Industrial big data governance method and system

Abstract

The invention discloses an industrial big data management method and system, and mainly aims at the field of semiconductor manufacturing. According to the method, multiple sensors are installed on equipment to collect technological parameters and equipment operation data, after preprocessing, fluctuation characteristics of photoetching link data are extracted through time sequence analysis and wavelet transformation, and a photoresist thickness fluctuation model is constructed in combination with historical data and a technological principle. According to the model, an improved PID control algorithm is adopted to formulate and execute a fluctuation suppression strategy, and the photoresist thickness is stabilized. And meanwhile, relevant data are classified and stored for query and calling. The problems of data management and process control in semiconductor manufacturing are effectively solved, the data quality is improved, the process stability is guaranteed, the production efficiency and the product quality are improved, and the intelligent development of the semiconductor manufacturing industry is promoted.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and specifically to an industrial big data governance method and system. Background Technique

[0002] At present, when the semiconductor manufacturing industry is deeply integrated into the wave of industrial big data development, its data governance faces many severe challenges. These problems restrict the high-quality development of the industry. In the field of semiconductor manufacturing industrial data processing, with the continuous progress of technology, the chip manufacturing process has become increasingly complex and precise, posing extremely high requirements for data processing and management in the production process. Lithography, as the core link of semiconductor manufacturing, the process stability directly determines the quality and performance of the chip, and the precise control of the photoresist thickness is the key to the lithography process.

[0003] In the traditional semiconductor manufacturing process, in the lithography link, for the key parameters that affect the photoresist thickness, such as the temperature of the lithography equipment close to the photoresist coating area and the pressure applied to the photoresist during coating, effective real-time monitoring and data collection are not carried out. This results in the inability to accurately grasp the change law of the photoresist thickness and makes it difficult to timely and accurately regulate the production process. In terms of the control of the photoresist thickness, the existing technical means cannot meet the high-precision process requirements. Since the photoresist thickness is affected by a variety of factors such as exposure energy, temperature, the characteristics of the photoresist itself, equipment aging, and environmental humidity, traditional control strategies are difficult to comprehensively consider these complex factors. For example, in the face of performance changes caused by equipment aging, traditional methods cannot timely adjust the control parameters, resulting in large fluctuations in the photoresist thickness, which in turn affects the yield rate of the chips. In view of the above problems, an industrial big data governance method and system are designed. Summary of the Invention

[0004] The purpose of the present invention is to provide an industrial big data governance method and system to solve the problems raised in the above background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An industrial big data governance method, including the following steps:

[0006] Step 1, data collection: By installing temperature sensors, pressure sensors, and lithography precision sensors on semiconductor manufacturing equipment, real-time collection of process parameter data during the manufacturing process;

[0007] Step 2, data preprocessing: Perform preliminary cleaning on the collected data, remove obvious error values, fill in a small number of missing values, and standardize the data to unify the data format to obtain preprocessed data;

[0008] Step 3, Fluctuation Feature Extraction: Receive the preprocessed data, extract the fluctuation period and amplitude features of the data of the exposure time and photoresist thickness changing with time in the lithography process. Through wavelet transform technology, decompose the data into different frequency components, and identify high-frequency noise fluctuations and low-frequency trend fluctuations;

[0009] Step 4, Establish a Fluctuation Model: According to the extracted fluctuation features, combined with the principles of semiconductor manufacturing processes, establish a process fluctuation prediction model. Specifically, based on historical data and physical models, construct a mathematical model for the fluctuation of photoresist thickness in the lithography process with respect to exposure energy and temperature. This model simulates the fluctuation of process parameters under different conditions;

[0010] Step 5, Formulation and Execution of Fluctuation Suppression Strategies: Based on the fluctuation prediction model, formulate corresponding fluctuation suppression strategies. That is, when it is predicted that the photoresist thickness will have a large fluctuation, automatically adjust the exposure energy and temperature control parameters, and intervene in the manufacturing process in real time through a feedback control system to suppress the fluctuation of process parameters and ensure the stability of semiconductor manufacturing processes;

[0011] Step 6, Data Storage: Store the preprocessed data in Step 2 and the fluctuation suppression strategy data in Step 5 into a distributed database, and classify and store them according to different process links, equipment types, and time series for quick query and call.

[0012] Preferably, the process parameter data in Step 1 specifically includes the following: Lithography process: Install a temperature sensor near the photoresist coating area of the lithography equipment to monitor the ambient temperature around the photoresist during lithography; Add a pressure sensor to the photoresist coating equipment to collect the pressure data applied to the photoresist during coating; Use a high-precision lithography accuracy sensor to record the accuracy deviation between the lithography pattern and the actual formation of the photoresist in real time; Association process: Install a flow sensor on the etching equipment to collect the etching gas flow data, and the etching gas flow affects the etching rate.

[0013] Preferably, the specific implementation steps of the fluctuation feature extraction are as follows:

[0014] Step S31, Data Preparation: Collect the exposure time and photoresist thickness data in the lithography process in Step 2 for a period of time, so as to obtain the exposure time series at consecutive time points and the photoresist thickness series

[0015] ;

[0016] Period Analysis: Use Fourier transform for period feature extraction. The Fourier transform formula is: , where is the original time series, i.e., the exposure time and the photoresist thickness , is the transformed frequency-domain representation. By performing a Fourier transform on the exposure time series , the amplitudes at different frequencies are obtained. In the frequency domain, the period corresponding to the frequency component with a larger amplitude is the fluctuation period of the exposure time; similarly applied to the photoresist thickness series , the fluctuation period is found;

[0017] Amplitude feature extraction: Extract the fluctuation amplitude feature and calculate the range of the time series, that is, for the exposure time series , calculate the range which represents the maximum fluctuation range of the exposure time within the observation period; for the photoresist thickness series , the range represents the fluctuation amplitude of the photoresist thickness;

[0018] Step 33, Wavelet transform for frequency component decomposition: Use discrete wavelet transform to process the exposure time and photoresist thickness data. The discrete wavelet transform selects an appropriate wavelet basis function , where is the scale parameter, is the translation parameter, and perform discrete wavelet transform on the exposure time series to obtain wavelet coefficients at different scales and positions. At high-frequency scales, the wavelet coefficients reflect the rapidly changing part in the exposure time series, that is, high-frequency noise fluctuations; similarly perform discrete wavelet transform on the photoresist thickness series to separate its high-frequency noise fluctuations and low-frequency trend fluctuations, that is, comprehensively grasp the fluctuation characteristics of the photolithography process parameter data and provide an accurate data basis for subsequent establishment of a fluctuation model.

[0019] Preferably, the specific implementation steps for establishing the fluctuation model are as follows:

[0020] Step 41, Data sorting and preparation: Collect historical data in the photolithography process, including the measured values of the photoresist thickness, the corresponding exposure energy setting values, and the environmental temperature or the temperature data of the key parts of the equipment during the production of different batches of products. Assume there are groups of data, denote the photoresist thickness as , the exposure energy as , and the temperature as ;

[0021] Step 42: Perform modeling using the multiple linear regression algorithm; based on the principle of the lithography process, the thickness of the photoresist is comprehensively affected by the exposure energy and temperature. The multiple linear regression algorithm is used for modeling, specifically as follows: , where is the dependent variable, that is, the thickness of the photoresist, and , is the intercept, and are the regression coefficients, is the random error term, where , and are estimated through historical data;

[0022] Step 43: Parameter estimation: The least squares method is used to estimate the regression coefficients to minimize the sum of the squared errors between the photoresist thickness and the model prediction value , that is is minimized; by taking the partial derivatives of with respect to , and respectively, and setting the partial derivatives to zero, a system of equations is obtained:

[0023]

[0024] Solving the above system of equations, the estimated values of , and are obtained as , and . Finally, the predicted model of the photoresist thickness fluctuation is , where is the predicted photoresist thickness, is the exposure energy, is the temperature;

[0025] Step 44: Model testing and optimization: Use the coefficient of determination to evaluate the predicted model of the photoresist thickness fluctuation, where , where is 's average value, and when is close to 1, the better the fitting effect of the model to the data.

[0026] Preferably, the specific working logic of formulating and implementing the fluctuation suppression strategy is as follows:

[0027] Threshold setting: Based on historical data and the quality standards of the lithography process, determine the normal fluctuation range of the photoresist thickness. Let the average value of the normal photoresist thickness be , the fluctuation range of the photoresist thickness is , where is a reasonable fluctuation threshold set according to the process precision requirements;

[0028] Prediction and judgment: Using the established photoresist thickness fluctuation prediction model , input the current exposure energy and temperature data in real time, predict the photoresist thickness, and when the predicted value exceeds the normal fluctuation range , it is determined that a large fluctuation will occur in the photoresist thickness;

[0029] Strategy formulation: Use the PID control algorithm to formulate a control strategy. Specifically, its output consists of the proportional term , the integral term and the derivative term , and the formula is: , is the error at the current moment, that is, the difference between the target value and the actual predicted value , , and are the proportional coefficient, integral coefficient, and derivative coefficient respectively. Among them, the proportional term instantaneously adjusts the control amount according to the magnitude of the current error; the integral term is used to eliminate the steady-state error of the system; the derivative term adjusts the control amount in advance according to the change trend of the error; the above coefficients are specifically improved according to the characteristics of the lithography process and the response of the equipment, and the improvement is based on the deviation degree and change rate of the photoresist thickness prediction value from the normal range, and the coefficients are dynamically adjusted;

[0030] Consider multi-factor interference compensation: The photoresist thickness is affected not only by exposure energy and temperature, but also by various factors such as the characteristics of the photoresist itself, equipment aging, and environmental humidity. A multi-factor interference model is established by introducing an interference compensation term into the PID algorithm; collect data on the influence of changes in different factors on the photoresist thickness, and analyze the relationship between each factor and the change in the photoresist thickness. Specifically, let the interference factor set be , and the corresponding compensation coefficient be , then the adjusted control amount is , where is the control amount after dynamically adjusting the coefficient;

[0031] Strategy execution: According to the control amount calculated by the PID algorithm, automatically adjust the control parameters of the exposure energy control device of the exposure equipment and the temperature adjustment equipment.

[0032] Preferably, the , and are specifically improved according to the characteristics of the lithography process and the response of the equipment. The specific implementation steps are as follows:

[0033] Step A, real-time data acquisition: During the lithography process, the thickness data of the photoresist is collected in real time through a sensor. Let the measured value of the photoresist thickness at the current moment be , the normal photoresist thickness range is , and at the same time, the exposure energy and temperature corresponding to the moment are obtained;

[0034] Step B, calculate the change rate: Calculate the change rate of the photoresist thickness error , where is the time interval of data acquisition, and the error change rate is calculated through the error values at adjacent moments, that is, the change trend of the photoresist thickness error;

[0035] Step C, construct a coefficient adjustment function:

[0036] Determine the function form: Use multiple linear regression to construct a coefficient adjustment function. The adjustment function of the proportional coefficient is . Let the function form be , where , and are coefficients to be determined; similarly, the adjustment function of the integral coefficient is set as , and the adjustment function of the differential coefficient is set as ;

[0037] Coefficient training: Train through the preset experimental data to determine the coefficients in the above functions. Prepare several groups of experimental data under different lithography process conditions. Each group of data includes the measured value of the photoresist thickness, the corresponding error , the error change rate and the best-performing , and values under the circumstances. Use the least squares method to solve for the coefficients; for the solution of the adjustment function coefficients of , assume there are groups of experimental data, and the goal is to make . Respectively take the partial derivatives of , and and set them to zero to obtain a system of equations:

[0038]

[0039] Solving this system of equations gives 、 and values, thereby determining adjustment function; similarly determine and coefficients 、 、 and 、 、 ;

[0040] Step D, dynamic coefficient update: Substitute the calculated error , error change rate into the determined coefficient adjustment function, calculate the adjusted coefficients 、 and to be adopted at the current moment, and substitute the adjusted coefficients into the PID control algorithm formula to calculate the control quantity .

[0041] Preferably, an industrial big data governance system, according to an industrial big data governance method, the system includes:

[0042] The data acquisition module collects process parameters and equipment operation status data in the manufacturing process in real time through sensors, and performs processing on the collected data such as cleaning error values, filling missing values, and standardizing the data format to obtain preprocessed data;

[0043] The semiconductor process fluctuation suppression module processes the process parameter data using time series algorithms, distinguishes high-frequency noise and low-frequency trend fluctuations by wavelet transform, and then constructs a prediction model according to the fluctuation characteristics and process principles, that is, a mathematical model of photoresist thickness, exposure energy, and temperature; according to the model prediction, automatically regulate the exposure energy parameters to stabilize the process;

[0044] The data storage module is used to store the preprocessed data in the acquisition module and the fluctuation suppression strategy data in the semiconductor process fluctuation suppression module into a distributed database, and classify and store them according to different process links, equipment types, and time series.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows: accurately grasping the fluctuation characteristics and optimizing the lithography process: by using time series analysis algorithms and wavelet transform techniques for extracting fluctuation characteristics, it is possible to accurately obtain the fluctuation period, amplitude, and frequency components of key data such as exposure time and photoresist thickness in the lithography process, clearly distinguishing high-frequency noise fluctuations from low-frequency trend fluctuations. This helps to deeply understand the dynamic change rules of the lithography process, provides accurate data support for establishing an accurate fluctuation model, thereby realizing precise control and optimization of the lithography process, and improving the stability and accuracy of lithography.

[0046] Establishing a reliable fluctuation model to predict process fluctuations: The mathematical model of the photoresist thickness fluctuation based on historical data and physical models between exposure energy and temperature fully considers the principles of semiconductor manufacturing processes and can effectively simulate the fluctuation situations of process parameters under different conditions. Through strict data sorting, multiple linear regression modeling, precise parameter estimation, and scientific model testing and optimization, the reliability and accuracy of the model are ensured. This model can predict the photoresist thickness fluctuation in advance, providing a reliable basis for the fluctuation suppression strategy, enabling the production process to anticipate potential process fluctuation problems in advance, and ensuring the stability of semiconductor manufacturing processes.

[0047] Implementing an effective fluctuation suppression strategy to ensure product quality: The fluctuation suppression strategy formulated based on the fluctuation prediction model, combined with the PID control algorithm and improved according to the characteristics of the lithography process, can quickly respond to abnormal fluctuations in the photoresist thickness. By dynamically adjusting coefficients to adapt to different process conditions and considering multi-factor interference compensation, it comprehensively addresses the complex situation where the photoresist thickness is affected by multiple factors. Automatically adjusting the exposure energy and temperature control parameters, intervening in the manufacturing process in real-time, effectively suppressing the fluctuation of the photoresist thickness, ensuring that the photoresist thickness always remains within the normal range, thereby ensuring the quality and performance of chip manufacturing and improving the product yield. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the method flow structure of the present invention;

[0049] Figure 2 It is a schematic diagram of the process flow of the fluctuation feature extraction step of the present invention;

[0050] Figure 3 It is a schematic diagram of the process flow of the step of establishing a fluctuation model of the present invention;

[0051] Figure 4 It is a schematic diagram of the work flow of formulating and implementing the fluctuation suppression strategy of the present invention;

[0052] Figure 5 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] Embodiment 1

[0055] Please refer to Figure 1 , the present invention provides a technical solution: an industrial big data governance method, including the following steps:

[0056] Step 1, data collection: By installing temperature sensors, pressure sensors, and lithography accuracy sensors on semiconductor manufacturing equipment, process parameter data during the manufacturing process is collected in real time;

[0057] Among them, the process parameter data specifically includes the following content: Lithography link: Install a temperature sensor near the photoresist coating area of the lithography equipment to monitor the temperature of the environment around the photoresist during lithography. Since the temperature change affects the reaction of the photoresist and thus changes the thickness; Add a pressure sensor to the photoresist coating equipment to collect the pressure data applied to the photoresist during coating. The pressure has a direct effect on the initial coating thickness of the photoresist; Through a high-precision lithography accuracy sensor, the accuracy deviation between the lithography pattern and the actual forming of the photoresist is recorded in real time, indirectly reflecting the influence of the photoresist thickness uniformity on the lithography accuracy; Association link: Install a flow sensor on the etching equipment to collect the etching gas flow data. The etching gas flow affects the etching rate and is indirectly associated with the change of the photoresist thickness during the etching process.

[0058] Step 2, data preprocessing: The collected data is preliminarily cleaned, obvious error values are removed, a small amount of missing values are filled, and the data is standardized to unify the data format to obtain preprocessed data;

[0059] Step 3, fluctuation feature extraction: Receive the preprocessed data, extract the fluctuation period and amplitude features of the data of the exposure time and the photoresist thickness changing with time in the lithography link. Through wavelet transform technology, the data is decomposed into different frequency components to identify high-frequency noise fluctuations and low-frequency trend fluctuations; Please refer to Figure 2 , and the specific implementation steps are as follows:

[0060] Step S31, data preparation: Collect the exposure time and photoresist thickness data in the lithography link for a period of time in Step 2, so as to obtain a continuous exposure time series at time points and the photoresist thickness series

[0061] Step S32, Time series analysis algorithm:

[0062] Period analysis: Use Fourier transform to extract period features. The Fourier transform formula is: , where is the original time series, i.e., the exposure time and the photoresist thickness , is the transformed frequency domain representation. By performing Fourier transform on the exposure time series , the amplitudes at different frequencies are obtained. In the frequency domain, the period corresponding to the frequency component with a larger amplitude is the fluctuation period of the exposure time; similarly, it is applied to the photoresist thickness series to find the fluctuation period;

[0063] For example: After calculation, it is found that the frequency corresponds to a larger amplitude, then its corresponding period is an important fluctuation period of the exposure time. The same method can be applied to the photoresist thickness series to find its fluctuation period.

[0064] Amplitude feature extraction: Extract the fluctuation amplitude feature and calculate the range of the time series, that is, for the exposure time series , calculate the range represents the maximum fluctuation range of the exposure time within the observation period; for the photoresist thickness series , the range represents the fluctuation amplitude of the photoresist thickness;

[0065] Step 33, Wavelet transform for frequency component decomposition: Use discrete wavelet transform to process the exposure time and photoresist thickness data. The discrete wavelet transform selects an appropriate wavelet basis function , where is the scale parameter, is the translation parameter, and perform discrete wavelet transform on the exposure time series to obtain wavelet coefficients at different scales and positions. At high frequency scales, the wavelet coefficients reflect the rapidly changing part in the exposure time series, that is, high frequency noise fluctuations; similarly, perform discrete wavelet transform on the photoresist thickness series to separate its high frequency noise fluctuations and low frequency trend fluctuations, that is, comprehensively grasp the fluctuation characteristics of the lithography process parameter data and provide an accurate data basis for subsequent establishment of a fluctuation model; for example, at the scale , the part where the absolute value of the wavelet coefficient is larger and more dispersed corresponds to the high frequency noise in the exposure time. And at low frequency scales, the wavelet coefficients reflect the trend fluctuations of the data.​​

[0066] Step 4. Establish a fluctuation model: Based on the extracted fluctuation characteristics and combined with the principles of semiconductor manufacturing processes, establish a process fluctuation prediction model. Specifically, based on historical data and physical models, construct a mathematical model for the fluctuation of photoresist thickness in the lithography process with respect to exposure energy and temperature. This model simulates the fluctuation of process parameters under different conditions; please refer to Figure 3 , and the specific implementation steps for establishing the fluctuation model are as follows:

[0067] Step 41. Data sorting and preparation: Collect historical data in the lithography process, including the measured values of photoresist thickness, the corresponding exposure energy settings, and the environmental temperature or the temperature data of key parts of the equipment during the production of different batches of products. Assume there are groups of data. Denote the photoresist thickness as , the exposure energy as , and the temperature as ;

[0068] Step 42. Use the multiple linear regression algorithm for modeling; Based on the principles of the lithography process, the photoresist thickness is affected by the combined influence of exposure energy and temperature. Use the multiple linear regression algorithm for modeling, specifically: , where is the dependent variable, that is, the photoresist thickness, and , is the intercept, and are the regression coefficients, is the random error term, where , and are estimated through historical data;

[0069] Step 43. Parameter estimation: Use the least squares method to estimate the regression coefficients to minimize the sum of the squares of the errors between the photoresist thickness and the model predicted value , that is, is minimized; By taking the partial derivatives of with respect to , and respectively and setting the partial derivatives to zero, a system of equations is obtained:

[0070]

[0071] Solve the above system of equations to obtain the estimated values , and of , and , the finally obtained prediction model for the thickness fluctuation of the photoresist is , where is the predicted photoresist thickness, is the exposure energy, is the temperature;

[0072] Step 44, Model inspection and optimization: Use the coefficient of determination to evaluate the prediction model for the thickness fluctuation of the photoresist, where , where is 's average value, and when is close to 1, the better the fitting effect of the model to the data.

[0073] Step 5, Formulation and implementation of the fluctuation suppression strategy: Based on the fluctuation prediction model, formulate the corresponding fluctuation suppression strategy, that is, when it is predicted that the photoresist thickness will have a large fluctuation, automatically adjust the exposure energy and temperature control parameters, and intervene in the manufacturing process in real time through the feedback control system to suppress the fluctuation of the process parameters and ensure the stability of the semiconductor manufacturing process; please refer to Figure 4 , the specific working logic is as follows:

[0074] Threshold setting: Based on historical data and the quality standards of the lithography process, determine the normal fluctuation range of the photoresist thickness. Let the average value of the normal photoresist thickness be , and the photoresist thickness fluctuation range is , where is a reasonable fluctuation threshold set according to the process accuracy requirements; for example, through a large amount of historical data statistical analysis, the average value of the photoresist thickness of a certain lithography process is 500nm. Combining with the process requirements, set , then the normal fluctuation range is ;

[0075] Prediction and judgment: Use the established prediction model for the thickness fluctuation of the photoresist , input the current exposure energy and temperature data in real time, predict the photoresist thickness, and when the predicted value exceeds the normal fluctuation range , it is determined that the photoresist thickness will have a large fluctuation;

[0076] Strategy formulation: Use the PID control algorithm to formulate the control strategy. Specifically, its output consists of the proportional term , the integral term and the differential term , and the formula is: , is the error at the current moment, that is, the difference between the target value and the actual predicted value , , and are the proportionality coefficient, integral coefficient, and derivative coefficient respectively. Among them, the proportional term instantly adjusts the control variable according to the magnitude of the current error; the integral term is used to eliminate the steady-state error of the system; the derivative term adjusts the control variable in advance according to the changing trend of the error; the above coefficients are specifically improved according to the characteristics of the lithography process and the response of the equipment. The improvement is based on the deviation degree and change rate of the predicted value of the photoresist thickness from the normal range, and the coefficients are dynamically adjusted;

[0077] It should be specifically noted that:

[0078] Function of the proportional term: For example, if the predicted photoresist thickness is higher than the upper limit , the proportional term will adjust the exposure energy and temperature control parameters in the direction of reducing the photoresist thickness, and the adjustment amplitude is proportional to the magnitude of the error;

[0079] Function of the integral term: The integral term accumulates over time. The integral term will continuously accumulate the error, prompting the control variable to be continuously adjusted until the error is zero. For example, during the lithography process, if there are some minor interference factors that continuously affect the photoresist thickness, the integral term can gradually correct the control parameters to bring the photoresist thickness back to the normal range;

[0080] Function of the derivative term: When the predicted photoresist thickness has a tendency to quickly deviate from the normal range, the derivative term will quickly increase or decrease the control variable to suppress this tendency. For example, if is positive and large, indicating that the photoresist thickness is rising rapidly, the derivative term will increase the adjustment intensity of the exposure energy and temperature control parameters to slow down the rising speed of the photoresist thickness.

[0081] Consider multi-factor interference compensation: The photoresist thickness is affected not only by the exposure energy and temperature, but also by various factors such as the characteristics of the photoresist itself, equipment aging, and environmental humidity. A disturbance compensation term is introduced into the PID algorithm to establish a multi-factor interference model; collect data on the influence of changes in different factors on the photoresist thickness, and analyze the relationship between each factor and the change in the photoresist thickness. Specifically, let the set of disturbance factors be , and the corresponding compensation coefficient be , then the adjusted control variable is , where is the control variable after dynamically adjusting the coefficient;

[0082] Strategy execution: According to the control variable calculated by the PID algorithm , automatically adjust the exposure energy control device of the exposure equipment and the control parameters of the temperature regulation equipment. For example, if it is calculated that the exposure energy needs to be reduced, the system will send an instruction to the energy regulation module of the exposure equipment to reduce the output exposure energy; for temperature control, if the temperature needs to be increased, the system will control the heating device to increase the power and raise the temperature of the lithography environment or the key parts of the equipment. Through real-time feedback control, the thickness of the photoresist is always maintained within the normal fluctuation range to ensure the stability of the semiconductor manufacturing process.

[0083] Specifically 、 and Specific improvements are made according to the characteristics of the lithography process and the response of the equipment. The specific implementation steps are as follows:

[0084] Step A, Real-time data acquisition: During the lithography process, the thickness data of the photoresist is collected in real time through sensors. Let the measured value of the photoresist thickness at the current moment be , the normal photoresist thickness range is , and at the same time, the exposure energy and temperature at the corresponding moment are obtained;

[0085] Step B, Calculate the change rate: Calculate the change rate of the photoresist thickness error , where is the time interval of data acquisition, and the error change rate is calculated through the error values at adjacent moments, that is, the change trend of the photoresist thickness error;

[0086] Step C, Construct a coefficient adjustment function:

[0087] Determine the function form: Use multiple linear regression to construct a coefficient adjustment function. The adjustment function of the proportional coefficient is , let the function form be , where 、 and are coefficients to be determined; similarly, the adjustment function of the integral coefficient is set as , and the adjustment function of the differential coefficient is set as ;

[0088] Coefficient training: Train through the preset experimental data to determine the coefficients in the above functions. Prepare several groups of experimental data under different lithography process conditions. Each group of data includes the measured value of the photoresist thickness, the corresponding error , the error change rate and the best performance under the circumstances , and values, and use the least squares method to solve for the coefficients; for to solve for the coefficients of the adjustment function, assume there are groups of experimental data, and the goal is to make . Respectively, take the partial derivatives of , and and set them to zero to obtain the system of equations:

[0089]

[0090] Solve this system of equations to obtain the values of , and , thereby determining the adjustment function; similarly, determine the coefficients and in the , , and , , ;

[0091] Step D, Dynamic Coefficient Update: Substitute the calculated error and the error change rate into the determined coefficient adjustment function to calculate the adjusted coefficients , and to be adopted at the current moment. Substitute the adjusted coefficients into the PID control algorithm formula to calculate the control quantity .

[0092] Among them, the system automatically adjusts the exposure energy and temperature control parameters according to the calculated control quantity , such as controlling the energy output device and temperature regulation device of the exposure equipment, to achieve precise control of the photoresist thickness. In the subsequent photolithography process, continuously repeat the above steps of data acquisition, coefficient calculation, and control quantity adjustment, and continuously adjust the coefficients of the PID control algorithm according to the actual situation of the photoresist thickness to ensure that the photoresist thickness is always stable within the normal range.

[0093] Step 6, Data Storage: Store the preprocessed data in Step 2 and the fluctuation suppression strategy data in Step 5 into the distributed database, and classify and store them according to different process links, equipment types, and time series for quick query and call.

[0094] Embodiment 2

[0095] Please refer toFigure 5 , an industrial big data governance system, according to an industrial big data governance method, the system includes:

[0096] The data acquisition module collects process parameters and equipment operation status data in the manufacturing process in real time through sensors, and cleans error values, fills in missing values, and standardizes the data format of the collected data to obtain preprocessed data;

[0097] The semiconductor process fluctuation suppression module processes the process parameter data using time series algorithms, differentiates high-frequency noise and low-frequency trend fluctuations through wavelet transform, and then constructs a prediction model based on the fluctuation characteristics and process principles, that is, a mathematical model of photoresist thickness, exposure energy, and temperature; according to the model prediction, automatically adjust the exposure energy parameter to stabilize the process;

[0098] The data storage module is used to store the preprocessed data in the acquisition module and the fluctuation suppression strategy data in the semiconductor process fluctuation suppression module into a distributed database, and classify and store them according to different process links, equipment types, and time series.

[0099] The present invention focuses on the field of industrial big data in semiconductor manufacturing, aims to solve the problems in data governance and process control in this industry, and proposes an innovative industrial big data governance method and system.

[0100] In the data processing flow, starting from data acquisition, by installing sensors such as temperature, pressure, and lithography accuracy on various semiconductor manufacturing equipment, the process parameters and equipment operation status data in the manufacturing process are collected in real time to ensure the comprehensiveness and real-time nature of the data. The collected data undergoes preliminary cleaning, filling in missing values, and standardization processing to unify the format, providing a high-quality data basis for subsequent analysis.

[0101] For the lithography process, using time series analysis algorithms and wavelet transform technology, extract the fluctuation period and amplitude characteristics of data such as exposure time and photoresist thickness, identify high-frequency noise and low-frequency trend fluctuations, and provide accurate data support for establishing a fluctuation model. Based on historical data and process principles, use the multiple linear regression algorithm to construct a mathematical model between photoresist thickness, exposure energy, and temperature, and perform parameter estimation through the least squares method, and use the coefficient of determination to evaluate and optimize the model to accurately simulate the fluctuation of process parameters under different conditions.

[0102] To suppress the fluctuations in photoresist thickness, a strategy is formulated based on a fluctuation prediction model. First, a threshold for the normal fluctuation range is set, and the model is used to predict the photoresist thickness in real time. When the predicted value exceeds the range, an improved PID control algorithm is adopted to adjust the exposure energy and temperature control parameters. This algorithm dynamically adjusts the coefficients according to the deviation degree and change rate of the predicted photoresist thickness from the normal range, and at the same time takes into account the interference of multiple factors such as the characteristics of the photoresist itself, equipment aging, and environmental humidity and compensates for them to ensure the stability of the photoresist thickness and the stability of the semiconductor manufacturing process.

[0103] In addition, the present invention classifies and stores the preprocessed data and the fluctuation suppression strategy data in a distributed database for convenient query and call.

[0104] Through a complete set of data governance processes, the present invention effectively improves the data quality in the semiconductor manufacturing process, accurately controls key process parameters such as photoresist thickness, enhances the process stability, improves the product quality and production efficiency, provides important technical support for the intelligent development of the semiconductor manufacturing industry, and has remarkable innovation and practicality.

[0105] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made in these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An industrial big data governance method, characterized in that: The following steps are involved: Step 1: Data collection: By installing temperature sensors, pressure sensors, and photolithography precision sensors on semiconductor manufacturing equipment, process parameter data in the manufacturing process can be collected in real time; Step 2: Data preprocessing: Perform preliminary cleaning on the collected data, remove obvious erroneous values, fill in a small number of missing values, standardize the data, unify the data format, and obtain preprocessed data; Step 3, fluctuation feature extraction: receiving pre-processed data, extracting the fluctuation period and amplitude characteristics of the exposure time and photoresist thickness in the photolithography process over time, decomposing the data into different frequency components through wavelet transform technology, and identifying high-frequency noise fluctuations and low-frequency trend fluctuations; Step 4: Establish a fluctuation model: Based on the extracted fluctuation characteristics and the principle of semiconductor manufacturing process, a process fluctuation prediction model is established. Specifically, based on historical data and physical models, a mathematical model between the fluctuation of photoresist thickness and exposure energy and temperature in the photolithography process is constructed. This model simulates the fluctuation of process parameters under different conditions. Step 5, formulation and implementation of fluctuation suppression strategy: formulate corresponding fluctuation suppression strategy based on the fluctuation prediction model, that is, when it is predicted that the photoresist thickness will fluctuate greatly, automatically adjust the exposure energy and temperature control parameters, and intervene in the manufacturing process in real time through the feedback control system to suppress the fluctuation of process parameters and ensure the stability of the semiconductor manufacturing process; Step 6, data storage: store the preprocessed data in step 2 and the fluctuation suppression strategy data in step 5 in a distributed database, and classify and store them according to different process links, equipment types and time series for quick query and call.

2. The industrial big data management method according to claim 1 is characterized in that: The process parameter data in step 1 specifically include the following contents: photolithography link: installing a temperature sensor near the photoresist coating area of ​​the photolithography equipment to monitor the ambient temperature of the photoresist during the photolithography process; adding a pressure sensor to the photoresist coating equipment to collect pressure data applied to the photoresist during coating; using a high-precision photolithography precision sensor to record the precision deviation between the photolithography pattern and the actual photoresist forming in real time; related link: installing a flow sensor on the etching equipment to collect etching gas flow data, as the etching gas flow affects the etching rate.

3. The industrial big data management method according to claim 1 is characterized in that: The specific steps of extracting the fluctuation characteristics are as follows: Step S31, data preparation: collect the exposure time and photoresist thickness data of the photolithography process in step 2 for a period of time, so as to obtain continuous Exposure time series at each time point and the photoresist thickness sequence ; Step S32, time series analysis algorithm: Periodic analysis: Use Fourier transform to extract periodic features. The Fourier transform formula is: ,in is the original time series, i.e., the exposure time and photoresist thickness , is the transformed frequency domain representation, through the exposure time series Perform Fourier transform to obtain its amplitude at different frequencies. In the frequency domain, the period corresponding to the frequency component with a larger amplitude is the fluctuation period of the exposure time. The same applies to the photoresist thickness sequence , find the fluctuation period; Amplitude feature extraction: Extract the fluctuation amplitude feature and calculate the range of the time series, that is, for the exposure time series , calculate the range Indicates the maximum fluctuation range of exposure time within the observation period; for the photoresist thickness series , very poor Indicates the fluctuation range of photoresist thickness; Step 33: Decomposing the frequency components by wavelet transform: The exposure time and photoresist thickness data are processed by discrete wavelet transform. The discrete wavelet transform selects a suitable wavelet basis function. ,in is the scale parameter, is the translation parameter, for the exposure time series Discrete Wavelet Transform , we get the wavelet coefficients at different scales and positions. At high-frequency scales, the wavelet coefficients reflect the fast-changing part in the exposure time series, that is, the high-frequency noise fluctuation. Similarly, for the photoresist thickness series Discrete Wavelet Transform , separate its high-frequency noise fluctuations and low-frequency trend fluctuations, that is, fully grasp the fluctuation characteristics of lithography process parameter data, and provide an accurate data basis for the subsequent establishment of a fluctuation model.

4. The industrial big data management method according to claim 1 is characterized in that: The specific execution steps of establishing the fluctuation model are as follows: Step 41: Data collation and preparation: Collect historical data from the photolithography process, including the photoresist thickness measurement values ​​during the production of different batches of products, the corresponding exposure energy setting values, and the ambient temperature or temperature data of key parts of the equipment. Set data and record the photoresist thickness as , the exposure energy is recorded as , the temperature is recorded as ; Step 42: Modeling using a multiple linear regression algorithm; Based on the principle of photolithography, the thickness of photoresist is affected by the combined influence of exposure energy and temperature. The multivariate linear regression algorithm is used for modeling, specifically: ,in is the dependent variable, i.e., photoresist thickness, and , is the intercept, and is the regression coefficient, is a random error term, where , and Make estimates using historical data; Step 43, parameter estimation: Use the least squares method to estimate the regression coefficient so that the photoresist thickness With the model prediction value The sum of squared errors between minimum; through About , and Find the partial derivatives and set them to zero to obtain a system of equations: Solving the above system of equations, we get , and Estimated value of , and The final photoresist thickness fluctuation prediction model is ,in is the predicted photoresist thickness, is the exposure energy, is temperature; Step 44: Model testing and optimization: using the coefficient of determination The photoresist thickness fluctuation prediction model is evaluated, including ,in yes The average value of The closer it is to 1, the better the model fits the data.

5. The industrial big data management method according to claim 1 is characterized in that: The specific working logic of the volatility suppression strategy formulation and execution is as follows: Threshold setting: Based on historical data and the quality standards of the photolithography process, determine the normal fluctuation range of the photoresist thickness, and set the average normal photoresist thickness to , the photoresist thickness fluctuation range is ,in A reasonable fluctuation threshold set according to the process accuracy requirements; Prediction and judgment: Use the established photoresist thickness fluctuation prediction model , input the current exposure energy and temperature data in real time, predict the photoresist thickness, and when the predicted value Extraordinary fluctuation range When the photoresist thickness is determined to fluctuate greatly; Strategy formulation: Use PID control algorithm to formulate control strategy, specifically, its output By proportional terms , integral term and the differential term , the formula is: , is the error at the current moment, that is, the difference between the target value and the actual predicted value The difference , , and are proportional coefficient, integral coefficient and differential coefficient respectively, among which the proportional term According to the current error, the control amount is adjusted immediately; Integral Item Used to eliminate the steady-state error of the system; differential term Adjust the control amount in advance according to the changing trend of the error; The above coefficients are specifically improved according to the characteristics of the photolithography process and the response of the equipment. The improvement dynamically adjusts the coefficients according to the deviation degree and change rate of the predicted value of the photoresist thickness from the normal range; Consider multi-factor interference compensation: introduce interference compensation terms into the PID algorithm and establish a multi-factor interference model; collect data on the impact of different factors on the thickness of the photoresist, and analyze the relationship between each factor and the change of the photoresist thickness. Specifically, assume that the interference factor set is , the corresponding compensation coefficient is , then the adjusted control quantity for ,in is the control quantity after dynamic adjustment of coefficient; Strategy execution: The control quantity calculated according to the PID algorithm , automatically adjust the exposure energy control device of the exposure equipment and the control parameters of the temperature regulation equipment.

6. The industrial big data management method according to claim 5 is characterized by: Said , and Specific improvements are made based on the characteristics of the lithography process and the response of the equipment. The specific implementation steps are as follows: Step A, real-time data collection: During the photolithography process, the photoresist thickness data is collected in real time through the sensor. The photoresist thickness is measured as , the normal photoresist thickness range is , and obtain the exposure energy at the corresponding moment and temperature ; Step B, calculate the rate of change: calculate the photoresist thickness error Rate of change ,in is the time interval for data collection, and the error change rate, i.e., the change trend of the photoresist thickness error, is calculated by the error values ​​at adjacent moments; Step C: Construct coefficient adjustment function: Determine the function form: Use multiple linear regression to construct a coefficient adjustment function, where the proportional coefficient The adjustment function is , let the function form be ,in , and is the coefficient to be determined; similarly, the integral coefficient The adjustment function is set to , differential coefficient The adjustment function is set to ; Coefficient training: Train with preset experimental data to determine the coefficients in the above function and prepare several sets of experimental data under different photolithography process conditions. Each set of data contains the measured value of photoresist thickness and the corresponding error. , Error change rate and the best performance in the case , and value, use the least square method to solve the coefficients; for Solve the adjustment function coefficients, assuming that the experimental data are group, target , respectively , and Taking the partial derivatives and setting them equal to zero, we obtain the system of equations: Solving this system of equations, we get , and The value of The adjustment function of and Adjusting coefficients in functions , , and , , ; Step D, dynamic coefficient update: calculate the error , Error change rate , substitute into the determined coefficient adjustment function, and calculate the adjusted coefficient to be used at the current moment , and , substitute the adjusted coefficient into the PID control algorithm formula , calculate the control quantity .

7. An industrial big data management system, characterized by: According to an industrial big data governance method according to any one of claims 1 to 6, the system comprises: The data acquisition module collects process parameters and equipment operation status data in the manufacturing process in real time through sensors, and cleans the collected data of error values, fills in missing values, and standardizes the data format to obtain pre-processed data; The semiconductor process fluctuation suppression module uses a time series algorithm to process process parameter data, distinguishes high-frequency noise from low-frequency trend fluctuations by wavelet transform, and then builds a prediction model based on the fluctuation characteristics and process principles, that is, a mathematical model of photoresist thickness, exposure energy, and temperature; according to the model prediction, the exposure energy parameters are automatically adjusted to stabilize the process; The data storage module is used to store the pre-processed data in the acquisition module and the fluctuation suppression strategy data in the semiconductor process fluctuation suppression module in a distributed database, and is classified and stored according to different process links, equipment types and time series.

Images