Wafer intrinsic temperature control method, device and equipment

Abstract

The invention relates to the technical field of semiconductors, in particular to a wafer intrinsic temperature control method, device and equipment. Performing feature extraction on the amplitude ratio spectrum and the phase difference spectrum based on a preset time sequence to obtain a corrected growth rate; calculating the process formula change sequence according to a preset thermal inertia equation and a preset growth dynamics relation to obtain feed-forward compensation power; calculating the corrected growth rate according to a preset multivariable PID control algorithm to obtain feedback regulation power; resolving the feedforward compensation power and the feedback adjustment power according to a preset decoupling formula to obtain a wafer intrinsic temperature power control instruction; feedback regulation power is generated by means of a multivariable PID control algorithm, temperature zone interference is reduced through decoupling, partition temperature control is achieved, the anti-interference capacity of the system is improved, stable quality of an epitaxial layer is guaranteed, the consistency and reliability of the MOCVD process are enhanced, and support is provided for large-scale production of semiconductor devices.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, and in particular to a method, apparatus and equipment for controlling the intrinsic temperature of a wafer. Background Technology

[0002] Metal-organic chemical vapor deposition (MOCVD) is a core technology in the field of compound semiconductor material preparation, widely supporting the mass production of epitaxial materials such as gallium nitride and gallium arsenide. Its temperature control accuracy and uniformity are crucial for ensuring the quality and compositional consistency of the epitaxial layer crystal, and are also key factors determining the performance, yield, and industrialization progress of high-end optoelectronic devices such as Micro-LEDs and RF devices. Currently, mainstream temperature control solutions for MOCVD equipment adopt indirect measurement feedback control modes, which face insurmountable technical bottlenecks: thermocouples and infrared temperature measurement methods either generate dynamic temperature differences due to contact thermal resistance or introduce measurement errors due to emissivity drift and cavity contamination, exhibiting inherent limitations. Furthermore, existing solutions indirectly control process quality by adjusting intermediate temperature variables such as heaters and trays, which are susceptible to process fluctuations due to the "black box" effect of multi-physical field coupling, including gas-phase mass transfer and surface reactions. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the present invention aims to provide a method, apparatus and equipment for controlling the intrinsic temperature of a wafer.

[0004] The first aspect of this invention provides a method for controlling the intrinsic temperature of a wafer, comprising: acquiring an amplitude ratio spectrum and a phase difference spectrum; extracting features from the amplitude ratio spectrum and the phase difference spectrum based on a preset time sequence to obtain a corrected growth rate; acquiring a process formulation change sequence and calculating the process formulation change sequence according to a preset thermal inertia equation and a preset growth kinetic relationship to obtain a feedforward compensation power; calculating the corrected growth rate according to a preset multivariable PID control algorithm to obtain a feedback regulation power; and solving the feedforward compensation power and the feedback regulation power according to a preset decoupling formula to obtain a wafer intrinsic temperature power control command.

[0005] Furthermore, the step of extracting features from the amplitude ratio spectrum and phase difference spectrum based on a preset time sequence to obtain the corrected growth rate includes: solving the amplitude ratio spectrum and phase difference spectrum based on a preset layered optical model to obtain multiple model parameters; sorting the multiple model parameters according to the time sequence to obtain a characteristic time sequence signal; acquiring the reflected light intensity signal and analyzing the reflected light intensity signal to obtain the reflected light intensity oscillation period; calculating the reflected light intensity oscillation period according to a preset growth rate calculation formula to obtain the growth rate; and correcting the growth rate according to the characteristic time sequence signal to obtain the corrected growth rate.

[0006] Furthermore, the step of calculating the process formula change sequence based on the preset thermal inertia equation and the preset growth kinetic relationship to obtain the feedforward compensation power includes: calculating the process formula change sequence based on the thermal inertia equation to obtain the wafer temperature disturbance value; calculating the wafer temperature disturbance value based on the growth kinetic relationship to obtain the growth rate disturbance value; and generating the feedforward compensation power based on the wafer temperature disturbance value and the growth rate disturbance value.

[0007] Further, the step of calculating the corrected growth rate according to a preset multivariable PID control algorithm to obtain feedback regulation power includes: performing deviation analysis between the corrected growth rate and the preset target growth rate to obtain the growth rate deviation; obtaining the alloy composition from the characteristic time-series signal; performing deviation analysis between the alloy composition and the preset target alloy composition to obtain the alloy composition deviation; calculating the growth rate deviation and the alloy composition deviation according to the multivariable PID control algorithm to obtain the preliminary regulation output; and processing the preliminary regulation output according to a preset dynamic decoupling network to obtain the feedback regulation power.

[0008] Further, the step of calculating the feedforward compensation power and feedback adjustment power according to a preset decoupling formula to obtain the wafer intrinsic temperature power control command includes: summing the feedforward compensation power and feedback adjustment power to obtain a power command vector; predicting the corrected growth rate and reflected light intensity signal according to a preset lightweight neural network to obtain the wafer intrinsic temperature; analyzing the power command vector and the wafer intrinsic temperature according to a preset recursive least squares algorithm to obtain a thermal coupling matrix; and calculating the thermal coupling matrix according to the decoupling formula to obtain the wafer intrinsic temperature power control command.

[0009] Furthermore, the step of predicting the corrected growth rate and reflected light intensity signal based on a preset lightweight neural network to obtain the intrinsic temperature of the wafer includes: extracting features from the reflected light intensity signal to obtain surface morphology trend features; correcting the surface morphology trend features based on the feature time sequence signal to obtain corrected trend features; predicting the corrected trend features and the reflected light intensity oscillation period based on the lightweight neural network to obtain a predicted temperature value; and calculating the corrected growth rate and the predicted temperature value to obtain the intrinsic temperature of the wafer.

[0010] Furthermore, the step of predicting the corrected trend features and the oscillation period of reflected light intensity based on the lightweight neural network to obtain the temperature prediction value includes: acquiring historical parameter data pairs; pre-training the lightweight neural network based on the historical parameter data pairs to obtain a pre-trained neural network model; obtaining the oscillation amplitude features from the corrected trend features; and predicting the oscillation amplitude features, the feature time-series signal, and the oscillation period of reflected light intensity based on the pre-trained neural network model to obtain the temperature prediction value.

[0011] Further, the calculation of the corrected growth rate and temperature prediction value to obtain the intrinsic temperature of the wafer includes: obtaining the current process parameters, and solving for the current process parameters, the corrected growth rate, and the alloy composition to obtain a preliminary estimated temperature; performing a deviation analysis on the preliminary estimated temperature and the temperature prediction value to obtain a temperature deviation value; determining whether the temperature deviation value is less than a preset tolerance threshold; and when the temperature deviation value is less than the tolerance threshold, performing a weighted calculation on the preliminary estimated temperature and the temperature prediction value to obtain the intrinsic temperature of the wafer.

[0012] Furthermore, a wafer intrinsic temperature control device includes: an acquisition module for acquiring amplitude ratio spectrum and phase difference spectrum; a feature extraction module for extracting features from the amplitude ratio spectrum and phase difference spectrum based on a preset time sequence to obtain a corrected growth rate; a first calculation module for acquiring a process recipe change sequence and calculating the process recipe change sequence according to a preset thermal inertia equation and a preset growth kinetic relationship to obtain a feedforward compensation power; a second calculation module for calculating the corrected growth rate according to a preset multivariable PID control algorithm to obtain a feedback regulation power; and a solution module for solving the feedforward compensation power and the feedback regulation power according to a preset decoupling formula to obtain a wafer intrinsic temperature power control command.

[0013] Furthermore, a wafer intrinsic temperature control device is provided, the wafer intrinsic temperature control device comprising: a memory and at least one processor, the memory storing instructions; at least one processor invokes the instructions in the memory to cause the computer device to execute the various steps of the wafer intrinsic temperature control method described above.

[0014] In the technical solution of this invention, amplitude ratio spectrum and phase difference spectrum are obtained by relying on a multi-wavelength ellipsometry and the corrected growth rate is extracted to provide high-precision data support for process control; the feedforward compensation power is calculated by combining the process formula sequence with the thermal inertia equation and the growth kinetic equation to offset temperature and rate disturbances in advance; feedback adjustment power is generated based on a multivariable PID algorithm to correct growth deviations in real time; the feedforward compensation power and feedback adjustment power are calculated by decoupling formula to reduce temperature coupling interference and achieve independent temperature control in different zones; the whole mechanism not only improves the dynamic response and anti-interference capability of the system, but also comprehensively ensures the uniformity of epitaxial layer thickness, the consistency of composition, and the stability of crystal quality, enhancing the consistency and reliability of MOCVD process, and laying a solid core technical support for the large-scale production of high-performance semiconductor devices. Attached Figure Description

[0015] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a first flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 2 This is a second flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 3 This is a third flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 4 This is a fourth flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 5 A fifth flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 6 A sixth flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 7 A seventh flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 8 The eighth flowchart of a wafer intrinsic temperature control method provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a wafer intrinsic temperature control device provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of a wafer intrinsic temperature control device provided in an embodiment of the present invention. Detailed Implementation

[0016] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of a wafer intrinsic temperature control method according to the present invention includes: 101. Obtain the amplitude ratio spectrum and phase difference spectrum; In this embodiment, the multi-wavelength elliptic polarimeter unit serves as an auxiliary calibration unit. It employs a broadband light source with a wavelength range of, for example, 250-850 nm, and is incident on the wafer surface at a fixed angle. This auxiliary calibration unit directly obtains the amplitude ratio spectrum Ψ(λ,t) and phase difference spectrum Δ(λ,t) as a function of wavelength λ and time t by measuring the change in polarization state of polarized light after reflection from the sample. 102. Based on a preset time sequence, feature extraction is performed on the amplitude ratio spectrum and phase difference spectrum to obtain the corrected growth rate; In this embodiment, based on the preset iterative optimization algorithm and time sequence, feature extraction is performed on the amplitude ratio spectrum and phase difference spectrum to accurately obtain the corrected growth rate. This provides high-precision and high-timeliness growth state data support for the MOCVD process, helps to calibrate growth deviations in real time, effectively ensures the uniformity of epitaxial layer thickness and composition consistency, improves process stability and repeatability, and lays a solid core technology foundation for the large-scale production of high-quality semiconductor devices. 103. Obtain the process formula change sequence, and calculate the process formula change sequence according to the preset thermal inertia equation and the preset growth kinetic relationship to obtain the feedforward compensation power. In this embodiment, by acquiring the process formula change sequence and combining the preset thermal inertia equation with the growth kinetics relationship, the feedforward compensation power is calculated. This can predict the temperature and growth rate disturbances caused by formula changes in advance, offset the interference, improve the dynamic response and anti-interference capability of the MOCVD system, ensure the stability of the epitaxial layer growth environment, lay a solid foundation for the consistency of crystal quality, composition and thickness, and help the production of high-performance semiconductor devices. 104. Calculate the corrected growth rate according to the preset multivariable PID control algorithm to obtain the feedback adjustment power; In this embodiment, based on the multivariable PID control algorithm, the growth rate is accurately calculated to generate feedback adjustment power. This method can respond to changes in growth rate in real time, accurately correct deviations, effectively ensure the uniformity of epitaxial layer thickness and growth stability, improve the consistency and reliability of MOCVD process, avoid device defects caused by rate fluctuations, and build a solid core support for the production of high-quality semiconductor devices. 105. The feedforward compensation power and feedback regulation power are calculated according to the preset decoupling formula to obtain the wafer intrinsic temperature power control command. In this embodiment, the feedforward compensation power and feedback adjustment power are calculated by decoupling formula to accurately generate wafer intrinsic temperature power control commands. This effectively counteracts temperature zone coupling interference, realizes independent and accurate temperature control of each heating zone, improves the uniformity and stability of temperature control of MOCVD equipment, lays a solid foundation for ensuring the quality and compositional consistency of epitaxial layer crystals, and helps the production of high-performance semiconductor devices. In this embodiment, amplitude ratio and phase difference spectra are obtained using a multi-wavelength ellipsometry, and the corrected growth rate is extracted to provide high-precision data support for process control. The feedforward compensation power is calculated by combining the process formulation sequence with the thermal inertia equation and growth kinetic equation to preemptively offset temperature and rate disturbances. Feedback adjustment power is generated based on a multivariable PID algorithm to correct growth deviations in real time. The feedforward compensation power and feedback adjustment power are calculated using decoupling formulas to reduce temperature coupling interference and achieve independent temperature control in different zones. This entire mechanism not only improves the system's dynamic response and anti-interference capabilities but also comprehensively ensures uniform epitaxial layer thickness, consistent composition, and stable crystal quality, enhancing the consistency and reliability of the MOCVD process and providing solid core technology support for the large-scale production of high-performance semiconductor devices.

[0018] Please see Figure 2 In a second embodiment of the wafer intrinsic temperature control method of the present invention, step 102 specifically includes: 201. Solve the amplitude ratio spectrum and phase difference spectrum based on the preset layered optical model to obtain multiple model parameters; In this embodiment, in the layered optical model, the complex refractive index of the epitaxial layer serves as a key intermediate variable connecting the alloy composition and the optical response, and is determined by a pre-stored functional relationship (n,k)=f(λ,x). In the calculation, n is the refractive index (the real part of the complex refractive index), which characterizes the change in phase velocity of light as it propagates in the material; k is the extinction coefficient (the imaginary part of the complex refractive index), which characterizes the absorption intensity of light by the material; the functional relationship f is obtained through prior calibration and stored in the system, and λ is the laser wavelength; at each sampling time point t, the amplitude ratio spectrum Ψ(λ,t) and phase difference spectrum Δ(λ,t) are used as fitting targets, and the parameters to be determined in the model (including epitaxial layer thickness d(t), alloy composition x(t), surface roughness parameters, etc.) are adjusted using an iterative optimization algorithm (such as the Levenberg-Marquardt algorithm). In each iteration, the system calculates the corresponding complex refractive index (n,k) through the functional relationship f based on the currently assumed composition x value, and then combines it with other model parameters in the iteration, including epitaxial layer thickness d, alloy composition x(t) and surface roughness parameters, etc., and finally calculates the theoretical spectrum based on these model parameters. The goal of the iterative optimization algorithm is to minimize the difference between the theoretical spectrum and the measured spectrum. 202. Sort multiple model parameters according to time order to obtain characteristic time series signals; In this embodiment, these model parameters are arranged in chronological order to form a high-precision static and compositional characteristic time series signal (for example, the sequence {x(t1),x(t2),...,x(tn)} is the time series signal of alloy component x). 203. Acquire the reflected light intensity signal and analyze the reflected light intensity signal to obtain the oscillation period of the reflected light intensity; In this embodiment, a high spatiotemporal resolution in-situ optical monitoring module is integrated into the MOCVD reaction chamber, located above or to the side of the substrate, ensuring that the probe spot can cover the area of ​​the wafer under test. Based on the principle of optical thin film interference, when a monochromatic laser is incident perpendicularly on the surface of the growing epitaxial layer, it will be reflected at the air-epitaxial layer interface and the epitaxial layer-substrate interface, and the two reflected beams will interfere. The laser reflectivity interferometer is the core monitoring unit. A laser source with a stable wavelength (e.g., λ=635nm) is incident perpendicularly on the center of the wafer and multiple radial feature points. A high-speed photodetector receives the reflected light intensity signal I(t) (i=1,2,3...), and performs time-domain or frequency-domain analysis on the reflected light intensity signal I(t) to measure the oscillation period T of the reflected light intensity (i.e., the time from one maximum value to the next maximum value). The reflected light intensity signal is the main signal source for dynamic temperature feedback. The relationship between the reflected light intensity signal I(t) and time t can be approximated as: In the formula, For the incident light intensity, The first interface reflection coefficient, Let be the reflection coefficient of the second interface, n be the refractive index of the epitaxial layer (known), and d(t) be the epitaxial layer thickness as a function of time. For fixed phase offset; 204. Calculate the oscillation period of the reflected light intensity according to the preset growth rate calculation formula to obtain the growth rate; In this embodiment, within one reflected light intensity oscillation period T, the optical thickness of the thin film increases by λ / (2n). The growth rate is calculated by the formula GR=λ / 2nT, where GR is the growth rate. This formula is the core basis for real-time and continuous calculation of the growth rate. 205. Correct the growth rate based on the characteristic time series signal to obtain the corrected growth rate; In this embodiment, the characteristic time-series signals such as epitaxial layer thickness d(t), alloy composition x(t), and roughness parameters obtained by the elliptic polarizer unit are used to periodically calibrate and correct the growth rate GR and surface morphology trend obtained by the laser reflectivity interferometer unit. In this embodiment, the amplitude ratio spectrum and phase difference spectrum are solved based on the iterative optimization algorithm and the layered optical model. Combined with the pre-stored functional relationship to associate the alloy composition and complex refractive index, the model parameters such as epitaxial layer thickness and alloy composition are accurately calculated and characteristic time-series signals are formed. At the same time, relying on the laser reflectivity interferometer of the in-situ optical monitoring module, the reflected light intensity signal is captured and the oscillation period of the reflected light intensity is analyzed. The growth rate is calculated in real time through the core growth rate calculation formula. Finally, the growth rate is periodically calibrated and corrected using the characteristic time-series signals to improve the accuracy of growth rate measurement and the reliability of surface morphology monitoring. This provides highly reliable data support for subsequent temperature inversion and precise temperature control, effectively ensuring the uniformity of epitaxial layer thickness and composition consistency, strengthening the stability and repeatability of MOCVD process, and laying a solid core foundation for the production of high-quality semiconductor devices.

[0019] Please see Figure 3 In a third embodiment of the wafer intrinsic temperature control method of the present invention, step 103 specifically includes: 301. Calculate the process formulation change sequence based on the thermal inertia equation to obtain the wafer temperature perturbation value; In this embodiment, the process formulation change sequence is a time series consisting of a series of predetermined process parameter adjustment events, including step or ramp changes in parameters such as precursor flow rate and reaction chamber pressure; the expression for the thermal inertia equation is: Using the process formulation change sequence ΔF(t) as input, the predicted temperature disturbance value ΔT is obtained by solving the first-order differential equation, where τ is the thermal time constant. For gain coefficient and The delay coefficient (a model parameter determined in advance through system identification). By calculating the process formula change sequence using the thermal inertia equation, the wafer temperature disturbance value can be accurately predicted. Based on this, compensation strategies can be formulated in advance to offset the temperature fluctuations caused by formula adjustments, improve the anti-interference capability and dynamic response speed of the MOCVD temperature control system, ensure the stability of the epitaxial layer growth environment, lay a solid foundation for crystal quality and composition consistency, and help the production of high-performance semiconductor devices. 302. Calculate the wafer temperature perturbation value based on the growth kinetics relationship to obtain the growth rate perturbation value; In this embodiment, the expression for the growth kinetic relationship is: In the formula, To obtain the predicted growth rate perturbation value through linear mapping; The temperature sensitivity coefficient is derived from the linearization of the Arrhenius equation at the operating point. These are the delay dynamic parameters; Based on the growth kinetics, the growth rate perturbation value can be accurately obtained by calculating the process formula change sequence. Based on this, control strategies can be planned in advance to offset the growth rate fluctuations caused by formula changes, enhance the anti-interference ability and stability of MOCVD process, ensure the uniformity of epitaxial layer thickness, and provide key support for improving the production yield and performance of semiconductor devices. 303. Generate feedforward compensation power based on wafer temperature perturbation and growth rate perturbation values; In this embodiment, the feedforward compensation power is applied to the heating system in advance to actively cancel the disturbance, improve the dynamic response performance and anti-interference capability of the system, and achieve true advance adjustment; In this embodiment, the process formulation change sequence is calculated based on the thermal inertia equation to accurately predict wafer temperature disturbance values ​​and formulate compensation strategies in advance to offset temperature fluctuations. Based on the growth kinetics relationship, the growth rate disturbance value is calculated through the process formulation change sequence, and a control strategy is planned to suppress growth rate fluctuations. The two types of disturbance values ​​are combined to generate feedforward compensation power and apply it to the heating system in advance to achieve true advance regulation. This offsets the temperature and growth rate disturbances caused by formulation adjustments, enhances the system's dynamic response performance and anti-interference capability, and ensures the stability of the epitaxial layer growth environment, uniform thickness, and consistency of crystal quality and composition. This provides key support for improving the yield and performance of semiconductor devices and helps to achieve efficient and stable production of high-performance semiconductor devices.

[0020] Please see Figure 4 In the fourth embodiment of the wafer intrinsic temperature control method of the present invention, step 104 specifically includes: 401. Perform a deviation analysis between the modified growth rate and the preset target growth rate to obtain the growth rate deviation; In this embodiment, by analyzing the deviation between the corrected growth rate and the target growth rate, the deviation of the epitaxial growth rate is accurately captured; using this as the core input, it provides a precise basis for subsequent multivariable PID control, helps to correct growth deviation in real time, ensures the uniformity of epitaxial layer thickness and growth stability, improves the consistency and reliability of MOCVD process, and lays a solid foundation for high-quality semiconductor device production. 402. Obtain the alloy composition from the characteristic time-series signal; 403. Perform deviation analysis on the alloy composition and the preset target alloy composition to obtain the alloy composition deviation; In this embodiment, by analyzing the deviation between the alloy composition and the target alloy composition, the deviation state of the composition during epitaxial growth is accurately captured; this deviation data provides a key input for multivariable PID control, helps to correct the composition deviation in real time, effectively ensures the consistency of the epitaxial layer composition, avoids device performance defects caused by composition fluctuations, improves the stability of MOCVD process and product yield, and provides core support for the production of high-performance semiconductor devices. 406. The growth rate deviation and alloy composition deviation are calculated based on the multivariable PID control algorithm to obtain the preliminary adjustment output; 407. Process the preliminary regulation output according to the preset dynamic decoupling network to obtain the feedback regulation power; In this embodiment, in each control cycle, based on the growth rate deviation and alloy composition deviation, the standard calculation of the multivariable PID control algorithm is performed using PID parameters dynamically optimized by reinforcement learning to generate a preliminary regulation output. This output is further processed by a simultaneously optimized dynamic decoupling network to actively cancel the coupling interference between multiple loops, and finally generate the feedback regulation power ΔPfb acting on each heating zone. In this embodiment, the deviation between the growth rate and the target growth rate is first analyzed and corrected to accurately capture the deviation. Simultaneously, the alloy composition is extracted from the characteristic time-series signal and compared with the target composition for deviation analysis. The deviation data provides a precise basis for subsequent control. Based on the PID parameters dynamically optimized by reinforcement learning, combined with the multivariable PID control algorithm and dynamic decoupling network, the two types of deviations are calculated in real time. A preliminary adjustment output is first generated, and then the multi-loop coupling interference is canceled by the dynamic decoupling network. Finally, the feedback adjustment power of each heating zone is output. The whole mechanism not only corrects the deviation between the growth rate and the alloy composition in real time, ensuring the uniformity of the epitaxial layer thickness and the consistency of the composition, and avoiding device performance defects, but also enhances the system's anti-interference capability, improves the stability, consistency and product yield of the MOCVD process, and lays a solid core support for the stable production of high-quality and high-performance semiconductor devices.

[0021] Please see Figure 5 The fifth embodiment of a wafer intrinsic temperature control method according to the present invention includes step 105 specifically comprising: 501. Summing the feedforward compensation power and the feedback regulation power yields the power command vector; In this embodiment, a power command vector is obtained by summing the feedforward compensation power and the feedback adjustment power. Combined with relevant decoupling and control mechanisms, precise temperature control of the MOCVD equipment is achieved. This not only offsets process disturbances in advance but also corrects growth deviations in real time, effectively suppressing temperature coupling interference, improving the dynamic response and anti-interference capability of the temperature control system, and providing core assurance for the stability of epitaxial layer quality. 502. Based on a preset lightweight neural network, the corrected growth rate and reflected light intensity signals are predicted to obtain the intrinsic temperature of the wafer. In this embodiment, a lightweight neural network is used to fuse the corrected growth rate and reflected light intensity signals to predict the intrinsic temperature of the wafer, accurately capture the actual growth temperature state of the wafer, improve the real-time performance and accuracy of the MOCVD temperature control system, and help ensure the quality and compositional consistency of the epitaxial layer crystal, thus facilitating the efficient production of high-performance semiconductor devices. 503. Analyze the power command vector and wafer intrinsic temperature according to the preset recursive least squares algorithm to obtain the thermal coupling matrix; In this embodiment, a system identification algorithm such as the online recursive least squares algorithm is used to analyze the dynamic relationship between the power command vector U(t) and the intrinsic temperature of the wafer Ts(t) in real time, thereby identifying and continuously updating a thermal coupling matrix C(t), which quantitatively describes the intensity of the interaction between the power changes of each heating zone and the temperature of itself and other regions. 504. Solve the thermal coupling matrix according to the decoupling formula to obtain the wafer intrinsic temperature power control command; In this embodiment, the decoupling formula uses matrix inversion to construct a compensator, which improves the performance of the compensated system. This approximately cancels out the original coupling effect, enabling dynamic decoupling of each control loop. The expression for the decoupling formula is as follows: In the formula, Let I be the inverse of the thermal coupling matrix C(t) that varies with time t, and let I be the identity matrix. The intrinsic temperature power control command for the wafer is to send the decoupled power command Ucomp(t) to each zone heater, thereby effectively suppressing mutual interference between temperature zones and laying the foundation for high-precision and high-uniformity independent temperature control. In this embodiment, the power command vector is obtained by summing the feedforward compensation power and the feedback adjustment power to offset process disturbances in advance. A lightweight neural network is used to fuse and correct the growth rate and reflected light intensity signals, accurately predicting the intrinsic temperature of the wafer and improving the real-time performance and accuracy of temperature control. An online recursive least squares algorithm is used to analyze the power command vector and the intrinsic temperature of the wafer, dynamically updating the thermal coupling matrix. The thermal coupling matrix inversion operation, combined with the decoupling formula, achieves dynamic decoupling of each control loop. This entire mechanism effectively suppresses temperature-zone coupling interference, enhances the system's dynamic response and anti-interference capabilities, lays the foundation for high-precision, high-uniformity independent temperature control, comprehensively ensures the quality and compositional consistency of the epitaxial layer crystal, and facilitates the efficient and stable production of high-performance semiconductor devices.

[0022] Please see Figure 6In the sixth embodiment of a wafer intrinsic temperature control method of the present invention, step 502 specifically includes: 601. Extract features from the reflected light intensity signal to obtain surface morphology trend features; 602. Correct the surface morphology trend features based on the characteristic time series signal to obtain the corrected trend features; In this embodiment, the surface morphology trend characteristics are corrected by characteristic time-series signals to improve the surface condition monitoring accuracy and provide accurate data support for wafer intrinsic temperature inversion; 603. Based on the lightweight neural network, the corrected trend features and the oscillation period of reflected light intensity are predicted to obtain the temperature prediction value; In this embodiment, the correction trend feature is associated with the change in wafer surface morphology, and the oscillation period of reflected light intensity corresponds to the epitaxial layer growth rate. Both are closely related to temperature. The lightweight neural network balances computational efficiency and prediction accuracy, meets the requirements of real-time temperature control, provides reliable data support for the accurate calculation of the intrinsic temperature of the wafer, and helps to improve the overall performance of the temperature control system of MOCVD equipment. 604. Calculate the corrected growth rate and temperature prediction values ​​to obtain the intrinsic temperature of the wafer; In this embodiment, the calculated intrinsic temperature of the wafer is used as the main feedback signal input to the multivariable controller to drive the heater actuator. This method replaces the traditional indirect temperature control mode, improves the real-time performance and accuracy of the MOCVD temperature control system, and lays a core foundation for ensuring the quality and compositional consistency of the epitaxial layer crystal. In this embodiment, surface morphology trend features are first extracted from the reflected light intensity signal, and then corrected using the characteristic time-series signal to improve the surface state monitoring accuracy and provide highly reliable data support for wafer intrinsic temperature inversion. Subsequently, relying on a lightweight neural network, the corrected trend features and the reflected light intensity oscillation period are fused to predict the temperature, which balances computational efficiency and prediction accuracy, and makes full use of the close correlation between morphology changes, growth rate and temperature to meet real-time temperature control requirements. Finally, the wafer intrinsic temperature is obtained by fusing the corrected growth rate and temperature prediction values, which is used as the main feedback signal to drive the heater, improving the real-time performance and accuracy of the MOCVD temperature control system. This lays a core foundation for ensuring the quality and compositional consistency of the epitaxial layer crystal and effectively contributes to the high-quality production of large-size, high-performance semiconductor devices.

[0023] Please see Figure 7 In the seventh embodiment of a wafer intrinsic temperature control method of the present invention, step 603 specifically includes: 701. Obtain historical parameter data pairs and pre-train the lightweight neural network based on the historical parameter data pairs to obtain a pre-trained neural network model; In this embodiment, a lightweight neural network (such as a Long Short-Term Memory network LSTM) is used to run multiple processes under known and uniform temperature conditions during the initial calibration phase of the system, collecting massive amounts of historical parameter data pairs, such as {optical signal features, traditional temperature measurement reference values, and process parameters}. This lightweight neural network is then pre-trained to learn the complex mapping relationship from complex optical features to temperature. 702. Obtain the oscillation amplitude characteristics from the corrected trend characteristics; In this embodiment, by extracting oscillation amplitude features from the correction trend features, key parameters strongly correlated with wafer surface morphology and temperature changes are obtained, providing accurate input for subsequent temperature prediction and improving the model prediction accuracy; 703. Based on the pre-trained neural network model, the oscillation amplitude characteristics, characteristic time series signals, and oscillation period of reflected light intensity are predicted to obtain the temperature prediction value; In this embodiment, based on a pre-trained neural network model, temperature prediction is performed by fusing oscillation amplitude features, characteristic time-series signals, and the oscillation period of reflected light intensity. This fully utilizes the correlation between multi-dimensional optical features and temperature, improving the accuracy of temperature prediction and providing reliable data support for subsequent wafer intrinsic temperature fusion calculations. In this embodiment, the pre-training stage relies on various processes under known uniform temperature conditions to collect massive amounts of optical signal features, traditional temperature measurement reference values, and process parameter data pairs. This allows the LSTM network to learn the precise mapping relationship between complex optical features and temperature, laying a solid foundation for subsequent predictions. By extracting oscillation amplitude features strongly correlated with wafer surface morphology and temperature changes from the correction trend features, high-value input parameters are provided for temperature prediction. Then, based on the pre-trained model, predictions are made by fusing oscillation amplitude features, feature time-series signals, and the oscillation period of reflected light intensity. This fully explores the intrinsic correlation between multi-dimensional features and temperature, improves the accuracy of temperature prediction, avoids the fundamental errors of traditional temperature measurement methods, and provides reliable data support for the subsequent weighted fusion of wafer intrinsic temperatures and model self-calibration, enhancing the robustness and adaptability of the temperature control system. Please see Figure 8 In the eighth embodiment of the wafer intrinsic temperature control method of the present invention, step 604 specifically includes: 801. Obtain the current process parameters, and solve for the current process parameters, corrected growth rate and alloy composition to obtain a preliminary estimated temperature; In this embodiment, during real-time operation, the physical driving layer obtains a preliminary estimated temperature by solving an inverse problem based on the current process parameters (P, flow rate, etc.) and the measured corrected growth rate and alloy composition. This method ensures that the temperature estimation closely matches the actual state of the current process, providing basic data for subsequent temperature fusion that is adapted to the process scenario, improving the process matching degree of wafer intrinsic temperature calculation, and helping to achieve the accuracy of MOCVD temperature control. 802. Perform a deviation analysis on the preliminary estimated temperature and the predicted temperature to obtain the temperature deviation value; In this embodiment, by performing a deviation analysis between the preliminary estimated temperature and the predicted temperature value, the difference between the two is quantified, providing a basis for subsequent temperature fusion or model correction, thereby improving the reliability and accuracy of wafer intrinsic temperature calculation. 803. Determine whether the temperature deviation value is less than the preset tolerance threshold; 804. When the temperature deviation is less than the tolerance threshold, the preliminary estimated temperature and the temperature prediction value are weighted and calculated to obtain the intrinsic temperature of the wafer. In this embodiment, when the temperature deviation value is less than the tolerance threshold, it is determined to be a high-confidence state, and weighted fusion is performed according to dynamic weights to output the intrinsic temperature of the wafer. In the formula, The intrinsic temperature of the wafer. As the first weight, The first and second weights can be dynamically calculated based on the recent errors of the two models (e.g., proportional allocation based on the inverse of the error). This is a predicted temperature value. For initial temperature estimation, a low-confidence alarm is triggered when the temperature deviation is greater than or equal to the tolerance threshold. When the low-confidence alarm is triggered, the system judges the potential cause based on the deviation characteristics. If it is determined that the physical model parameters are inaccurate, the online update process of the physical model parameters is initiated; if it is determined that the data model operating conditions are out of limit, the online incremental learning process of the data model is initiated to achieve model self-correction. MOCVD temperature control is optimized by differentially processing temperature deviation: when the temperature deviation is less than the tolerance threshold, the temperature value is dynamically weighted and fused to ensure high-confidence output of the intrinsic temperature of the wafer; when the temperature deviation is greater than or equal to the tolerance threshold, an alarm is triggered, and the physical model update or data model incremental learning is initiated in a targeted manner, which not only improves the temperature control accuracy but also achieves model self-correction, enhancing the robustness and reliability of the system. In this embodiment, the preliminary estimated temperature obtained by solving an inverse problem, combined with the current process parameters, corrected growth rate, and alloy composition, closely matches the actual state of the current process, providing basic data for process adaptability in temperature fusion. By analyzing the deviation between the preliminary estimated temperature and the predicted temperature, the difference between the two is quantified, providing a precise basis for subsequent processing. Finally, differentiated operations are performed based on the deviation threshold. When the temperature deviation is less than the tolerance threshold, dynamic weighted fusion is used to ensure high confidence in the temperature output. When the temperature deviation is greater than or equal to the tolerance threshold, an alarm is triggered, and targeted updates to the physical model parameters or incremental learning of the data model are initiated to achieve model self-correction. This strategy improves the accuracy and reliability of wafer intrinsic temperature calculation and enhances the system's anti-interference and adaptive capabilities, laying a core foundation for ensuring the quality of epitaxial layer growth.

[0024] The above describes a wafer intrinsic temperature control method according to an embodiment of the present invention. The following describes a wafer intrinsic temperature control device according to an embodiment of the present invention. Please refer to [link / reference]. Figure 9 One embodiment of the wafer intrinsic temperature control device of the present invention includes: Module 1 is used to acquire amplitude ratio spectrum and phase difference spectrum; Feature extraction module 2 is used to extract features from amplitude ratio spectrum and phase difference spectrum based on a preset time sequence to obtain the corrected growth rate; The first calculation module 3 is used to obtain the process formula change sequence and calculate the process formula change sequence according to the preset thermal inertia equation and the preset growth kinetic relationship to obtain the feedforward compensation power. The second calculation module 4 is used to calculate the corrected growth rate according to the preset multivariable PID control algorithm in order to obtain the feedback adjustment power. The calculation module 5 is used to calculate the feedforward compensation power and feedback adjustment power according to the preset decoupling formula to obtain the wafer intrinsic temperature power control command. In this embodiment, amplitude ratio and phase difference spectra are obtained using a multi-wavelength ellipsometry, and the corrected growth rate is extracted to provide high-precision data support for process control. The feedforward compensation power is calculated by combining the process formulation sequence with the thermal inertia equation and growth kinetic equation to preemptively offset temperature and rate disturbances. Feedback adjustment power is generated based on a multivariable PID algorithm to correct growth deviations in real time. The feedforward compensation power and feedback adjustment power are calculated using decoupling formulas to reduce temperature coupling interference and achieve independent temperature control in different zones. This entire mechanism not only improves the system's dynamic response and anti-interference capabilities but also comprehensively ensures uniform epitaxial layer thickness, consistent composition, and stable crystal quality, enhancing the consistency and reliability of the MOCVD process and providing solid core technology support for the large-scale production of high-performance semiconductor devices.

[0025] Figure 10 This is a schematic diagram of a wafer intrinsic temperature control device 900 provided in an embodiment of the present invention. This wafer intrinsic temperature control device 900 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 910 (e.g., one or more processors) and a memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the wafer intrinsic temperature control device 900. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute a series of instruction operations in the storage media 930 on the wafer intrinsic temperature control device 900 to implement the steps of the wafer intrinsic temperature control method provided in the above-described method embodiments.

[0026] A wafer intrinsic temperature control device 900 may further include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, MacOSX, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 10 The illustrated wafer intrinsic temperature control device structure does not constitute a limitation on a wafer intrinsic temperature control device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0027] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0028] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling the intrinsic temperature of a wafer, characterized in that, include: Obtain the amplitude ratio spectrum and phase difference spectrum; Based on a preset time sequence, features are extracted from the amplitude ratio spectrum and phase difference spectrum to obtain the corrected growth rate; The process formulation change sequence is obtained, and the process formulation change sequence is calculated according to the preset thermal inertia equation and the preset growth kinetic relationship to obtain the feedforward compensation power. The modified growth rate is calculated based on a preset multivariable PID control algorithm to obtain the feedback adjustment power; The feedforward compensation power and feedback regulation power are calculated according to the preset decoupling formula to obtain the wafer intrinsic temperature power control command.

2. The wafer intrinsic temperature control method as described in claim 1, characterized in that, The step of extracting features from the amplitude ratio spectrum and phase difference spectrum based on a preset time sequence to obtain the corrected growth rate includes: The amplitude ratio spectrum and phase difference spectrum are solved based on a pre-defined layered optical model to obtain multiple model parameters; Multiple model parameters are sorted according to time sequence to obtain characteristic time-series signals; The reflected light intensity signal is acquired and analyzed to obtain the oscillation period of the reflected light intensity. The growth rate is obtained by calculating the oscillation period of the reflected light intensity according to the preset growth rate calculation formula. The growth rate is corrected based on the characteristic time-series signal to obtain the corrected growth rate.

3. The wafer intrinsic temperature control method as described in claim 1, characterized in that, The step of calculating the process formulation change sequence based on a preset thermal inertia equation and a preset growth kinetic relationship to obtain the feedforward compensation power includes: The wafer temperature perturbation value is obtained by calculating the process formulation change sequence based on the thermal inertia equation. The wafer temperature perturbation value is calculated based on the growth kinetics relationship to obtain the growth rate perturbation value; Feedforward compensation power is generated based on wafer temperature perturbation and growth rate perturbation values.

4. The wafer intrinsic temperature control method as described in claim 2, characterized in that, The step of calculating the corrected growth rate according to a preset multivariable PID control algorithm to obtain the feedback adjustment power includes: A deviation analysis was performed between the corrected growth rate and the preset target growth rate to obtain the growth rate deviation. The alloy composition is obtained from the characteristic time-series signal; Deviation analysis is performed on the alloy composition and the preset target alloy composition to obtain the alloy composition deviation; The growth rate deviation and alloy composition deviation are calculated based on the multivariable PID control algorithm to obtain the preliminary adjustment output; The initial regulation output is processed according to the preset dynamic decoupling network to obtain the feedback regulation power.

5. The wafer intrinsic temperature control method as described in claim 4, characterized in that, The step of calculating the feedforward compensation power and feedback adjustment power according to a preset decoupling formula to obtain the wafer intrinsic temperature power control command includes: The feedforward compensation power and the feedback regulation power are summed to obtain the power command vector; The wafer intrinsic temperature is obtained by predicting the corrected growth rate and reflected light intensity signals based on a preset lightweight neural network. The power command vector and wafer intrinsic temperature are analyzed using a pre-defined recursive least squares algorithm to obtain the thermal coupling matrix. The thermal coupling matrix is ​​solved according to the decoupling formula to obtain the wafer intrinsic temperature power control command.

6. The wafer intrinsic temperature control method as described in claim 5, characterized in that, The step of predicting the corrected growth rate and reflected light intensity signals based on a preset lightweight neural network to obtain the intrinsic temperature of the wafer includes: Feature extraction is performed on the reflected light intensity signal to obtain surface morphology trend features; The surface morphology trend features are corrected based on the characteristic time series signals to obtain the corrected trend features; The temperature prediction value is obtained by predicting the corrected trend characteristics and the oscillation period of reflected light intensity using a lightweight neural network. The corrected growth rate and temperature predictions are calculated to obtain the intrinsic temperature of the wafer.

7. The wafer intrinsic temperature control method as described in claim 6, characterized in that, The step of predicting the temperature prediction value based on the modified trend features and the oscillation period of reflected light intensity using a lightweight neural network includes: Obtain historical parameter data pairs and pre-train the lightweight neural network based on the historical parameter data pairs to obtain a pre-trained neural network model; The oscillation amplitude characteristics are obtained from the corrected trend characteristics; The temperature prediction value is obtained by predicting the oscillation amplitude characteristics, characteristic time sequence signals, and oscillation period of reflected light intensity based on the pre-trained neural network model.

8. The wafer intrinsic temperature control method as described in claim 6, characterized in that, The calculation of the corrected growth rate and temperature prediction values ​​to obtain the intrinsic temperature of the wafer includes: Obtain the current process parameters, and solve for the current process parameters, corrected growth rate, and alloy composition to obtain a preliminary estimated temperature; A deviation analysis was performed between the preliminary estimated temperature and the predicted temperature to obtain the temperature deviation value; Determine whether the temperature deviation value is less than the preset tolerance threshold; When the temperature deviation is less than the tolerance threshold, the preliminary estimated temperature and the temperature prediction value are weighted and calculated to obtain the intrinsic temperature of the wafer.

9. A wafer intrinsic temperature control device, characterized in that, include: The acquisition module is used to acquire amplitude ratio spectrum and phase difference spectrum; The feature extraction module is used to extract features from the amplitude ratio spectrum and phase difference spectrum based on a preset time sequence to obtain the corrected growth rate; The first calculation module is used to obtain the process formula change sequence and calculate the process formula change sequence according to the preset thermal inertia equation and the preset growth kinetic relationship to obtain the feedforward compensation power. The second calculation module is used to calculate the corrected growth rate according to the preset multivariable PID control algorithm in order to obtain the feedback adjustment power. The calculation module is used to calculate the feedforward compensation power and feedback regulation power according to the preset decoupling formula to obtain the wafer intrinsic temperature power control command.

10. A wafer intrinsic temperature control device, characterized in that, include: A memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the wafer intrinsic temperature control device to perform the steps of the wafer intrinsic temperature control method as claimed in any one of claims 1-8.

Images