Monocrystalline silicon thermal field multi-section dynamic cooperative heating control method and system

Abstract

The invention relates to a monocrystalline silicon thermal field multi-section dynamic cooperative heating control method and system, and the method comprises the following steps: constructing a thermal field multi-section temperature coupling model, dividing a thermal field into n axial heating sections, and building a dynamic relation between power input and temperature gradient; a deep reinforcement learning agent is designed, the state space of the deep reinforcement learning agent integrates the temperature of each heating section, the crystal lifting speed, the crucible lifting speed and the temperature gradient, and the action space outputs the power adjusting quantity of each heating section; defining a multi-target reward function, and jointly optimizing temperature gradient tracking, total power change and crucible bottom safety threshold; training the intelligent agent offline based on the thermal field model, and optimizing the strategy network; an intelligent agent is deployed online, and a power adjusting instruction is generated in real time; the control strategy is dynamically corrected, and online fine adjustment is performed through setting trigger adjustment. According to the method, the defects of a traditional method in the aspects of temperature gradient control precision, energy consumption economy and abnormal working condition self-adaptability are overcome, and the growth quality and efficiency of monocrystalline silicon are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of single crystal silicon thermal field, and in particular to a single crystal silicon thermal field multi-section dynamic cooperative heating control method and system. Background Art

[0002] During the growth of single crystal silicon, the temperature gradient control of the thermal field directly determines the crystal quality. Traditional technologies mainly use the following control schemes: 1. PID control: The heating power is adjusted through the proportional-integral-differential algorithm, but there are the following problems: Multivariable coupling is difficult to decouple: the heating sections of the thermal field affect each other, and the fixed PID parameters lead to serious overshoot under dynamic conditions, and the temperature gradient fluctuation range reaches ±10°C / cm; Delayed response defect: The heat conduction delay effect is significant (typical delay is 3-10 seconds), and PID feedback control cannot predict the temperature change trend.

[0003] 2. Fuzzy control: relying on expert experience to set the rule base, lack of adaptability: Strong reliance on experience: different furnace types and silicon material purity require re-adjustment, and the deployment cycle is long (>2 weeks); Lack of multi-objective optimization: It is difficult to take into account temperature gradient, energy consumption and equipment safety threshold at the same time.

[0004] 3. Static zone heating: Divide the heat field into fixed heating sections and manually set the power of each section: Poor coordination: The power adjustment of each section is isolated from each other, resulting in uneven distribution of thermal field energy (local temperature difference > 50°C); High energy consumption: In order to compensate for thermal field distortion, it is often necessary to increase the total power by an excess amount (15%-20%).

[0005] In summary, the existing technology has the following problems with the thermal field of single crystal silicon: insufficient dynamic coordination capability: unable to respond in real time to the disturbance of the thermal field caused by the crystal pulling speed and the change of the melt level; single control target: only focusing on the temperature gradient or energy consumption, ignoring the comprehensive optimization of multiple constraints; rigid handling of abnormal conditions: relying on manual intervention, resulting in high frequency of shutdowns (an average of 3-5 times per furnace) and a high risk of crystal growth interruption. Summary of the invention

[0006] To this end, the technical problem to be solved by the present invention is to overcome the problems existing in the single crystal silicon thermal field in the prior art, and to provide a single crystal silicon thermal field multi-segment dynamic collaborative heating control method and system, which solves the shortcomings of traditional methods in temperature gradient control accuracy, energy consumption economy and adaptability to abnormal working conditions, and can improve the growth quality and efficiency of single crystal silicon.

[0007] In order to solve the above technical problems, the present invention provides a method for controlling multi-segment dynamic cooperative heating of a single crystal silicon thermal field, comprising the following steps: A multi-segment temperature coupling model of the thermal field is constructed, the thermal field is divided into n axial heating segments, and the dynamic relationship between the power input of each heating segment and the crystal growth temperature gradient is established; Design a deep reinforcement learning agent whose state space includes: the real-time temperature values ​​of each heating section T1~T n , crystal pulling speed v, crucible lifting speed s, thermal field longitudinal temperature gradient measured value G; the action space is the power adjustment amount of each heating section ΔP1~ΔP n ; Define the reward function: ; Among them, α, β, γ are weight coefficients related to G, ΔP and T respectively. t : Target temperature gradient, ΔP is the total power change, T c : Crucible bottom temperature, T th : Safety temperature threshold; Based on the thermal field multi-segment temperature coupling model and reward function, the whole process of crystal growth is simulated and the parameters of the agent strategy network are optimized; Connect the trained agent to the thermal field, read sensor data in real time, and output power adjustment instructions for each section; Dynamically modify control strategies, set trigger conditions, automatically trigger strategy updates, and modify control parameters.

[0008] In one embodiment of the present invention, constructing a thermal field multi-segment temperature coupling model includes: Based on the finite element analysis method, the dynamic relationship between the power input and the temperature gradient of each heating section is established to meet the following requirements: G(t)=f(P1(t-τ1),P2(t-τ2),...,Pn(t-τ n ))+ε(t); Among them, τ1~τ n is the heat conduction delay time of each heating section, and ε(t) is the environmental noise disturbance term.

[0009] In one embodiment of the present invention, a plurality of independent temperature-controlled heaters are arranged along the axial direction of the thermal field, including: The upper section ring heater covers the area 50 to 100 mm above the solid-liquid interface and isolates the top water-cooling area; A multi-stage gradient heater in the middle section, comprising at least three concentric heating rings with independently controllable temperatures; Auxiliary heater at the bottom of the lower section to isolate the bottom area.

[0010] In one embodiment of the present invention, an infrared temperature measurement array is arranged at the crystal growth interface, the bottom of the crucible and multiple axial positions, including: Laser thermometer: detect the temperature of each heating section; Contact temperature measurement module: used to measure the bottom temperature of the crucible; Distributed fiber optic temperature sensor, used to collect axial temperature gradient data: In one embodiment of the present invention, the crystal pulling speed v and the crucible rising speed s satisfy a dynamic synchronization relationship: s(t)=v(t)×(1+k×dD(t) / dt) Among them, k=0.1~0.3 is the diameter compensation coefficient, D(t) is the real-time crystal diameter, and dD(t) / dt is the diameter change rate of the crystal.

[0011] In one embodiment of the present invention, the weight coefficient in the reward function is dynamically adjusted according to the crystal growth stage: Seeding stage: increase the α value and give priority to stabilizing the temperature gradient; Shoulder release phase: increase the β value and suppress power mutation; Equal diameter stage: balance coefficient, simultaneous optimization of temperature gradient and energy consumption.

[0012] In one embodiment of the present invention, based on the thermal field multi-segment temperature coupling model and the reward function, the whole crystal growth process is simulated, which at least includes: The dynamic response process of the thermal field includes: the evolution of multi-segment temperature fields and the formation of longitudinal temperature gradients; The crystal morphology evolution process includes: seed growth in the seeding stage, diameter expansion in the shouldering stage, and steady-state growth in the equal diameter stage; Thermo-mechanical-electrical coupling process, including: interaction between pulling speed v and thermal field, crucible lifting speed s and melt level control, as well as power regulation and energy consumption accumulation; Abnormal operating conditions and fault-tolerant processes, including equipment failure simulation, process disturbance simulation, and safety boundary testing; Long-term evolution and device lifetime, including: heater aging simulation, quartz crucible corrosion.

[0013] In one embodiment of the present invention, the dynamic correction control strategy includes: First-level response: diameter deviation > 0.5mm and ≤ 1mm, adjust the crystal pulling speed and give priority to adjusting the power of the top heating section; Secondary response: diameter deviation> 1mm, and melt level deviation ≤ 1mm, adjust the crystal pulling speed, and give priority to adjusting the power of the middle heating section; Level 3 response: If the melt level deviation is greater than 1mm and the crucible bottom temperature fluctuation is greater than 15°C, the pulling action, crucible lifting action, crystal rotation and crucible rotation actions are suspended, the top middle heating power is set to zero, and the constant temperature protection mode of the bottom heating section is started.

[0014] In one embodiment of the present invention, a PID closed-loop compensation module is provided, and the power adjustment amount output by the reinforcement learning is used as a feedforward signal, which is superimposed with the PID feedback control amount and then output to the temperature-controlled heater.

[0015] In one embodiment of the present invention, after each crystal growth is completed, the agent performs the following operations: Drive the temperature-controlled heaters in each heating section to increase the load in steps of 10%-95% of the rated power, and record the temperature rise curve; Comparing with historical data, if the temperature rise rate deviation of a section is greater than 15%, the temperature control heater of this section is marked as waiting for maintenance.

[0016] In order to solve the above technical problems, the present invention also provides a single crystal silicon thermal field multi-segment dynamic cooperative heating system, comprising: Thermal field coupling modeling module: including: storage unit, used to preset the thermal field multi-segment temperature coupling model, define the dynamic relationship between the power input and the temperature gradient of n axial heating sections; processing unit: calculate the power-temperature gradient mapping relationship based on the finite element analysis equation; Intelligent control module: used to receive the temperature of each heating section T1~T n , crystal pulling speed v, crucible lifting speed s, thermal field longitudinal temperature gradient measured value G, and output power adjustment value ΔP1~ΔP for each heating section n ; Reward calculation module: preset target temperature gradient G and safety temperature threshold T th , execute the reward function operation: ; Strategy optimization module: simulates the entire crystal growth process based on the thermal field coupling model and optimizes the agent strategy network parameters; Real-time control module: collects temperature, speed, and gradient data in real time, deploys the trained strategy network, generates power adjustment instructions, and sends ΔP1~ΔP to the temperature control heater n The regulatory signal; Strategy update module: used to automatically trigger strategy updates according to preset trigger conditions to correct control parameters.

[0017] The above technical solution of the present invention has the following advantages compared with the prior art: Compared with the traditional partitioned PID control, the method for dynamic collaborative heating control of multiple sections of the thermal field of single crystal silicon described in the present invention realizes the collaborative adjustment of the power of multiple sections of the thermal field through dynamic coupling modeling and multi-objective reinforcement learning control, designs a deep reinforcement learning agent, synchronously perceives the temperature, pulling speed, crucible position and real-time temperature gradient of all sections in its state space, directly outputs the power adjustment amount of each heating section in the action space, and automatically learns the collaborative rules of multi-section power increase and decrease through the neural network strategy, breaking through the limitations of manual experience parameter adjustment, setting the temperature gradient tracking item and the power change penalty item in the reward function, driving the agent to avoid drastic power adjustment while reducing gradient fluctuations; introducing an online strategy fine-tuning module, setting trigger conditions, triggering the re-weighting of the local reward function, and realizing dynamic compensation for the thermal field; through the synergistic effect of the above-mentioned technical features, theoretically solves the inherent defects of traditional methods in dynamic collaboration, multi-objective optimization and adaptive control. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein: Figure 1 It is a flowchart of the steps of the method for controlling the multi-segment dynamic cooperative heating of the single crystal silicon thermal field of the present invention; Figure 2 It is a framework diagram of the single crystal silicon thermal field multi-segment dynamic cooperative heating control system of the present invention. DETAILED DESCRIPTION

[0019] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention. Embodiment 1

[0020] Reference Figure 1 As shown, the present invention discloses a method for controlling multi-segment dynamic cooperative heating of a single crystal silicon thermal field, comprising the following steps: S10. Construct a multi-segment temperature coupling model of the thermal field, divide the thermal field into n axial heating segments, and establish a dynamic relationship between the power input of each heating segment and the crystal growth temperature gradient.

[0021] It should be noted that during the growth of single crystal silicon, the temperature gradient requirements near the solid-liquid interface (upper section), the crystal body (middle section) and the bottom of the crucible (lower section) are different. In this embodiment, the thermal field is divided into axial independent temperature control zones, and the temperature of each section can be controlled. However, in the thermal field of single crystal silicon, the power input of each heating section will affect each other through heat conduction, radiation, etc., forming a complex multi-variable coupling system. It is difficult for traditional methods to decouple this nonlinear relationship, resulting in low temperature gradient control accuracy. For this reason, the present application establishes a multi-section temperature coupling model of the thermal field, which can represent the dynamic relationship between the power input of each heating section and the temperature gradient of crystal growth.

[0022] Specifically, in this embodiment, based on the finite element analysis method, a dynamic relationship between the power input and the temperature gradient of each heating section is established to meet the following requirements: G(t)=f(P1(t-τ1),P2(t-τ2),...,Pn(t-τ n ))+ε(t); Among them, τ1~τ n is the heat conduction delay time of each heating section, and ε(t) is the environmental noise disturbance term.

[0023] Discretize the thermal field into finite elements, describe the transfer relationship between power and temperature gradient in each section through matrix equations, clarify the weight coefficients between variables, and realize mathematical decoupling of the coupled system; Through experiments, it is found that there is a significant delay in heat transfer in the thermal field. The experimentally measured delay time is 3 to 10 seconds. Traditional feedback control produces overshoot due to the inability to predict the delay effect. The temperature gradient fluctuates by ±10°C / cm. In the dynamic equation of this embodiment, the delay terms τ1~τ n , through the pre-compensation mechanism to offset the lag effect, so that the control instructions take effect in advance; Moreover, in the thermal field, there are also the influences of environmental noises such as melt convection and cooling water flow fluctuations, which will destroy the stability of the thermal field. It is difficult for the empirical model to distinguish between noise and real thermal field changes. In order to solve this problem, the dynamic equation of this embodiment is also combined with Kalman filtering to perform online estimation and suppression of the noise term ε(t) to improve the model's anti-interference ability.

[0024] When actually dividing the axial heating sections, it is necessary to divide them according to the size of the thermal field, and at least three sections are set along the axial direction of the thermal field. Each section is correspondingly provided with an independent temperature control heater, which includes: The upper section ring heater covers the area 50-100mm above the solid-liquid interface and isolates the top water-cooling area. On the one hand, it can quickly adjust the temperature gradient of the solid-liquid interface, and on the other hand, it can isolate the top water-cooling area to reduce heat loss; The multi-stage gradient heater in the middle section includes at least 3 concentric heating rings with independent temperature control. The multi-stage gradient heater in the middle section plays a vital role in the temperature gradient regulation. The more heaters are set in this section, the more stable the temperature gradient regulation effect will be, and the longitudinal temperature can be maintained uniformly. The auxiliary heater at the bottom of the lower section isolates the bottom area. On the one hand, it can stably heat the melt and stabilize the melt convection. On the other hand, it can prevent the bottom of the crucible from being overcooled and reduce heat loss.

[0025] S20, design a deep reinforcement learning agent, whose state space includes: the real-time temperature values ​​T1~T of each heating section n , crystal pulling speed v, crucible lifting speed s, thermal field longitudinal temperature gradient measured value G; the action space is the power adjustment amount of each heating section ΔP1~ΔP n .

[0026] In reinforcement learning, the state space is the set of all possible states of the agent's perceived environment, while the action space is the set of all possible actions that the agent can perform. The parameters in the state space need to fully describe the state of the current environment so that the agent can make decisions, while the action space is the action selected by the agent based on the current state.

[0027] In this embodiment, the state space introduces the real-time temperature values ​​T1 to T2 of each heating zone. n , which reflects the instantaneous state of the axial energy distribution of the thermal field, is the basis for realizing independent temperature control of different zones and can suppress local temperature mutations.

[0028] The state space introduces crystal pulling speed v and crucible rising speed s: mechanical control quantities used to characterize the crystal growth rate. During the crystal pulling process, as the silicon crystal is slowly pulled, its lifting speed is v, the melt level will gradually drop, and the crucible will rise accordingly. The role of the crucible rising speed s is to synchronously raise the crucible position, compensate for the drop in the melt level, and avoid the position shift of the solid-liquid interface. At the same time, the rise of the crucible will change the relative position of the heater and the melt, directly affecting the thermal field radiation heat transfer path. The crucible rising speed s is strongly related to the crystal growth stage. The intelligent agent needs to identify the current growth stage through the crucible rising speed s, so as to adapt the control strategies of different stages. Specifically, the crystal pulling speed v and the crucible rising speed s must satisfy the dynamic mass conservation equation: s(t)=v(t)×(1+k×dD(t) / dt) Where k=0.1~0.3 is the diameter compensation coefficient, D(t) is the real-time crystal diameter, and dD(t) / dt is the diameter change rate of the crystal; if s≠v×(1+k×dD / dt), the melt level will continue to rise or fall, resulting in changes in the thermal field boundary conditions (such as the displacement of the solid-liquid interface position), causing a temperature gradient G Loss of control, for example: when the crystal diameter increases (dD / dt>0), the crucible rising speed s needs to be increased to replenish the melt consumption and avoid the temperature in the lower section rising suddenly due to the drop in liquid level; In addition, the pulling speed v directly affects the crystal growth rate, and the change in growth rate will change the heat flow demand at the solid-liquid interface, and the power adjustment amount ΔP1~ΔP of the heating section needs to be adjusted synchronously. n In order to maintain the target temperature gradient G, the pulling speed v is incorporated into the state space of the agent, and the agent can learn the pulling speed v→ΔP1~ΔP n The mapping relationship can control the power adjustment amount ΔP1~ΔP of each heating section according to the actual pulling speed v. n .

[0029] The measured value G of the longitudinal temperature gradient of the thermal field is introduced into the state space to quantify the core indicator of the thermodynamic stability of the solid-liquid interface. G is directly measured and used as the state input to improve the real-time control.

[0030] In this embodiment, the action space is defined as the power adjustment amount of each heating section ΔP1~ΔP n Multiple heating sections are distributed to meet different power regulation requirements. For example, the upper section ΔP1 needs to respond quickly to the change of heat flow at the solid-liquid interface and suppress diameter fluctuations. The middle section ΔP2~ΔP n-1 It is necessary to maintain longitudinal temperature uniformity and reduce temperature fluctuations. The lower section ΔP n The temperature of the bottom of the crucible needs to be controlled to prevent the bottom of the crucible from being overcooled / overheated.

[0031] Specifically, corresponding to the division of the heating sections, in order to realize the temperature detection needs of each heating section, an infrared temperature measurement array is set at the crystal growth interface, the bottom of the crucible and multiple axial positions, including: Laser thermometer: detects the temperature of each heating section and can measure the surface temperature of each heating section non-contactly; Contact temperature measurement module: used to measure the bottom temperature of the crucible; Distributed optical fiber temperature sensor is used to collect axial temperature gradient data and can realize continuous measurement of temperature gradient.

[0032] S30. Define the reward function: ; Among them, α, β, γ are weight coefficients related to G, ΔP and T respectively.t : Target temperature gradient, ΔP is the total power change, T c : Crucible bottom temperature, T th : Safety temperature threshold.

[0033] It should be noted that the growth of single crystal silicon must simultaneously meet the goals of temperature gradient accuracy, energy economy, equipment safety, etc. Traditional PID or fuzzy control is difficult to balance these mutually restrictive factors. Therefore, in this embodiment, a reward function is introduced to play the role of a policy optimization guide in deep reinforcement learning. The reward function can transform the complex single crystal silicon thermal field control problem into a multi-objective optimization task. In the reward function, multiple objectives are integrated into a single scalar signal through weighted summation to drive the intelligent agent to learn the optimal strategy.

[0034] Specifically, the parameters of the reward function include: a temperature gradient tracking term, which is used to penalize the measured gradient G and the deviation from the target G t , ensuring low-defect growth of crystals; power variation penalty term, used to suppress drastic power fluctuations, reduce energy consumption and extend heater life; safety constraint term, used when the crucible bottom temperature T c Super threshold T th Punishment is imposed in time to avoid melting loss of the quartz crucible.

[0035] In this embodiment, the agent continuously generates actions, and the reward function provides real-time feedback on the quality of the actions, including: Positive reward: When G→G t And when ΔP is small, the R value increases, strengthening the current strategy; Negative penalty: If T c >T th , R drops sharply, prompting the agent to avoid dangerous actions; For example, in the diameter mutation scenario, the agent may try to increase the power of the upper segment and reduce the power of the lower segment at the same time. The reward function is expressed by α(G t A comprehensive evaluation of β(-G)² and β(ΔP)² guides it to find the balance point.

[0036] Specifically, the weight coefficients α, β, and γ actually define the priority of the control objectives. In different production stages, the control objectives are different, so different weight coefficient ratios need to be set. The intelligent agent learns the optimal strategy under different weights through offline training, and dynamically switches the strategy according to the growth stage during online deployment.

[0037] In actual production, crystal growth is divided into three stages, including seeding stage, shoulder release stage and equal diameter stage. The weight coefficient in the reward function is dynamically adjusted according to the crystal growth stage: Seeding stage: The seed crystal and the melt make initial contact and a solid-liquid interface is formed. At this time, the seed crystal has a small diameter (4.5-5.5 mm) and a low heat capacity. A slight fluctuation in the temperature gradient can lead to melting or dislocation proliferation. The α value is increased to allow short-term power mutations to quickly respond to gradient deviations, sacrificing some energy consumption economy and giving priority to growth startup stability. At this time, the weight distribution is α=0.6, β=0.3, and γ=0.1. Gradient accuracy is prioritized at this stage.

[0038] Shoulder release stage: The crystal diameter increases rapidly, resulting in a sharp decrease in heat demand. The drastic power adjustment is prone to cause mechanical vibration and thermal stress cracks. The β value is increased to limit the power adjustment range of each section. At this time, the weight distribution is α=0.2, β=0.7, γ=0.1 to prevent power mutation; Constant diameter stage: The diameter remains constant. Long-term steady-state growth requires consideration of both energy consumption (cost sensitivity) and equipment life (accumulated damage from power cycles). At this time, the weight distribution is α=0.4, β=0.4, γ=0.2, and the balance coefficient optimizes the temperature gradient and energy consumption simultaneously. In addition, the safety weight γ is increased to prevent the bottom temperature of the crucible from drifting during long-term operation (such as slow rise caused by aging of the quartz crucible).

[0039] S40. Based on the thermal field multi-segment temperature coupling model and reward function, the whole process of crystal growth is simulated and the parameters of the intelligent agent strategy network are optimized.

[0040] Specifically, in the above step S10, the dynamic relationship between the power input and the temperature gradient of each heating section is established through finite element analysis to provide a physical basis for the simulation. During the simulation, the input and output of the intelligent agent must be strictly aligned according to step S20 to ensure that the policy network structure is compatible with the actual control. The reward function of step S30 is used as the optimization criterion to drive the intelligent agent to explore the optimal solution set.

[0041] Specifically, in order to cover as many situations and problems as possible in the actual process, the simulation of the entire crystal growth process should at least include: The dynamic response process of the thermal field includes: the evolution of multi-segment temperature fields and the formation of longitudinal temperature gradients, such as the influence of the upper segment ring heater on the heat flow at the solid-liquid interface, the coordinated temperature control of the gradient heating ring in the middle segment, the regulating effect of the lower segment auxiliary heater on the crucible bottom temperature, as well as gradient tracking under normal operating conditions and gradient distortion caused by abnormal disturbances (such as sudden changes in melt convection).

[0042] The crystal morphology evolution process includes: seed growth in the seeding stage, diameter expansion in the shouldering stage, and steady-state growth in the equal-diameter stage. It simulates the initial crystallization process after the seed crystal is immersed in the melt, the rapid growth process of the crystal diameter from millimeters to the target size (such as 200mm), and simulates the crystal growth process under constant diameter.

[0043] The thermal-mechanical-electrical coupling process includes: the interaction between the pulling speed v and the thermal field, the crucible rising speed s and the melt level control, as well as the power regulation and energy consumption accumulation. It simulates the feedback effect of the change in crystal pulling speed on the heat flow at the solid-liquid interface, and the dynamic synchronization relationship between the pulling speed v and the crucible rising speed s. It simulates the offset of the thermal field boundary conditions caused by the liquid level deviation and the risk of melt depletion caused by the abnormal crucible rising speed (such as sticking s=0). It simulates the total energy consumption under different control strategies and optimizes the β term in the reward function.

[0044] Abnormal operating conditions and fault-tolerant processes, including: equipment failure simulation, process disturbance simulation, safety boundary testing, for example: thermal field compensation strategy when heater failure causes ΔP=0 in a certain section, melt composition changes and cooling water movement, emergency response when the bottom of the crucible is overheated or the diameter is out of control.

[0045] Long-term evolution and equipment life, including: heater aging simulation, quartz crucible corrosion, for example: simulating the power attenuation effect caused by the increase in resistivity of the temperature-controlled heater over time, training the intelligent agent to adaptively adjust the initial power regulation; simulating the impact of the reduction in crucible wall thickness on the thermal field boundary, and determining under what circumstances to replace the crucible.

[0046] S50, connecting the trained intelligent agent to the thermal field, reading sensor data in real time, and outputting power adjustment instructions for each section.

[0047] After ensuring that the strategy network has learned the typical working conditions of the thermal field, the intelligent agent is applied in actual process production to control the power adjustment instructions of each section to achieve the expected technical effect, realize the coordinated adjustment of the power of multiple sections of the thermal field, and improve crystal quality and production efficiency through real-time closed-loop optimization.

[0048] S60, dynamically modify the control strategy, set the trigger conditions, automatically trigger the strategy update, and modify the control parameters.

[0049] In this embodiment, the purpose of the dynamic correction control strategy is to ensure the continuous optimal performance of the monocrystalline silicon thermal field control system in a complex dynamic environment by sensing the changes in system status in real time and adaptively adjusting the control parameters and strategies. The trigger conditions can be set to different types and different thresholds according to actual production requirements, including: process parameter deviation, equipment performance degradation, energy consumption economy degradation, and safety boundary approach, etc., and different levels of corresponding mechanisms can also be set according to the actual triggering situation.

[0050] In this embodiment, two levels of corresponding mechanisms are set up to produce different control strategies respectively, wherein: Level 1 response: When the diameter deviation is greater than 0.5 mm and less than or equal to 1 mm, the crystal pulling speed is adjusted, and the power of the top heating section is adjusted first. The level 1 response is an adjustment response mechanism, that is, the strategy is automatically adjusted according to the simulation training of the intelligent agent, which is an early warning of process deviation. Secondary response: If the diameter deviation is greater than 1mm and the melt level deviation is less than or equal to 1mm, the crystal pulling speed is adjusted, and the power of the middle heating section is adjusted preferentially. The secondary response is an adjustment response mechanism, that is, the strategy is automatically adjusted according to the simulation training of the intelligent agent, which is a process deviation warning in the production stage. Level 3 response: If the melt level line deviation is greater than 1mm and the crucible bottom temperature fluctuation is greater than 15°C, the pulling action, crucible lifting action, crystal rotation and crucible rotation action are suspended, the top middle heating power is set to zero, and the bottom heating section constant temperature protection mode is started. The level 3 response is an emergency response mechanism, which is independent of the intelligent body, that is, emergency handling of major process abnormalities to avoid equipment damage and safety accidents.

[0051] Specifically, after each crystal growth is completed, the agent performs the following operations: Drive the temperature-controlled heaters in each heating section to increase the load in steps of 10%-95% of the rated power, and record the temperature rise curve; Comparing with historical data, if the temperature rise rate deviation of a section is greater than 15%, the temperature control heater of this section is marked as waiting for maintenance.

[0052] In this embodiment, through step load test and temperature rise rate analysis, online diagnosis and predictive maintenance of the health status of the temperature-controlled heater are realized, and hidden faults such as aging of the heater resistance wire and degradation of the insulation layer can be detected, the degree of performance degradation can be quantified, and anomalies (such as local hot spots and poor contact) can be identified before the heater fails completely, so as to avoid sudden failures in the production process that would cause the entire furnace of crystals to be scrapped. Embodiment 2

[0053] Specifically, based on the above embodiment 1, in order to further illustrate how to implement the above heating control method, refer to Figure 2 As shown, the present invention also discloses a single crystal silicon thermal field multi-segment dynamic cooperative heating system for implementing the method described in the first embodiment, including: Thermal field coupling modeling module: including: storage unit, used to preset the thermal field multi-segment temperature coupling model, define the dynamic relationship between the power input and the temperature gradient of n axial heating sections; processing unit: calculate the power-temperature gradient mapping relationship based on the finite element analysis equation; Intelligent control module: used to receive the temperature of each heating section T1~T n , crystal pulling speed v, crucible lifting speed s, thermal field longitudinal temperature gradient measured value G, and output power adjustment value ΔP1~ΔP for each heating section n ; Reward calculation module: preset target temperature gradient G and safety temperature threshold T th , execute the reward function operation: ; Strategy optimization module: simulates the entire crystal growth process based on the thermal field coupling model and optimizes the agent strategy network parameters; Real-time control module: collects temperature, speed, and gradient data in real time, deploys the trained strategy network, generates power adjustment instructions, and sends ΔP1~ΔP to the temperature control heater n The regulatory signal; Strategy update module: used to automatically trigger strategy updates according to preset trigger conditions to correct control parameters.

[0054] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A method for controlling multi-segment dynamic cooperative heating of a single crystal silicon thermal field, characterized in that: The following steps are involved: A multi-segment temperature coupling model of the thermal field is constructed, the thermal field is divided into n axial heating segments, and the dynamic relationship between the power input of each heating segment and the crystal growth temperature gradient is established; Design a deep reinforcement learning agent whose state space includes: the real-time temperature values ​​of each heating section T1~T n , crystal pulling speed v, crucible lifting speed s, thermal field longitudinal temperature gradient measured value G; the action space is the power adjustment amount of each heating section ΔP1~ΔP n ; Define the reward function: ； Among them, α, β, γ are weight coefficients related to G, ΔP and T respectively. t : Target temperature gradient, ΔP is the total power change, T c : Crucible bottom temperature, T th : Safety temperature threshold; Based on the thermal field multi-segment temperature coupling model and reward function, the whole process of crystal growth is simulated and the parameters of the agent strategy network are optimized; Connect the trained agent to the thermal field, read sensor data in real time, and output power adjustment instructions for each section; Dynamically modify control strategies, set trigger conditions, automatically trigger strategy updates, and modify control parameters.

2. The method for controlling the multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: The construction of the thermal field multi-segment temperature coupling model includes: Based on the finite element analysis method, the dynamic relationship between the power input and the temperature gradient of each heating section is established to meet the following requirements: G(t)=f(P1(t-τ1),P2(t-τ2),...,Pn(t-τ n ))+ε(t); Among them, τ1~τ n is the heat conduction delay time of each heating section, and ε(t) is the environmental noise disturbance term.

3. The method for controlling the multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: Multiple independent temperature-controlled heaters are arranged along the axial distribution of the thermal field, including: The upper section ring heater covers the area 50 to 100 mm above the solid-liquid interface and isolates the top water-cooling area; A multi-stage gradient heater in the middle section, comprising at least three concentric heating rings with independently controllable temperatures; Auxiliary heater at the bottom of the lower section to isolate the bottom area.

4. The method for controlling the multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: An infrared temperature measurement array is set up at the crystal growth interface, crucible bottom and multiple axial positions, including: Laser thermometer: detect the temperature of each heating section; Contact temperature measurement module: used to measure the bottom temperature of the crucible; Distributed fiber optic temperature sensor, used to collect axial temperature gradient data.

5. The method for controlling the multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: The crystal pulling speed v and the crucible rising speed s satisfy the dynamic synchronization relationship: s(t)=v(t)×(1+k×dD(t) / dt) Among them, k=0.1~0.3 is the diameter compensation coefficient, D(t) is the real-time crystal diameter, and dD(t) / dt is the diameter change rate of the crystal.

6. The method for controlling multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: The weight coefficient in the reward function is dynamically adjusted according to the crystal growth stage: Seeding stage: increase the α value and give priority to stabilizing the temperature gradient; Shoulder release phase: increase the β value and suppress power mutation; Equal diameter stage: balance coefficient, simultaneous optimization of temperature gradient and energy consumption.

7. The method for controlling multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: Based on the thermal field multi-segment temperature coupling model and reward function, the whole crystal growth process is simulated, including at least: The dynamic response process of the thermal field includes: the evolution of multi-segment temperature fields and the formation of longitudinal temperature gradients; The crystal morphology evolution process includes: seed growth in the seeding stage, diameter expansion in the shouldering stage, and steady-state growth in the equal diameter stage; Thermo-mechanical-electrical coupling process, including: interaction between pulling speed v and thermal field, crucible lifting speed s and melt level control, as well as power regulation and energy consumption accumulation; Abnormal operating conditions and fault-tolerant processes, including equipment failure simulation, process disturbance simulation, and safety boundary testing; Long-term evolution and device lifetime, including: heater aging simulation, quartz crucible corrosion.

8. The method for controlling multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: The dynamic correction control strategy includes: First-level response: diameter deviation > 0.5mm and ≤ 1mm, adjust the crystal pulling speed and give priority to adjusting the power of the top heating section; Secondary response: diameter deviation> 1mm, and melt level deviation ≤ 1mm, adjust the crystal pulling speed, and give priority to adjusting the power of the middle heating section; Level 3 response: If the melt level deviation is greater than 1mm and the crucible bottom temperature fluctuation is greater than 15°C, the pulling action, crucible lifting action, crystal rotation and crucible rotation actions are suspended, the top middle heating power is set to zero, and the constant temperature protection mode of the bottom heating section is started.

9. The method for controlling multi-segment dynamic cooperative heating of a single crystal silicon thermal field according to claim 1, characterized in that: After each crystal growth is completed, the agent performs the following operations: Drive the temperature-controlled heaters in each heating section to increase the load in steps of 10%-95% of the rated power, and record the temperature rise curve; Comparing with historical data, if the temperature rise rate deviation of a section is greater than 15%, the temperature control heater of this section is marked as waiting for maintenance.

10. A single crystal silicon thermal field multi-segment dynamic cooperative heating system, characterized in that: include: Thermal field coupling modeling module: including: storage unit, used to preset the thermal field multi-segment temperature coupling model, define the dynamic relationship between the power input and the temperature gradient of n axial heating sections; processing unit: calculate the power-temperature gradient mapping relationship based on the finite element analysis equation; Intelligent control module: used to receive the temperature of each heating section T1~T n , crystal pulling speed v, crucible lifting speed s, thermal field longitudinal temperature gradient measured value G, and output power adjustment value ΔP1~ΔP for each heating section n ; Reward calculation module: preset target temperature gradient G and safety temperature threshold T th , execute the reward function operation: ； Strategy optimization module: simulates the entire crystal growth process based on the thermal field coupling model and optimizes the agent strategy network parameters; Real-time control module: collects temperature, speed, and gradient data in real time, deploys the trained strategy network, generates power adjustment instructions, and sends ΔP1~ΔP to the temperature control heater n The regulatory signal; Strategy update module: used to automatically trigger strategy updates according to preset trigger conditions to correct control parameters.

Images