface (using OCR), reading the messages sent by the machine communication interface (such as SECS-II), or parsing the machine log file, the wafer batch number currently being processed or about to be processed by the machine is determined.

[0057] 208、Send a query request to the EAP third-party system according to the wafer batch number;

[0058] In this embodiment, the system sends a structured query request message to the EAP (Equipment Automation Program) system according to the identified wafer batch number through a standardized communication protocol (such as SECS / GEM or Web Service API over TCP / IP), and requests to obtain the detailed process recipe data corresponding to the batch number.

[0059] 209、Receive the process recipe file returned by the EAP third-party system, and parse the process recipe file to obtain process requirement data;

[0060] In this embodiment, the system receives the process recipe file returned by the EAP system (which can be XML, JSON, or a specific format text), parses the file, extracts the key requirement data related to the ion implantation process of the wafer batch, including the target beam setting value, the allowed upper and lower limit tolerance range, the process step requirements, etc., and converts it into internal available structured process requirement data.

[0061] In the embodiment of the present application, the KVM resource allocation is optimized by using the time sequence wheel patrol strategy, the efficient and centralized monitoring of the operation interfaces of multiple IMP machines is realized, the region positioning technology based on the pre-defined interface template is combined with the mixed visual recognition scheme of CRNN and YOLOv5 (CRNN processes digital text sequence, YOLOv5 recognizes state icon color and type), which ensures the high precision and robustness of the extraction of key beam parameters and state identification, the non-structured recognition results are converted into JSON data conforming to SEMIE125 standard, the standardization and structuring of data are realized, the wafer batch number is automatically identified and integrated with the EAP system for interaction, the accurate and dynamic acquisition of target process requirement data is realized, which provides accurate and real-time data basis for subsequent intelligent decision-making, and solves the data acquisition and correlation problem in the multi-machine and multi-batch scenario.

[0062] Please refer to Figure 3 The third embodiment of the intelligent beam adjustment method based on IMP in the embodiment of the present application includes:

[0063] 301、Align the parameter and state data with the process requirement data by using the device ID of the IMP machine and the nanosecond-level timestamp as the association key to obtain integrated data;

[0064] In this embodiment, each real-time collected parameter and state data is marked with a high-precision (nanosecond-level) timestamp and a device unique ID tag. Similarly, the process requirement data obtained from the EAP is also marked with its corresponding device ID and acquisition timestamp (or the batch number effective time). By using the device ID and timestamp (accurately aligned to the same moment or adjacent effective window) as the association key, the two are accurately matched and aligned to form integrated data records containing real-time state and target requirements.

[0065] 302、Remove duplicate data in the integrated data and fill in missing values in the integrated data to obtain data cleaning integrated data;

[0066] In this embodiment, the integrated data is cleaned, specifically, completely repeated records are identified and deleted, and for partial field missing (such as EAP data not returned at a certain moment) caused by collection or communication delay, a reasonable filling strategy is used, such as using the value of the previous valid record to fill, using the historical average value of the device for this batch to fill, or marking as a special value, to ensure data integrity.

[0067] 303、Normalize the data cleaning integrated data to obtain a preprocessed data set;

[0068] In this embodiment, the numerical features (such as current beam value, target beam value, upper tolerance limit, lower tolerance limit) in the cleaned integrated data are normalized. Min-Max normalization (scaled to [0, 1] interval) or Z-Score standardization (converted to a distribution with mean of 0 and standard deviation of 1) can be used. The processed data constitutes the feature vector required for model input, forming a preprocessed data set.

[0069] 304、Use a neural network algorithm to build an AI decision model, collect historical beam tuning success records of the IMP machine as a training set, and the historical beam tuning success records include pre-tuning parameters, executed actions, post-tuning parameters, and tuning effect;

[0070] In this embodiment, an appropriate neural network architecture (such as multi-layer perception MLP, long short-term memory network LSTM, or Transformer) is selected to build an AI decision model, and historical tuning operation records are collected as a training set. Each record contains machine parameter state before tuning, tuning action (such as parameter adjustment), parameter state after tuning, and final tuning effect evaluation (such as whether it meets the standard or yield change).

[0071] 305. Train the AI decision model with the training set to make the AI decision model learn the mapping relationship between the parameters and the tuning effect;

[0072] In this embodiment, the AI decision model is supervised learning trained with the historical training set collected in step 304, and the training target is to make the model learn the complex mapping relationship between the "difference between the current state and the target" and the "optimal tuning action" and the prediction of the "tuning effect". Gradient descent algorithm is used to optimize the model parameters to minimize the loss between the predicted action or effect and the true value.

[0073] 306. Convert the preprocessed data set into a feature vector conforming to the model input;

[0074] In this embodiment, the preprocessed data set (containing normalized feature values) obtained in step 303 is converted and organized according to the input format (such as specific tensor shape, feature order) required by the AI decision model to form a feature vector that can be directly input into the model.

[0075] 307. Input the preprocessed data set into the preset AI decision model, and use the AI decision model to analyze the difference information between the current beam state and the target process requirement based on the preprocessed data set, and calculate the optimal beam tuning parameters according to the difference information;

[0076] In this embodiment, the converted feature vector is input into the trained AI decision model, and the internal calculation layer of the model analyzes the difference (Delta) between the current beam state (such as actual value, stability) and the target process requirement (such as set value, tolerance) based on the input features, and based on the learned mapping relationship, the model infers and calculates a set of optimal beam tuning parameters (adjustment suggestion value) that can minimize the difference and achieve or approach the target state.

[0077] 308. Compliance verification of the optimal beam tuning parameters, including safety threshold verification and process specification verification;

[0078] In this embodiment, before applying the optimal tuning parameters output by the model, strict compliance verification is performed, including safety threshold verification, i.e. checking whether the parameter adjustment will cause the key physical quantities of the equipment (such as high voltage, temperature, vacuum degree) to exceed the absolute safety limit value specified in the hardware design or safety regulations, and process specification verification, i.e. checking whether the adjusted parameters still conform to the allowed operation range (such as maximum / minimum beam limit) and process rules (such as parameter constraints for specific steps) defined in the process recipe for this wafer batch. If any of the verifications fails, the set of parameters will be rejected.

[0079] In the embodiment of the present application, through the precise association key design of the device ID and the nanosecond timestamp, the high-precision space-time alignment of the real-time state data and the process requirement data is ensured, through data cleaning (de-duplication, missing value processing) and normalization processing, a high-quality, standardized model input data set (preprocessed data set) is constructed, through constructing and training an AI decision model based on a neural network, the mapping relationship between parameters and effects is learned by using historical successful tuning records, intelligent calculation of optimal tuning parameters according to the current state and target difference is realized, through the introduction of a double compliance checking mechanism including safety threshold checking (to protect the safety of device hardware) and process specification checking (to protect the process compliance), the operation risk is strictly avoided while the AI optimization capability is exerted, the reliability and executability of the tuning suggestion are ensured, and a core decision engine and safety guarantee for intelligent tuning are provided.

[0080] Please refer to Figure 4 The fourth embodiment of the intelligent beam current tuning method based on IMP in the embodiment of the present application includes:

[0081] 401. When the verification passes, JSON data conforming to the SEMI E84 standard is generated according to the optimal beam current tuning parameters, and a control instruction is obtained;

[0082] In this embodiment, when the optimal beam current tuning parameters pass the compliance check, they are encapsulated into structured JSON data packets according to the format and field requirements specified in the SEMI E84 device control message standard, forming the final control instruction that can be sent to the KVM.

[0083] 402. The control instruction is sent to the KVM switch through the KVM API, so that the KVM switch converts the control instruction into USB HID signals simulating keyboard and mouse operations;

[0084] In this embodiment, by calling the programming interface (KVM API) provided by the KVM switch manufacturer, the encapsulated JSON control instruction is sent to the KVM switch. The control signal conversion module inside the KVM switch parses the instruction and converts it into a series of USB HID protocol signals simulating real keyboard (such as specific key sequences: Tab, direction keys, number keys, Enter) and mouse operations (such as moving to a specific coordinate and clicking).

[0085] 403. The USB HID signals are transmitted to the IMP machine through a physically isolated channel, so that the IMP machine adjusts the relevant hardware module parameters according to the USB HID signals;

[0086] In this embodiment, the KVM switch transmits the generated USB HID signals to the IMP machine through its dedicated physical USB cable (constituting a physically isolated channel, separated from the factory control network, ensuring security), and the operating system of the IMP machine receives and processes these HID signals, just as if the operator were performing operations on the local keyboard and mouse, driving its control software to adjust the parameter settings of the corresponding hardware modules (such as ion source controllers, beam optical assembly power supplies).

[0087] 404、Monitor and record parameter changes, execution time and effect data during the tuning process, obtain monitoring record information, and synchronize the monitoring record information to the EAP third-party system;

[0088] In this embodiment, during and after the execution of the tuning instruction, the system continuously monitors the feedback of the IMP machine, specifically by obtaining interface changes through KVM, reading real-time parameters through machine data interfaces (such as SECS), recording key monitoring information, including parameter change curves during tuning, actual execution time of tuning actions, steady-state parameter values achieved after tuning, and calculated effect indicators (such as the degree of reduction in deviation from the target), and synchronously returning these monitoring record information to the EAP third-party system in real time or near real time through standard interfaces for production tracking and data analysis.

[0089] 405、When a fault alarm exception or parameter overrun exception occurs in the monitoring record information, a preset fault program is started and alarm information is generated;

[0090] In this embodiment, during the monitoring process, if a fault alarm signal from the IMP machine is detected (identified through status identification or directly obtained through communication), or any key parameter (beam current, vacuum level, etc.) monitored exceeds the preset safety or process warning threshold, the system immediately triggers the preset fault handling program (such as emergency stop, restore default parameters, notify maintenance), and generates alarm information containing detailed exception information.

[0091] 406、Send the alarm information to the management terminal;

[0092] In this embodiment, the alarm information generated in step 405 is sent in real time to the designated management terminal (such as an engineer's workstation, a factory MES system dashboard, a mobile App) through message queues, API calls, or emails, etc. to notify relevant personnel to handle it.

[0093] 407、Through the KVM switch, continuously collect the latest operation interface images of the IMP machine after adjusting the parameters of the related hardware modules, and use OCR visual algorithms to recognize and extract the latest beam parameters and the latest status identification in the latest operation interface images;

[0094] In this embodiment, after the execution of the tuning instruction, the system continues to poll the target IMP machine through the KVM switch to collect the latest operation interface image of the target IMP machine, and uses the same OCR visual algorithm process (CRNN + YOLOv5) as step 101 to identify and extract the latest beam parameter values and the latest device status identifier displayed in the interface.

[0095] 408. Determine whether the tuning effect meets the standard according to the latest beam parameters and the latest status identifier;

[0096] In this embodiment, according to the latest beam parameters obtained in step 407, the absolute deviation and the relative deviation of the target process set value are calculated, combined with the latest status identifier (confirming no alarm and normal state), and referring to the process allowed tolerance range, the effect of this tuning is judged whether it reaches the predetermined standard (such as the deviation is less than the upper limit of the tolerance).

[0097] 409. If not, the AI decision model is used to recalculate the optimal beam tuning parameters;

[0098] In this embodiment, if the tuning effect is not up to standard (the deviation is too large or the state is abnormal) in step 408, the system automatically takes the current latest state data, target requirements and non-compliance results as new inputs, and calls the AI decision model again. The AI decision model recalculates a set of optimal beam tuning parameters based on the latest working conditions, and repeats the subsequent compliance verification and execution process.

[0099] 410. If yes, generate a current beam tuning success record as a new training set, and use the new training set to iteratively optimize the AI decision model;

[0100] In this embodiment, if the tuning effect meets the standard in step 408, the system records the successful tuning process of this round, including the state before tuning (input features), the tuning parameters applied (AI output), the final state after tuning (effect features) and the result of meeting the standard. This successful record is added to the historical training set of the AI decision model as a new high-quality sample, and the expanded training set containing the new sample is used to incrementally train or fine-tune the AI model periodically or online, so that the AI decision model can continuously learn and adapt to changes in device state and new process requirements, and continuously improve the decision accuracy.

[0101] In the embodiment of the present application, by encapsulating the compliant optimal tuning parameters as control instructions conforming to the SEMI E84 standard, and converting them into USB HID signals by using KVM API, and then transmitting them to the IMP machine through a physically isolated channel, the standardized, safe and automated execution of the tuning instructions is realized, completely simulating manual operation and avoiding the risk of network attacks, through real-time monitoring and recording of tuning process parameters, time and effect, and synchronous feedback to the EAP system, the traceability of the tuning process and the integrity of the production data are realized, through the establishment of an abnormality detection and alarm mechanism (fault alarm, parameter overrun), the safety of the tuning process is ensured, through continuous collection of the latest interface information after tuning and evaluation of the tuning effect, a closed-loop control is formed, if the effect is not up to standard, the AI model is triggered to recalculate the optimal parameters, if the effect is up to standard, this successful record is used as a new training set to iteratively optimize the AI model, realizing continuous self-learning and performance improvement of the system, building a complete intelligent tuning closed loop, ensuring execution safety and effect, and endowing the system with self-adaptive evolution ability.

[0102] The intelligent beam tuning method based on IMP in the embodiment of the present application is described above, and the intelligent beam tuning device based on IMP in the embodiment of the present application is described below, please refer to Figure 5 An embodiment of the intelligent beam tuning device based on IMP in the embodiment of the present application includes:

[0103] The acquisition and recognition extraction module 501 is used to acquire the operation interface image of each IMP machine in real time through the KVM switch, and recognize and extract the current beam parameters and real-time state identifiers in the operation interface image by using an OCR visual algorithm;

[0104] The conversion acquisition module 502 is used to convert the current beam parameters and real-time state identifiers into structured parameter and state data, and acquire the process requirement data of the current wafer batch number from the EAP third-party system;

[0105] The integration and cleaning normalization module 503 is used to integrate the parameter and state data and the process requirement data to obtain integrated data, and perform data cleaning and normalization processing on the integrated data to obtain a preprocessed data set;

[0106] The analysis and calculation verification module 504 is used to input the preprocessed data set into a preset AI decision model, analyze the difference information between the current machine beam state and the target process requirement based on the preprocessed data set by using the AI decision model, and calculate the optimal beam tuning parameters according to the difference information, and perform compliance verification on the optimal beam tuning parameters, including safety threshold verification and process specification verification;

[0107] The generating and sending module 505 is configured to generate a control instruction according to the optimal beam current adjustment parameter when the verification passes, and send the control instruction to the IMP machine through the KVM switch, so that the IMP machine adjusts the relevant hardware module parameters according to the control instruction.

[0108] In this embodiment, the operation interface image of each IMP machine is acquired in real time through the KVM switch, the current beam current parameter and the real-time state identifier in the operation interface image are recognized and extracted by using an OCR visual algorithm, the current beam current parameter and the real-time state identifier are converted into structured parameter and state data, the process requirement data of the current wafer batch number is acquired from an EAP third-party system, the parameter and state data and the process requirement data are integrated, the integrated data is subjected to data cleaning and normalization processing to obtain a preprocessed data set, the preprocessed data set is input into a preset AI decision model, the AI decision model is used to analyze the difference information between the current machine beam current state and the target process requirement based on the preprocessed data set, and the optimal beam current adjustment parameter is calculated according to the difference information, so as to improve the adjustment accuracy, adapt to complex process requirements, perform compliance verification on the optimal beam current adjustment parameter, generate a control instruction according to the optimal beam current adjustment parameter when the verification passes, and send the control instruction to the IMP machine through the KVM switch, so that the IMP machine adjusts the relevant hardware module parameters according to the control instruction, improves the intelligent and automatic level, reduces human errors, improves the machine operation efficiency, and shortens the process waiting time.

[0109] Please refer to Figure 6 Another embodiment of the intelligent beam current adjustment device based on the IMP in the embodiment of the application includes:

[0110] The acquiring, recognizing and extracting module 501 is configured to acquire the operation interface image of each IMP machine in real time through the KVM switch, and recognize and extract the current beam current parameter and the real-time state identifier in the operation interface image by using an OCR visual algorithm.

[0111] The conversion and acquisition module 502 is configured to convert the current beam current parameter and the real-time state identifier into structured parameter and state data, and acquire the process requirement data of the current wafer batch number from an EAP third-party system.

[0112] The integration, cleaning and normalization module 503 is configured to integrate the parameter and state data and the process requirement data to obtain integrated data, and perform data cleaning and normalization processing on the integrated data to obtain a preprocessed data set.

[0113] The analysis and calculation verification module 504 is configured to input the preprocessed data set into a preset AI decision model, analyze the difference information between the current beam current state and the target process requirement based on the preprocessed data set by using the AI decision model, and calculate optimal beam current adjustment parameters according to the difference information. The optimal beam current adjustment parameters are subjected to compliance verification, including safety threshold verification and process specification verification.

[0114] The generation and sending module 505 is configured to generate a control instruction according to the optimal beam current adjustment parameters when the verification passes, and send the control instruction to the IMP machine through the KVM switcher, so that the IMP machine adjusts the related hardware module parameters according to the control instruction.

[0115] In this embodiment, the acquisition, recognition and extraction module 501 includes: an acquisition unit 5011 configured to automatically switch video channels by the KVM switcher in a preset time sequence, and acquire the operation interface real-time picture of each IMP machine in turn; a positioning unit 5012 configured to position the beam current parameter display area and the state lamp area in the operation interface real-time picture based on a predefined machine interface template; a preprocessing unit 5013 configured to preprocess the beam current parameter display area and the state lamp area, including grayscale processing, binarization processing and denoising filtering processing; a recognition output unit 5014 configured to recognize the numbers and units in the beam current parameter display area using a pre-trained CRNN convolutional recurrent neural network model, and output the current beam current parameter according to the numbers and units; and a detection output unit 5015 configured to detect the color and icon type of the state lamp area using a pre-trained YOLOv5 target detection model, and output the real-time state identifier according to the color and icon type.

[0116] In this embodiment, the conversion and acquisition module 502 includes: a first generation unit 5021 configured to generate JSON data conforming to the SEMI E125 standard according to the current beam current parameter and the real-time state identifier, to obtain parameter and state data; a recognition unit 5022 configured to recognize the wafer batch number loaded by the current IMP machine; a first sending unit 5023 configured to send a query request to an EAP third-party system according to the wafer batch number; and a receiving and analyzing unit 5024 configured to receive the process recipe file returned by the EAP third-party system, analyze the process recipe file, and obtain process requirement data.

[0117] In the embodiment, the integrated cleaning and normalization module 503 includes an alignment unit 5031 configured to align the parameter and state data and the process requirement data with the device ID and the nanosecond-level timestamp of the IMP machine as the association key to obtain integrated data; a removal and filling unit 5032 configured to remove the repeated data in the integrated data and fill in the missing values in the integrated data to obtain the integrated data after data cleaning; and a normalization unit 5033 configured to normalize the integrated data after data cleaning to obtain a preprocessed data set.

[0118] In the embodiment, the analysis and calculation verification module 504 includes a construction and collection unit 5041 configured to construct an AI decision model by using a neural network algorithm, collect historical beam tuning success records of the IMP machine as a training set, and collect the historical beam tuning success records including parameters before tuning, execution actions, parameters after tuning, and tuning effects; a training unit 5042 configured to train the AI decision model by using the training set to enable the AI decision model to learn the mapping relationship between the parameters and the tuning effects; a conversion unit 5043 configured to convert the preprocessed data set into a feature vector conforming to the model input; an analysis and calculation unit 5044 configured to input the preprocessed data set into the preset AI decision model, analyze the difference information between the current machine beam state and the target process requirement based on the preprocessed data set by using the AI decision model, and calculate optimal beam tuning parameters according to the difference information; and a verification unit 5045 configured to perform compliance verification on the optimal beam tuning parameters, and the compliance verification includes safety threshold verification and process specification verification.

[0119] In the embodiment, the generation and transmission module 505 includes a second generation unit 5051 configured to generate JSON data conforming to the SEMI E84 standard according to the optimal beam tuning parameters when the verification passes to obtain a control instruction; a second transmission unit 5052 configured to transmit the control instruction to the KVM switch through the KVM API to enable the KVM switch to convert the control instruction into a USB HID signal of an analog keyboard and mouse operation; and a transmission unit 5053 configured to transmit the USB HID signal to the IMP machine through a physically isolated channel to enable the IMP machine to adjust the related hardware module parameters according to the USB HID signal.

[0120] In the embodiment, the monitoring and recording feedback module 506 is further included for monitoring and recording parameter changes, execution time and effect data in the adjustment process to obtain monitoring and recording information, and synchronously feeding back the monitoring and recording information to the EAP third-party system; the start generation module 507 is used for starting a preset fault program and generating an alarm information when a fault alarm exception or a parameter overrun exception occurs in the monitoring and recording information; the sending module 508 is used for sending the alarm information to a management terminal; the acquisition and recognition extraction module 509 is used for continuously acquiring the latest operation interface image of the IMP machine adjustment related hardware module parameter through the KVM switch, recognizing and extracting the latest beam parameters and the latest state identifier in the latest operation interface image by using an OCR visual algorithm; the judgment module 510 is used for judging whether the adjustment effect meets the standard according to the latest beam parameters and the latest state identifier; the calculation module 511 is used for recalculating the optimal beam adjustment parameter by using an AI decision model if not; and the generation optimization module 512 is used for generating a current beam adjustment success record as a new training set and iteratively optimizing the AI decision model by using the new training set if yes.

[0121] The above Figure 5 And Figure 6 The IMP-based intelligent beam adjustment device in the embodiment of the application is described in detail from the perspective of a modular functional entity, and the IMP-based intelligent beam adjustment device in the embodiment of the application is described in detail from the perspective of hardware processing.

[0122] Figure 7 is a structural schematic diagram of an IMP-based intelligent beam adjustment device provided by the embodiment of the application. The IMP-based intelligent beam adjustment device 600 can have great differences due to different configurations or performances, and can include one or more than one processor (central processing unit, CPU) 610 (for example, one or more than one processor) and a memory 620, one or more than one storage medium 630 (for example, one or more than one mass storage device) storing an application program 633 or data 632. The memory 620 and the storage medium 630 can be temporary storage or persistent storage. The program stored in the storage medium 630 can include one or more than one module (not shown in the figure), and each module can include a series of instruction operations in the IMP-based intelligent beam adjustment device 600. Further, the processor 610 can be configured to communicate with the storage medium 630, execute a series of instruction operations in the storage medium 630 on the IMP-based intelligent beam adjustment device 600, so as to realize the steps of the IMP-based intelligent beam adjustment method provided by each method embodiment.

[0123] The IMP-based intelligent beam current trimming apparatus 600 can also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that, Figure 7 The illustrated IMP-based intelligent beam current trimming apparatus structure does not constitute a limitation on the IMP-based intelligent beam current trimming apparatus, and can include more or fewer components than illustrated, or combine certain components, or different component arrangements.

[0124] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium, or a volatile computer readable storage medium, and the computer readable storage medium stores instructions, which, when executed on a computer, cause the computer to perform the steps of the IMP-based intelligent beam current trimming method.

[0125] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system or device, unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0126] The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0127] Finally, it should be noted that the above only describes the preferred examples of the present application, and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art will appreciate that modifications can be made to the technical solutions described in the foregoing embodiments, or some of the technical features thereof can be replaced equivalently, without departing from the spirit and principle of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An IMP-based intelligent beam current trimming method, characterized in that, The application comprises the following steps: Real-time acquisition of the operation interface image of each IMP machine through a KVM switch, identification and extraction of the current beam parameters and real-time state identifiers in the operation interface image using an OCR visual algorithm; Conversion of the current beam parameters and real-time state identifiers into structured parameter and state data, acquisition of process requirement data of the current wafer batch number from an EAP third-party system; Integration of the parameter and state data and the process requirement data to obtain integrated data, data cleaning and normalization processing of the integrated data to obtain a preprocessed data set; Input of the preprocessed data set into a preset AI decision model, analysis of the difference information between the current machine beam state and the target process requirement based on the preprocessed data set using the AI decision model, and calculation of optimal beam tuning parameters according to the difference information, compliance verification of the optimal beam tuning parameters, including safety threshold verification and process specification verification; When the verification is passed, control instructions are generated according to the optimal beam tuning parameters, and the control instructions are sent to the IMP machine through the KVM switch to enable the IMP machine to adjust the relevant hardware module parameters according to the control instructions.

2. The IMP-based intelligent beam current trim method of claim 1, wherein, The real-time acquisition of the operation interface image of each IMP machine through a KVM switch, the identification and extraction of the current beam parameters and real-time state identifiers in the operation interface image using an OCR visual algorithm, comprises the following steps: Automatic switching of video channels through a KVM switch according to a preset time sequence, and sequential acquisition of the operation interface real-time picture of each IMP machine; Positioning of the beam parameter display area and the state lamp area in the operation interface real-time picture based on a predefined machine interface template; Preprocessing of the beam parameter display area and the state lamp area, including grayscale processing, binarization processing and denoising filter processing; Identification of the numbers and units in the beam parameter display area using a pre-trained CRNN convolutional recurrent neural network model, and output of the current beam parameters according to the numbers and units; Detection of the color and icon type of the state lamp area using a pre-trained YOLOv5 target detection model, and output of the real-time state identifier according to the color and icon type.

3. The IMP-based intelligent beam current trim method of claim 1, wherein, The conversion of the current beam parameters and real-time state identifiers into structured parameter and state data, and the acquisition of process requirement data of the current wafer batch number from an EAP third-party system, comprises the following steps: Generation of JSON data conforming to the SEMI E125 standard according to the current beam parameters and real-time state identifiers to obtain parameter and state data; Identification of the wafer batch number currently loaded by the IMP machine; Sending of a query request to the EAP third-party system according to the wafer batch number; Receiving of the process recipe file returned by the EAP third-party system, and analysis of the process recipe file to obtain process requirement data.

4. The IMP-based intelligent beam current trim method of claim 1, wherein, The integration of the parameter and state data and the process requirement data to obtain integrated data, the data cleaning and normalization processing of the integrated data to obtain a preprocessed data set, comprises the following steps: Align the parameter and state data and the process requirement data by taking the equipment ID of the IMP machine and the nanosecond-level timestamp as the association key to obtain integrated data; Remove the repeated data in the integrated data and fill in the missing values in the integrated data to obtain the integrated data after data cleaning; Perform normalization processing on the integrated data after data cleaning to obtain a pretreatment data set.

5. The IMP-based intelligent beam current trim method of claim 1, wherein, The pretreatment data set is input into a preset AI decision model, the AI decision model analyzes the difference information between the current machine beam current state and the target process requirement based on the pretreatment data set, and calculates the optimal beam current adjustment parameter according to the difference information, and the optimal beam current adjustment parameter is subjected to compliance verification, and the compliance verification includes safety threshold verification and process specification verification, including: An AI decision model is constructed by using a neural network algorithm, historical beam current adjustment success records of the IMP machine are collected as a training set, the historical beam current adjustment success records include pre-adjustment parameters, execution actions, post-adjustment parameters and adjustment effects; The training set is used to train the AI decision model, so that the AI decision model learns the mapping relationship between the parameters and the adjustment effects; The pretreatment data set is converted into a feature vector conforming to the model input; The pretreatment data set is input into a preset AI decision model, the AI decision model analyzes the difference information between the current machine beam current state and the target process requirement based on the pretreatment data set, and calculates the optimal beam current adjustment parameter according to the difference information; The optimal beam current adjustment parameter is subjected to compliance verification, and the compliance verification includes safety threshold verification and process specification verification.

6. The IMP-based intelligent beam current trim method of claim 1, wherein, When the verification passes, a control instruction is generated according to the optimal beam current adjustment parameter, the control instruction is sent to the IMP machine through the KVM switch, so that the IMP machine adjusts the related hardware module parameters according to the control instruction, including: When the verification passes, a control instruction is generated according to the optimal beam current adjustment parameter, the control instruction is sent to the IMP machine through the KVM switch, so that the IMP machine adjusts the related hardware module parameters according to the control instruction, including: When the verification passes, a control instruction is generated according to the optimal beam current adjustment parameter, the control instruction is sent to the IMP machine through the KVM switch, so that the IMP machine adjusts the related hardware module parameters according to the control instruction, including: When the verification passes, a control instruction is generated according to the optimal beam current adjustment parameter, the control instruction is sent to the IMP machine through the KVM switch, so that the IMP machine adjusts the related hardware module parameters according to the control instruction, including:

7. The IMP-based intelligent beam current trim method of claim 1, wherein, Monitor and record the parameter changes, execution time and effect data in the adjustment process to obtain monitoring record information, and synchronize the monitoring record information to the EAP third-party system; ​ When a fault alarm exception or a parameter overrun exception occurs in the monitoring record information, a preset fault program is started and alarm information is generated; The alarm information is sent to a management terminal; The KVM switch continuously acquires the latest operation interface image of the IMP machine after the related hardware module parameters are adjusted, and uses an OCR visual algorithm to identify and extract the latest beam parameters and the latest state identifier in the latest operation interface image; It is judged whether the adjustment effect meets the standard according to the latest beam parameters and the latest state identifier; If not, the AI decision model is used to recalculate the optimal beam adjustment parameters; If yes, a current beam adjustment success record is generated as a new training set, and the AI decision model is iteratively optimized using the new training set.

8. An IMP-based intelligent beam current trimming device, characterized in that, It comprises: An acquisition and identification extraction module is configured to acquire the operation interface image of each IMP machine in real time through a KVM switch, and use an OCR visual algorithm to identify and extract the current beam parameters and real-time state identifier in the operation interface image; A conversion acquisition module is configured to convert the current beam parameters and real-time state identifier into structured parameter and state data, and acquire the process requirement data of the current wafer batch number from an EAP third-party system; An integration, cleaning and normalization module is configured to integrate the parameter and state data and the process requirement data to obtain integrated data, and perform data cleaning and normalization processing on the integrated data to obtain a preprocessed data set; An analysis, calculation and verification module is configured to input the preprocessed data set into a preset AI decision model, use the AI decision model to analyze the difference information between the current machine beam state and the target process requirement based on the preprocessed data set, and calculate the optimal beam adjustment parameters according to the difference information, and perform compliance verification on the optimal beam adjustment parameters, wherein the compliance verification includes safety threshold verification and process specification verification; A generation and sending module is configured to generate a control instruction according to the optimal beam adjustment parameters when the verification passes, and send the control instruction to the IMP machine through the KVM switch, so that the IMP machine adjusts the related hardware module parameters according to the control instruction.

9. An IMP-based intelligent beam current trimming device, characterized in that, The IMP-based intelligent beam adjustment device comprises a memory and at least one processor, and the memory stores instructions; At least one processor calls the instructions in the memory to enable the IMP-based intelligent beam adjustment device to perform each step of the IMP-based intelligent beam adjustment method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement each step of the IMP-based intelligent beam adjustment method according to any one of claims 1-7.

Images