CN117672928B

CN117672928B - Box opening method

Info

Publication number: CN117672928B
Application number: CN202311361941.3A
Authority: CN
Inventors: 郭华欣; 曹立春; 古加林; 马格林; 康宏
Original assignee: Chongqing Eagle Valley Optoelectronic Ltd
Current assignee: Chongqing Eagle Valley Optoelectronic Ltd
Priority date: 2023-10-19
Filing date: 2023-10-19
Publication date: 2024-06-25
Anticipated expiration: 2043-10-19
Also published as: CN117672928A

Abstract

The application discloses a box opening method for controlling EFEM box opening, which relates to the technical field of semiconductor preparation and comprises the following steps: collecting gas data and temperature and humidity data; judging whether the gas concentration or the gas composition exceeds a threshold value, and if so, performing gas filtration according to a preset first gas filtration strategy; calculating the deviation between the received temperature and humidity data and the target temperature and humidity, generating a first temperature control strategy according to the deviation through a PID algorithm, and regulating the temperature and humidity according to the first temperature control strategy; calculating a gas cleanliness index and a temperature and humidity stability index; judging whether the index exceeds a threshold value, and if so, generating a second gas filtering strategy and a second temperature control strategy by the main control module based on the machine learning model; judging whether the index meets the box opening condition, if so, sending a box opening instruction to the manipulator module; the manipulator module receives and executes the box opening instruction. Aiming at the problem of low wafer yield in the prior art, the application improves the wafer manufacturing yield through multistage gas filtration, accurate temperature and humidity control and the like.

Description

Box opening method

Technical Field

The application relates to the technical field of semiconductor preparation, in particular to a box opening method for controlling EFEM box opening.

Background

With the continuous shrinking of integrated circuit manufacturing processes, the cleanliness requirements of wafer manufacturing environments are increasing. The wafer is extremely susceptible to tiny particles, chemical pollution and the like in the manufacturing, processing and packaging links, so that the surface defects of the wafer are increased, and the yield is reduced. How to improve the manufacturing yield of wafers is a problem to be solved in the semiconductor industry.

Wafer cassettes are important carriers and barriers to wafers during manufacturing and shipping. At present, the process of opening the wafer box mainly depends on experience of operators, and the real-time monitoring of gas cleanliness and temperature and humidity is lacking. And tiny particle pollution in the gas and unstable temperature and humidity directly affect the box opening environment, which may lead to the reduction of the surface quality of the wafers after box opening.

In the related art, for example, chinese patent document CN111090295A provides a control method and a control system for environmental parameters in EFEM. The control method comprises the following steps: disposing a clean space having a set volume outside the EFEM; removing predetermined impurities in the clean space; heating the temperature in the clean space to a preset temperature and then blowing air into the EFEM; the heating temperature in the clean space is dynamically adjusted according to the temperature and humidity in the EFEM so as to make the temperature and humidity in the EFEM constant. The application sets a clean space outside the EFEM, and conveys dust-free high-temperature gas into the EFEM after filtering and removing impurities such as acid, alkali, volatile organic compounds, particulate matters and the like in the clean space, and dynamically adjusts the temperature in the clean space to enable the EFEM to be in a constant temperature and humidity environment, so that the condition that the gas is not condensed on the surface of the wafer in the process of conveying the EFEM is ensured, but the application at least has the following steps: the lack of intelligent optimization control strategies cannot accommodate complex changes in the environment within the EFEM, resulting in further improvements in wafer manufacturing yields.

Disclosure of Invention

1. Technical problem to be solved

Aiming at the problem of low yield of wafers in the prior art, the application provides a box opening method which is used for controlling the opening of an EFEM, and can realize that the gas cleanliness and the temperature and humidity of the box opening environment are stable in ideal states through multistage gas filtration, accurate temperature and humidity control, self-adaptive filtration strategies and the like, thereby effectively improving the manufacturing yield of the wafers.

2. Technical proposal

The application aims to realize a box opening method by the following technical scheme that: step one, a gas monitoring module collects gas data containing gas concentration and gas components and sends the gas data to a gas control module; step two, the temperature and humidity monitoring module collects temperature and humidity data comprising temperature and humidity and sends the temperature and humidity data to the temperature control module; step three, the gas control module receives the gas data, judges whether the gas concentration or the gas composition exceeds a threshold value, if so, performs gas filtration according to a preset first gas filtration strategy, and sends the gas data in the gas filtration to the main control module; the temperature control module receives temperature and humidity data, calculates deviation between the received temperature and humidity data and target temperature and humidity, generates a first temperature control strategy according to the deviation through a PID algorithm, adjusts the temperature and humidity according to the first temperature control strategy, and sends the temperature and humidity data in the adjusting process to the main control module; step five, the main control module receives the gas data and the temperature and humidity data, and calculates a gas stability evaluation index KPI1 and a temperature and humidity stability index KPI2; judging whether the index exceeds a threshold value, if so, calling a second gas filtering strategy generated by the SVM model, calling a BP network model to generate a second temperature control strategy, and respectively transmitting the second temperature control strategy to the gas control module and the temperature control module; step six, the gas control module receives and filters the gas according to a second gas filtering strategy; the temperature control module receives and adjusts the temperature and the humidity according to a second temperature control strategy; step seven, the main control module judges whether the gas stability evaluation index KPI1 and the temperature and humidity stability index KPI2 reach the box opening condition, and if so, a box opening instruction is sent to the manipulator module; and step eight, the manipulator module receives and executes a box opening instruction.

Further, the first gas filtration strategy comprises: when the particle size of the gas pollutant is larger than 5 mu m, setting the gas to enter from the outer side of the filter element cylinder at a speed of 0.1m/s, and filtering by using a polypropylene filter material with a pore size of 5 mu m from inside to outside to remove the particle impurity larger than 5 mu m in the gas; the second-stage efficient filtration, when the particle size of the gas pollutant is 0.3-5 μm, setting the gas to enter through the inner side of the filter element cylinder at the speed of 0.05m/s, firstly passing through the 0.3 μm glass microfiber filter material layer to remove 0.3-5 μm fine particles, and then passing through the active carbon layer to adsorb volatile organic pollutants in the gas; the third stage of activated carbon adsorption, when the concentration of the volatile organic pollutants is higher than 10ppb, the gas is utilized to pass through the two layers of activated carbon filter element cylinders, the activated carbon adopts oval particles, and the activated carbon with the specific surface area of 1700m < 2 >/g is utilized to adsorb the volatile organic pollutants in the gas; fourth-stage gas purification, wherein when the concentration of residual pollutants in the gas is higher than 100ppt, non-thermal plasma generated by exciting the gas through an induction electrode is utilized to oxidatively decompose the residual pollutants in the gas by utilizing oxygen ions in the plasma; and fifth-stage negative pressure adsorption, namely after four-stage filtration, when the concentration of residual pollutants in the gas is higher than 10ppt, applying 0.05MPa negative pressure, and utilizing the negative pressure to enhance the adsorption capacity of the activated carbon so as to purify the residual pollutants in the gas.

Further, the first step includes: setting a plurality of groups of gas sensors, and collecting gas data, wherein the gas data comprises gas concentration and a plurality of gas components; inputting the acquired gas data into a pre-established data fusion module, and outputting the fused gas data by the data fusion module based on a particle filtering algorithm; inputting the fused data into a data correction model, and correcting the gas data by the data correction model by using a preset dynamic correction strategy, and outputting the corrected gas data; inputting the calibrated gas data into a gas control model, establishing the gas control model based on an MRAC algorithm, and updating control parameters of a first gas filtering strategy by using the calibrated gas data; and sending the calibrated gas data to a gas control module.

Further, the step of generating the second gas filtration strategy and the second temperature control strategy comprises: acquiring space coordinate information of a gas sensor and acquired gas data; according to the acquired space coordinate information and gas concentration of the gas sensor, calculating the statistical mean value and standard deviation of the gas concentration in space distribution, and taking the statistical mean value and standard deviation as a gas stability evaluation index KPI1; establishing a three-dimensional geometric model of an open box environment, performing grid division by adopting Fluent software, and performing CFD simulation of a temperature and humidity field; according to CFD simulation, extracting the variation ranges of a temperature field and a humidity field, and calculating the variation range to be used as a temperature and humidity stability index KPI2; constructing a LASSO regression model, and adjusting the weight of each gas component in the KPI1 according to the space coordinate information of the gas sensor; establishing a gas diffusion and temperature and humidity coupling model based on CFD simulation, and respectively calculating overrun probabilities P1 and P2 of KPI1 and KPI2 by adopting a Bayes network; when the overrun probability P1 or P2 exceeds a threshold value, the main control module calls an SVM model, and generates a second gas filtering strategy according to the gas data; the main control module calls a BP network model and generates a second temperature control strategy according to CFD temperature and humidity data; and sending the second gas filtering strategy to the gas control module, and sending the second temperature control strategy to the temperature control module.

Further, step seven includes: setting a gas cleanliness box opening threshold alpha and a temperature and humidity box opening threshold beta; constructing a double-layer LSTM network as a temperature and humidity prediction model, wherein the LSTM network comprises an encoder and a decoder; acquiring a temperature and humidity time sequence by using a Pt100 and humidity sensor, inputting the temperature and humidity time sequence into an encoder, extracting the characteristics of the time sequence by the encoder according to an LSTM network, and outputting a first characteristic expression vector; inputting the CFD simulation result to an encoder to generate a second feature expression vector; the second characteristic expression vector is input into a decoder, the decoder predicts the temperature and humidity state according to the LSTM network, and a third characteristic expression vector is output; judging whether the temperature and humidity state represented by the third characteristic expression vector is within the allowable temperature range of 20-25 ℃ and the allowable humidity range of 40-50% RH; judging whether the gas stability evaluation index KPI1 is lower than a gas box opening threshold alpha or not; judging whether the temperature and humidity stability evaluation index KPI2 is lower than a temperature and humidity box opening threshold value beta or not; if the conditions are satisfied, judging that the box opening condition is satisfied; and the main control module issues a Profinet box opening execution instruction to the manipulator.

Further, the step of fusing the collected gas data includes: taking gas data acquired by a gas sensor as an observed quantity of a particle filtering algorithm, wherein the type of the gas sensor comprises an infrared gas sensor and a gas chromatograph; initializing a particle state by using a sampling method based on Kullback Leibler divergence, and generating a first particle state of posterior probability distribution; carrying out state prediction on the first particle state by applying a Kalman filtering algorithm of a noise covariance matrix in an online self-adaptive adjustment process, and outputting a predicted state; when the weight of the first particle state is calculated, introducing a space constraint relation based on a CFD gas propagation model, and combining depth features of gas data extracted by using a pretrained convolutional neural network to serve as priori knowledge of weight calculation; resampling the particles by an Auxiliary PARTICLE FILTER algorithm according to the depth characteristics to generate a second particle state; correcting the second particle state obtained by sampling by using a Kalman filtering algorithm, and outputting the corrected second particle state; carrying out weighted average on the corrected second particle state to obtain a gas concentration fusion value; calculating the error between the gas concentration fusion value and the acquired gas data, and adopting a Gauss Newton method to adaptively adjust the process noise parameters of the Kalman filtering algorithm according to the error; and recursively executing the steps and outputting the fused gas concentration and gas composition.

Further, the step of calibrating the fused gas data includes: acquiring gas data acquired by a gas sensor, and acquiring a signal drift mode of the gas sensor through a time sequence analysis algorithm; according to the acquired signal drift mode, calculating zero calibration parameters and sensitivity calibration parameters of the gas sensor by adopting an incremental differential algorithm; calculating compensation control quantity of the gas data according to the acquired gas data, the zero calibration parameter and the sensitivity calibration parameter; and performing closed-loop PID control on the acquired gas data by using the obtained compensation control quantity to generate calibrated gas data.

Further, the step of updating the first gas filtration strategy with the calibrated gas data comprises: constructing a gas control model based on an MRAC algorithm, wherein the gas control model comprises a preset first gas filtering strategy F1; inputting the calibrated gas data into a gas control model, and generating a filtering strategy F1' based on an RLS algorithm; calculating errors of the first gas filtering strategy F1 and the filtering strategy F1'; based on the self-adaptive control algorithm, calculating a control input quantity delta u of gas filtration according to the error; adjusting the first gas filtering strategy according to the control input quantity delta u; the MRAC algorithm is recursively executed to update the first gas filtering strategy.

Further, the third step includes: a gas threshold database is arranged in the gas control module, and the thresholds of various gas components are preset in the database; the gas control module receives the gas data output by the gas monitoring module; judging whether the gas data exceeds a threshold value by utilizing a fuzzy control algorithm according to the threshold value of the gas components in the database, and if so, performing gas filtration according to a preset first gas filtration strategy; monitoring the gas pressure by adopting an MEMS pressure sensor, and inputting the monitored gas pressure into a gas flow intelligent regulating device; the intelligent gas flow regulator adopts adaptive PID algorithm to control the electronic RF valve to regulate the gas flow.

Further, the fourth step includes: acquiring temperature and humidity data of the collected gas; calculating the deviation e (t) between the acquired temperature and humidity data and the target temperature and humidity according to an incremental PID algorithm; generating a temperature control output u (t) by utilizing a PID algorithm according to the deviation e (t); converting u (t) into a voltage signal of 0V to 10V through a DAC (digital-to-analog converter), and taking the voltage signal as a control signal of the three-way heat exchange valve; the three-way heat exchange valve accurately adjusts the proportion of gas flowing through the electric heater and the Peltier refrigerator, and controls the temperature and the humidity of the gas; and acquiring the temperature adjustment times and the humidity adjustment times in the temperature control process by adopting an electronic counter.

3. Advantageous effects

Compared with the prior art, the application has the advantages that:

(1) By using a PID algorithm and a three-way heat exchanger, the scheme can realize rapid and accurate control of the temperature and the humidity of the box opening environment, which is of great importance, because the wafer manufacturing has extremely high requirements on the high stability of the temperature and the humidity of the environment, and the wafer manufacturing yield is improved by preventing the adverse effect of temperature and humidity fluctuation on the wafer performance;

(2) And a machine learning model is introduced to generate a temperature control strategy and a gas filtering strategy, so that the system has intelligence and self-adaptability. Compared with the traditional static preset strategy, the intelligent decision system can learn and optimize according to actual conditions, so that the intelligent decision system is better suitable for complex and variable open box environment conditions, the performance and efficiency of the system are improved, and the manufacturing yield of wafers is improved;

(3) The linkage judgment mechanism of the gas cleanliness and the temperature and humidity is set, the box opening operation is only executed when two key indexes meet the conditions, the mechanism ensures the perfection of the box opening environment, avoids the adverse effect of executing the box opening operation on the wafer manufacturing under unsuitable environmental conditions, and improves the manufacturing yield of the wafers.

Drawings

The present specification will be further described by way of exemplary embodiments, which will be described in detail by way of the accompanying drawings. The embodiments are not limiting, in which like numerals represent like structures, wherein:

FIG. 1 is an exemplary flow chart of a method of opening a box according to some embodiments of the present description;

FIG. 2 is an exemplary flow chart of processing collected gas data according to some embodiments of the present description;

FIG. 3 is an exemplary flow diagram illustrating gas filtration using a first gas filtration strategy according to some embodiments of the present disclosure;

FIG. 4 is an exemplary flow chart for regulating temperature and humidity using a first temperature control strategy according to some embodiments of the present disclosure;

FIG. 5 is an exemplary flow chart for generating a second gas filtration strategy and a second temperature control strategy according to some embodiments of the present description;

FIG. 6 is an exemplary flow chart for generating an open box instruction according to some embodiments of the present description.

Detailed Description

The method and system provided in the embodiments of the present specification are described in detail below with reference to the accompanying drawings.

Fig. 1 is an exemplary flow chart of a method of opening a box, as shown in fig. 1, according to some embodiments of the present description, the method of opening a box comprising the steps of:

S100, a gas monitoring module collects gas data containing gas concentration and gas components and sends the gas data to a gas control module; in the present application, the gas monitoring module may employ: thermal gas sensor: the gas components such as O2, CO2 and the like in the air can be detected; optical gas sensors, such as non-dispersive infrared gas sensors, can detect CO, CH4, etc.; semiconductor gas sensor: toxic gases such as H2, NH3, NO2 and the like can be detected; the electrochemical sensor can detect gases such as O3, SO2, NO and the like; during wafer fabrication, the monitored main gas components include: hydrocarbon: such as methane, acetylene, propylene, etc., can produce chemical contamination; nitrogen oxides, such as nitrous oxide, nitrogen dioxide, can corrode material surfaces; sulfur compounds, such as sulfur dioxide, hydrogen sulfide, can cause corrosion; chlorides, such as chlorine, hydrogen chloride, are highly corrosive; fluorides, such as sulfur hexafluoride, hydrogen fluoride, can attack the wafer; carbon dioxide, which affects material properties; moisture content: can cause electrostatic reaction and reduce the manufacturing yield; ammonia gas, which causes chemical pollution; volatile organics: such as formaldehyde, acetone, etc., can contaminate the wafer surface; the particles, various dust particles, can cause mechanical damage. Gas concentration refers to the concentration of a particular gas in an environment, typically expressed in ppm (parts per million), ppb (parts per billion) or ppt (parts per trillion) per unit volume, for measuring the trace or ultra trace presence of the gas. The gas control module is a computerized device or system for managing and controlling the gas treatment system. It receives gas data from the gas monitoring module and controls the gas treatment steps according to predetermined algorithms and logic, for example, to initiate different levels of filtration strategies to ensure gas cleanliness. The gas control module may include sensor interfaces, control algorithms, actuator control, and communication functions with other systems.

S200, the temperature and humidity monitoring module collects temperature and humidity data comprising temperature and humidity and sends the temperature and humidity data to the temperature control module; a temperature control module is a device or system for monitoring and regulating the temperature in an environment. In the application, the main task of the temperature control module is to receive temperature data from the temperature and humidity monitoring module and take measures to maintain or adjust the temperature of the environment according to the data. Typically, the temperature control module includes a temperature sensor, a control algorithm, and an actuator (e.g., a heater or cooler) that can control the temperature in the environment based on measured temperature data. In semiconductor manufacturing, many processes and equipment are very temperature sensitive, and wafer fabrication processes need to be performed within a specific temperature range to ensure product quality and performance.

S300, the gas control module receives the gas data, judges whether the gas concentration or the gas composition exceeds a threshold value, if so, performs gas filtration according to a preset first gas filtration strategy, and feeds back the pressure value and the gas composition in the gas filtration to the main control module; wherein the first gas filtering strategy is a core of the EFEM control method comprising: first stage prefiltering: when the particle size of the contaminants in the gas is greater than 5 μm, the gas enters through the outside of the cartridge at a velocity of 0.1 m/s. At this level, a 5 μm pore size polypropylene filter material was used which screens the gas from inside to outside to remove particulate impurities greater than 5 μm from the gas. Second-stage efficient filtration: when the particle size of the contaminants in the gas is between 0.3 μm and 5 μm, the gas enters through the inside of the cartridge at a velocity of 0.05 m/s. This grade first removes fine particles of 0.3 μm to 5 μm through a 0.3 μm glass microfiber filter layer, and then adsorbs volatile organic contaminants in the gas through an activated carbon layer. Third stage activated carbon adsorption: when the concentration of volatile organic contaminants in the gas is higher than 10ppb, the gas passes through a two-layer activated carbon filter cartridge. The activated carbon layers adopt oval particles and have specific surface areas of up to 1700m < 2 >/g so as to efficiently adsorb volatile organic pollutants in gas. Fourth stage gas purification: when the residual contaminant concentration in the gas is higher than 100ppt, the system introduces a non-thermal plasma excited by the sensing electrode. The oxygen ions in this plasma are used to oxidatively decompose residual contaminants in the gas, thereby further improving the cleanliness of the gas. Fifth-stage negative pressure adsorption: after the first four stages of filtration, if residual contaminants are still present in the gas, the system will apply a negative pressure of 0.05 MPa. This negative pressure helps to enhance the adsorption capacity of the activated carbon, further decontaminating the gas of residual contaminants, and ensuring the cleanliness of the ambient gas. The method of the EFEM gas filtering strategy adopts a multi-stage filtering and adsorbing technology, and ensures high cleanliness and stable temperature and humidity environment in the wafer manufacturing process. By using a machine learning model and a linkage judgment mechanism, the system can adaptively adjust the filtering and temperature and humidity control strategies according to real-time data, so that the quality and yield of wafer manufacturing are improved.

S400, the temperature control module receives temperature and humidity data, calculates the deviation between the received temperature and humidity data and the target temperature and humidity, generates a first temperature control strategy according to the deviation through a PID algorithm, adjusts the temperature and humidity according to the first temperature control strategy, and feeds back the temperature adjustment times and the humidity adjustment times in the adjustment process to the main control module; the specific technical scheme is as follows.

S500, the main control module receives the gas data and the temperature and humidity data, and calculates a gas stability evaluation index KPI1 and a temperature and humidity stability index KPI2; judging whether the index exceeds a threshold value, if so, calling a second gas filtering strategy generated by the SVM model, calling a BP network model to generate a second temperature control strategy, and respectively transmitting the second temperature control strategy to the gas control module and the temperature control module; the specific technical scheme is as follows.

S600, the gas control module receives and filters the gas according to a second gas filtering strategy; the temperature control module receives and adjusts the temperature and the humidity according to a second temperature control strategy; the specific technical scheme is as follows.

And S700, the main control module judges whether the gas stability evaluation index KPI1 and the temperature and humidity stability index KPI2 reach the box opening condition, and if so, sends a box opening instruction to the manipulator module.

In summary, the gas monitoring module comprehensively monitors the gas components in the EFEM case to obtain data such as gas concentration; the temperature and humidity monitoring module acquires temperature and humidity parameters. The gas control module purifies gas by utilizing a multistage filtering strategy and feeds data such as pressure back to the main control module; the temperature control module adopts a PID algorithm to accurately adjust the temperature and the humidity, and feeds back data such as adjustment times. The main control module calculates the gas cleanliness and temperature and humidity stability indexes, and generates a new gas filtering strategy and a new temperature control strategy by optimizing a machine learning model, so that the box opening environment is optimized. When the gas cleanliness and the temperature and humidity index reach the standards, a box opening instruction is given, and the box opening is executed by the manipulator module. The box opening method effectively improves the yield of wafer manufacturing through accurate gas monitoring and control, temperature and humidity monitoring and control, intelligent machine learning optimization and linkage judgment mechanisms.

FIG. 2 is an exemplary flow chart of processing acquired gas data according to some embodiments of the present disclosure, where the current EFEM open box control system cannot obtain high accuracy gas monitoring data, resulting in poor accuracy of gas cleanliness control and reduced wafer manufacturing yield, and where the gas monitoring module of the present application acquires gas data comprising gas concentration and gas composition and sends the gas data to the gas control module as shown in FIG. 2, the steps of:

S110 first, a plurality of sets of gas sensors are installed in the EFEM open box control system. These sensors are capable of collecting gas data comprising gas concentrations and various gas components. Specifically, a sensor array network is formed by installing a plurality of types of gas sensors such as a thermal sensor, a gas sensor and the like at key positions such as an air inlet, an air outlet, a box door periphery and the like of the EFEM; by arranging multiple sets of gas sensors, gas data can be acquired from different positions to comprehensively monitor the gas conditions inside the EFEM. Meanwhile, the multiple types of sensors can mutually verify, so that the data reliability is improved; by adopting a sensor array network and combining a data fusion algorithm, the monitoring data which accurately reflects the condition of the whole EFEM gas can be obtained, and support is provided for subsequent gas cleanliness control.

The gas data collected in S120 is input into a pre-established data fusion module. The data fusion module is based on a particle filtering algorithm, and can process and analyze the acquired gas data and output the fused gas data. This step helps to improve the accuracy and reliability of the data.

The fused gas data is further input into the data correction model S130. The data correction model uses a preset dynamic calibration strategy to calibrate the gas data so as to ensure the accuracy and reliability of the data. The calibrated gas data will be passed on to the next step.

S140 in the present application, the gas control model is established based on an MRAC (model reference adaptive control) algorithm. It uses the calibrated gas data to dynamically update the control parameters of the first gas filtration strategy. The self-adaptive control method can ensure that the gas cleanliness is controlled and kept in an ideal state, thereby improving the yield of wafer manufacturing;

and S150, finally, the calibrated gas data is sent to a gas control module to realize accurate gas cleanliness control, so that the manufacturing yield of the wafer is effectively improved. The technical scheme of the application solves the problem of inaccurate gas monitoring data in the traditional EFEM box opening control system, ensures that the gas cleanliness and the temperature and the humidity of the box opening environment can be stabilized in an ideal state, and finally achieves the aim of obviously improving the wafer manufacturing yield.

In summary, by providing multiple sets of gas sensors, a comprehensive monitoring of the internal gas state of the EFEM is achieved. And the particle filter algorithm is used for data fusion, so that the accuracy and reliability of gas data are effectively improved. And the dynamic calibration strategy is applied to calibrate the gas data, so that the data precision is further improved. Based on the calibrated high-quality gas data, the gas filtering strategy is optimized and updated, and the accurate control of the gas cleanliness is realized. And finally, outputting the calibrated and controlled optimized gas data, and providing support for controlling the gas cleanliness. By multiprocessing and controlling the gas monitoring data, the application obviously improves the quality and precision of controlling the gas cleanliness in the process of opening the EFEM, effectively solves the problem of inaccurate gas monitoring in the prior art, and achieves the purpose of improving the wafer manufacturing yield.

Specifically, the step of fusing the collected gas data includes: the gas data collected by the gas sensor is used as the observed quantity of a particle filtering algorithm, and the gas sensor type comprises an infrared gas sensor and a gas chromatograph. Particle filtering requires observables for importance sampling and weight updating; the application uses the gas data obtained by the infrared gas sensor and the gas chromatograph as the observed quantity; the infrared gas sensor can rapidly collect gas concentration. The gas chromatograph can accurately give the concentration of each gas component; the data of the two types of sensors are integrated to be used as the observed quantity of particle filtering, so that the advantages of the two types of sensors can be integrated; the data frequency of the infrared sensor can ensure the timeliness of the observed quantity. The data precision of the gas chromatograph is beneficial to improving the observation effect; the fusion of the multi-source heterogeneous observables enables the particle filtering to observe richer and more accurate gas state information; this is advantageous for the convergence speed and estimation performance of the particle filter algorithm.

The particle state is initialized by using a sampling method based on the Kullback Leibler divergence, and the sampling method based on the KL divergence is adopted, wherein the Kullback Leibler divergence (Kullback Leibler divergence), also called relative entropy (relative entropy), is an index for describing the difference of two probability distributions, and the sampling result of the optimal approximate target distribution can be obtained by minimizing the KL divergence of the target distribution and the sampling distribution, so that the initial particle state conforming to the posterior probability distribution is generated, and a foundation is laid for the subsequent Kalman prediction. In particle filtering, it is necessary to initialize a first particle state to reflect the posterior probability distribution of gas concentration; the conventional random sampling method may not reflect the probability distribution property of the gas concentration well; the importance sampling method based on the Kullback Leibler divergence is adopted in the application; the method comprises the steps of converting gas sensor data into probability density functions, and setting the probability density functions as target distribution; calculating KL divergence of the sampling distribution and the target distribution as an objective function of sampling optimization; obtaining the sampling distribution with the smallest KL divergence through iterative sampling and weighting; the sampling distribution is used for initializing the first particle state, so that the posterior probability distribution of the gas concentration can be reflected better; compared with random sampling, the first particle state generated by the method is more representative and effective; a good foundation is laid for subsequent particle filtering, and the accuracy of gas concentration estimation is improved.

And carrying out state prediction on the first particle state by applying an improved Kalman filtering method capable of carrying out online self-adaptive adjustment on the noise covariance of the process, and outputting a predicted state. In EFEM open-box gas detection, the gas state estimation is affected by process noise, and direct application of the traditional Kalman filtering algorithm can deviate predictions from a true state. The first particle state is gas concentration state distribution obtained by initializing a particle filtering algorithm; carrying out state prediction by applying Kalman filtering to the model, so as to obtain an ideal prediction state; but the prediction effect depends on the value of the process noise covariance matrix Q; the fixed Q is difficult to adapt to the change of the box opening environment; the application adopts an online self-adaptive Q matrix adjustment method; according to the Kalman prediction error, continuously iterating and optimizing Q by using a Gaussian Newton method; the self-adaptive adjustment enables the Q matrix to reflect the statistical rule of the process noise in real time; therefore, the state prediction effect of the Kalman filtering is better, and the prediction state is more accurate; compared with fixed Q, the method improves the accuracy and the robustness of prediction; is beneficial to the subsequent gas state control and realizes the purpose of improving the wafer manufacturing yield.

When the particle weight is calculated, spatial constraint knowledge of a CFD gas propagation model and depth characteristics of gas data extracted by a convolutional neural network are introduced to be used as priori information of the calculated weight. In the EFEM open box environment, due to the limitation of a gas sensor, errors exist in directly calculating particle weights from sensing data, and the spatial constraint relation and depth feature priori knowledge based on a CFD model, which are introduced when the particle weights are calculated, are interpreted by the effect of reducing particle filtering: the CFD gas propagation model can simulate the flow rule of gas in the open box environment; the model predicts the gas concentration distribution of different positions, and establishes the space correlation constraint between the gas concentrations; the accuracy of weight calculation in the particle filter algorithm can be improved by using the priori constraint knowledge; the convolutional neural network can extract representative depth features in the gas data and reflect the internal rules of the data; the depth characteristics are used as priori knowledge of particle weight calculation, so that the rationality of weight distribution can be improved; when the weight is calculated, the space constraint and depth feature priori of the CFD model are simultaneously utilized to carry out constraint optimization; therefore, the effectiveness of the particle state can be greatly improved, and the actual gas distribution can be reflected more accurately; therefore, in the subsequent particle filter fusion, more accurate and reliable gas concentration estimation can be obtained; effectively improves the gas monitoring and controlling effect and improves the wafer manufacturing yield.

Based on the depth features, the particles are sampled using an Auxiliary PARTICLE FILTER algorithm, generating a second particle state. The Auxiliary PARTICLE FILTER (APF) algorithm and the Auxiliary particle filtering algorithm are nonlinear filtering methods for estimating states, and are particularly suitable for the situation of noise with nonlinear dynamic systems and non-Gaussian distribution. In the particle filter algorithm, the problem of particle degradation can occur due to uneven weight, and the state estimation performance is reduced; to suppress particle degradation, the process of resampling the particle state by the Auxiliary PARTICLE FILTER (APF) algorithm is explained in the present application: resampling of particles is required in particle filtering to reduce particle degradation problems. Standard resampling methods tend to result in sample redundancy; the APF algorithm introduces auxiliary variables, namely depth characteristics and importance sampling, during resampling; in the gas monitoring system, the depth features are representative features in the gas sensor data extracted through a CNN model; the APF designs new importance weight distribution by utilizing depth characteristics, and resamples the particle state according to the weight; therefore, the second particle state generated after resampling can keep more effective particles, and sample redundancy is avoided; the newly generated second particle state can more comprehensively reflect the gas concentration distribution information, and is favorable for state correction tracking of Kalman filtering; APF resampling and depth feature extraction are combined, so that the core innovation of multi-source heterogeneous gas data fusion is realized; compared with the traditional method, the method improves the calculation efficiency and the state estimation precision of the particle filtering, is beneficial to improving the performance of subsequent gas control, and achieves the purpose of improving the wafer manufacturing yield.

Correcting the state of the second particles by using a Kalman filtering algorithm, outputting the corrected state, and improving the precision of state estimation; kalman filtering is a recursive algorithm that can optimally predict and correct noisy system conditions. In gas data fusion, the second particle state is the gas concentration particle state obtained by resampling. The Kalman filtering is applied to the particle state estimation method for correcting the particle state, so that noise in the particle state can be eliminated, and a more accurate state estimation value can be output.

Carrying out weighted average on the corrected second particle state to obtain a fusion value of the gas concentration; the corrected second particle state reflects the gas concentration more accurately, and the weighted average is carried out on the particle states, so that the integral fusion value of the gas concentration can be obtained, and the data fusion of multiple sensors is realized. In the application, after resampling by a particle filter algorithm, a second set of particle states { x1, x2,., xn }, representing n gas concentration state particles, is obtained; carrying out Kalman filtering correction on each state particle xi to remove noise and obtain corrected states { x1', x2',..xn ' }; calculating the weight wi of each state particle, wherein the weight is determined according to the prior probability of the particle, namely the prior estimation of gas concentration distribution; carrying out normalization processing on the weight wi to obtain a normalized weight wi'; and (3) carrying out weighted average by using the normalized weight wi 'and the corresponding correction state xi' to obtain a fusion estimated value of the gas concentration: x_fusion= Σi=1 nwi '·xi'; the application comprehensively utilizes the information of a plurality of particle states, absorbs the data advantages of sensors at different positions, can effectively eliminate the influence of noise and improves the accuracy of gas concentration estimation; compared with a single sensor, the data of the multi-source heterogeneous sensor is used for realizing fusion, so that more reliable and accurate gas concentration values can be obtained, and stable and dependable state input is provided for subsequent gas control.

And according to the fusion value error, adopting a Gauss Newton method to adaptively adjust the process noise parameters in the Kalman filtering. And (3) comparing the fused concentration value with the original sensor data, and when the error is overlarge, adjusting parameters of a process noise covariance matrix in a Kalman filtering algorithm. The gauss newton method can effectively perform parameter optimization. The process noise covariance matrix Q in Kalman filtering is a key parameter, which reflects the statistical characteristics of the system process noise; in a gas monitoring system, Q directly affects the tracking effect of a Kalman filtering algorithm on a gas state. If the Q is set improperly, the error of the filtering result is larger; the on-line iterative optimization can be carried out on Q by adopting a Gaussian Newton method, and an objective function is set as the mean square error between a Kalman filtering result and actual observation of a sensor; by solving the Gaussian Newton equation of the objective function, the optimized search direction of Q can be obtained. Determining a proper step length by using a line searching method, and iteratively updating the Q matrix; when the observation error is reduced below a threshold value, the Q is optimized to a proper value, and the process noise statistical characteristics are accurately reflected, so that the state estimation robustness of Kalman filtering is improved; the whole optimization iterative process runs on line, Q can be adaptively adjusted to adapt to the change of the gas monitoring environment, and the Kalman filtering effect is ensured; compared with the traditional constant value setting Q, the self-adaptive adjustment method can remarkably improve the accuracy of gas state estimation and prediction, thereby being beneficial to subsequent gas control and achieving the purpose of improving the wafer manufacturing yield; the adaptive adjustment of the process noise parameters is realized, and the robustness of Kalman filtering state estimation is improved.

The above procedure is repeated, and the fused gas concentration and gas composition are continuously output. By means of the parameter self-adaptive adjustment, the accuracy of the Kalman filtering algorithm on gas state estimation can be improved, subsequent gas monitoring and control are facilitated, and the purpose of improving the wafer manufacturing yield is achieved. By fusing a plurality of advanced algorithms, the method can effectively improve the accuracy and the adaptivity of gas data fusion, provide a stable and reliable data source for subsequent gas control, and meet the high-precision requirement of wafer manufacturing.

Specifically, the conventional EFEM box-opening environment control has the problems of unstable gas cleanliness and large temperature and humidity fluctuation, which directly causes defects on wafers in manufacturing and reduces the product yield. While the accuracy of the gas data has a direct impact on the quality of the environmental control. The gas sensor can generate signal drift phenomenon after long-time work, so that errors exist in measurement, and further the accuracy of gas monitoring is reduced, and the subsequent gas control effect is affected. In the application, the fused data is input into a data correction model, the data correction model corrects the gas data by utilizing a preset dynamic calibration strategy, and the step of outputting the corrected gas data comprises the following steps:

First, raw gas data is acquired from one or more gas sensors. These sensors may be devices dispersed in the EFEM system to monitor gas compositions and characteristics in the environment; the data generated by the gas sensor is typically time series data, recording gas measurements over a period of time. These data include time stamps and gas concentration values. The timing analysis algorithm analyzes the pattern of changes in the gas data to detect signal drift. Signal drift refers to systematic deviation of sensor output over time. By pattern recognition and trend analysis of the gas data, the algorithm can identify the type, frequency and amplitude of signal drift. This helps to understand the instability of the sensor data. Based on the analysis results, the timing analysis algorithm will construct a signal drift model. This model describes how the signal drift varies with time, including the direction and speed of the drift. Providing important information for subsequent calibration and control steps. This helps to ensure that the gas cleanliness and temperature and humidity in the open box environment remain at ideal conditions, thereby improving the manufacturing yield of wafers.

According to the acquired signal drift mode, calculating zero calibration parameters and sensitivity calibration parameters of the gas sensor by adopting an incremental differential algorithm; the signal drift pattern of the gas sensor obtained in the above. This pattern describes the variation of the sensor output signal over time, including the nature and trend of drift. The step adopts an incremental differential algorithm to calculate zero calibration and sensitivity calibration parameters. The incremental differential algorithm is a method for real-time data processing and calibration that combines the sensor's drift pattern and gas measurement data to determine calibration parameters. The incremental differentiation algorithm first selects an appropriate reference point, typically the value of the output of the gas sensor at steady state. Based on the signal drift pattern and the measurement data, an algorithm calculates parameters for zero calibration. These parameters will be used to adjust the sensor output to eliminate zero drift. The incremental differentiation algorithm also analyzes the trend of the signal drift to determine the nature of the drift, such as linear drift or nonlinear drift. Based on the drift trend and the measurement data, the algorithm calculates parameters for sensitivity calibration. These parameters will be used to adjust the sensitivity of the sensor to eliminate measurement errors caused by signal drift.

Calculating compensation control quantity of the gas data according to the acquired gas data, the zero calibration parameter and the sensitivity calibration parameter; the compensation control quantity is calculated by using the acquired gas data and the calibration parameters calculated previously. The purpose of this control is to adjust the control elements in the system to maintain the gas cleanliness and temperature and humidity of the open box environment at ideal conditions. Zero compensation can be performed on the acquired gas data using zero calibration parameters to eliminate errors caused by sensor zero drift. This ensures baseline accuracy of the gas measurement. With the sensitivity calibration parameters, sensitivity compensation can be performed to correct measurement bias due to sensor sensitivity errors. This helps to ensure that the sensitivity of the gas measurement is consistent with ideal conditions. The compensation control quantity for controlling the gas environment can be calculated according to the collected gas data and the calibration parameters. The accurate calculation and application of the control amounts are beneficial to keeping the gas cleanliness and the temperature and the humidity of the box opening environment in an ideal state, so that the manufacturing yield of the wafer is effectively improved.

And performing closed-loop PID control on the acquired gas data by using the obtained compensation control quantity to generate calibrated gas data. Core principle of PID control: the proportion P, the proportion control part adjusts the control output according to the error between the current measured value and the target value. In this case, the error is the difference between the acquired gas data and the target gas data. The integral I, the integral control part considers the situation that the error is accumulated with time, so as to eliminate the static error of the system. The differential D, differential control section is used for predicting the future change trend of the error to prevent the overshoot of the system. By combining the compensation control amount with the PID control method, the collected gas data can be accurately calibrated and adjusted. The calibrated gas data represents the gas quality and characteristics in the environment, and has been processed for zero crossing calibration and sensitivity calibration. These data more accurately reflect the actual gas conditions in the environment, thereby improving wafer fabrication yield. The closed-loop PID control is a continuous process, and continuously adjusts and controls output according to gas data acquired in real time so as to keep the gas cleanliness and the temperature and humidity in the environment in an ideal state. This continuous control ensures the stability and consistency of the wafer fabrication process.

In conclusion, through carrying out dynamic calibration to the gas monitoring data, detection errors caused by sensor signal drift can be eliminated, the accuracy of the gas data is ensured, and an accurate control basis is provided for subsequent gas control, so that the gas control effect is improved, the EFEM box opening environment achieves the required gas cleanliness, the influence on wafer manufacturing is reduced, and the product yield is improved.

Specifically, the preset gas filtering strategy cannot be updated in real time, and when the environmental parameters change, the environmental parameters cannot be correspondingly adjusted, so that the gas filtering effect is reduced, therefore, the method inputs calibrated gas data into the gas control model, the gas control model is established based on the MRAC algorithm, and the step of updating the first gas filtering strategy by using the calibrated gas data comprises the following steps:

Constructing a gas control model based on an MRAC algorithm, wherein the gas control model comprises a preset first gas filtering strategy F1; MRAC (model reference adaptive control) algorithm, in the present application, the application of the MRAC algorithm includes: constructing a gas control model comprising a first gas filtration strategy; inputting calibrated gas data, adapting to model parameters, and generating a new filtering strategy; calculating errors of a preset strategy and a new strategy; adjusting control parameters of an actual filtering system, namely a first gas filtering strategy, according to the errors; and circularly executing the steps to realize the self-adaptive optimization of the gas filtering strategy. The application of the MRAC algorithm can enable the gas filtering strategy to be continuously optimized and adapted to environmental changes, so that the gas purifying effect is improved, the stability of the gas cleanliness inside the EFEM is ensured, and the yield of wafer manufacturing is improved.

Inputting the calibrated gas data into a gas control model, and generating a filtering strategy F1' based on an RLS algorithm; the RLS algorithm is a self-adaptive filtering algorithm and is mainly used in the field of on-line identification and control, and in the application, the RLS algorithm is applied to: in the gas control model, the gas control model is used for recursively determining model parameters based on the latest calibrated data. In the recursive process, parameter estimation is optimized by the least squares method. Thus, the model can be updated in real time, and a new gas filtering strategy is output. The application of the RLS algorithm enables the generation of the gas filtering strategy to be more flexible and intelligent, and can be quickly adapted to data change, so that the gas purifying effect and the EFEM environment control quality can be further improved.

Calculating errors of the first gas filtering strategy F1 and the filtering strategy F1'; in the application, a group of fuzzy rules are established based on fuzzy concepts, and the application degree of the filtering strategy under different environmental conditions is described. And performing fuzzy logic operation by using the fuzzy rule and the blurred data to obtain the relative performance of two filtering strategies F1 and F1'. The relative properties of F1 and F1' are compared and the error between them is calculated to quantify their performance differences.

Based on the self-adaptive control algorithm, calculating a control input quantity delta u of gas filtration according to the error; the self-adaptive control algorithm means that the control system can adjust control parameters according to environmental changes and changes of self states, so as to realize optimal control of the process; in the present application, the application of the adaptive control algorithm includes: comparing the error between the first gas filtering strategy F1 and the updated filtering strategy F1'; according to the error, the control input quantity Deltau of the gas filtration is calculated in real time by adopting an on-line calibration method. Deltau reflects the effect of the current environmental change on gas filtration. And correspondingly adjusting control parameters of a filtering device in the first gas filtering strategy, such as air quantity, air pressure and the like, according to the delta u. Thus, the self-adaptive optimal control of the gas filtering process is realized. The key to adaptive control is error feedback and control input calculation. Compared with the traditional fixed control, the air filter can make the air filter respond to environmental changes quickly, and ensure the air purifying effect, thereby improving the stability of the air cleanliness of the EFEM.

Adjusting the first gas filtering strategy according to the control input quantity delta u; Δu is a control amount obtained by calculation in real time, and can accurately reflect the influence of the current environmental change on gas filtration. Specific control parameters in the filtering strategy, such as air volume, air pressure intensity and the like, are correspondingly adjusted according to the value of Deltau, so that the self-adaptive optimization of the air filtering on the environmental change is realized.

The MRAC algorithm is recursively executed to update the first gas filtering strategy. An iteration process of the MRAC algorithm of the above step. The gas control system may periodically recursively execute the MRAC algorithm, input the most current gas data, and regenerate the optimized filtering strategy. Through continuous recursion optimization, the first gas filtering strategy is updated in real time along with environmental changes, so that the continuous improvement of the gas filtering effect is ensured, and the cleanliness of the environmental gas in the EFEM is stabilized in an optimal state. The dynamic optimization generation of the gas filtering strategy is realized through a recursive MRAC algorithm, and the method is one of key technical means for improving the control effect of the EFEM gas cleanliness.

In summary, the MRAC algorithm is applied to enable the gas filtering strategy to be adaptively adjusted so as to correspond to environmental changes, so that the gas filtering system can still provide an optimized and accurate filtering effect under the condition of fluctuation of environmental parameters, thereby improving the gas cleaning quality of the environment in the EFEM, reducing the influence on wafer manufacturing and improving the product yield.

FIG. 3 is an exemplary flow chart for gas filtration using a first gas filtration strategy, as shown in FIG. 3, according to some embodiments of the present description, for establishing a gas threshold database within a gas control module. The database is used for presetting thresholds of various gas components so as to judge the gas concentration or the gas components in the real-time monitoring process.

Step three, the gas control module receives the gas data, judges whether the gas concentration or the gas composition exceeds a threshold value, if so, performs gas filtration according to a preset first gas filtration strategy, and feeds back the pressure value and the gas composition in the gas filtration to the main control module, wherein the step comprises the following steps:

S310, a gas threshold database is arranged in the gas control module, and thresholds of various gas components are preset in the database; in the present application, the database may be: the database storage medium, the gas threshold database may be stored in a non-volatile memory of the gas control module, such as flash memory, solid state disk, and the like. The database structure can adopt a relational database structure, comprises a gas composition table, a threshold value table and the like, and establishes an association relationship between the gas composition table and the threshold value table. And selecting key gas components such as particles, various corrosive gases and the like which influence the wafer manufacturing yield according to the gas component range and the requirements of the process flow and the product on the environmental cleanliness. And determining a threshold value, namely determining the concentration limit value of each gas component according to product quality standards, environment monitoring specifications and the like, so that the process requirements are met, and a certain allowance is considered. And maintaining the database, namely backing up the database according to a rule, providing an interface for maintaining and updating the threshold value, and ensuring the reliability of the database. The design makes the gas threshold management systemized, and is helpful for the gas control module to judge the accuracy of gas exceeding standard, thereby ensuring the stability of the gas cleanliness inside the EFEM.

S320, the gas control module receives the gas data output by the gas monitoring module; in the EFEM open box control system, a gas control module needs to control the gas cleanliness according to the gas data collected by a gas monitoring module; the gas monitoring module monitors and collects data such as gas concentration, gas composition and the like, and sends the data to the gas control module in the form of digital or analog signals; the gas control module receives the gas data output by the gas monitoring module, and obtains the gas data signal sent by the gas monitoring module through a communication interface or an input module in the gas control module.

S330, judging whether the gas data exceeds a threshold value by utilizing a fuzzy control algorithm according to the threshold value of the gas components in the database, and if so, performing gas filtration according to a preset first gas filtration strategy; fuzzy control is a control method for handling fuzzy, uncertainty and complexity problems. It can generate fuzzy outputs for decision making and control based on fuzzy rules and fuzzy sets of input data. And (3) carrying out fuzzy comparison on the gas monitoring data and standard values in a threshold database to obtain language description of gas conditions, wherein a fuzzy rule base stores experience control rules such as 'if the concentration is slightly higher', and the fuzzy rule base is called to carry out reasoning according to a fuzzy judgment result to obtain language control quantity of gas filtering. Finally, the language control quantity is converted into a determined numerical control quantity so as to adjust the gas filtering system. The fuzzy control considers the influence of errors in gas sensing, avoids the problem caused by simple Boolean judgment, improves the reliability of judgment, and is beneficial to the stable control of a gas filtering system.

S340, monitoring the gas pressure by adopting an MEMS pressure sensor, and inputting the monitored gas pressure into a gas flow intelligent regulation device; MEMS pressure sensor MEMS is an abbreviation of micro-electro-mechanical system, and the MEMS pressure sensor is integrated with an electronic circuit through mechanical components, so that high-precision and rapid-response pressure measurement can be realized. In the application, the MEMS pressure sensor is used for monitoring the pressure change in the gas filtering system in real time, and the application has the advantages of small volume, high precision and sensitive response. The intelligent gas flow regulating device can adopt a PID control algorithm, calculate the regulating variable of the gas flow according to the monitoring data of the MEMS pressure sensor, and control a regulating valve to regulate the flow of the gas in the filtering system in real time so as to ensure the pressure stability of the gas filtering process. PID regulation realizes closed-loop control of gas flow, can prevent pressure abnormality, and can optimize the flow of gas in the filtering system, thereby improving the gas purifying effect.

S350, the intelligent gas flow regulating device adopts a self-adaptive PID algorithm to control the electronic radio frequency valve so as to regulate the gas flow. The self-adaptive PID algorithm can dynamically adjust PID parameters according to the real-time change of the gas flow and noise interference, so that the gas flow control is optimized, and the proportional, integral and differential coefficients of the PID controller are adjusted in real time by an online calibration method, so that the self-adaptive PID algorithm adapts to the change of the gas parameters, and the flow control precision is improved. The electronic radio frequency valve is a fast-responding precise gas flow regulating device. The method adopts radio frequency electromagnetic waves to heat and ionize the ionized gas, thereby rapidly opening or cutting off a gas flow channel and realizing fine modulation of gas flow. Compared with the traditional mechanical valve, the valve has the characteristics of short response time and accurate flow regulation. The application integrates the self-adaptive PID algorithm and the electronic radio frequency valve to realize the closed-loop accurate control of the gas flow. The PID algorithm generates a control signal to drive the opening degree of the electronic radio frequency valve to change, and the gas flow is regulated in real time to enable the gas flow to reach the set point rapidly.

FIG. 4 is an exemplary flow chart for regulating temperature and humidity using a first temperature control strategy according to some embodiments of the present disclosure, including, as shown in FIG. 4:

S410, acquiring temperature and humidity data of the collected gas; s420, calculating the deviation e (t) between the acquired temperature and humidity data and the target temperature and humidity according to an incremental PID algorithm; the incremental PID algorithm is improved and designed aiming at the characteristics of temperature and humidity control. The PID output is calculated according to the tiny variation of the temperature and the humidity in each sampling period instead of directly using the absolute temperature and humidity value. This can effectively suppress large oscillations due to measurement noise. The increment PID can realize high-precision temperature and humidity tracking control. The following factors need to be comprehensively considered in the setting of the target temperature and humidity: the requirement of the open box environment on the wafer material is 22+/-0.1 ℃; the refrigerating temperature range of the equipment liquid cooling system; spray technical parameters of the humidification system; heat recovery and humidity control functions of the precise air conditioning system. Therefore, the target temperature and humidity of the temperature control system can be reasonably set on the premise of meeting the requirements of the wafer process, and the optimal embodiment of the application is set at 22 ℃ and 45%RH. The over-increment PID algorithm accurately controls the gas to reach the set target temperature and humidity, so that the stability of the box opening environment can be improved, the thermal stress and electrostatic damage of the wafer are reduced, and the yield of wafer manufacturing is effectively improved.

S430, generating a temperature control output u (t) by utilizing a PID algorithm according to the deviation e (t); e (t) represents the deviation between the acquired temperature and humidity data and the set target temperature and humidity, and is also called the input quantity of a control system; the PID controller takes e (t) as input, and generates a control output u (t) through three calculation of proportion P, integral I and differential D; the proportional term P is controlled according to the real-time value of the deviation e (t) to enable the temperature control system to reach the target; the integral term I eliminates the steady-state error of the system according to the integral sum of the deviation e (t); differentiation item D: predicting a control trend according to the change rate of the deviation e (t), and improving dynamic response; the three weights are linearly combined to form a final control output u (t) for driving the executing mechanism; u (t) is converted into an analog voltage signal through a DAC, and a three-way heat exchange valve is controlled to accurately adjust the temperature and the humidity; through PID calculation, the temperature and humidity closed-loop control with high precision is realized according to the real-time dynamic adjustment of the deviation e (t), so as to adapt to the change of the external environment and improve the temperature and humidity stability of the opened box.

S440, u (t) is converted into a voltage signal of 0V to 10V through a DAC (digital-to-analog converter), and the voltage signal is used as a control signal of the three-way heat exchange valve; u (t) is a control quantity output by a digital PID algorithm, and represents a control quantity of the three-way heat exchange valve, and the range is 0-1000; the digital quantity u (t) is converted into an analog voltage signal of 0-10V to drive the three-way heat exchange valve; converting by using a digital-to-analog converter (DAC) circuit; the DAC circuit is composed of a precision resistor voltage division network, and u (t) is mapped to different voltage outputs; when u (t) =0, the corresponding voltage output is 0V; when u (t) =1000, it corresponds to 10V; the 0V signal indicates that the three-way valve is fully open to the heater; 10V indicates full on to the refrigerator; the intermediate voltage represents different opening ratios, and the quantity of the gas passing through the cold and hot device is accurately controlled; in the EFEM open box temperature and humidity control system, a digital PID controller (i.e. DIGIPID controller) is adopted to realize accurate temperature and humidity regulation. Compared with an analog PID controller, the digital PID controller can perform high-speed digital signal processing, and control accuracy and response speed are improved. Meanwhile, the digital PID controller can also modify control parameters as required to realize self-adaptive control. In the application, DIGIPID controller outputs PWM control signal of digital quantity according to temperature and humidity feedback. Then the opening degree of the three-way heat exchange valve is accurately regulated by converting the control signal into a 0-10V control signal of analog quantity through an AC (digital-analog converter). Therefore, the conversion from digital control quantity to analog execution quantity is realized by utilizing the high-speed digital control advantage of DIGIPID controllers and the high-precision conversion of the DAC, so that the temperature and humidity control realizes high stability, high precision and high response speed.

S450, the three-way heat exchange valve accurately adjusts the proportion of gas flowing through the electric heater and the Peltier refrigerator, and controls the temperature and the humidity of the gas; the three-way heat exchange valve is a precise flow regulating device and can distribute gas to flow through the heater and the refrigerator in proportion. A Peltier cooler, which is a thermoelectric refrigerating device in a solid state, for achieving the purpose of refrigeration by absorbing and releasing heat using a primary effect; the basic structure comprises two semiconductor materials with different conductivity types, and the absorption or release of heat, namely refrigeration or heating, can be realized by changing the current direction; in the EFEM open box control system, a Peltier refrigerator and an electric heater are used together, and the temperature of gas is accurately controlled through flow regulation of a three-way heat exchange valve; compared with a common refrigeration compressor, the Peltier refrigerator has the advantages of small volume, quick response and no moving parts, and is very suitable for accurately controlling the low temperature of gases such as air and the like; therefore, the application selects the Peltier refrigerator as the refrigerating equipment, and is matched with the electric heater for use, thereby realizing the rapid and accurate adjustment of the gas temperature in the EFEM and further ensuring the temperature and humidity stability of the box opening environment. The opening and proportion of the three-way valve are regulated by 0-10V control voltage, and the voltage is output from the PID controller; when the voltage is 0V, the three-way valve is fully opened to the heater, and the gas is fully heated by flowing through the heater; when the voltage is 10V, the three-way valve is fully opened to the refrigerator, and the gas is fully cooled by flowing through the refrigerator; in the case of intermediate voltage, the opening ratio of the three-way valve is different, and the quantity of gas split flow passing through the heater and the refrigerator is different; the heat exchange quantity of the gas can be accurately controlled by accurately controlling the split ratio, and the gas can be adjusted to the target temperature and humidity in real time; the quick response and the high stability control of the temperature and the humidity of the gas are realized by utilizing the variable opening degree proportional control of the three-way heat exchange valve; the control accuracy of the three-way valve directly influences the regulation accuracy and control stability of the temperature and the humidity of the gas.

S460, acquiring the temperature adjustment times and the humidity adjustment times in the temperature control process by adopting an electronic counter. The application adopts a 16-bit timing counter built in an STC89C52 singlechip as an electronic counter; in the temperature control process, when the electric heater works and the temperature is adjusted once, the temperature adjustment frequency counter is increased by 1; when the refrigerator works and the humidity is adjusted once, the humidity adjustment frequency counter is increased by 1; the singlechip counts the starting times of the electric heater and the refrigerator in real time by detecting the working signals of the electric heater and the refrigerator; the counter uses a timing/counter function register group of the singlechip, can automatically count, and automatically returns to zero after overflowing; outputting an instruction through a serial port of the singlechip, and regularly reading the current value of the counter to obtain the temperature adjustment times and the humidity adjustment times; the number of temperature and humidity adjustment times reflects the stability and control precision of temperature control; feeding the counted adjustment times back to the main control module for evaluating the current temperature control strategy; the singlechip counter realizes digital statistics of the temperature control process, and obtains important temperature and humidity control evaluation parameters.

In sum, the temperature control module can generate temperature control output quantity by utilizing a PID algorithm according to real-time temperature and humidity data, and the temperature control output quantity is converted into a control signal through a DAC, so that the temperature and humidity of gas can be accurately regulated through a three-way heat exchange valve. The temperature adjustment times and the humidity adjustment times are recorded to provide detailed feedback for the temperature control process for the system, so that the temperature and humidity of the box opening environment are ensured to be stable in an ideal state, and finally the manufacturing yield of the wafer is improved.

FIG. 5 is an exemplary flow chart for generating a second gas filtration strategy and a second temperature control strategy according to some embodiments of the present disclosure, as shown in FIG. 5, the steps of generating the second gas filtration strategy and the second temperature control strategy comprising:

S510, acquiring space coordinate information of a gas sensor and acquired gas data. Three-dimensional point cloud data of the sensor can be acquired by using a three-dimensional scanner, and then three-dimensional space coordinate information is extracted.

S520, calculating the statistical mean value and standard deviation of the gas concentration in the space distribution according to the acquired space coordinate information and the gas concentration of the gas sensor, and taking the statistical mean value and standard deviation as a gas stability evaluation index KPI1; acquiring space coordinate information of a plurality of groups of gas sensors, for example, placing the gas sensors at xyz positions of an open box environment; simultaneously acquiring gas concentration data acquired by a sensor; the concentration data and the coordinate information are corresponding to reflect the spatial distribution condition of the gas concentration field; calculating the statistical average of the data acquired at all positions, and representing the overall average level of the gas concentration; calculating standard deviation, and reflecting the fluctuation range of concentration distribution; the mean value and the standard deviation are comprehensively used as a gas stability evaluation index KPI1; KPI1 intuitively reflects the overall level and spatial uniformity of the current gas concentration; the smaller the KPI1, the more stable the gas concentration.

S530, establishing a three-dimensional geometric model of an open box environment, performing grid division by adopting Fluent software, and performing CFD simulation of a temperature and humidity field; according to CFD simulation, extracting the variation ranges of a temperature field and a humidity field, and calculating the variation range to be used as a temperature and humidity stability index KPI2; establishing a three-dimensional geometric model of an open box environment, wherein the three-dimensional geometric model comprises details of a chassis structure, an air duct component and the like; performing grid division before numerical simulation by using CFD software Fluent, and discretizing a model space; on the discrete grid, solving a control equation by a numerical method, and simulating the evolution of a temperature field and a humidity field; setting boundary conditions such as fan parameters, wall conditions and the like; selecting a thermodynamic model, and considering multi-physical field coupling of gas flow and heat and mass transfer; performing time sequence iterative computation, and outputting the distribution and variation trend of the temperature and the humidity at each position in the space; the CFD simulation can observe the temperature and humidity effect of the box opening environment, and is an important means of virtual test; the numerical simulation technology can explore the temperature and humidity regulation effect of various control schemes; and a theoretical basis is provided for formulating a temperature and humidity control strategy. After CFD simulation calculation is finished, data of a temperature field and a humidity field can be extracted; analyzing the variation range of the temperature and the humidity at each point in the space; for example, the difference between the maximum and minimum values of temperature, represents the magnitude of the temperature change; the difference between the maximum relative humidity and the minimum relative humidity of the humidity indicates the variation range of the humidity; the temperature variation amplitude and the humidity variation amplitude are comprehensively used as temperature and humidity stability indexes KPI2; the smaller the KPI2 is, the more stable the temperature and humidity change is, and the better the control effect is; when KPI2 is larger than a set threshold, the temperature and humidity control is unstable; the application can quantitatively evaluate the stability of temperature and humidity control; providing basis for adjusting temperature and humidity control strategy.

S540, constructing a LASSO regression model, and adjusting the weight of each gas component in the KPI1 according to the space coordinate information of the gas sensor; constructing a LASSO (Least Absolute SHRINKAGE AND Selection Operator) regression model; the independent variable is the space coordinate information of the gas sensor; the dependent variable is the weight of each gas component in KPI 1; through LASSO regression training, the rule of influence of coordinate information on weight can be obtained; for example, a sensor near the loading element that measures the contribution of gas concentration changes to KPI1 should be weighted up; LASSO regression utilizes L1 norm regularization, and can automatically filter unimportant coordinate features; thus, only coordinate information with obvious influence on the weight can be reserved; according to the regression model, the weight of the corresponding gas component in the KPI1 can be adjusted according to the sensor coordinates; the application can reasonably set the weight, so that the KPI1 can reflect the gas stability more accurately.

S550, on the basis of CFD simulation, establishing a multi-physical field model considering the mutual coupling of gas diffusion and temperature and humidity; comprehensively considering gas concentration distribution, temperature and humidity distribution and dynamic change of the gas concentration distribution and the temperature and humidity distribution by the model; carrying out probability prediction by using a Bayes network, namely, giving a current state, and predicting the probability P1 and P2 that the KPI1 and the KPI2 exceed limit values in the future; the Bayes network can handle uncertainty and probability relationships between various influencing factors; training a Bayes network according to a gas diffusion rule, a temperature and humidity coupling mechanism and historical data; then, during online monitoring, current KPI1 and KPI2 values are input, and the Bayes network predicts the overrun probability; p1 and P2 reflect the pre-judging results of gas stability and temperature and humidity stability; when P1 or P2 is too high, prompting that a control strategy needs to be adjusted; the model fuses CFD simulation and Bayes prediction, and improves the manufacturing yield of the wafer.

S560 when the previous Bayes network prediction result P1 or P2 exceeds the set threshold; indicating that the gas concentration or the temperature and humidity control effect may be reduced; at this time, the main control module calls a pre-trained SVM model; inputting current gas monitoring data, and predicting and generating a second gas filtering strategy by the SVM model according to the principle of a support vector machine; meanwhile, the main control module calls a BP neural network model; inputting temperature and humidity data obtained by CFD simulation, training a BP neural network by using the CFD data and a corresponding strategy as a sample, and learning a complex mapping relation between a temperature and humidity state and optimal control by the BP neural network; the BP network predicts and generates a second temperature control strategy according to the learned temperature control law; the new gas filtering strategy and the temperature control strategy can obtain better control effect in the current state; the new strategy is sent to a gas control module and a temperature control module for implementation; the application can realize closed-loop optimization of gas filtration and temperature and humidity control. And sending the second gas filtering strategy to the gas control module, and sending the second temperature control strategy to the temperature control module.

As another embodiment of the present application, generating the second gas filtration strategy and the second temperature control strategy may further comprise: the main control module receives the pressure value, the gas components, the temperature adjustment times and the humidity adjustment times, and calculates a gas cleanliness index and a temperature and humidity stability index; judging whether the index exceeds a threshold value, if so, generating a second gas filtering strategy and a second temperature control strategy by the main control module based on the machine learning model, and respectively sending the second gas filtering strategy and the second temperature control strategy to the gas control module and the temperature control module, wherein the steps of the main control module comprise: receiving the gas pressure and a plurality of groups of gas composition data output by the gas control module; calculating the ratio of the concentration of the gas component to the standard concentration to generate a gas cleanliness evaluation index KPI1, wherein KPI1 is calculated by the following formula:

Wherein, C _i is the collection concentration of the ith gas component, and C _0i is the industry standard upper concentration limit of the ith gas component. Presetting according to a cleanliness standard of a semiconductor process; the actual concentration Ci of each gas component is calculated as a ratio of the corresponding industry standard upper concentration limit C _0i, i.e., C _i/C_0i. This calculated value represents a multiple of the actual concentration of the i-th gas component with respect to the upper limit of the standard concentration; by using KPI ₁, the system can evaluate the status of gas cleanliness in the environment in real time. If the value of KPI ₁ exceeds a predetermined threshold, the system may take appropriate control strategies, such as increasing the gas filtration level, adjusting the temperature and humidity, or initiating adaptive filtration strategies, to minimize the value of KPI ₁. Through the control measures, a clean and stable gas environment can be maintained, and the cleanliness requirement of the semiconductor process is met, so that the manufacturing yield of the wafer is effectively improved.

Receiving the temperature adjustment times n ₁ and the humidity adjustment times n ₂ output by the temperature control module; the calculated temperature and humidity stability index KPI ₂,KPI₂ is calculated by the following formula:

KPI₂＝(n₁+n₂)/t

Wherein t is a temperature control time period, and the period represents the condition of temperature and humidity adjustment in a certain time range, and is usually in units of seconds or minutes; the temperature and humidity adjustment times are combined with the time period to measure the stability of the temperature and humidity. The lower the KPI ₂ value is, the more stable the temperature and humidity are, and the environment is more suitable for manufacturing the wafer; the method for controlling the open box of the EFEM can realize real-time monitoring and control of the temperature and the humidity in a manufacturing environment so as to improve the manufacturing yield of wafers by calculating the ratio of the temperature and humidity adjustment times to the time period to generate the temperature and humidity stability index KPI ₂.

The KPI ₁ is compared with a preset gas cleanliness evaluation threshold alpha, and the KPI ₂ is compared with a preset temperature and humidity stability threshold beta. In the present application, determination of the gas cleanliness evaluation threshold α: the cleanliness requirements of the gas environment for wafer fabrication are determined with reference to semiconductor international standards, for example, the 0.1 μm particle content in the gas is not more than 10/L. Through testing KPI ₁ values under different strategies of the gas filtering system for a plurality of times; the preferred embodiment of the present application is to determine that when KPI ₁ exceeds 15%, indicating that gas cleanliness is below the required level, the filtration level needs to be raised, then α=15%. Determination of a temperature and humidity stability threshold value beta: with reference to the semiconductor process standard, the temperature response is controlled at + -0.1 ℃ and the humidity is + -2% RH. Testing KPI ₂ values under different temperature control strategies; the optimal embodiment of the application is to determine that when the sum of the temperature adjustment times and the humidity adjustment times exceeds 10 times/h, the temperature and the humidity tend to be unstable. Then take β=10 times/h.

And inputting and comparing the determined values of alpha and beta in a main control module, and starting a new intelligent optimization strategy when the KPI exceeds a threshold value, and adjusting in real time to ensure that the gas cleanliness and the temperature and humidity stability meet the requirements. When KPI ₁ is more than or equal to alpha or KPI ₂ is more than or equal to beta, the judgment index exceeds a threshold value; the condition settings allow the system to monitor in real time the critical parameters of the manufacturing environment, namely gas cleanliness (KPI ₁) and temperature and humidity stability (KPI ₂). Once KPI ₁ or KPI ₂ exceeds a preset threshold α or β, the system reacts immediately without manual intervention. The condition settings introduce automation control into the manufacturing environment. This means that the system can automatically take measures to maintain or optimize gas cleanliness and temperature and humidity without operator intervention. This improves the stability and reliability of the manufacturing process.

Based on an SVM algorithm, generating a second gas filtering strategy according to the gas data; the SVM is a support vector machine, is a supervised learning model, can be used for classification and regression analysis, and can process linear and nonlinear data sets. Based on the BP neural network, a second temperature control strategy is generated according to the temperature and humidity data. The BP neural network is a multi-layer feedforward network comprising an input layer, a hidden layer and an output layer, and can approximate any nonlinear function by training the network through an error back propagation algorithm. The SVM algorithm and the BP neural network based intelligent algorithm work cooperatively, so that the system can learn and evolve on line, dynamically adjust the control strategy in a closed loop, continuously optimize the gas cleanliness and the temperature and humidity stability, adapt to the change of the external environment, and improve the yield of wafer manufacturing.

When KPI ₁ < alpha and KPI ₂ < beta, judging that the gas cleanliness and the temperature and the humidity reach the box opening conditions; the system judges that the gas cleanliness and the temperature and the humidity reach the box opening condition. This means that the manufacturing environment has satisfied the open box requirements of semiconductor manufacturing, and that wafer processing operations can be performed; once the system determines that the open box condition has been met, the master control module begins generating an open box instruction. The main control module plans and generates proper box opening operation instructions based on the satisfied conditions. This includes determining key parameters such as timing, manner and location of opening the box. The master control module is communicated with the PLC of the manipulator module through the Profinet communication module. The master control module sends the generated box opening instruction to the PLC of the manipulator module through the Profinet communication module so as to trigger the box opening operation of the EFEM. After receiving the box opening instruction, the PLC of the manipulator module executes corresponding actions and opens the box opening mechanism of the EFEM so that the wafer can be processed or conveyed. The main control module generates a box opening instruction and sends the box opening instruction to the PLC of the manipulator module through the Profinet communication module. The Profinet is a real-time Ethernet communication protocol, is mainly used in the field of industrial automation, and can realize high-speed and reliable data exchange. By adopting Profinet communication, real-time reliable transmission of instructions can be ensured. The manipulator module is an actuating mechanism for executing the box opening mechanical movement and comprises a manipulator, a transmission device and the like, and can unlock the box door and open the box door according to the instruction. The PLC is a programmable logic controller and can carry out logic operation on input quantity and send a control instruction. The application improves the efficiency of the wafer manufacturing process by automatically judging the box opening conditions and generating the corresponding box opening instruction.

Fig. 6 is an exemplary flow chart for generating an open box instruction, as shown in fig. 6,

S710, setting a gas cleanliness box opening threshold alpha and a temperature and humidity box opening threshold beta; in the present application, the setting of the gas cleanliness opening threshold α: preliminarily determining a constraint concentration limit value L of the gas pollutant by referring to industry standards and technical specifications; determining the weight w of the key gas pollutants by combining the process requirements; and predicting a gas diffusion rule by using CFD simulation, and determining a safety concentration compensation value c.

Combining the two to obtain a gas cleanliness opening threshold value: α=w1l1+w2l2+ & gt. Setting a temperature and humidity box opening threshold value beta: according to the requirements of the technological process on the temperature and the humidity, the ranges of the temperature and the humidity [ Tmin, tmax ], [ RHmin, RHmax ] are preliminarily determined; based on CFD heat conduction simulation, determining a temperature and humidity uniformity tolerance d; combining product quality analysis to determine a temperature and humidity stability requirement K; obtaining a temperature and humidity box opening threshold value: beta= [ tmin+d, tmax-d ], [ RHmin +k, RHmax-K ].

S720, constructing a double-layer LSTM network as a temperature and humidity prediction model, wherein the LSTM network comprises an encoder and a decoder; LSTM stands for "Long Short-Term Memory", a variant of Recurrent Neural Network (RNN) for processing sequence data. LSTM networks have memory capabilities that capture long-term dependencies and are therefore very effective in the processing of time-series data. Constructing a double-layer LSTM network comprising an encoder and decoder structure; the encoder part uses an LSTM network, so that time sequence characteristics and internal rules in the temperature and humidity time sequence data can be learned and extracted; inputting the feature vector output by the encoder and the CFD simulation result into a decoder; the decoder is also based on an LSTM network to predict the temperature and humidity state; the LSTM network can capture the long-term dependency of the time sequence due to the gating structure; the double-layer LSTM encoding-decoding structure enhances the modeling capability of the network on the dynamic temperature and humidity change rule; compared with single-layer LSTM and unidirectional LSTM, the bidirectional LSTM network with the encoder and the decoder can improve the effect of temperature and humidity prediction; the CFD simulation data provides spatial information of temperature and humidity distribution for the network; the method improves the accuracy of temperature and humidity state prediction and provides support for the subsequent judgment of the box breaking condition.

Acquiring a temperature and humidity time sequence by using a Pt100 and humidity sensor, inputting the temperature and humidity time sequence into an encoder, extracting the characteristics of the time sequence by the encoder according to an LSTM network, and outputting a first characteristic expression vector; pt100 is a commonly used precision resistance temperature sensor that can provide high-precision temperature measurement; pt100 uses platinum as a sensitive element and has the characteristics of high accuracy, stability and reliability. In the wafer manufacturing process, the temperature and humidity stability of the EFEM box opening environment is important to ensuring the yield; the temperature and humidity change condition in the box opening environment can be monitored through the time sequence acquisition of the Pt100 and the humidity sensor; the characteristics of the temperature and humidity sequence are extracted by using the encoder and the LSTM network, so that the temperature and humidity change rule can be deeply analyzed; the first characteristic expression vector contains key information such as time sequence mode, stability, relativity and the like of temperature and humidity fluctuation; based on the time sequence characteristics, the temperature and humidity stability of the box opening environment can be evaluated, and the box opening environment can be adjusted in time; the intelligent analysis of temperature and humidity data is realized through the encoder and the LSTM, so that the temperature control effect can be greatly improved; the effective control of temperature and humidity directly influences the improvement of the manufacturing quality and yield of the wafer.

S730, inputting the CFD simulation result to an encoder to generate a second feature expression vector; the second characteristic expression vector is input into a decoder, the decoder predicts the temperature and humidity state according to the LSTM network, and a third characteristic expression vector is output; the CFD simulation can predict the spatial distribution and dynamic evolution of an air flow field, a temperature field and a humidity field; the temperature and humidity field data obtained by CFD simulation are used as input and are input into an encoder of an LSTM network; the encoder learns and extracts features in the CFD data by utilizing the LSTM structure, and generates a second feature expression vector representing temperature and humidity distribution information; the second feature vector fuses the spatial correspondence of the temperature and humidity field contained in the CFD data; inputting the second feature vector into the LSTM decoder together with the sequence data; the decoder predicts the temperature and humidity state at the future moment based on the LSTM network, the comprehensive sequence time sequence information and the CFD space information; and outputting a third characteristic expression vector representing the predicted temperature and humidity state; thus, the encoder encodes CFD data in the LSTM network, and the decoder decodes the predictions to form an end-to-end model; the CFD data provides useful space constraint knowledge for the network, and the effect of temperature and humidity prediction is improved.

S740, judging whether the temperature and humidity state represented by the third characteristic expression vector is within the allowable temperature range of 20-25 ℃ and the allowable humidity range of 40-50%RH; less than 40% RH increases the risk of electrostatic damage. Greater than 50% RH can affect lithographic imaging quality. The temperature range of 20 ℃ to 25 ℃ can ensure the stability of the air flow in the EFEM and avoid wafer pollution caused by temperature change. The humidity range from 40% RH to 50% RH can effectively inhibit electrostatic discharge and prevent equipment from being damaged, and the risk of electrostatic discharge is increased due to the fact that the humidity is too low below 40% RH; over 50% RH, too high humidity can generate water vapor condensation, which affects the imaging quality of the precise photoetching machine; therefore, the allowable range of the temperature and humidity is 20-25 ℃ and 40-50% RH, so that the process requirement can be met, and the matching of equipment and the environment is considered; by judging whether the temperature and humidity state is within the allowable range, the negative influence of temperature and humidity abnormality on the process and equipment can be effectively avoided; thereby ensuring the yield, improving the equipment use environment and improving the production efficiency.

S750, judging whether the gas stability evaluation index KPI1 is lower than a gas box opening threshold alpha; judging whether the temperature and humidity stability evaluation index KPI2 is lower than a temperature and humidity box opening threshold value beta or not; the gas box opening threshold value alpha is a preset value according to the requirement of the process flow on the quality of the gas environment. The temperature and humidity box opening threshold value beta is a preset value according to the requirements of equipment and technology on the temperature and humidity environment. By setting the box opening threshold value of the gas and the temperature and the humidity and judging whether the evaluation index meets the standard, the comprehensive monitoring and intelligent judgment of the box opening environment are realized, the box opening operation under the environment abnormal state is effectively avoided, the stability of the process flow is ensured, and the yield of the process is improved

S760, if the conditions are met, judging that the box opening condition is met; the main control module issues a Profinet opening execution instruction to the manipulator; profinet is an open industrial ethernet communication protocol for automation, which can provide deterministic network services; the main control module is connected with the manipulator through a Profinet industrial Ethernet to form an automatic control system; after judging that the box opening condition is met, the main control module sends a box opening execution instruction to the manipulator; in the Profinet communication network, the communication between the master control module and the manipulator follows the master station-slave station mode; the master control module is used as a master station for Profinet communication, and the manipulator is used as a slave station for Profinet communication; the master station is responsible for sending control instructions, and the slave station is responsible for receiving the instructions; after the manipulator receives the instruction, the manipulator calls an internal PLC program to decode the instruction content; the PLC program starts a motion control unit of the manipulator to drive the manipulator to finish the box opening action; after the box opening is finished, the manipulator feeds back the execution state to the main control module; the accurate control of the manipulator box opening action by the main control module is realized through industrial Ethernet communication and deterministic protocol of Profinet.

In summary, the application constructs a gas multi-sensor monitoring and data fusion module, which can provide accurate and reliable gas data and ensure the input basis of gas control; the gas threshold and fuzzy control are set, so that the gas state can be judged in real time, and the gas filtering strategy can be flexibly adjusted. The MEMS pressure sensor and the self-adaptive gas flow control can accurately adjust gas parameters; the temperature and humidity closed-loop PID control is matched with the three-way heat exchanger, so that the temperature and humidity can be accurately tracked and regulated, and the temperature and humidity control quality is improved; the main control module is used for multi-source heterogeneous data fusion, a machine learning model is adopted to generate a self-adaptive regulation strategy, and the collaborative optimization of gas filtration and temperature and humidity control is realized; setting a box opening threshold of gas and temperature and humidity, and judging the time for opening the box by an LSTM prediction model, so as to avoid the influence of environmental abnormality on the process; the whole system information flows smoothly, the control strategy is adaptively optimized, and the long-term stability of the box opening environment in an optimal state can be ensured; the gas cleanliness and the temperature and humidity stability of the box opening environment are key for ensuring the wafer yield, and the intelligent level of the box opening environment control is improved; therefore, the application can effectively improve the controllability, stability and optimizability of the EFEM box opening environment, and minimize the negative influence of the box opening environment on the wafer manufacturing yield, thereby playing a key technical effect of improving the wafer manufacturing yield.

Claims

1. A method of opening a pod for use in controlling the opening of an EFEM, comprising:

step one, a gas monitoring module collects gas data containing gas concentration and gas components and sends the gas data to a gas control module; the gas monitoring module is used for collecting the internal gas state of the EFEM;

step two, the temperature and humidity monitoring module collects temperature and humidity data comprising temperature and humidity and sends the temperature and humidity data to the temperature control module;

Step three, the gas control module receives the gas data, judges whether the gas concentration or the gas composition exceeds a threshold value, if so, performs gas filtration according to a preset first gas filtration strategy, and sends the gas data in the gas filtration to the main control module;

the temperature control module receives temperature and humidity data, calculates deviation between the received temperature and humidity data and target temperature and humidity, generates a first temperature control strategy according to the deviation through a PID algorithm, adjusts the temperature and humidity according to the first temperature control strategy, and sends the temperature and humidity data in the adjusting process to the main control module;

Step five, the main control module receives the gas data and the temperature and humidity data, and calculates a gas stability evaluation index KPI1 and a temperature and humidity stability index KPI2; judging whether the index exceeds a threshold value, if so, calling a second gas filtering strategy generated by the SVM model, calling a BP network model to generate a second temperature control strategy, and respectively transmitting the second temperature control strategy to the gas control module and the temperature control module;

Step six, the gas control module receives and filters the gas according to a second gas filtering strategy; the temperature control module receives and adjusts the temperature and the humidity according to a second temperature control strategy;

step seven, the main control module judges whether the gas stability evaluation index KPI1 and the temperature and humidity stability index KPI2 reach the box opening condition, and if so, a box opening instruction is sent to the manipulator module;

And step eight, the manipulator module receives and executes a box opening instruction.

2. The method of opening a box according to claim 1, characterized in that:

the first gas filtration strategy comprises:

when the particle size of the gas pollutant is larger than 5 mu m, setting the gas to enter from the outer side of the filter element cylinder at a speed of 0.1m/s, and filtering by using a polypropylene filter material with a pore size of 5 mu m from inside to outside to remove the particle impurity larger than 5 mu m in the gas;

The second-stage efficient filtration, when the particle size of the gas pollutant is 0.3-5 μm, setting the gas to enter through the inner side of the filter element cylinder at the speed of 0.05m/s, firstly passing through the 0.3 μm glass microfiber filter material layer to remove 0.3-5 μm fine particles, and then passing through the active carbon layer to adsorb volatile organic pollutants in the gas;

the third stage of activated carbon adsorption, when the concentration of the volatile organic pollutants is higher than 10ppb, the gas is utilized to pass through the two layers of activated carbon filter element cylinders, the activated carbon adopts oval particles, and the activated carbon with the specific surface area of 1700m < 2 >/g is utilized to adsorb the volatile organic pollutants in the gas;

Fourth-stage gas purification, wherein when the concentration of residual pollutants in the gas is higher than 100ppt, non-thermal plasma generated by exciting the gas through an induction electrode is utilized to oxidatively decompose the residual pollutants in the gas by utilizing oxygen ions in the plasma;

And fifth-stage negative pressure adsorption, namely after four-stage filtration, when the concentration of residual pollutants in the gas is higher than 10ppt, applying 0.05MPa negative pressure, and utilizing the negative pressure to enhance the adsorption capacity of the activated carbon so as to purify the residual pollutants in the gas.

3. The method of opening a box according to claim 1, characterized in that:

The first step comprises the following steps:

Setting a plurality of groups of gas sensors, and collecting gas data, wherein the gas data comprises gas concentration and a plurality of gas components;

Inputting the acquired gas data into a pre-established data fusion module, and outputting the fused gas data by the data fusion module based on a particle filtering algorithm;

inputting the fused data into a data correction model, and correcting the gas data by the data correction model by using a preset dynamic correction strategy, and outputting the corrected gas data;

Inputting the calibrated gas data into a gas control model, establishing the gas control model based on an MRAC algorithm, and updating control parameters of a first gas filtering strategy by using the calibrated gas data;

and sending the calibrated gas data to a gas control module.

4. A method of opening a box according to claim 3, characterized in that:

the step of generating a second gas filtration strategy and a second temperature control strategy comprises:

Acquiring space coordinate information of a gas sensor and acquired gas data;

According to the acquired space coordinate information and gas concentration of the gas sensor, calculating the statistical mean value and standard deviation of the gas concentration in space distribution, and taking the statistical mean value and standard deviation as a gas stability evaluation index KPI1;

Establishing a three-dimensional geometric model of an open box environment, performing grid division by adopting Fluent software, and performing CFD simulation of a temperature and humidity field;

According to CFD simulation, extracting the variation ranges of a temperature field and a humidity field, and calculating the variation range to be used as a temperature and humidity stability index KPI2;

constructing a LASSO regression model, and adjusting the weight of each gas component in the KPI1 according to the space coordinate information of the gas sensor;

establishing a gas diffusion and temperature and humidity coupling model based on CFD simulation, and respectively calculating overrun probabilities P1 and P2 of KPI1 and KPI2 by adopting a Bayes network;

When the overrun probability P1 or P2 exceeds a threshold value, the main control module calls an SVM model, and generates a second gas filtering strategy according to the gas data; the main control module calls a BP network model and generates a second temperature control strategy according to CFD temperature and humidity data;

and sending the second gas filtering strategy to the gas control module, and sending the second temperature control strategy to the temperature control module.

5. The method for opening a box according to claim 4, wherein:

The seventh step includes:

Setting a gas cleanliness box opening threshold alpha and a temperature and humidity box opening threshold beta;

constructing a double-layer LSTM network as a temperature and humidity prediction model, wherein the LSTM network comprises an encoder and a decoder;

Acquiring a temperature and humidity time sequence by using a Pt100 and humidity sensor, inputting the temperature and humidity time sequence into an encoder, extracting the characteristics of the time sequence by the encoder according to an LSTM network, and outputting a first characteristic expression vector;

Inputting the CFD simulation result to an encoder to generate a second feature expression vector;

the second characteristic expression vector is input into a decoder, the decoder predicts the temperature and humidity state according to the LSTM network, and a third characteristic expression vector is output;

judging whether the temperature and humidity state represented by the third characteristic expression vector is within the allowable temperature range of 20-25 ℃ and the allowable humidity range of 40-50% RH;

Judging whether the gas stability evaluation index KPI1 is lower than a gas box opening threshold alpha or not;

Judging whether the temperature and humidity stability evaluation index KPI2 is lower than a temperature and humidity box opening threshold value beta or not;

if the conditions are satisfied, judging that the box opening condition is satisfied;

And the main control module issues a Profinet box opening execution instruction to the manipulator.

6. A method of opening a box according to claim 3, characterized in that:

The step of fusing the acquired gas data includes:

Taking gas data acquired by a gas sensor as an observed quantity of a particle filtering algorithm, wherein the type of the gas sensor comprises an infrared gas sensor and a gas chromatograph;

initializing a particle state by using a sampling method based on Kullback Leibler divergence, and generating a first particle state of posterior probability distribution;

Carrying out state prediction on the first particle state by applying a Kalman filtering algorithm of a noise covariance matrix in an online self-adaptive adjustment process, and outputting a predicted state;

When the weight of the first particle state is calculated, introducing a space constraint relation based on a CFD gas propagation model, and combining depth features of gas data extracted by using a pretrained convolutional neural network to serve as priori knowledge of weight calculation;

resampling the particles by an Auxiliary PARTICLE FILTER algorithm according to the depth characteristics to generate a second particle state;

Correcting the second particle state obtained by sampling by using a Kalman filtering algorithm, and outputting the corrected second particle state;

Carrying out weighted average on the corrected second particle state to obtain a gas concentration fusion value;

calculating the error between the gas concentration fusion value and the acquired gas data, and adopting a Gauss Newton method to adaptively adjust the process noise parameters of the Kalman filtering algorithm according to the error;

And recursively executing the steps and outputting the fused gas concentration and gas composition.

7. A method of opening a box according to claim 3, characterized in that:

the step of calibrating the fused gas data includes:

Acquiring gas data acquired by a gas sensor, and acquiring a signal drift mode of the gas sensor through a time sequence analysis algorithm;

According to the acquired signal drift mode, calculating zero calibration parameters and sensitivity calibration parameters of the gas sensor by adopting an incremental differential algorithm;

calculating compensation control quantity of the gas data according to the acquired gas data, the zero calibration parameter and the sensitivity calibration parameter;

And performing closed-loop PID control on the acquired gas data by using the obtained compensation control quantity to generate calibrated gas data.

8. A method of opening a box according to claim 3, characterized in that:

The step of updating the first gas filtration strategy using the calibrated gas data comprises:

Constructing a gas control model based on an MRAC algorithm, wherein the gas control model comprises a preset first gas filtering strategy F1;

Inputting the calibrated gas data into a gas control model, and generating a filtering strategy F1' based on an RLS algorithm;

calculating errors of the first gas filtering strategy F1 and the filtering strategy F1';

Based on the self-adaptive control algorithm, calculating a control input quantity delta u of gas filtration according to the error;

Adjusting the first gas filtering strategy according to the control input quantity delta u;

The MRAC algorithm is recursively executed to update the first gas filtering strategy.

9. The method of opening a box according to claim 1, characterized in that:

The third step comprises:

A gas threshold database is arranged in the gas control module, and the thresholds of various gas components are preset in the database;

The gas control module receives the gas data output by the gas monitoring module;

judging whether the gas data exceeds a threshold value by utilizing a fuzzy control algorithm according to the threshold value of the gas components in the database, and if so, performing gas filtration according to a preset first gas filtration strategy;

monitoring the gas pressure by adopting an MEMS pressure sensor, and inputting the monitored gas pressure into a gas flow intelligent regulating device;

the intelligent gas flow regulator adopts adaptive PID algorithm to control the electronic RF valve to regulate the gas flow.

10. The method of opening a box according to claim 1, characterized in that:

The fourth step comprises:

acquiring temperature and humidity data of the collected gas;

calculating the deviation e (t) between the acquired temperature and humidity data and the target temperature and humidity according to an incremental PID algorithm;

generating a temperature control output u (t) by utilizing a PID algorithm according to the deviation e (t);

converting u (t) into a voltage signal of 0V to 10V through a DAC (digital-to-analog converter), and taking the voltage signal as a control signal of the three-way heat exchange valve;

the three-way heat exchange valve accurately adjusts the proportion of gas flowing through the electric heater and the Peltier refrigerator, and controls the temperature and the humidity of the gas;

And acquiring the temperature adjustment times and the humidity adjustment times in the temperature control process by adopting an electronic counter.