Intelligent energy-saving regulation and control system and method for silicon single crystal thermal field

Abstract

The invention discloses an intelligent energy-saving regulation and control system and method for a silicon single crystal thermal field. The system comprises a multi-physical field acquisition unit, a thermal field simulation calculation module, a gradient thermal insulation assembly, an AI regulation and control module, a feedback execution control module and an energy consumption evaluation module. The system constructs a coupling temperature field / flow field model by collecting temperature, flow velocity and pressure data in a furnace in real time, generates a dynamic control strategy by using an AI algorithm, and drives a heater, a cooling system and an adjustable graphite felt assembly to cooperatively respond; the gradient heat insulation assembly adapts to heat flow distribution through the axial heat conductivity decreasing design, and heat loss is reduced. The method comprises the steps of data acquisition, simulation modeling, AI strategy generation, execution control, quality evaluation and model correction, and closed-loop optimization is formed. The method realizes intelligent dynamic regulation and control of thermal field distribution, remarkably improves crystal growth quality and energy efficiency, and is suitable for the field of semiconductor silicon single crystal preparation.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor silicon single crystal preparation, and in particular to an intelligent energy-saving control system and method for a silicon single crystal thermal field. Background Art

[0002] During silicon single crystal growth, precise control of thermal field distribution directly impacts crystal quality and energy efficiency. Traditional thermal field control methods rely primarily on empirical parameter settings and static thermal insulation materials, which present the following problems:

[0003] Response hysteresis: Control strategies based on fixed rules are difficult to adapt to dynamic changes in the thermal field, resulting in large temperature fluctuations and high crystal defect rates;

[0004] Excessive energy consumption: A single thermal insulation material cannot match the heat flux gradient distribution, resulting in severe heat loss and high energy consumption per unit crystal growth;

[0005] Lack of intelligence: There is a lack of multi-physics field collaborative simulation and real-time feedback mechanism, and the adjustment of control parameters relies on manual intervention, which is inefficient.

[0006] While existing technologies employ PID control or simple thermal insulation structure optimization, none achieve dynamic, coordinated optimization of the thermal field, materials, and algorithms, making it difficult to achieve both crystal quality and energy efficiency improvements. Therefore, a thermal field control system integrating intelligent algorithms, gradient materials, and closed-loop feedback is urgently needed. Summary of the Invention

[0007] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0008] Therefore, in order to overcome the problems of high energy consumption, slow response, lack of intelligent adjustment mechanism, and single thermal field material performance in the thermal field control process of existing silicon single crystal growth equipment, a silicon single crystal thermal field intelligent energy-saving control system and method are proposed. It combines the dynamic thermal performance adjustment capability of adaptive thermal field materials, a joint optimization design method based on multi-physics field simulation and AI algorithm, and a machine learning real-time feedback control mechanism for thermal field steady-state response to achieve intelligent control of thermal field distribution and significantly improve energy efficiency during silicon single crystal growth.

[0009] To achieve the above object, the present invention proposes the following technical solutions:

[0010] A silicon single crystal thermal field intelligent energy-saving control system, comprising:

[0011] Multi-physics field acquisition unit, used to collect real-time temperature (measurement range 0-2000°C, accuracy ±0.1°C), flow rate (accuracy better than 5%), pressure (accuracy ±0.01Pa) and heat flux data in the silicon single crystal growth furnace;

[0012] A thermal field simulation calculation module is used to construct a coupled temperature field and flow field model based on the data obtained by the acquisition unit and perform dynamic heat loss prediction;

[0013] Gradient thermal insulation components, including graphite felt materials with different thermal conductivity zones (typically ranging from 0.3 to 1.5 W / m·K), whose arrangement gradually changes along the radial or axial direction to adapt to the changing heat flow gradient. This component material is synthesized through a composite mixing process with different graphitization degrees and carbon contents, and the temperature-dependent thermal conductivity characteristics can be adjusted.

[0014] The AI ​​control module is used to integrate thermal field simulation results with historical control data and generate thermal field control strategies based on reinforcement learning or deep learning algorithms. This module has an adaptive update function and dynamically adjusts model weights according to the latest control effects.

[0015] A feedback execution control module is connected to the crystal heater, the crucible drive assembly, the cooling system and the thermal field insulation assembly, and is used to adjust each execution unit according to the control strategy;

[0016] The energy consumption assessment and strategy optimization module is used to periodically analyze the energy consumption value corresponding to the unit crystal growth quality and correct the control priority of the AI ​​model; this module dynamically adjusts the loss function weight and control priority of the AI ​​model based on the assessment results to form an optimization closed loop.

[0017] The system periodically performs thermal field control tasks according to the following time sequence:

[0018] Data is collected at T0, simulation and AI calculation are completed at T1, control adjustments are implemented from T2 to T3, and energy consumption and crystal quality evaluation results are fed back at T4.

[0019] As a preferred solution of the intelligent energy-saving control system for the thermal field of a silicon single crystal described in the present invention, the gradient thermal insulation component includes at least two graphite felt areas with different thermal conductivities, whose thermal conductivities decrease step by step along the axial direction, and are arranged in sequence from the furnace wall to the crystal core to achieve a gradient balanced distribution of heat flow.

[0020] As an optimal solution of the intelligent energy-saving control system for the thermal field of a silicon single crystal described in the present invention, the thermal field simulation calculation module adopts a finite element multi-physics field simulation method, integrating electrothermal coupling, radiation heat transfer and gas flow field disturbance factors, the simulation time step is set to 10 to 30 seconds, and the simulation frequency is synchronized with the control instruction output.

[0021] As an optimal solution of the intelligent energy-saving control system for the thermal field of a silicon single crystal described in the present invention, the AI ​​control module uses historical growth data to construct a thermal field control model, and dynamically updates the strategy based on the current crystal diameter change rate and dislocation density evolution trend.

[0022] As a preferred solution of the intelligent energy-saving control system for the thermal field of a silicon single crystal described in the present invention, the feedback execution control module has a control logic closed-loop detection function. If the crystal quality does not improve significantly within two consecutive control cycles, the AI ​​model retraining process is started and the control strategy output interval is adjusted.

[0023] This solution also provides the above-mentioned method for intelligent energy-saving control of silicon single crystal thermal field, which is characterized by comprising the following steps:

[0024] S1: Real-time data acquisition, using a multi-physics field acquisition unit to acquire temperature, airflow, pressure, and heat flux data during crystal growth every 10 to 30 seconds;

[0025] S2: Build a multi-physics field simulation model based on the data collected in step S1, perform coupled thermal field simulation based on the collected data, consider the location of the electric heat source, electromagnetic field changes, cooling airflow pattern, etc., and simulate and output a thermal field equivalent diagram and heat loss prediction value (i.e., output temperature distribution and heat loss estimation results);

[0026] S3: AI intelligent strategy generation, integrating simulation output with historical control data, and using the AI ​​model (Transformer) to generate thermal field control parameters for the next period;

[0027] S4: Control instructions are issued and executed. Specifically, the parameters generated by AI are used to control each execution component to adjust the heating power, crucible position, cooling rate and deformation configuration of the gradient insulation structure within the control cycle.

[0028] S5: Crystal growth quality assessment, collecting indicators such as current crystal diameter, defect density, dislocation index, and total power consumption, and evaluating the current crystal diameter uniformity, defect density, and unit energy consumption;

[0029] S6: Strategy feedback and model correction, update the AI ​​model parameter weights based on the results of S5, and determine whether the control cycle T needs to be adjusted n+1 .

[0030] As a preferred solution of the intelligent energy-saving control method of the silicon single crystal thermal field described in the present invention, in which: in step S3, the AI ​​model uses the Transformer structure and the historical control sequence to perform attention fusion to determine the optimal control parameter distribution.

[0031] As a preferred scheme of the silicon single crystal thermal field intelligent energy-saving regulation method, in step S4, the adjustable structure of the gradient thermal insulation assembly controls its position or form through a servo mechanism to adapt to the optimization requirements of the heat flow direction in different growth stages.

[0032] As a preferred scheme of the silicon single crystal thermal field intelligent energy-saving regulation method, in step S5, the energy consumption evaluation index includes total power consumption corresponding to unit crystal quality, average power curve smoothness and crystal defect number.

[0033] As a preferred scheme of the silicon single crystal thermal field intelligent energy-saving regulation method, in step S6, if the unit energy consumption does not decrease significantly in two consecutive growth cycles, the system triggers the AI model backtracking mechanism, re-trains the model and preferentially selects different regulation parameter combinations in the next cycle.

[0034] The beneficial effects of the present application are:

[0035] 1. The present application constructs a high-precision temperature field / flow field model through real-time acquisition and coupling simulation of multiple physical fields, dynamically predicts heat loss and generates regulation strategies, and significantly improves the uniformity of thermal field distribution.

[0036] 2. The present application adopts a graphite felt assembly arranged in an axial gradient, and the thermal conductivity gradually decreases from the furnace wall to the crystal core, effectively balances the heat flow gradient, and reduces radial heat loss.

[0037] 3. The AI regulation module of the present application fuses simulation data and historical working conditions, generates optimal control parameters based on the Transformer structure, and dynamically corrects the model weight through closed-loop detection, realizing self-adaptive adjustment.

[0038] 4. The present application periodically evaluates the unit crystal energy consumption and defect density, triggers the model backtracking mechanism to optimize the control strategy, and simultaneously reduces the energy consumption and dislocation density.

[0039] 5. The present application forms a closed loop from data acquisition, simulation calculation to execution control, reduces manual intervention, and improves process stability and production efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor. Among them:

[0041] Figure 1 The overall workflow diagram of the present application.

[0042] Figure 2 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0044] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0046] Example 1

[0047] Reference Figures 1-2 , which is the first embodiment of the present invention, provides a silicon single crystal thermal field intelligent energy-saving control system, specifically as follows:

[0048] 1. System configuration;

[0049] 1.1. Multi-physics field acquisition unit:

[0050] Temperature sensors (measuring range 0-2000°C, accuracy ±0.1°C) are distributed along the radial and axial directions of the crucible;

[0051] Piezoresistive pressure sensor (accuracy ±0.01Pa) and hot-wire flow sensor (accuracy ±3%);

[0052] The data collection cycle is 10 to 30 seconds and is transmitted to the central controller via a high-speed bus.

[0053] 1.2. Gradient thermal insulation components:

[0054] This gradient thermal insulation component consists of three layers of graphite felt with a thermal conductivity gradient distribution, with a thermal conductivity range of 0.3 to 1.5 W / m·K. The component has an axial displacement adjustable function with an adjustable range of ±15mm;

[0055] The gradient material structure optimization process is as follows:

[0056] Thermal field simulation: Through thermal field simulation technology, the target heat flow distribution function F(x) is generated, which describes the heat flow distribution of the component under ideal conditions.

[0057] Material gradient vector definition: Define the material gradient vector G = [g1, g2, g 3, g 4, g5];

[0058] Where g i represents the degree of graphitization (or thermal conductivity, carbon content, etc.) of the i-th layer of graphite felt, which directly affects the thermal conductivity and in turn affects the thermal insulation performance of the assembly;

[0059] Optimization algorithm: Genetic algorithm is used to optimize the material gradient vector G; the optimization goal is to minimize the loss function Loss = ∑ x |F(x)-f material (x; G)|, where f material (x; G) is the actual heat flow distribution function generated by the assembly under the given material gradient vector G.

[0060] Output and update: After optimization algorithm calculation, output the optimal material gradient vector G * , and update the configuration of the graphite felt according to the optimal vector;

[0061] This optimization process is performed every 5 cycles to ensure that the assembly always maintains good thermal insulation performance;

[0062] Specifically: After each material gradient optimization, update G * to the thermal field simulation model, and the AI control module generates a control strategy that adapts to the new heat flow distribution.

[0063] 1.3, Thermal field simulation calculation module:

[0064] Adopt finite element multi-physical field simulation method, combine electric heating coupling, radiation heat exchange and gas flow field disturbance;

[0065] Simulation step 10-30 seconds, synchronized with control period, output temperature field distribution and heat loss prediction value.

[0066] 1.4, AI control module:

[0067] Based on the Transformer architecture, input multi-physical field data (temperature, pressure, flow rate), output heating power, cooling rate and graphite felt displacement instruction;

[0068] Dynamic strategy update: When the crystal diameter change rate deviation > 2% or dislocation density growth rate > 5% / h, trigger online model optimization.

[0069] 1.5, Feedback execution control module:

[0070] Control heater power (baseline ±10% adjustable), crucible lifting motor, cooling gas flow valve (20-30 SLPM);

[0071] Closed-loop detection: If the crystal quality does not improve for two consecutive periods (diameter fluctuation > ±0.5 mm or energy consumption decrease <3%), extend the control period and retrain the model.

[0072] 1.6, Energy consumption evaluation and strategy optimization module:

[0073] Evaluation indicators: unit crystal quality energy consumption (kWh / kg), diameter uniformity (fluctuation range ±0.3 mm), dislocation density (<1×10 3 / cm 2 );

[0074] To evaluate the energy consumption of the assembly, the unit energy consumption calculation formula is used:

[0075] Where P(t) is the total power of the heater and cooling system (kW);

[0076] m is the crystal mass increment in this period (kg);

[0077] Through this formula, the unit energy consumption of the assembly during operation can be accurately calculated, providing a quantitative basis for energy consumption evaluation;

[0078] Diameter fluctuation and dislocation density are evaluated as follows:

[0079] Diameter fluctuation: defined as the difference between the maximum and minimum values in the crystal end diameter data sequence obtained in an evaluation period; that is, the actual diameter value sequence of the crystal end is extracted from the continuously collected high-temperature images, and the difference between the maximum and minimum diameter values in the sequence is calculated, which is the diameter fluctuation index in this period.

[0080] Dislocation density: use etching method combined with microscopic imaging to obtain etching pit image, count by image recognition, and calculate dislocation density per unit area: dislocation density = image recognition count divided by unit area.

[0081] Optimization logic: dynamically adjust the AI model loss function weight according to the evaluation results (such as increasing the energy consumption weight from 0.6 to 0.7).

[0082] 2, Control flow;

[0083] S1 (T0 time): Real-time acquisition of furnace temperature, pressure, flow rate and heat flux data.

[0084] S2 (T1 time): Build multi-physics simulation model to predict heat loss peak at crucible sidewall.

[0085] S3 (T1): The AI ​​model generates control instructions, for example:

[0086] a. Before the Transformer model outputs a control signal based on the predicted temperature difference ΔT, it must first calculate the following linear adjustment terms:

[0087] The heating power adjustment formula is as follows (the formula uses a simplified PID structure):

[0088]

[0089] In the above formula, the first term is the proportional adjustment term, the second term is the integral term, and the third term is the differential term, which reflects the rate of change of temperature difference.

[0090] Where ΔT = T 目标 -T 实测 is the difference between the target crystal interface temperature and the measured temperature;

[0091] k p 、k i 、k d These coefficients are control coefficients, which can be obtained through experiments or simulations, and are usually called adjustment parameters or weight factors; they determine the speed, smoothness and anti-interference ability of the adjustment response; in this embodiment, in a typical single crystal growth process, the above proportional / integral / differential coefficients are assumed to be 0.5, 0.2 and 0.1 respectively, and can also be adjusted in the range of [0.2-1.0], [0.1-0.5], [0.05-0.3] according to the target temperature control accuracy and system inertia to achieve fast and stable thermal field adjustment.

[0092] The above parameters can also be optimized online through simulation modeling or experimental feedback to meet the thermal field stability requirements under different crystal growth strategies;

[0093] T c is the control period, which is 25 seconds in this embodiment and can be adjusted according to the system inertia;

[0094] b. Cooling rate adjustment formula (non-linear mapping):

[0095] Where: k is the adjustment sensitivity coefficient, which is typically 0.3 and can be optimized in the range of [0.1, 1.0] according to the crystal size and system inertia;

[0096] When ΔT>0, the system rapidly increases the cooling rate to suppress the crystal temperature rise;

[0097] When ΔT<0, the system slows down the cooling regulation to avoid overcooling;

[0098] The adjustment parameters in the above two groups of formulas are calculated by the distribution module as the heating zone power instruction and the argon gas flow adjustment amount, and are executed synchronously within the control period T2-T3.

[0099] For the convenience of actual execution, the output value Q cool is converted into the actual physical control parameter by the following method:

[0100] Cooling gas flow = Q cool ; max ;

[0101] Wherein, Q max is the upper limit of the cooling flow rate set by the system (such as 30 SLPM);

[0102] S4 (during T2-T3): execute the control instruction, adjust the heating power to 53 kW, the cooling gas flow to 23 SLPM, and the middle layer graphite felt to the crystal center displacement of 5 mm;

[0103] S5 (at T4): evaluate the crystal quality in the current period, the unit energy consumption is 210 kWh / kg (decreased by 12%), and the diameter fluctuation is ±0.3 mm;

[0104] S6 (at T4): if the energy consumption decrease does not reach the preset threshold (such as <3%), update the AI model parameter weight (such as energy consumption weight 0.6→0.7), and extend the next control period to 35 seconds;

[0105] Input: historical data set, real-time data, model parameters.

[0106] Embodiment 2

[0107] For the second embodiment of the application, which is different from the first embodiment, the embodiment further discloses the processing case of the above abnormal working condition:

[0108] 1. Abnormal condition:

[0109] In the last two control periods, the dislocation density increase rate is >5% / h (from 1.2×10 3 / cm 2 to 1.5×10 3 / cm 2 );

[0110] The unit energy consumption has not decreased (225 kWh / kg).

[0111] 2. Closed-loop response mechanism:

[0112] Step 1: trigger the AI model backtracking mechanism, load the last 100 groups of normal working condition data to retrain the model;

[0113] Step 2: extend the control cycle from 25 seconds to 35 seconds, reduce the adjustment frequency to enhance stability;

[0114] Step 3: prioritize different combinations of control parameters (such as reducing the heating power gradient, increasing the cooling rate upper limit).

[0115] The execution effect of the above technical solution is:

[0116] Third cycle: dislocation density drops to 0.8 x 10 3 / cm 2 (decrease of 47%);

[0117] Unit energy consumption drops to 205 kWh / kg (decrease of 9%), diameter fluctuation stabilizes at ±0.3mm.

[0118] In summary, this scheme uses dynamic model backtracking and parameter optimization, the system quickly responds to abnormal working conditions, avoids the spread of crystal defects; by extending the control cycle, it effectively balances the adjustment frequency and the stability of the thermal field.

[0119] It should be noted that the above examples are used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A silicon single crystal thermal field intelligent energy-saving control system, characterized by: include Multi-physics field acquisition unit, used to collect real-time temperature, flow rate, pressure and heat flux data in the silicon single crystal growth furnace; A thermal field simulation calculation module is used to construct a coupled temperature field and flow field model based on the data obtained by the acquisition unit and perform dynamic heat loss prediction; Gradient thermal insulation components, including graphite felt materials with different thermal conductivity zones, whose arrangement gradually changes along the radial or axial direction to adapt to the change of heat flow gradient; AI control module, used to integrate thermal field simulation results with historical control data and generate thermal field control strategies based on reinforcement learning or deep learning algorithms; A feedback execution control module is connected to the crystal heater, the crucible drive assembly, the cooling system and the thermal field insulation assembly, and is used to adjust each execution unit according to the control strategy; Energy consumption evaluation and strategy optimization module, used to periodically analyze the energy consumption value corresponding to unit crystal growth quality and correct the control priority of the AI ​​model; The system periodically performs thermal field control tasks according to the following time sequence: Data is collected at T0, simulation and AI calculation are completed at T1, control adjustments are implemented from T2 to T3, and energy consumption and crystal quality evaluation results are fed back at T4.

2. The silicon single crystal thermal field intelligent energy-saving control system according to claim 1, characterized in that: The gradient thermal insulation assembly includes at least two graphite felt areas with different thermal conductivities, whose thermal conductivities decrease step by step along the axial direction and are arranged in sequence from the furnace wall to the crystal core to achieve a gradient balanced distribution of heat flow.

3. The silicon single crystal thermal field intelligent energy-saving control system according to claim 2, characterized in that: The thermal field simulation calculation module adopts the finite element multi-physics field simulation method, integrating electrothermal coupling, radiation heat transfer and gas flow field disturbance factors. The simulation time step is set to 10 to 30 seconds, and the simulation frequency is synchronized with the control instruction output.

4. The silicon single crystal thermal field intelligent energy-saving control system according to claim 3, characterized in that: The AI ​​control module uses historical growth data to build a thermal field control model and dynamically updates the strategy based on the current crystal diameter change rate and dislocation density evolution trend.

5. The silicon single crystal thermal field intelligent energy-saving control system according to claim 4, characterized in that: The feedback execution control module has a control logic closed-loop detection function. If the crystal quality does not significantly improve within two consecutive control cycles, the AI ​​model retraining process is started and the control strategy output interval is adjusted.

6. The method for intelligent energy-saving control of a silicon single crystal thermal field according to claim 5, characterized in that: The following steps are involved: S1: Real-time data acquisition, acquiring temperature, airflow, pressure, and heat flux data during crystal growth every 10 to 30 seconds; S2: Multi-physics field simulation construction, coupled thermal field simulation based on collected data, output temperature distribution and heat loss estimation results; S3: AI intelligent strategy generation, integrating simulation output with historical control data, and using AI models to generate thermal field control parameters for the next period; S4: Control instructions are issued and executed to adjust the heating power, crucible position, cooling rate, and deformation configuration of the gradient insulation structure within the control cycle; S5: Crystal growth quality assessment, evaluating the current crystal diameter uniformity, defect density and specific energy consumption; S6: Strategy feedback and model correction, update the AI ​​model parameter weights based on the results of S5, and determine whether the control cycle T needs to be adjusted n+1 .

7. The method for intelligent energy-saving control of a silicon single crystal thermal field according to claim 6, characterized in that: In step S3, the AI ​​model uses the Transformer structure and the historical control sequence to perform attention fusion to determine the optimal control parameter distribution.

8. The method for intelligent energy-saving control of a silicon single crystal thermal field according to claim 7, characterized in that: In step S4, the adjustable structure of the gradient thermal insulation component controls its position or shape through a servo mechanism to adapt to the optimization requirements of the heat flow direction at different growth stages.

9. The method for intelligent energy-saving control of a silicon single crystal thermal field according to claim 8, characterized in that: In step S5 , the energy consumption evaluation indicators include the total power consumption per unit crystal mass, the average power curve smoothness, and the number of crystal defects.

10. The method for intelligent energy-saving control of a silicon single crystal thermal field according to claim 9, characterized in that: In step S6, if the unit energy consumption does not decrease significantly in two consecutive growth cycles, the system triggers the AI ​​model backtracking mechanism, retrains the model and prioritizes different control parameter combinations in the next cycle.

Images