Semiconductor photoetching system based on intelligent industrial robot

Abstract

The invention relates to the field of industrial automation, and particularly provides a semiconductor photoetching system based on an intelligent industrial robot. And a multi-sensor fusion and real-time compensation algorithm is adopted, so that photoetching positioning errors are accurately sensed and corrected, the positioning precision is greatly improved, and the stable performance of the chip is ensured. According to a multi-physical field monitoring and self-adaptive control algorithm, process parameters are dynamically regulated and controlled according to a photoetching environment and a photoresist state, pattern transfer is optimized, pattern defects are avoided, and photoetching pattern quality is improved. According to the multi-modal data fusion and deep learning algorithm, photoresist coating parameters are automatically adjusted according to the surface characteristics of the wafer, uniform coating is realized, and the stability of the photoetching process is enhanced. The reinforcement learning algorithm assists the multiple robots in efficient collaboration, optimizes task allocation, improves the overall photoetching efficiency and the resource utilization rate, and promotes the progress of the semiconductor photoetching technology in all directions.

Description

Technical Field

[0001] The present invention belongs to the field of industrial automation, and relates to a semiconductor lithography system based on an intelligent industrial robot.

[0002] The semiconductor lithography process is a key step in transferring the circuit pattern on the mask plate to the surface of the semiconductor wafer coated with photoresist, which determines the feature size and performance of the chip. With the continuous development of semiconductor technology, the integration degree of chips is getting higher and higher, and the requirements for the accuracy and quality of the lithography process are also becoming more and more stringent. Industrial robots play an important role in the operation and control of semiconductor lithography equipment. However, there are still many problems in the existing industrial robot technology in the lithography process, which seriously restricts the further development of semiconductor lithography technology.

[0003] It is difficult to ensure the lithography positioning accuracy: During the lithography process, it is necessary to accurately align the pattern on the mask plate to a specific position on the wafer, and the positioning accuracy directly affects the performance and yield of the chip. The traditional industrial robot lithography positioning system mainly relies on simple vision positioning technology and preset alignment parameters, and cannot accurately adapt to the influence of factors such as the micro-undulation, thermal expansion, and mechanical vibration of the wafer surface. Moreover, due to the long-term operation of the lithography equipment and the change of environmental factors, the accuracy of the positioning system will gradually decrease, resulting in an increase in the alignment deviation of the lithography pattern, affecting the performance and reliability of the chip.

[0004] The accuracy of lithography pattern transfer is poor: The accuracy of lithography pattern transfer refers to the ability to accurately copy the pattern on the mask plate to the photoresist layer. In the existing industrial robot lithography system, during the pattern transfer process, it is easily affected by factors such as uneven thickness of the photoresist, inconsistent exposure dose, and fluctuations in developing process parameters, resulting in problems such as deformation, blurring, and distortion of the lithography pattern. These problems will not only affect the performance of the chip, but also increase the manufacturing cost of the chip.

[0005] It is difficult to control the uniformity of photoresist coating: The uniformity of photoresist coating is crucial for the quality of the lithography pattern and the stability of the lithography process. At present, during the photoresist coating process by industrial robots, fixed coating parameters and preset coating paths are mainly used, and cannot be adjusted in real time according to the size, shape, and surface characteristics of the wafer. Moreover, due to the structural and technological limitations of the coating equipment, it is easy to cause uneven coating thickness of the photoresist on the wafer surface, affecting the quality of the lithography pattern and the stability of the lithography process. Summary of the Invention

[0006] The present invention provides a semiconductor lithography system based on intelligent industrial robots, mainly aiming at the key technical problems existing in the semiconductor lithography process. By developing innovative algorithms and intelligent control strategies, it aims to improve key indicators such as the positioning accuracy, pattern transfer accuracy, and photoresist coating uniformity of industrial robots during the lithography process, so as to meet the increasing requirements of the semiconductor industry for high-precision, high-efficiency, and high-quality lithography processes, including a high-precision lithography positioning algorithm based on multi-sensor fusion and real-time compensation. This algorithm deploys various sensors such as laser interferometers, capacitive sensors, and vision sensors in the industrial robot lithography positioning system, uses multi-sensor fusion technology to construct a real-time state model for lithography positioning, and combines a real-time compensation algorithm to adjust the positioning parameters in real time to correct the lithography positioning error. In this algorithm, the positioning error compensation is achieved through the following formula: ΔP = K·f(S1, S2, …, Sn), where ΔP represents the adjustment amount of the positioning parameters, K is the compensation coefficient matrix, and its element values are determined according to the weights of different sensor data on the positioning impact and are obtained by fitting a large amount of experimental data; f(S1, S2, …, Sn) is an error calculation function based on multi-sensor fusion data. S1, S2, …, Sn respectively represent the original data collected by various sensors such as laser interferometers, capacitive sensors, and vision sensors. After processing these data through multi-sensor fusion technology and inputting them into this function, the quantization value of the positioning error is obtained, and then the adjustment amount of the positioning parameters is calculated according to the formula to achieve real-time compensation for the lithography positioning error and improve the lithography positioning accuracy.

[0007] Furthermore, the multi-sensor data fusion adopts a fusion algorithm based on deep learning, such as a fusion network based on the attention mechanism, to process the data collected by different sensors to construct a real-time state model.

[0008] Furthermore, an optimized algorithm for lithography pattern transfer based on multi-physical field monitoring and adaptive control installs temperature sensors, pressure sensors, electric field sensors, magnetic field sensors, and spectrometers in the lithography equipment, uses multi-source information fusion technology to construct a multi-physical field model for lithography pattern transfer, and combines an adaptive control algorithm to adjust the lithography process parameters in real time to adapt to the changes in the lithography process.

[0009] Furthermore, the multi-source information fusion adopts an algorithm based on deep learning, and uses a convolutional neural network to analyze and process the fused data to extract key feature information for constructing the multi-physical field model.

[0010] Furthermore, a photoresist coating uniformity control algorithm based on multi-modal data fusion and deep learning obtains multi-modal data on the wafer surface by deploying a laser rangefinder, an optical microscope, and an infrared thermal imager on the photoresist coating equipment. The multi-modal data fusion technology is used to construct a comprehensive information model of the wafer surface, and a mapping relationship model between the wafer surface characteristics and the photoresist coating parameters is established in combination with the deep learning algorithm to adjust the coating parameters in real time.

[0011] Furthermore, the deep learning algorithm uses a combination of a convolutional neural network and a recurrent neural network to learn and train the fused multi-modal data to establish a mapping relationship model.

[0012] Furthermore, a multi-robot collaborative lithography scheduling algorithm based on reinforcement learning regards multiple industrial robots in a semiconductor lithography workshop as a multi-agent system, defines the state space, action space, and reward function, and uses the reinforcement learning algorithm to enable the robots to learn the optimal collaborative lithography scheduling strategy, and dynamically adjusts the task allocation and production plan according to the real-time production situation and equipment status.

[0013] Furthermore, the reinforcement learning algorithm uses a deep Q-network or a proximal policy optimization algorithm for model training and policy optimization.

[0014] Furthermore, a semiconductor lithography process optimization algorithm based on digital twin and virtual simulation establishes a digital twin model of the semiconductor lithography system, uses virtual simulation technology to simulate and optimize the lithography process, and optimizes the lithography process parameters and procedures according to the simulation results to formulate an optimal process plan.

[0015] Furthermore, the digital twin model is constructed through modeling software and sensor data to achieve real-time interaction and synchronization between the physical entity and the virtual model for lithography process optimization.

[0016] Beneficial effects

[0017] The high-precision lithography positioning algorithm based on multi-sensor fusion and real-time compensation endows the system with the ability to keenly perceive and instantaneously correct lithography positioning errors. Multiple sensors work together to comprehensively capture various types of information in lithography positioning, construct a real-time state model, enabling the system to accurately grasp the relative position between the wafer and the mask, greatly improving the accuracy of lithography positioning, laying a solid foundation for chip manufacturing, and ensuring more stable and reliable chip performance.

[0018] The lithography pattern transfer optimization algorithm for multi-physical field monitoring and adaptive control is like equipping the lithography process with an intelligent "housekeeper". Multi-physical field sensors monitor the changes in the lithography environment in real time, and spectrometers accurately grasp the state of the photoresist, and then a multi-physical field model is constructed. The adaptive control algorithm dynamically adjusts the lithography process parameters according to the model and the real-time information of the photoresist, effectively avoiding problems such as deformation, blurring, and distortion during the pattern transfer process, significantly improving the quality of the lithography pattern, and ensuring the stability of the lithography process.

[0019] The photoresist coating uniformity control algorithm based on multi-modal data fusion and deep learning has completely revolutionized the photoresist coating method. Multi-modal detection equipment obtains comprehensive information on the wafer surface, constructs a comprehensive information model, and the deep learning algorithm establishes an accurate mapping between the wafer surface characteristics and the coating parameters based on this. During the coating process, the system can automatically and accurately adjust the coating parameters according to the real-time data, achieving uniform coating of the photoresist, providing strong guarantee for high-quality lithography patterns, and further improving the stability of the lithography process.

[0020] The multi-robot collaborative lithography scheduling algorithm based on reinforcement learning enables more efficient cooperation of industrial robots in the lithography workshop. By defining a comprehensive state, action space, and reward function, the robots optimize the collaborative strategy in continuous practice. Facing complex and changing production situations, they can quickly and reasonably adjust the task allocation and production plan, greatly improving the overall lithography efficiency and resource utilization rate.

[0021] The semiconductor lithography process optimization algorithm based on digital twin and virtual simulation has opened up a new path for lithography process optimization. The digital twin model accurately replicates the physical entity, and virtual simulation can anticipate potential problems in advance, helping to optimize process parameters and processes, formulate the optimal plan, effectively reduce the trial-and-error cost, shorten the process optimization cycle, and promote a significant increase in lithography efficiency and product yield. Description of the Drawings

[0022] Figure 1 Flow chart of the system operation principle. Detailed Implementation Manner

[0023] Example 1

[0024] A well-known semiconductor manufacturing enterprise is committed to the production of high-end chips, and its lithography workshop undertakes the key task of accurately transferring complex circuit patterns onto wafers. With the continuous improvement of market requirements for chip performance and integration, the enterprise faces problems such as insufficient lithography process accuracy, unstable pattern transfer quality, and poor photoresist coating uniformity, which seriously affect the product yield and production efficiency. To solve these problems, the enterprise introduced a semiconductor lithography system based on intelligent industrial robots.

[0025] II. System Implementation Process

[0026] (1) Implementation of a High-Precision Lithography Positioning Algorithm Based on Multi-Sensor Fusion and Real-Time Compensation

[0027] Sensor Deployment: In the lithography positioning system of an industrial robot, a high-precision laser interferometer is installed. Its displacement measurement accuracy can reach ±0.1 nm, and it can accurately measure the displacement information of the wafer in the X, Y, and Z-axis directions. A capacitive sensor is used to monitor the microscopic undulations on the wafer surface in real time, with a resolution of 0.01 nm. The vision sensor is equipped with a high-resolution lens with a pixel count of 6000×5000, and it can clearly obtain the image information of the wafer and the mask. These sensors collect data in real time at a frequency of 100 times per second and transmit it to the data processing unit through a high-speed network.

[0028] Data Fusion and Model Construction: The data processing unit uses a fusion network based on the attention mechanism to fuse the data collected by the laser interferometer, capacitive sensor, and vision sensor. For example, when processing a batch of 12-inch wafers, the laser interferometer detects a 0.5-nm displacement of the wafer in the X-axis direction, the capacitive sensor detects a 0.03-nm microscopic undulation in a certain area of the wafer surface, and the vision sensor identifies the relative position deviation between the wafer and the mask. The fusion network assigns corresponding weights according to the importance of the data from each sensor and constructs a real-time state model of the lithography positioning process. A convolutional neural network is used to analyze the fused data and extract key features, such as the precise position, attitude, and surface microscopic features of the wafer.

[0029] Real-Time Compensation and Positioning Execution: According to the constructed real-time state model, the real-time compensation algorithm calculates the positioning parameter adjustment amount according to the formula

[0030] ΔP = K·f(S1, S2, …, Sn). Among them, the compensation coefficient matrix K is determined by analyzing a large amount of historical data and experimental results, and the error calculation function f(S1, S2, …, Sn)

[0031] accurately obtains the quantization value of the positioning error based on the sensor fusion data. For example, when it is calculated that the wafer needs to be adjusted by 0.3 nm in the Y-axis direction, the robot control system quickly adjusts the positioning parameters of the robot to ensure the precise alignment of the mask and the wafer. During the entire lithography positioning process, the sensor data is continuously monitored in real time, and the positioning parameters are dynamically adjusted, improving the lithography positioning accuracy from ±10 nm of the traditional method to within ±1 nm, greatly improving the performance and yield of the chip.

[0032] (2) Implementation of a Lithography Pattern Transfer Optimization Algorithm Based on Multi-Physical Field Monitoring and Adaptive Control

[0033] Sensor Deployment: In a lithography device, temperature sensors are installed with an accuracy of up to ±0.1°C to monitor the temperature changes during the lithography process in real time. The pressure sensor has a measurement accuracy of ±0.01 Pa and can monitor the pressure conditions during exposure. Electric field sensors and magnetic field sensors can detect the changes in the electric and magnetic field intensities around the lithography area. The spectrometer is used to monitor the exposure dose and development degree of the photoresist in real time, with an accuracy of ±0.1%. These sensors collect data at a frequency of 50 times per second and transmit it to the data processing unit.

[0034] Multi-source Information Fusion and Model Construction: The data processing unit uses a multi-source information fusion algorithm based on deep learning to fuse data such as temperature, pressure, electric field, magnetic field, and photoresist exposure dose. For example, during a lithography process, the temperature sensor detects that the temperature in the lithography area has increased by 0.5°C, the pressure sensor finds that the exposure pressure has decreased by 0.02 Pa, and the spectrometer detects insufficient photoresist exposure dose. Through multi-source information fusion technology, a multi-physical field model of the lithography pattern transfer process is constructed. A convolutional neural network is used to analyze the fused data and extract the key features affecting lithography pattern transfer, such as the influence of temperature and pressure on the photoresist performance, and the interference of electric and magnetic fields on the exposure process.

[0035] Adaptive Control and Lithography Execution: According to the constructed multi-physical field model and the real-time change information of the photoresist, the adaptive control algorithm adjusts the lithography process parameters of the industrial robot in real time. For example, when the temperature rises, the exposure intensity is automatically reduced to prevent overexposure of the photoresist; when the spectrometer detects insufficient exposure dose, the exposure time is extended. During the lithography process, the sensor data is continuously monitored in real time, and the lithography process parameters are dynamically adjusted, significantly improving the accuracy of lithography pattern transfer, effectively improving problems such as deformation, blurring, and distortion of the lithography pattern, and greatly enhancing the quality of the lithography pattern and the stability of the lithography process.

[0036] (3) Implementation of the Photoresist Coating Uniformity Control Algorithm Based on Multi-modal Data Fusion and Deep Learning

[0037] Detection Equipment Deployment: On the photoresist coating equipment, a laser rangefinder is installed with a measurement accuracy of up to ±0.01 μm to obtain the height information of the wafer surface. The optical microscope can magnify 2000 times to observe the microscopic morphology of the wafer surface. The infrared thermal imager can detect the temperature distribution on the wafer surface with an accuracy of ±0.2°C. These detection equipment collect data at a frequency of 30 times per second and transmit it to the data processing center.

[0038] Multi-modal Data Fusion and Feature Extraction: The data processing center uses a deep learning-based multi-modal data fusion algorithm to fuse the data collected by the laser rangefinder, optical microscope, and infrared thermal imager. For example, the laser rangefinder detects that the height of a certain area on the edge of the wafer is 0.05 μm higher than that of the central area, the optical microscope observes that there are tiny particles in this area, and the infrared thermal imager shows that the temperature of this area is 0.3 °C higher than the average temperature. Through multi-modal data fusion technology, a comprehensive information model of the wafer surface is constructed. A convolutional neural network is used to analyze the fused data to extract key features of the wafer surface, such as surface topography, temperature distribution, and photoresist thickness information, etc.

[0039] Deep Learning Model Training and Coating Execution: Surface data and photoresist coating data of more than 100,000 different types of wafers are collected to construct a training dataset. A deep learning algorithm combining convolutional neural network and recurrent neural network is used to learn and train the training dataset to establish a mapping relationship model between the wafer surface characteristics and photoresist coating parameters. During the actual coating process, the real-time collected multi-modal data is input into the trained model, and the model outputs the corresponding adjustment amount of photoresist coating parameters according to the wafer surface characteristics. For example, when the laser rangefinder detects a change in the height of the wafer surface, the deep learning model automatically adjusts the coating speed and pressure of the photoresist to ensure the uniformity of the photoresist coating. During the coating process, the data of the detection equipment is continuously monitored in real time, and the coating parameters are dynamically adjusted, greatly improving the uniformity of the photoresist coating, and further improving the quality of the photolithography pattern and the stability of the photolithography process.

[0040] (4) Implementation of a Multi-robot Cooperative Lithography Scheduling Algorithm Based on Reinforcement Learning

[0041] Agent Definition and Communication Mechanism Establishment: Ten industrial robots in the lithography workshop are defined as ten agents, and each agent has independent decision-making ability and goals. A high-speed communication network based on 5G technology is established to achieve real-time information interaction between agents. For example, when a robot completes the current lithography task, it sends the task completion information and its own status information to other agents through the communication network so that other agents can adjust the task allocation and production plan according to the real-time situation.

[0042] Reinforcement Learning Model Training and Application: Define the state space of reinforcement learning, including the robot's position, task progress, working status, and environmental information of the workshop, etc.; the action space includes various motion actions of the robot and task assignment decisions; the reward function takes lithography efficiency, pattern transfer accuracy, photoresist coating uniformity, etc. as reward indicators. Collect the historical operation data of the lithography workshop in the past year, and use the Proximal Policy Optimization algorithm to train and optimize the reinforcement learning model. In actual production, the agent selects actions based on real-time state information and adjusts the policy according to the reward feedback after executing the actions. For example, when the lithography task volume in a certain area of the workshop suddenly increases, the agent calculates the optimal task assignment plan through the reinforcement learning model, assigns some tasks to idle or more efficient robots, realizes efficient collaborative work among multiple robots, the overall lithography efficiency is increased by more than 30%, and the resource utilization rate is also significantly improved.

[0043] (5) Implementation of Semiconductor Lithography Process Optimization Algorithm Based on Digital Twin and Virtual Simulation

[0044] Digital Twin Model Construction: Use professional modeling software, combine the structural parameters, physical characteristics of the lithography equipment, and the data collected by sensors to establish a digital twin model of the semiconductor lithography system. Precisely map physical entities such as industrial robots, lithography equipment, and wafers into the virtual space to achieve real-time interaction and synchronization between physical entities and virtual models. For example, when the temperature of a certain component in the actual lithography equipment changes, the corresponding component in the digital twin model will also update the temperature information in real time.

[0045] Virtual Simulation and Process Optimization: Use virtual simulation technology to simulate and optimize the semiconductor lithography process. In the virtual environment, set various different lithography process parameters and scenarios, and predict and analyze in advance possible problems that may occur during the lithography process, such as lithography positioning deviation, inaccurate pattern transfer, uneven photoresist coating, etc. According to the results of the virtual simulation, optimize and adjust the parameters and processes of the semiconductor lithography process. For example, through virtual simulation, it is found that a certain photoresist is prone to pattern deformation at a specific temperature and exposure time, so in actual production, the type and process parameters of the photoresist are adjusted to formulate the optimal lithography process plan. Then apply the optimized plan to actual production to achieve the optimization and improvement of the semiconductor lithography process, the lithography efficiency is increased by 25%, and the product yield is increased from the original 80% to over 90%.

[0046] The high-precision lithography positioning algorithm based on multi-sensor fusion and real-time compensation endows the lithography positioning system with unprecedented precision. By integrating laser interferometers, capacitive sensors and vision sensors, it comprehensively captures the information of wafers and masks, and constructs a real-time state model. This enables the system to be no longer limited by the helplessness of traditional methods in dealing with factors such as microscopic undulations and thermal expansions, and can keenly sense and quickly correct positioning errors. In actual production, the lithography positioning accuracy has been improved from ±10nm to within ±1nm, providing a solid position foundation for chip manufacturing, ensuring the precise alignment of circuit patterns, greatly enhancing the stability and reliability of chip performance, and effectively improving the product yield.

[0047] The lithography pattern transfer optimization algorithm based on multi-physical field monitoring and adaptive control is like installing intelligent "monitoring eyes" and "control hands" for the lithography process. Multi-physical field sensors such as temperature, pressure, electric field and magnetic field sensors can sense the subtle changes in the lithography environment in real time. Combining with the precise monitoring of photoresist exposure and development by spectrometers, a multi-physical field model is constructed. Based on this, the adaptive control algorithm dynamically adjusts the lithography process parameters, effectively overcoming problems such as pattern deformation, blurring and distortion caused by uneven photoresist thickness and inconsistent exposure dose. The quality of lithography patterns has been significantly improved, and the stability of the lithography process has been greatly enhanced, providing high-quality pattern transfer guarantee for chip manufacturing and promoting the lithography process to a higher level.

[0048] The photoresist coating uniformity control algorithm based on multi-modal data fusion and deep learning has brought revolutionary changes to photoresist coating. Laser rangefinders, optical microscopes and infrared thermal imagers work together to comprehensively obtain multi-modal data on the wafer surface and construct a comprehensive information model. Through learning a large amount of data, the deep learning algorithm establishes an accurate mapping relationship between the wafer surface characteristics and photoresist coating parameters. During the coating process, the system automatically and precisely adjusts the coating parameters according to the real-time collected data to ensure that the photoresist evenly covers the wafer surface. This greatly improves the quality of lithography patterns, enhances the stability of the lithography process, reduces lithography defects caused by uneven coating, and lays a foundation for the fine manufacturing of semiconductor lithography.

[0049] The multi-robot collaborative lithography scheduling algorithm based on reinforcement learning transforms multiple industrial robots in the lithography workshop into an intelligent agent team with efficient collaboration. By defining a comprehensive state space, action space and reward function, the robots continuously optimize their collaborative strategies in the continuous interaction with the environment. When the task volume distribution is uneven or the equipment status changes, the robots can quickly adjust the task allocation and production plan according to the reinforcement learning model. This realizes the efficient collaboration among multiple robots, greatly improves the overall lithography efficiency, significantly improves the resource utilization rate, makes the production process in the lithography workshop more smooth and efficient, and gives full play to the maximum efficiency of the equipment.

[0050] The semiconductor lithography process optimization algorithm based on digital twin and virtual simulation has opened up a brand-new path for lithography process optimization. The digital twin model accurately replicates the physical entity, enabling real-time interactive synchronization between the physical and virtual spaces. In the virtual simulation environment, various lithography process parameters and scenarios are simulated in advance to gain in-depth insights into potential problems. According to the simulation results, the process parameters and flow are optimized specifically, and the optimal solution is formulated and applied to actual production. This effectively reduces the trial-and-error costs in actual production, shortens the process optimization cycle, significantly improves the lithography efficiency, enhances the product yield rate, promotes the continuous innovation of semiconductor lithography processes, and helps semiconductor manufacturing enterprises gain an advantageous position in the fierce market competition.

Claims

1. An intelligent industrial robot-based semiconductor lithography system, characterized in that, It includes a high-precision lithography positioning algorithm based on multi-sensor fusion and real-time compensation. This algorithm deploys multiple sensors such as laser interferometers, capacitive sensors, and vision sensors in the industrial robot lithography positioning system, uses multi-sensor fusion technology to construct a real-time state model for lithography positioning, and combines a real-time compensation algorithm to adjust positioning parameters in real time to correct lithography positioning errors. In this algorithm, the positioning error compensation is achieved through the following formula: ΔP = K·f(S1, S2, …, Sn) where ΔP represents the adjustment amount of the positioning parameter, K is the compensation coefficient matrix, the element values of which are determined according to the weights of the influence of different sensor data on positioning and are obtained by fitting a large amount of experimental data; f(S1, S2, …, Sn) is an error calculation function based on multi-sensor fusion data. S1, S2, …, Sn respectively represent the original data collected by multiple sensors such as laser interferometers, capacitive sensors, and vision sensors. After processing these data through multi-sensor fusion technology and inputting them into this function, the quantization value of the positioning error is obtained, and then the adjustment amount of the positioning parameter is calculated according to the formula to achieve real-time compensation for lithography positioning errors and improve lithography positioning accuracy.

2. The semiconductor lithography system based on an intelligent industrial robot according to claim 1, wherein The multi-sensor data fusion adopts a fusion algorithm based on deep learning, such as a fusion network based on the attention mechanism, to process the data collected by different sensors to construct a real-time state model.

3. The semiconductor lithography system based on an intelligent industrial robot according to claim 1, wherein A lithography pattern transfer optimization algorithm based on multi-physical field monitoring and adaptive control installs temperature sensors, pressure sensors, electric field sensors, magnetic field sensors, and spectrometers in the lithography equipment, uses multi-source information fusion technology to construct a multi-physical field model for lithography pattern transfer, and combines an adaptive control algorithm to adjust lithography process parameters in real time to adapt to the changes in the lithography process.

4. The semiconductor lithography system based on an intelligent industrial robot according to claim 3, wherein The multi-source information fusion adopts an algorithm based on deep learning, and uses a convolutional neural network to analyze and process the fused data to extract key feature information for constructing a multi-physical field model.

5. The semiconductor lithography system based on an intelligent industrial robot according to claim 1, characterized in that, A lithography resist coating uniformity control algorithm based on multi-modal data fusion and deep learning deploys a laser rangefinder, an optical microscope, and an infrared thermal imager on the lithography resist coating equipment to obtain multi-modal data on the wafer surface, uses multi-modal data fusion technology to construct a comprehensive information model of the wafer surface, and combines a deep learning algorithm to establish a mapping relationship model between the wafer surface characteristics and the lithography resist coating parameters to adjust the coating parameters in real time.

6. The semiconductor lithography system based on an intelligent industrial robot according to claim 5, characterized in that, The deep learning algorithm adopts a combination of a convolutional neural network and a recurrent neural network to learn and train the fused multi-modal data to establish a mapping relationship model.

7. The semiconductor lithography system based on an intelligent industrial robot according to claim 1, wherein, A multi-robot collaborative lithography scheduling algorithm based on reinforcement learning regards multiple industrial robots in a semiconductor lithography workshop as a multi-agent system, defines a state space, an action space, and a reward function, and uses the reinforcement learning algorithm to enable the robots to learn the optimal collaborative lithography scheduling strategy and dynamically adjust task allocation and production plans according to the real-time production situation and equipment status.

8. The semiconductor lithography system based on an intelligent industrial robot according to claim 7, characterized in that, The reinforcement learning algorithm uses a deep Q network or a proximal policy optimization algorithm for model training and policy optimization.

9. The semiconductor lithography system based on an intelligent industrial robot according to claim 1, characterized in that, An optimization algorithm for semiconductor lithography process based on digital twin and virtual simulation. By establishing a digital twin model of the semiconductor lithography system, using virtual simulation technology to simulate and optimize the lithography process, and optimizing the lithography process parameters and procedures according to the simulation results to formulate the optimal process plan.

10. The semiconductor lithography system based on an intelligent industrial robot according to claim 9, characterized in that, The digital twin model is constructed through modeling software and sensor data, realizing real-time interaction and synchronization between the physical entity and the virtual model for lithography process optimization.

Images