Wafer annealing thermal management control system

Abstract

The invention discloses a wafer annealing thermal management control system, and relates to the technical field of annealing thermal management. The method comprises the steps of calculating an error variable by acquiring temperature information of a target wafer in real time; determining a fuzzy image according to the error variable and a preset membership function; calculating a coefficient adjustment value according to the fuzzy image; setting the reference coefficient according to the coefficient adjustment value to obtain a target scale factor; and a PID control algorithm is adopted, and the heating power is calculated according to the target scale factor and the temperature difference value. Through combination of temperature acquisition, fuzzy control and self-adaptive PID adjustment, the problems of temperature response delay, overshoot oscillation, poor anti-interference capability and the like in traditional control are effectively solved. The system performs fuzzy reasoning based on the temperature difference and the change rate thereof, dynamically adjusts PID parameters, and optimizes an initial coefficient through a particle swarm and simulated annealing hybrid optimization algorithm, so that high-precision and high-robustness temperature control adjustment is realized, and the temperature control precision is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of annealing thermal management, and in particular to a wafer annealing thermal management control system. Background Art

[0002] In the semiconductor manufacturing process, wafer annealing is a critical step in optimizing device performance, restructuring material structures, and relieving stress. This process typically involves heating the wafer to hundreds or even thousands of degrees Celsius under precisely controlled temperature conditions, followed by maintaining or cooling the wafer using specific temperature control strategies to improve crystal defects, adjust doping profiles, or enhance interface stability. However, because the annealing process places extremely high demands on temperature control precision and stability, the thermal management and control systems involved in this process face numerous challenges in practical applications.

[0003] In actual production environments, heating systems often suffer from large thermal inertia and delayed response, making it difficult for temperature changes to respond quickly to control commands. This in turn causes system overshoot and undershoot, resulting in temperature inaccuracy and impacting annealing results and product consistency. During operation, wafer annealing systems are often affected by external environmental disturbances, such as changes in furnace thermal radiation and power supply fluctuations. These factors can cause changes in the system's dynamic characteristics. Traditional control strategies, such as PID control, struggle to adapt to these complex and changing interference conditions when parameters are fixed. Improper system parameter settings can easily lead to excessive temperature fluctuations, oscillations in the adjustment process, and even system instability, further impacting process yield. Summary of the Invention

[0004] The purpose of the present invention is to solve the problem of inaccurate temperature mentioned in the above background technology and to provide a wafer annealing thermal management control system.

[0005] A first aspect of the present invention provides a wafer annealing thermal management control system, the system comprising:

[0006] Temperature measurement module, used to obtain temperature information of the target wafer in real time;

[0007] an error calculation module, configured to obtain an error variable based on the temperature information and the target temperature; the error variable includes a normalized temperature difference and a temperature difference change rate;

[0008] A fuzzification module, configured to determine a fuzzy atlas according to the error variable and a preset membership function;

[0009] a defuzzification module, configured to calculate coefficient adjustment values ​​according to the fuzzy atlas; the coefficient adjustment values ​​include proportional coefficient adjustment values, integral coefficient adjustment values, and differential coefficient adjustment values;

[0010] A parameter tuning module is used to tune the preset reference coefficients according to the coefficient adjustment values ​​to obtain target proportional factors; the reference coefficients include initial values ​​of the proportional coefficient, the integral coefficient, and the differential coefficient; before deployment, the reference coefficients are optimized using a hybrid algorithm combining a particle swarm optimization algorithm and a simulated annealing algorithm;

[0011] The power determination module is used to calculate the heating power according to the target proportional factor and the temperature difference using a PID control algorithm.

[0012] Preferably, the system stores a plurality of preset rule mapping tables; the operation process of the fuzzification module includes:

[0013] Inputting the target variable into a preset membership function to obtain its membership in each fuzzy language; the target variable is any one of the error variables; the membership function is a triangular membership function;

[0014] Take the fuzzy language with a membership degree greater than zero to obtain the membership label set corresponding to the target variable;

[0015] According to the membership label sets of the two error variables, each element is matched with each other to obtain multiple membership pairs;

[0016] According to the plurality of membership pairs, searching a target rule mapping table to determine an activated rule and its activation strength; the target rule mapping table is any one of the plurality of rule mapping tables; the activation strength is determined by the membership label with the lower membership degree in the membership pair;

[0017] The rules with the same consequent are grouped together, and the membership function of the consequent is truncated using the maximum activation intensity to obtain a fuzzy subgraph.

[0018] Merge all fuzzy subgraphs to obtain the target fuzzy graph;

[0019] Multiple target fuzzy graphs obtained from multiple rule mapping tables are combined to obtain a fuzzy graph set.

[0020] Preferably, the operation process of the defuzzification module includes:

[0021] According to the function expression of the target fuzzy graph, a discrete point set is obtained;

[0022] According to the discrete point set, the target coefficient adjustment value is obtained:

[0023]

[0024] Wherein, ΔY is the target coefficient adjustment value, corresponding to any one of ΔKp, ΔKi and ΔKd, representing the proportional coefficient adjustment value, the integral coefficient adjustment value and the differential coefficient adjustment value respectively; fY is the function expression of the target fuzzy graph; X i is the horizontal coordinate of the i-th discrete point; N is the total number of discrete points.

[0025] Preferably, the process of determining the reference coefficient includes:

[0026] Step 1: Use the reverse learning strategy to generate the initial position of each particle and set the initial velocity of each particle to zero; each particle position is a three-dimensional vector, representing a solution;

[0027] Step 2: Substitute the particle position into the fuzzy PID controller in the simulation environment, simulate the temperature change for a period of time, and obtain a temperature change curve; calculate the fitness value based on the temperature change curve and the preset expected temperature curve; the smaller the fitness value, the better;

[0028] Step 3: Update the global optimal position and the individual optimal position of each particle according to the fitness value of each particle;

[0029] Step 4: Add a small disturbance to the global optimal position to obtain a new candidate solution;

[0030] Step 5: Calculate the fitness value of the new candidate solution; if the fitness value of the new candidate solution is less than the fitness value of the global optimal position, update the global optimal position to the new candidate solution and proceed to step 7; otherwise, proceed to step 6;

[0031] Step 6: Calculate the acceptance probability based on the current number of iterations and randomly generate a random number in [0, 1]. If the random number is less than the acceptance probability, update the global optimal position to the new candidate solution; otherwise, proceed directly to step 7.

[0032] Step 7, updating the number of iterations; if the number of iterations reaches the maximum number of iterations, the optimization process ends and the global optimal position or the individual optimal position with the smallest fitness value is output as the benchmark coefficient, otherwise proceed to step 8;

[0033] Step 8: Update the speed and position of each particle based on the global optimal position and the individual optimal position of each particle; return to step 2.

[0034] Preferably, the method of generating the initial position of each particle by adopting a reverse learning strategy includes:

[0035] In the preset search space, M position vectors are randomly generated;

[0036] Inverting the M position vectors according to the maximum and minimum values ​​of the positions in the search space to obtain another M position vectors;

[0037] Calculate the fitness of 2M position vectors, arrange them from small to large, and take the first M position vectors as the initial position of each particle in the population.

[0038] Preferably, the calculating the fitness value according to the temperature change curve and the expected temperature curve includes:

[0039]

[0040] Among them, f is the fitness value; T set (t) is the expected temperature at time step step, given by the expected temperature curve; T sim (t) is the response temperature at the time step step, which is given by the temperature change curve.

[0041] Preferably, calculating the reception probability according to the current number of iterations includes:

[0042]

[0043] Among them, P acc is the probability of receiving; e is a natural constant; COUNT is the maximum number of iterations; t is the current number of iterations; f NEW and f CUR are the fitness values ​​of the new candidate solution and the global optimal position, respectively.

[0044] Preferably, updating the speed and position of each particle according to the global optimal position and the individual optimal position of each particle includes:

[0045]

[0046] Among them, w i is the inertia weight of the i-th particle; f i is the fitness of the i-th particle; f avg is the mean fitness of all particles; f max is the maximum fitness of all particles; w max and w min is the maximum and minimum value of the inertia weight; c1 and c2 are learning factors; c 1max and c 1min is the maximum and minimum value of the learning factor c1; c 2max and c 2min is the maximum and minimum value of the learning factor c2; t is the current number of iterations; COUNT is the maximum number of iterations; and is the particle velocity and particle position of the i-th particle at iteration round t; PB iis the individual optimal position of the i-th particle; GB is the global optimal position; r1 and r2 are random numbers between (0, 1); and are the updated particle speed and particle position.

[0047] Beneficial effects of the present invention:

[0048] The present invention proposes a wafer annealing thermal management control system, which includes: a temperature measurement module for acquiring temperature information of a target wafer in real time; an error calculation module for obtaining an error variable based on the temperature information and the target temperature; the error variable includes a normalized temperature difference and a temperature difference change rate; a fuzzification module for determining a fuzzy atlas based on the error variable and a preset membership function; a defuzzification module for calculating a coefficient adjustment value based on the fuzzy atlas; the coefficient adjustment value includes a proportional coefficient adjustment value, an integral coefficient adjustment value, and a differential coefficient adjustment value; a parameter setting module for setting a preset reference coefficient based on the coefficient adjustment value to obtain a target proportional factor; the reference coefficient includes an initial value of the proportional coefficient, an initial value of the integral coefficient, and an initial value of the differential coefficient; before deployment, the reference coefficient is optimized and determined using a hybrid algorithm combining a particle swarm optimization algorithm and a simulated annealing algorithm; a power determination module for calculating the heating power based on the target proportional factor and the temperature difference using a PID control algorithm.

[0049] By combining real-time temperature acquisition, fuzzy control, and adaptive PID regulation, the system effectively addresses the temperature response hysteresis, overshoot oscillation, and poor anti-interference capabilities inherent in traditional control. The system dynamically adjusts PID parameters using fuzzy reasoning based on the temperature difference and its rate of change. It also optimizes initial coefficients using a hybrid particle swarm optimization algorithm and simulated annealing, achieving high-precision, robust temperature control and significantly improving temperature control accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 An architecture diagram of a wafer annealing thermal management control system is provided for an embodiment of the present invention;

[0051] Figure 2 A flowchart of iterative optimization of a benchmark coefficient is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] The embodiment of the present invention provides a wafer annealing thermal management control system. Figure 1 , Figure 1 This is an architecture diagram of a wafer annealing thermal management control system provided by an embodiment of the present invention. The system includes a temperature measurement module, an error calculation module, a fuzzification module, a defuzzification module, a parameter setting module, and a power determination module, wherein:

[0054] The temperature measurement module is used to obtain the temperature information of the target wafer in real time.

[0055] The error calculation module is used to obtain an error variable according to the temperature information and the target temperature.

[0056] The fuzzification module is used to determine the fuzzy atlas according to the error variable and the preset membership function.

[0057] The defuzzification module is used to calculate the coefficient adjustment value according to the fuzzy atlas.

[0058] The parameter setting module is used to set the preset reference coefficient according to the coefficient adjustment value to obtain the target proportional factor.

[0059] The power determination module is used to calculate the heating power according to the target proportional factor and the temperature difference using a PID control algorithm.

[0060] The error variables include the normalized temperature difference and the temperature difference change rate. The coefficient adjustment values ​​include the proportional coefficient adjustment value ΔKP, the integral coefficient adjustment value ΔKI, and the differential coefficient adjustment value ΔKD. The baseline coefficients include the initial values ​​of the proportional coefficient KP0, the integral coefficient KI0, and the differential coefficient KD0. Prior to deployment, a hybrid algorithm combining particle swarm optimization and simulated annealing was used to optimize the baseline coefficients.

[0061] A wafer annealing thermal management control system, based on an embodiment of the present invention, effectively addresses the problems of temperature response hysteresis, overshoot oscillation, and poor anti-interference capability found in traditional control systems by combining real-time temperature acquisition, fuzzy control, and adaptive PID regulation. The system dynamically adjusts PID parameters using fuzzy inference based on temperature differences and their rates of change. It also optimizes initial coefficients using a hybrid particle swarm optimization algorithm and simulated annealing, achieving high-precision and robust temperature control and significantly improving temperature control accuracy.

[0062] In one embodiment, the system stores a plurality of preset rule mapping tables, which correspond to proportional coefficient adjustment, integral coefficient adjustment and differential coefficient adjustment respectively. According to the actual experience of technicians, fuzzy language can be used to describe error variables and coping methods. For example, when the absolute value of the temperature difference is large, the actual temperature differs greatly from the expected temperature, and a larger proportional coefficient Kp should be selected to make the temperature respond quickly. When the absolute value of the temperature difference change rate is large, a smaller differential coefficient should be selected to avoid oscillation. When the absolute value of the temperature difference is small, a larger integral coefficient should be selected to improve the steady-state performance of the system. The larger and smaller here are fuzzy descriptions. Taking the rule mapping table for proportional coefficient adjustment as an example, see Table 1. Table 1 shows a mapping relationship from error variable fuzzy statements to coping method fuzzy statements, totaling 49 items.

[0063] Table 1

[0064]

[0065] Among them, the fuzzy languages ​​NB, NM, NS, ZO, PS, PM, and PB represent large negative, medium negative, small negative, zero, small positive, medium positive, and large positive, respectively, and are used to describe errors and countermeasures.

[0066] The operation process of the fuzzification module includes:

[0067] Input the target variable into the preset membership function to obtain its membership in each fuzzy language; the target variable is any one of the error variables. Specifically, the membership function adopts a triangular membership function. The fuzzy domains of the normalized temperature difference e and the temperature difference change rate ec are [-6, 6] and [-3, 3], and the fuzzy domain of the adjustment value is [-6, 6]. Normalization refers to mapping the calculated temperature difference to the interval [-6, 6] and mapping the temperature difference change rate to the interval [-3, 3]. The form of the triangular membership function is:

[0068]

[0069] For each language value (NB, NM, NS, ZO, PS, PM, PB), the parameter settings of the membership function are shown in Table 2.

[0070] Table 2

[0071]

[0072] We select fuzzy linguistics with membership greater than zero to obtain the set of membership labels corresponding to the target variable. Specifically, the membership function and Table 2 show that the target variable only has two fuzzy linguistics with membership greater than zero. For example, when the variable e is -2.5, the membership labels are NM and NS, with memberships of 0.25 and 0.75, respectively. When the variable ec is 1.5, the membership labels are PS and PM, with memberships of 0.5 and 0.5, respectively.

[0073] Based on the membership label sets of the two error variables, each element is matched pairwise to obtain multiple membership pairs. Specifically, one variable corresponds to two labels, and the cross combination has a total of four membership pairs, such as (NM, PS), (NM, PM), (NS, PS), and (NS, PM).

[0074] Based on multiple membership pairs, the target rule mapping table is searched to determine the activated rule and its activation strength. The target rule mapping table is any one of the multiple rule mapping tables. The activation strength is determined by the membership label with the lower membership degree in the membership pair. Specifically, taking the rule mapping table with proportional coefficient adjustment as an example, see Table 1. The membership pair (NM, PS) corresponds to ZO, that is, the activation rule (NM, PS) -> ZO. The membership degree of NM is 0.25, and the membership degree of PS is 0.5, so the activation strength is 0.25.

[0075] Rules with the same consequent are grouped together, and the membership function of the consequent is truncated using the maximum activation strength to obtain a fuzzy subgraph. For example, when the consequent of the rule is ZO and the activation strength is 0.25, the function expression u of the truncated fuzzy subgraph is ZO (x) is:

[0076]

[0077] Merge all fuzzy subgraphs to get the target fuzzy graph. Specifically, take the larger value of the overlapping domains.

[0078] Multiple target fuzzy graphs obtained from multiple rule mapping tables are combined to obtain a fuzzy graph set.

[0079] This embodiment uses the smaller degree of membership in a membership pair as the rule activation strength. This prevents high membership in a single dimension from misleading overall judgment during rule activation, improving the credibility and stability of inference results and enhancing the system's fault tolerance for fuzzy input. By truncating the maximum activation strength of multiple rule results for the same output language, it preserves the strongest control trends among all possible rules, preventing effective rules from being "diluted" by weaker ones. This enhances the system's response sensitivity and control direction consistency during fuzzy graph synthesis, improving the accuracy and control effectiveness of the final defuzzification results.

[0080] In one embodiment, the operation process of the defuzzification module includes:

[0081] According to the function expression of the target fuzzy graph, a discrete point set is obtained.

[0082] According to the discrete point set, the target coefficient adjustment value is obtained:

[0083]

[0084] Wherein, ΔY is the target coefficient adjustment value, corresponding to any one of ΔKp, ΔKi and ΔKd, representing the proportional coefficient adjustment value, the integral coefficient adjustment value and the differential coefficient adjustment value respectively; f Y is the function expression of the target fuzzy graph; X i is the horizontal coordinate of the i-th discrete point; N is the total number of discrete points.

[0085] This embodiment uses the centroid method to defuzzify the fuzzy graph and calculates the coefficient adjustment value through weighted averaging of discrete points. This ensures smooth and continuous output, avoids jumps, and improves the stability and response consistency of the control system. This method preserves the combined effects of multiple rules in fuzzy inference, has good globality and accuracy, and can reflect the combined contributions of all activated rules.

[0086] In one implementation, the target coefficient adjustment value is mapped to the adjustment interval of the corresponding coefficient and then added to the reference parameter to obtain the tuned PID parameter.

[0087] In one embodiment, see Figure 2 , Figure 2 A flowchart of iterative optimization of a benchmark coefficient is provided for an embodiment of the present invention. The process of determining the benchmark coefficient includes:

[0088] In step 1, a reverse learning strategy is used to generate the initial position of each particle and the initial velocity of each particle is set to zero; each particle position is a three-dimensional vector (KP, KI, KD), representing a solution.

[0089] Step 2: Substitute the particle position into the fuzzy PID controller in the simulation environment, simulate the temperature change for a period of time, and obtain the temperature change curve; calculate the fitness value based on the temperature change curve and the preset expected temperature curve; the smaller the fitness value, the better.

[0090] Step 3: Update the global optimal position and the individual optimal position of each particle according to the fitness value of each particle.

[0091] Step 4: Add a small perturbation to the global optimal position to obtain a new candidate solution.

[0092] Step 5: Calculate the fitness value of the new candidate solution. If the fitness value of the new candidate solution is less than the fitness value of the global optimal position, update the global optimal position to the new candidate solution and proceed to step 7. Otherwise, proceed to step 6.

[0093] Step 6: Calculate the acceptance probability based on the current number of iterations and randomly generate a random number in [0, 1]. If the random number is less than the acceptance probability, update the global optimal position to the new candidate solution, otherwise proceed directly to step 7.

[0094] Step seven, update the number of iterations; if the number of iterations reaches the maximum number of iterations, end the optimization process and output the global optimal position or the individual optimal position with the smallest fitness value as the benchmark coefficient, otherwise go to step eight.

[0095] Step 8: Update the speed and position of each particle based on the global optimal position and the individual optimal position of each particle; return to step 2.

[0096] This embodiment utilizes a hybrid algorithm combining particle swarm optimization and simulated annealing to balance global search capabilities with local convergence, effectively avoiding being trapped in local optima. The introduction of reverse learning initialization enhances population diversity, while the perturbation mechanism of simulated annealing increases the probability of escaping local optima. This provides a high-quality initial value foundation for fuzzy parameter adjustment, improving the overall system control performance and robustness.

[0097] In one implementation, the reverse learning strategy is used to generate the initial position of each particle, including:

[0098] Step 1: Randomly generate M position vectors in a preset search space.

[0099] Step 2: According to the maximum and minimum values ​​of the position in the search space, the M position vectors are inverted to obtain another M position vectors. Specifically, for the position vector (KPj, KIj, KDj), the inversion is (KPMAX+KPMIN-KPj, KIMAX+KIMIN-KIj, KDMAX+KDMIN-KDj). Among them, KPMAX and KPMIN are the maximum and minimum values ​​of KP; KIMAX and KIMIN are the maximum and minimum values ​​of KI; KDMAX and KDMIN are the maximum and minimum values ​​of KD;

[0100] Step 3: Calculate the fitness of 2M position vectors, arrange them from small to large, and take the first M position vectors as the initial position of each particle in the population.

[0101] This implementation reverse-maps the initial particles to generate the original solution and its symmetric solution, which together form an initial population that covers a wider range of the search space, effectively improving the diversity and uniformity of the population. Combined with fitness screening, retaining the optimal initial solution helps particle swarm optimization start from more promising areas, avoiding local optimality caused by excessive initial position concentration, enhancing global search capabilities and convergence efficiency, and providing a more robust initial foundation for subsequent parameter optimization.

[0102] In one implementation, calculating the fitness value according to the temperature change curve and the expected temperature curve includes:

[0103]

[0104] Among them, f is the fitness value; T set (t) is the expected temperature at time step step, given by the expected temperature curve; T sim (t) is the response temperature at the time step step, which is given by the temperature change curve.

[0105] The fitness function calculates the sum of squares of the errors between the temperature response and the expected curve at each time point. The larger the value, the worse the effect, and the smaller the value, the better the effect, which effectively feeds back the particle position evaluation.

[0106] In one implementation, calculating the reception probability according to the current number of iterations includes:

[0107]

[0108] Among them, P acc is the probability of receiving; e is a natural constant; COUNT is the maximum number of iterations; t is the current number of iterations; f NEW and f CUR are the fitness values ​​of the new candidate solution and the global optimal position, respectively.

[0109] This acceptance probability design incorporates the temperature-decreasing principle of the simulated annealing algorithm. As the number of iterations increases, the acceptance probability gradually decreases. A high acceptance probability in the early stages of an iteration allows the algorithm to more frequently accept poor solutions, helping it escape local optima and explore a wider range of potential solutions. As the iterations progress, the acceptance probability decreases, and the algorithm becomes more inclined to reject poor solutions and focus its search around the currently optimal solution, thus achieving a smooth transition from a coarse global search to a finer local search.

[0110] In one implementation, updating the velocity and position of each particle based on the global optimal position and the individual optimal position of each particle includes:

[0111]

[0112] Among them, w i is the inertia weight of the i-th particle; f i is the fitness of the i-th particle; f avg is the mean fitness of all particles; f max is the maximum fitness of all particles; w max and w min is the maximum and minimum value of the inertia weight; c1 and c2 are learning factors; c 1max and c 1min is the maximum and minimum value of the learning factor c1; c 2max and c 2min is the maximum and minimum value of the learning factor c2; t is the current number of iterations; COUNT is the maximum number of iterations; and is the particle velocity and particle position of the i-th particle at iteration round t; PB i is the individual optimal position of the i-th particle; GB is the global optimal position; r1 and r2 are random numbers between (0, 1); and are the updated particle speed and particle position.

[0113] In this implementation, particles with above-average fitness have their inertia weight reduced, favoring local exploration. Particles with below-average fitness maintain a high inertia weight, enhancing global exploration. This mechanism effectively balances the algorithm's global and local search, preventing premature entrapment in local optima. Initially, a large learning factor is assigned to encourage particles to explore toward the individual optimum (PB) and the global optimum (GB). As iterations proceed, the learning factor is gradually reduced to reduce randomness and enhance the refined exploration of high-quality solutions. This phased strategy balances early exploration with later convergence efficiency.

[0114] It should be noted that, in this document, terms such as "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements that are inherent to such process, method, article or apparatus.

[0115] The embodiments of the present invention are described in detail above, but the contents described are only preferred embodiments of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A wafer annealing thermal management control system, characterized in that: The system comprises: Temperature measurement module, used to obtain temperature information of the target wafer in real time; an error calculation module, configured to obtain an error variable based on the temperature information and the target temperature; the error variable includes a normalized temperature difference and a temperature difference change rate; A fuzzification module, configured to determine a fuzzy atlas according to the error variable and a preset membership function; a defuzzification module, configured to calculate coefficient adjustment values ​​according to the fuzzy atlas; the coefficient adjustment values ​​include proportional coefficient adjustment values, integral coefficient adjustment values, and differential coefficient adjustment values; A parameter tuning module is used to tune the preset reference coefficients according to the coefficient adjustment values ​​to obtain target proportional factors; the reference coefficients include initial values ​​of the proportional coefficient, the integral coefficient, and the differential coefficient; before deployment, the reference coefficients are optimized using a hybrid algorithm combining a particle swarm optimization algorithm and a simulated annealing algorithm; The power determination module is used to calculate the heating power according to the target proportional factor and the temperature difference using a PID control algorithm.

2. A wafer annealing thermal management control system according to claim 1, characterized in that: The system stores a plurality of preset rule mapping tables; the operation process of the fuzzification module includes: Inputting the target variable into a preset membership function to obtain its membership in each fuzzy language; the target variable is any one of the error variables; the membership function is a triangular membership function; Take the fuzzy language with a membership degree greater than zero to obtain the membership label set corresponding to the target variable; According to the membership label sets of the two error variables, each element is matched with each other to obtain multiple membership pairs; According to the plurality of membership pairs, searching a target rule mapping table to determine an activated rule and its activation strength; the target rule mapping table is any one of the plurality of rule mapping tables; the activation strength is determined by the membership label with the lower membership degree in the membership pair; The rules with the same consequent are grouped together, and the membership function of the consequent is truncated using the maximum activation intensity to obtain a fuzzy subgraph. Merge all fuzzy subgraphs to obtain the target fuzzy graph; Multiple target fuzzy graphs obtained from multiple rule mapping tables are combined to obtain a fuzzy graph set.

3. The wafer annealing thermal management control system according to claim 2, characterized in that: The operation process of the defuzzification module includes: According to the function expression of the target fuzzy graph, a discrete point set is obtained; According to the discrete point set, the target coefficient adjustment value is obtained: Wherein, ΔY is the target coefficient adjustment value, corresponding to any one of ΔKp, ΔKi and ΔKd, representing the proportional coefficient adjustment value, the integral coefficient adjustment value and the differential coefficient adjustment value respectively; f Y is the function expression of the target fuzzy graph; X i is the horizontal coordinate of the i-th discrete point; N is the total number of discrete points.

4. The wafer annealing thermal management control system according to claim 1, characterized in that: The process of determining the reference coefficient includes: Step 1: Use the reverse learning strategy to generate the initial position of each particle and set the initial velocity of each particle to zero; each particle position is a three-dimensional vector, representing a solution; Step 2: Substitute the particle position into the fuzzy PID controller in the simulation environment, simulate the temperature change for a period of time, and obtain a temperature change curve; calculate the fitness value based on the temperature change curve and the preset expected temperature curve; the smaller the fitness value, the better; Step 3: Update the global optimal position and the individual optimal position of each particle according to the fitness value of each particle; Step 4: Add a small disturbance to the global optimal position to obtain a new candidate solution; Step 5: Calculate the fitness value of the new candidate solution; if the fitness value of the new candidate solution is less than the fitness value of the global optimal position, update the global optimal position to the new candidate solution and proceed to step 7; otherwise, proceed to step 6; Step 6: Calculate the acceptance probability based on the current number of iterations and randomly generate a random number in [0, 1]. If the random number is less than the acceptance probability, update the global optimal position to the new candidate solution; otherwise, proceed directly to step 7. Step 7, updating the number of iterations; if the number of iterations reaches the maximum number of iterations, the optimization process ends and the global optimal position or the individual optimal position with the smallest fitness value is output as the benchmark coefficient, otherwise proceed to step 8; Step 8: Update the speed and position of each particle based on the global optimal position and the individual optimal position of each particle; return to step 2.

5. The wafer annealing thermal management control system according to claim 4, characterized in that: The method of generating the initial position of each particle by adopting the reverse learning strategy includes: In the preset search space, M position vectors are randomly generated; Inverting the M position vectors according to the maximum and minimum values ​​of the positions in the search space to obtain another M position vectors; Calculate the fitness of 2M position vectors, arrange them from small to large, and take the first M position vectors as the initial position of each particle in the population.

6. The wafer annealing thermal management control system according to claim 4, characterized in that: The calculating the fitness value according to the temperature change curve and the expected temperature curve includes: Among them, f is the fitness value; T set (t) is the expected temperature at time step step, given by the expected temperature curve; T sim (t) is the response temperature at the time step step, which is given by the temperature change curve.

7. The wafer annealing thermal management control system according to claim 4, characterized in that: Calculating the reception probability according to the current number of iterations includes: Among them, P acc is the probability of receiving; e is a natural constant; COUNT is the maximum number of iterations; t is the current number of iterations; f NEW and f CUR are the fitness values ​​of the new candidate solution and the global optimal position, respectively.

8. The wafer annealing thermal management control system according to claim 4, characterized in that: The updating of the speed and position of each particle according to the global optimal position and the individual optimal position of each particle includes: Among them, w i is the inertia weight of the i-th particle; f i is the fitness of the i-th particle; f avg is the mean fitness of all particles; f max is the maximum fitness of all particles; w max and w min is the maximum and minimum value of the inertia weight; c1 and c2 are learning factors; c 1max and c 1min is the maximum and minimum value of the learning factor c1; c 2max and c 2min is the maximum and minimum value of the learning factor c2; t is the current number of iterations; COUNT is the maximum number of iterations; and is the particle velocity and particle position of the i-th particle at iteration round t; PB i is the individual optimal position of the i-th particle; GB is the global optimal position; r1 and r2 are random numbers between (0, 1); and are the updated particle speed and particle position.

Images