Optimization method and device of spectral data, computer equipment and storage medium

Abstract

The application provides an optimization method and device of spectral data, a computer device and a storage medium, relates to the technical field of semiconductor detection, and the application obtains original spectral data output by a spectrometer, calculates a center wavelength value of each frame of original spectral data, and queries a corresponding instrument line function from a preset database according to the center wavelength value; then, an inverse convolution algorithm is used to process each frame of original spectral data and the corresponding instrument line function, so that optimized real spectral data are obtained. By using the above scheme, the application can accurately compensate for the system distortion introduced by the spectrometer itself in the full spectral range through algorithm compensation of each frame of spectral data, greatly improves the extraction accuracy of spectral characteristics, and realizes the order-of-magnitude improvement of the measurement accuracy. In addition, the optimized spectral data more truly reflect the physical characteristics of the measured sample, reduce the dependence of the optical model on the empirical parameters, and improve the model generalization ability and prediction reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor detection, and in particular to a spectrum data optimization method and device, a computer device and a storage medium. BACKGROUND

[0002] In the field of semiconductors, epitaxial growth technologies such as metal organic chemical vapor deposition (MOCVD) and molecular beam epitaxy (MBE) are core processes for preparing high-performance optoelectronic and microelectronic devices. From gallium nitride (GaN) based power electronic devices and micro-LEDs to indium phosphide based high-speed lasers, the outstanding performance of the devices is due to the precise controllability of epitaxial materials at the atomic scale.

[0003] In in-situ monitoring of epitaxial growth processes, due to factors such as optical aberration, slit diffraction, and discrete sampling of the detector, the instrument line shape function (ILS) of the spectrometer itself is convolved with the true spectrum, causing the spectrum to be broadened, distorted, and even shifted, resulting in the apparent spectrum measured by the spectrometer not being the true physical spectrum of the sample. Traditional techniques usually directly use the spectrum data output by the spectrometer for fitting analysis, which ignores the distortion effect of the ILS, resulting in low accuracy of the key parameters such as growth rate, thickness and composition inverted out; at the same time, for data such as Fabry-Perot cavity modes and sharp features of steep stop band that need to be accurately measured in VCSEL growth, this instrument effect also introduces a non-negligible system error, severely limiting the further improvement of measurement accuracy. SUMMARY

[0004] The purpose of the present application is to provide a spectrum data optimization method and device, a computer device and a storage medium, to overcome the defect that the instrument line shape function of the spectrometer itself is convolved with the true spectrum in traditional technology, resulting in distorted output spectrum data and limiting the measurement accuracy.

[0005] In a first aspect, the present application provides a spectrum data optimization method, applicable to an in-situ monitoring system of a semiconductor epitaxial growth device, the in-situ monitoring system comprising a spectrometer, the method comprising:

[0006] obtaining original spectrum data output by the spectrometer;

[0007] for each frame of original spectrum data, calculating a central wavelength value thereof;

[0008] according to the central wavelength values, querying corresponding instrument line shape functions from a pre-set database; wherein the instrument line shape functions comprise model parameters, and the model parameters are parameter functions that vary continuously with wavelength, the model parameters comprising amplitude, Gaussian half-width parameter, Lorentz half-width parameter and background offset;

[0009] The deconvolution algorithm is used to process each frame of original spectral data and the corresponding instrument line function to obtain optimized real spectral data, which is used as input into the optical model to obtain the epitaxial growth parameters of the wafer.

[0010] In one of the embodiments, the database is pre-configured in the following manner:

[0011] A plurality of discrete calibration wavelength points are selected within the measurement wavelength range of the spectrometer.

[0012] A standard reference light source is used to measure at each calibration wavelength point to obtain measurement data, wherein the emission linewidth of the standard reference light source at each calibration wavelength point is less than the resolution of the spectrometer.

[0013] The measurement data at each calibration wavelength point is fitted based on the Voigt function model to determine the fitted parameter value of the model parameter of the Voigt function model at the calibration wavelength point.

[0014] An interpolation algorithm is used to continuously interpolate the fitted parameter values of all model parameters to generate a target parameter function that continuously varies with wavelength within the measurement wavelength range.

[0015] The instrument line function that continuously varies with wavelength is generated according to each target parameter function and stored as the database.

[0016] In one of the embodiments, the fitting of the measurement data at each calibration wavelength point based on the Voigt function model to determine the fitted parameter value of the model parameter of the Voigt function model at the calibration wavelength point includes:

[0017] For each calibration wavelength point, the calibration wavelength point and an initial value of the model parameter are taken as input to obtain the output result of the Voigt function model.

[0018] A nonlinear least squares method is used to calculate the fitting difference between the output result and the measurement data.

[0019] An optimization algorithm is used to iteratively adjust the model parameter according to the fitting difference until the fitting difference reaches a preset tolerance range to obtain the fitted parameter value of the model parameter at the calibration wavelength point.

[0020] In one of the embodiments, the deconvolution algorithm is used to process each frame of original spectral data and the corresponding instrument line function to obtain optimized real spectral data, including:

[0021] The deconvolution algorithm is used to deconvolve and reconstruct each frame of original spectral data by solving a convolution equation to remove instrument effects and random noise to obtain optimized real spectral data, and the expression of the convolution equation is:

[0022]

[0023] wherein, represents original spectral data; represents true spectral data; represents a convolution operation; represents an instrument line shape function; represents random noise introduced in the acquisition process.

[0024] In one of the embodiments, in the case that the deconvolution algorithm comprises a Richardson-Lucy algorithm, the employing a deconvolution algorithm, by solving the convolution equation, deconvolves each frame of original spectral data to remove the instrument effect and the random noise, and obtains the optimized true spectral data, comprising:

[0025] Based on the maximum likelihood estimation principle, an iterative optimization method is employed, and the following steps are repeatedly executed until a preset convergence condition is met:

[0026] Convolving the current true spectral estimate with the instrument line shape function to obtain a simulated measured spectrum;

[0027] Dividing the original spectral data by the simulated measured spectrum to obtain a correction coefficient;

[0028] Performing a deconvolution operation on the correction coefficient and the instrument line shape function, and multiplying the operation result by the current true spectral estimate to update the estimate;

[0029] Taking the estimate when the convergence condition is met as the optimized true spectral data.

[0030] In one of the embodiments, in the case that the deconvolution algorithm comprises a Wiener filter algorithm, the employing a deconvolution algorithm, by solving the convolution equation, deconvolves each frame of original spectral data to remove the instrument effect and the random noise, and obtains the optimized true spectral data, comprising:

[0031] Based on the instrument line shape function and the signal-to-noise ratio prior information, a reference filter is constructed in the frequency domain;

[0032] Performing a Fourier transform on each frame of original spectral data, and multiplying the transform result with the reference filter to obtain a weighted frequency domain signal to suppress the influence of random noise through the weight;

[0033] Performing an inverse Fourier transform on the frequency domain signal to obtain the optimized true spectral data.

[0034] In one of the embodiments, a central wavelength value of each frame of original spectral data is calculated, comprising:

[0035] a characteristic spectrum line is obtained according to each frame of original spectral data; the characteristic spectrum line is used to represent the corresponding relationship between light intensity and wavelength, and the characteristic spectrum line includes at least one wave peak;

[0036] a weight is assigned to each wave peak according to the light intensity of the wave peak, and a central wavelength value is obtained by weighted average calculation of the wavelengths of all wave peaks according to the weights.

[0037] In a second aspect, the present application provides an optimization device for spectral data, which is arranged in an in-situ monitoring system of a semiconductor epitaxial growth device, and the in-situ monitoring system includes a spectrometer, and the device includes:

[0038] an acquisition module, configured to acquire original spectral data output by the spectrometer;

[0039] a query module, configured to calculate a central wavelength value for each frame of original spectral data, and query a corresponding instrument line function from a preset database according to the central wavelength value; the instrument line function includes model parameters, and the model parameters are parameter functions that vary continuously with wavelength, and the model parameters include amplitude, Gaussian half-width parameter, Lorentz half-width parameter and background offset;

[0040] an optimization module, configured to process each frame of original spectral data and the corresponding instrument line function by using a deconvolution algorithm to obtain optimized real spectral data, and the real spectral data is used as input into an optical model to obtain epitaxial growth parameters of a wafer by inversion.

[0041] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the method steps in the first aspect when executing the computer program.

[0042] In a fourth aspect, the present application further provides a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method steps in the first aspect.

[0043] The above-mentioned spectral data optimization method and device, computer device and storage medium have at least the following advantages:

[0044] The application obtains original spectral data output by a spectrometer, calculates a central wavelength value of each frame of original spectral data, and queries a corresponding instrument line function from a preset database according to the central wavelength value; and then adopts a deconvolution algorithm to process each frame of original spectral data and the corresponding instrument line function, to obtain optimized real spectral data. By adopting the above scheme, the application can accurately compensate for system distortion introduced by the spectrometer itself in the full spectral range through algorithm compensation of each frame of spectral data, greatly improves the extraction accuracy of spectral characteristics, and realizes a magnitude improvement in measurement accuracy. In addition, the optimized spectral data more truly reflects the physical characteristics of the measured sample, reduces the dependence of the optical model on empirical parameters, and improves the model generalization ability and prediction reliability. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 An application environment diagram of the spectral data optimization method in an embodiment;

[0046] Figure 2 A flowchart of the spectral data optimization method in an embodiment;

[0047] Figure 3 A flowchart of the pre-set database step in an embodiment;

[0048] Figure 4 A structural block diagram of the spectral data optimization device in an embodiment;

[0049] Figure 5 An internal structure diagram of the computer device in an embodiment. DETAILED DESCRIPTION

[0050] The embodiments of the application will be described in detail below with reference to the drawings. Those skilled in the art can easily understand other advantages and effects of the application from the content disclosed in the specification. The application can also be implemented or applied in other different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0051] Some exemplary embodiments of the application are described for illustrative purposes, and it should be understood that the application can be implemented in other ways not specifically shown in the drawings.

[0052] The spectral data optimization method provided by the embodiments of the application can be applied in an application environment as shown in Figure 1 The reaction chamber 1 is provided with a pedestal for placing a wafer, the pedestal is made of high-purity graphite, is supported by a rotating shaft driven by a motor, and can be continuously rotated during epitaxial growth of the wafer.

[0053] Semiconductor epitaxial growth equipment refers to a process equipment for epitaxial growth of thin films or multilayer structures on a substrate, usually including metal organic chemical vapor deposition (MOCVD), hydride vapor phase epitaxy (HVPE), molecular beam epitaxy (MBE) and various chemical vapor deposition (CVD).

[0054] The in-situ monitoring system 2 refers to a measurement sub-system for real-time measurement and data processing of the epitaxial growth state of the wafer in the reaction chamber without destroying the process environment and interrupting the growth.

[0055] The in-situ monitoring system 2 is arranged above the reaction chamber 1 and includes a laser light source, a spectrometer and an optimized device for spectral data.

[0056] The laser light source is used to output laser signals to the wafer surface during epitaxial growth.

[0057] The spectrometer is used to collect the laser signals reflected by the wafer surface and output spectral data representing the correspondence between wavelength and intensity.

[0058] Specifically, the spectrometer captures the light intensity information of different wavelengths by analyzing the laser signals reflected by the wafer surface. The wavelength information of these reflected light signals can reveal the material properties and the state of the wafer surface during growth, including the thickness, composition and lattice matching of the epitaxial layer. Further, by analyzing the spectral characteristics of these reflected light signals, the epitaxial growth process can be monitored in real time to ensure precise control of the process.

[0059] The optimized device for spectral data is used to obtain the original spectral data output by the spectrometer; for each frame of original spectral data, the central wavelength value is calculated; according to the central wavelength values, the corresponding instrument line function is obtained from the pre-set database; wherein the instrument line function includes model parameters, and the model parameters are parameter functions that vary continuously with wavelength, and the model parameters include amplitude, Gaussian half-width parameter, Lorentz half-width parameter and background offset; using deconvolution algorithm, each frame of original spectral data and its corresponding instrument line function are processed to obtain optimized real spectral data, which is used as input into the optical model to obtain the epitaxial growth parameters of the wafer.

[0060] The optimization method of the above-mentioned spectral data obtains the original spectral data output by the spectrometer, calculates the center wavelength value of each frame of original spectral data, and queries the corresponding instrument line function from the preset database according to the center wavelength value; then, the deconvolution algorithm is used to process each frame of original spectral data and the corresponding instrument line function to obtain the optimized real spectral data. By using the above-mentioned scheme, the present application can accurately compensate the system distortion introduced by the spectrometer itself in the full spectral range through algorithm compensation of each frame of spectral data, greatly improves the extraction accuracy of spectral characteristics, and realizes the order of magnitude improvement of the measurement accuracy. In addition, the optimized spectral data more truly reflects the physical characteristics of the measured sample, reduces the dependence of the optical model on the empirical parameters, and improves the model generalization ability and prediction reliability.

[0061] Please refer to Figure 2 In one exemplary embodiment, the present embodiment provides an optimization method of spectral data, which is suitable for an in-situ monitoring system of a semiconductor epitaxial growth device, and the in-situ monitoring system includes a spectrometer, and specifically includes the following steps:

[0062] Step 202, obtaining original spectral data output by the spectrometer.

[0063] Specifically, the spectrometer collects the reflected light signal of the wafer surface in the epitaxial growth process under the current set slit width and optical path parameters, disperses and separates the reflected light signal, analyzes different wavelength components to form corresponding light intensity distribution signals, converts the light intensity distribution signals at each wavelength into electrical signals, and performs data processing such as filtering, normalization and calibration, to obtain a data set of light intensity change with wavelength synchronized with time, which is the spectral data.

[0064] Step 204, calculating the center wavelength value of each frame of original spectral data.

[0065] Specifically, the center wavelength value is a characteristic wavelength parameter obtained by calculating the light intensity distribution of the reflected light signal in a predetermined wavelength range, which is used to represent the main energy distribution position of the spectrum, and usually corresponds to the center position of the target characteristic spectral line or a group of related peaks on the wavelength axis.

[0066] Step 206, querying the corresponding instrument line function from the preset database according to each center wavelength value; wherein the instrument line function includes model parameters, and the model parameters are parameter functions that change continuously with wavelength, and the model parameters include amplitude, Gaussian half-width parameter, Lorentz half-width parameter and background offset.

[0067] Specifically, the instrument line shape (ILS) refers to a function describing the point spread of a spectrometer in the wavelength domain. An ideal infinitely narrow monochromatic spectral line will be broadened and deformed after passing through the spectrometer, and the broadened shape is the ILS. The measured spectral data is approximately equal to the convolution of the true spectral data and the ILS. In the growth of a VCSEL (vertical cavity surface emitting laser) device, the high-performance distributed Bragg reflector (DBR) and Fabry-Perot (F-P) resonant cavity are extremely sensitive to spectral characteristics. The ILS effect will cause the steepness of the stop band edge to decrease, resulting in the failure of the algorithm for reversing the composition and thickness based on the edge mode slope; at the same time, the resonant mode is broadened and the peak wavelength is shifted. If the distorted apparent spectrum is directly used for fitting and process decision, the growth endpoint of the DBR will be incorrectly judged, resulting in a mismatch between the actual grown cavity mode wavelength and the design target, ultimately causing the threshold current of the device to rise, the slope efficiency to decrease, and the product yield to be damaged; secondly, in the growth of the AlGaN barrier layer of a high electron mobility transistor (HEMT) or a SiGe heterostructure in ultra-thin layer epitaxy and interface atomic level control, its thickness monitoring relies on the weak interference oscillation in the reflection spectrum. The low-pass filtering effect of the ILS will significantly attenuate the amplitude of these high-frequency oscillation signals, making it extremely difficult to determine the growth endpoint in real time and accurately, and it is easy to cause layer thickness to be out of tolerance or interface atoms to interdiffuse, directly degrading the two-dimensional electron gas transport characteristics of the HEMT.

[0068] The fundamental limitation of the traditional technology is that, for the purpose of simplifying the model or reducing the calculation complexity, the apparent spectrum directly output by the spectrometer is generally equated to the true physical spectrum, and is directly put into the subsequent physical model such as the transfer matrix model for fitting analysis. This processing method essentially ignores the defects of the instrument's own response in the field of advanced semiconductor manufacturing with single atomic layer precision and sub-nanometer wavelength control. This makes the upper limit of the precision of the physical parameters such as growth rate, thickness, and composition inverted by all cannot meet the stringent requirements of realizing nanoscale uniformity and consistency on a large-size wafer, and has become a key bottleneck restricting the precision of epitaxial process to a new height.

[0069] Based on this, the purpose of the present application is to compensate for the instrument effect introduced by the spectrometer by using the ILS to obtain the true spectral data, wherein the instrument effect includes broadening and deformation.

[0070] A parameter function refers to a function of a parameter varying with wavelength, and the value of the parameter changes accordingly when the center wavelength λ changes; for any given center wavelength λ, the function can give a certain value of the parameter at the wavelength. Take "amplitude A" as an example: at λ = 450 nm, A has a certain value; at λ = 650 nm, A has another certain value. The correspondence between λ and A is represented by the function A(λ). Amplitude A(λ) is used to represent the scaling amount of the overall intensity or energy; Gaussian half-width parameter σ(λ) is used to determine the width and smoothness of the peak core; Lorentz half-width parameter γ(λ) is used to determine the strength of the peak tail wing; background offset B(λ) is used to compensate for stray light, dark current, background scattering or slow baseline.

[0071] It should be understood that, according to the center wavelength value, the corresponding ILS is obtained by querying the value set of the model parameters at the center wavelength, and then uniquely determining and generating the ILS at the wavelength based on the value set.

[0072] In step 208, the deconvolution algorithm is used to process each frame of original spectral data and its corresponding instrument line shape function to obtain optimized real spectral data, which is used as input to the optical model to obtain the epitaxial growth parameters of the wafer.

[0073] Specifically, the deconvolution algorithm refers to a calculation process for removing the instrument effects caused by the ILS and suppressing random noise, which regards the measured original spectral data as the result of the real spectral data after the ILS effect and superimposed noise, and restores the real spectral data closer to the physical intrinsic by iterative optimization or frequency domain filtering to solve the inverse problem.

[0074] The epitaxial growth parameters refer to physical information, material information and geometric quantities used to characterize the epitaxial growth process and results, which are usually used for process monitoring and closed-loop control. The epitaxial growth parameters include at least one of film thickness and thickness uniformity, refractive index, extinction coefficient, doping type, concentration, carrier density, stress, growth temperature, growth rate, DBR related parameters.

[0075] The optical model refers to a physical model that establishes the mapping between the real spectral data and the epitaxial growth parameters, which is used to obtain the epitaxial growth parameters from the real spectrum.

[0076] The optimization method of the above-mentioned spectral data acquires the original spectral data output by the spectrometer, calculates the center wavelength value of each frame of original spectral data, and queries the corresponding instrument line function from the preset database according to the center wavelength value; and then adopts the deconvolution algorithm to process each frame of original spectral data and the corresponding instrument line function to obtain the optimized real spectral data. By adopting the above-mentioned scheme, the present application can accurately compensate the system distortion introduced by the spectrometer itself in the full spectral range through algorithm compensation of each frame of spectral data, greatly improves the extraction accuracy of spectral characteristics, and realizes the order of magnitude improvement of the measurement accuracy. In addition, the optimized spectral data more truly reflects the physical characteristics of the measured sample, reduces the dependence of the optical model on the empirical parameters, and improves the model generalization ability and prediction reliability.

[0077] Please refer to Figure 3 Optionally, the pre-set manner of the database comprises:

[0078] Step 302, a plurality of discrete calibration wavelength points are selected in the measurement wavelength range of the spectrometer.

[0079] Step 304, a standard reference light source is used to measure at each calibration wavelength point to obtain measurement data; wherein the emission linewidth of the standard reference light source at each calibration wavelength point is less than the resolution of the spectrometer.

[0080] Step 306, based on the Voigt function model, the measurement data of each calibration wavelength point is fitted to determine the fitting parameter value of the model parameter of the Voigt function model at the calibration wavelength point.

[0081] Step 308, an interpolation algorithm is adopted to continuously interpolate the fitting parameter values of all model parameters to generate a target parameter function that continuously changes with wavelength in the measurement wavelength range.

[0082] Step 310, according to each target parameter function, an instrument line function that continuously changes with wavelength is generated and stored as a database.

[0083] Specifically, since the measurement wavelength coverage range of the spectrometer is large, usually from several hundred to several thousand nanometers, and the dispersion and aberration characteristics of the optical system are different in different wavelength bands, if only one wavelength is calibrated, it will be severely distorted, therefore, the present application selects several representative center wavelengths as calibration wavelength points. Exemplarily, a plurality of points can be selected from the short wavelength band, the intermediate wavelength band and the long wavelength band as calibration wavelength points.

[0084] A standard reference light source refers to a light source emitting a very narrow spectral line at a known and traceable central wavelength, which is used to calibrate the wavelength axis of the spectrometer and / or characterize the ILS. In particular to the present fact example, the line width of the standard reference light source is narrower than the resolution of the spectrometer, so that the spectral data output by the spectrometer is mainly determined by the instrument effects, which facilitates subsequent fitting of an accurate ILS. Exemplarily, the standard reference light source uses a tunable laser or a monochromator.

[0085] At each calibration wavelength point, the spectral data output by the spectrometer is measured to obtain measurement data.

[0086] Since the Voigt function V(x;σ,γ) is a convolution of the Gaussian function G(x;σ) and the Lorentz function L(x;γ), it can accurately characterize the ILS line shape composed of Gaussian-type effects caused by, for example, optical aberrations, slit diffraction, and Lorentz-type effects caused by, for example, detector response, therefore, the Voigt function is adopted as the reference model in the present embodiment, and its expression is:

[0087] V(x;σ,γ) = ∫G(x';σ) L(x-x';γ)dx'

[0088] wherein x =λ-λ0 represents the offset relative to the central wavelength λ0, σ represents the Gaussian half-width parameter, γ represents the Lorentz half-width parameter, and x' represents all possible relative positions or offsets traversed in the integration process.

[0089] Optionally, based on the Voigt function model, the measurement data of each calibration wavelength point is fitted to determine the fitting parameter value of the model parameter of the Voigt function model at the calibration wavelength point, including:

[0090] For each calibration wavelength point, the calibration wavelength point and the initial value of the model parameter are taken as inputs to obtain the output result of the Voigt function model;

[0091] The non-linear least squares method is used to calculate the fitting difference between the output result and the measurement data;

[0092] An optimization algorithm is used to iteratively adjust the model parameter according to the fitting difference until the fitting difference reaches a preset tolerance range, to obtain the fitting parameter value of the model parameter at the calibration wavelength point.

[0093] Specifically, the model parameter can be represented as p_i = [A_i, σ_i, γ_i, B_i], wherein A_i is the amplitude; σ_i is the Gaussian half-width; γ_i is the Lorentz half-width; and B_i is the background offset. The initial value of the model parameter can be a random value or an empirical value.

[0094] Further, the expression of the instrument line function is:

[0095] ILS_model(λ) = A_i · V(λ - λ_i; σ_i, γ_i) + B_i

[0096] For one calibration wavelength point, input the calibration wavelength point and initial values of model parameters into the expression of the above instrument line function, and obtain the output result; use the nonlinear least squares method to calculate the fitting difference between the output result and the actual measurement data, use the optimization algorithm to optimize the model parameters p_i, and use the numerical value of the optimized model parameters for the next optimization. Repeat the above steps until the fitting difference reaches the tolerance range, and the numerical value of the model parameters p_i at this time is used as the fitting parameter value of the calibration wavelength point, and then the fitting parameter values of all calibration wavelength points are obtained. It should be understood that the fitting parameter values of all calibration wavelength points are a set of discrete (λ_i, p_i) data pairs. Exemplarily, the optimization algorithm of the embodiment of the present application uses Levenberg-Marquardt, and in actual application, other optimization algorithms such as Gauss-Newton, trust region reflection method, Dogleg can also be selected according to the needs.

[0097] Further, in order to achieve the corresponding ILS at any center wavelength λ, the embodiment of the present application further converts the discrete parameter set p_i into a continuous parameter function p(λ) through a difference algorithm. Exemplarily, three spline interpolations are respectively performed on each parameter (σ, γ, A, B) in the model parameter to generate continuous interpolation functions σ(λ), γ(λ), A(λ), B(λ).

[0098] Thus, the complete database is a set of the above interpolation functions, that is, a parameterized ILS model database. For any center wavelength λ_c, the corresponding instrument line function can be given by the following formula:

[0099] ILS(λ_c) = A(λ_c) · V(λ - λ_c; σ(λ_c), γ(λ_c)) + B(λ_c)

[0100] Through the above formula, the high-order function or data structure corresponding to the high-fidelity ILS function can be calculated and returned in real time, which provides core algorithm support for the subsequent deconvolution reconstruction step.

[0101] The above scheme is used to sample multiple points in the measurement wavelength range of the spectrometer and use a standard reference light source with a line width less than the resolution of the instrument. The measured spectral shape is determined by the instrument response, and the sample factor is excluded, so the calibration result has high accuracy. The Voigt model is used for fitting, which can accurately characterize the peak center and tail wing, improve the accuracy of the description of the instrument broadening mechanism, and provide reliable parameters for subsequent compensation. The interpolation algorithm is used to continuously segment the fitting parameters into the target parameter function varying with the wavelength, which can eliminate the discontinuity and boundary artifacts of the segmented splicing, and the ILS at any wavelength can be uniquely determined.

[0102] Optionally, the deconvolution algorithm is used to process each frame of original spectral data and its corresponding instrument line shape function to obtain the optimized real spectral data, including:

[0103] The deconvolution algorithm is used to solve the convolution equation to deconvolve and reconstruct each frame of original spectral data, remove the instrument effect and random noise, and obtain the optimized real spectral data. The expression of the convolution equation is:

[0104]

[0105] wherein, S_measured(λ) represents the original spectral data; S_real(λ) represents the real spectral data; represents a convolution operation; ILS(λ) represents the instrument line shape function; represents random noise introduced in the acquisition process. Specifically, the purpose of optimization is to remove the ILS caused by the instrument effect and the random noise introduced in the spectral data acquisition process from the original spectral data S_measured(λ), and finally reconstruct the high-fidelity physical spectrum S_real(λ). To achieve stable solution of this ill-posed problem, the deconvolution algorithm based on prior constraints is used in the embodiments of the present application. Specifically, the deconvolution algorithm is the Richardson-Lucy algorithm or the Wiener filtering algorithm.

[0106] Optionally, in the case where the deconvolution algorithm includes the Richardson-Lucy algorithm, the deconvolution algorithm is used to solve the convolution equation to deconvolve and reconstruct each frame of original spectral data, remove the instrument effect and random noise, and obtain the optimized real spectral data, including:

[0107] Based on the maximum likelihood estimation principle, an iterative optimization method is used to repeatedly execute the following steps until the preset convergence condition is met:

[0108] The current real spectrum estimation value is convoluted with the instrument line shape function to obtain a simulated measurement spectrum; the original spectrum data is divided by the simulated measurement spectrum to obtain a correction coefficient; the correction coefficient is subjected to an inverse convolution operation with the instrument line shape function, and the operation result is multiplied by the current real spectrum estimation value to update the estimation value; and the estimation value meeting the convergence condition is taken as the optimized real spectrum data.

[0109] Specifically, the Richardson-Lucy algorithm is based on the maximum likelihood estimation principle of Poisson noise statistics, and iteratively approaches the real solution, and the iteration process is as follows:

[0110] First, based on the original spectrum data An initial estimation of the real spectrum is set , wherein represents the real physical spectrum; the superscript 0 represents the starting point of the iteration process, that is, the estimation value at the 0th iteration (or initial state); (λ) indicates that the estimation value is a function related to the center wavelength λ.

[0111] Secondly, the estimation value is updated according to the following iteration formula:

[0112]

[0113] , wherein k is the iteration number, represents the flipped instrument line shape function. The algorithm naturally maintains the non-negativity of the spectrum value in the iteration process, gradually sharpens the spectral characteristics, and meets the preset convergence condition.

[0114] Optionally, in the case that the deconvolution algorithm includes a Wiener filter algorithm, the deconvolution algorithm is adopted, the convolution equation is solved, each frame of original spectrum data is deconvoluted and reconstructed, the instrument effect and random noise are removed, and the optimized real spectrum data is obtained, including:

[0115] Based on the instrument line shape function and the signal-to-noise ratio prior information, a reference filter is constructed in the frequency domain; each frame of original spectrum data is subjected to Fourier transform, and the transformed result is multiplied by the reference filter to obtain a weighted frequency domain signal, so as to suppress the influence of random noise through the weight; the inverse Fourier transform is performed on the frequency domain signal to obtain the optimized real spectrum data.

[0116] Specifically, the Wiener filter algorithm constructs an optimal filter in the frequency domain, and uses the optimal filter to simultaneously realize deconvolution and noise suppression under the minimum mean square error criterion, and the core lies in designing the transfer function of the Wiener filter :

[0117]

[0118] , wherein denotes Fourier transform, denotes complex conjugate, denotes power spectrum, K is a regularization parameter, and the value thereof is related to the ratio of noise power and signal power. The optimized spectral data is obtained by the following formula:

[0119]

[0120] wherein, denotes inverse Fourier transform. By adjusting the parameter K, an optimal balance between spectral fidelity and noise suppression level can be achieved.

[0121] By using the above scheme, after execution by any of the above algorithms, the reconstructed high-fidelity physical spectrum S_real(λ) can be output, which maximally strips the broadening and distortion effects caused by the instrument line function, laying a foundation for accurate inversion of subsequent physical parameters.

[0122] Optionally, the center wavelength value of each frame of original spectral data is calculated, comprising:

[0123] According to each frame of original spectral data, a characteristic spectral line is obtained; the characteristic spectral line is used to represent the corresponding relationship between light intensity and wavelength, and the characteristic spectral line includes at least one peak;

[0124] According to the light intensity of each peak, a weight is assigned to the peak, and the wavelengths of all peaks are weighted and averaged according to the weights to obtain the center wavelength value.

[0125] Specifically, the spectral data is preprocessed by smoothing and denoising and baseline correction to suppress random noise and background drift. Within a set wavelength window, a characteristic spectral line is extracted, peaks are identified based on threshold comparison and morphological conditions, and abnormal peaks are deleted. The threshold comparison rules include minimum peak height, minimum peak spacing and minimum signal-to-noise ratio; the abnormal peaks include saturated and isolated noise peaks. The wavelengths of all peaks are weighted and averaged according to the weights to obtain the center wavelength value.

[0126] By using the above scheme, the preprocessing and threshold screening can retain the reliable peaks in the spectral data. By using the light intensity as the weight of the weighted strategy, the disturbance of the secondary peaks to the result is automatically weakened, so that the center wavelength calculation is more robust to noise and can better represent the true spectral state.

[0127] The optimization method of the above-mentioned spectral data acquires the original spectral data output by the spectrometer, calculates the center wavelength value of each frame of original spectral data, and queries the corresponding instrument line function from the preset database according to the center wavelength value; and then adopts the deconvolution algorithm to process each frame of original spectral data and the corresponding instrument line function to obtain the optimized real spectral data. By adopting the above-mentioned scheme, the present application can accurately compensate the system distortion introduced by the spectrometer itself in the full spectral range through algorithm compensation of each frame of spectral data, greatly improves the extraction accuracy of spectral characteristics, and realizes the order of magnitude improvement of the measurement accuracy. In addition, the optimized spectral data more truly reflects the physical characteristics of the measured sample, reduces the dependence of the optical model on the empirical parameters, and improves the model generalization ability and prediction reliability.

[0128] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0129] Based on the same inventive concept, the present embodiment also provides an optimization device for spectral data. The device is suitable for the above-mentioned optimization method for spectral data. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above-mentioned method. Therefore, the specific limitations in one or more device embodiments provided below can refer to the limitations of the method described above, and will not be repeated here.

[0130] Please refer to Figure 4 In one embodiment, the present embodiment provides an optimization device for spectral data. The device is arranged in an in-situ monitoring system of a semiconductor epitaxial growth device. The in-situ monitoring system includes a spectrometer. The device specifically includes an acquisition module, a query module, and an optimization module.

[0131] The acquisition module is configured to acquire original spectral data output by the spectrometer.

[0132] The query module is configured to calculate a central wavelength value for each frame of original spectral data, and query a corresponding instrument line function from a preset database according to the central wavelength value, wherein the instrument line function includes model parameters, and the model parameters are parameter functions that vary continuously with wavelength, and the model parameters include an amplitude, a Gaussian half-width parameter, a Lorentz half-width parameter, and a background offset.

[0133] The optimization module is configured to process each frame of original spectral data and the corresponding instrument line function by using a deconvolution algorithm to obtain optimized real spectral data, and the real spectral data is used as input into an optical model to obtain epitaxial growth parameters of the wafer.

[0134] Optionally, the spectral data optimization device further includes a database setting module.

[0135] The database setting module is configured to pre-set the database in the following manner: selecting a plurality of discrete calibration wavelength points within a measurement wavelength range of the spectrometer; measuring at each calibration wavelength point by using a standard reference light source to obtain measurement data, wherein the emission linewidth of the standard reference light source at each calibration wavelength point is less than the resolution of the spectrometer; fitting the measurement data at each calibration wavelength point based on a Voigt function model to determine a fitting parameter value of a model parameter of the Voigt function model at the calibration wavelength point; continuously interpolating the fitting parameter values of all model parameters by using an interpolation algorithm to generate a target parameter function that varies continuously with wavelength within the measurement wavelength range; generating an instrument line function that varies continuously with wavelength according to the target parameter functions, and storing the instrument line function as the database.

[0136] Optionally, the database setting module fits the measurement data at each calibration wavelength point based on the Voigt function model to determine a fitting parameter value of a model parameter of the Voigt function model at the calibration wavelength point, including: taking the calibration wavelength point and an initial value of the model parameter as input to obtain an output result of the Voigt function model for each calibration wavelength point; calculating a fitting difference between the output result and the measurement data by using a nonlinear least squares method; and adjusting the model parameter iteratively according to the fitting difference by using an optimization algorithm until the fitting difference reaches a preset tolerance range to obtain the fitting parameter value of the model parameter at the calibration wavelength point.

[0137] Optionally, the optimization module processes each frame of original spectral data and the corresponding instrument line function by using a deconvolution algorithm to obtain optimized real spectral data, including: performing deconvolution reconstruction on each frame of original spectral data by solving a convolution equation by using the deconvolution algorithm to remove instrument effects and random noise to obtain the optimized real spectral data, and the expression of the convolution equation is:

[0138]

[0139] wherein, represents original spectral data; represents true spectral data; represents a convolution operation; represents an instrument line function; represents random noise introduced in the acquisition process.

[0140] Optionally, the optimization module, in the case that the deconvolution algorithm comprises a Richardson-Lucy algorithm, adopts the deconvolution algorithm to perform deconvolution reconstruction on each frame of original spectral data by solving the convolution equation, removes the instrument effect and the random noise, and obtains the optimized true spectral data, comprising: based on a maximum likelihood estimation principle, repeatedly performing the following steps until a preset convergence condition is met by using an iterative optimization method: performing convolution on a current true spectral estimation value and the instrument line function to obtain a simulated measurement spectrum; dividing the original spectral data by the simulated measurement spectrum to obtain a correction coefficient; performing deconvolution operation on the correction coefficient and the instrument line function, and multiplying the operation result by the current true spectral estimation value to update the estimation value; taking the estimation value when the convergence condition is met as the optimized true spectral data.

[0141] Optionally, the optimization module, in the case that the deconvolution algorithm comprises a Wiener filter algorithm, adopts the deconvolution algorithm to perform deconvolution reconstruction on each frame of original spectral data by solving the convolution equation, removes the instrument effect and the random noise, and obtains the optimized true spectral data, comprising: based on the instrument line function and the signal-to-noise ratio prior information, constructing a reference filter in the frequency domain; performing Fourier transform on each frame of original spectral data, and multiplying the transform result by the reference filter to obtain a weighted frequency domain signal to suppress the influence of random noise through the weight; performing inverse Fourier transform on the frequency domain signal to obtain the optimized true spectral data.

[0142] Optionally, the query module calculates a center wavelength value of each frame of original spectral data, comprising: obtaining a characteristic spectral line according to each frame of original spectral data; the characteristic spectral line is used to represent the corresponding relationship between light intensity and wavelength, and the characteristic spectral line comprises at least one wave peak; assigning a weight to each wave peak according to the light intensity of the wave peak, and calculating the center wavelength value by weighted average of the wavelengths of all wave peaks according to the weights.

[0143] The optimization device of the above-mentioned spectral data acquires original spectral data output by a spectrometer, calculates a central wavelength value of each frame of original spectral data, and queries a corresponding instrument line function from a preset database according to the central wavelength value; then, an inverse convolution algorithm is used to process each frame of original spectral data and the corresponding instrument line function, to obtain optimized real spectral data. By using the above scheme, the present application can accurately compensate for the system distortion introduced by the spectrometer itself in the full spectral range through algorithm compensation of each frame of spectral data, greatly improves the extraction accuracy of spectral characteristics, and realizes a magnitude improvement in measurement accuracy. In addition, the optimized spectral data more truly reflects the physical characteristics of the measured sample, reduces the dependence of the optical model on empirical parameters, and improves the model generalization ability and prediction reliability.

[0144] Each module in the above-mentioned optimization device of spectral data can be realized by software, hardware, and a combination thereof, in whole or in part. The above-mentioned modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform the operations corresponding to the above-mentioned modules.

[0145] In a feasible embodiment, a computer device, which can be a terminal, can have an internal structure diagram as shown in Figure 5 The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, mobile cellular network, NFC (near field communication), or other technologies. The computer program is executed by the processor to implement the above-mentioned spectral data optimization method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or can be a key, trackball, or touchpad arranged on the shell of the computer device. It can also be an external keyboard, touchpad, or mouse, etc.

[0146] Those skilled in the art can understand that Figure 5The structure shown in the foregoing is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the foregoing, or combine certain components, or have a different arrangement of components. Figure 5

[0147] In an implementable embodiment, a computer device is provided, including a memory and a processor, the memory having stored therein a computer program, the processor implementing the method steps in the above method for optimizing spectral data when executing the computer program.

[0148] In an implementable embodiment, a computer readable storage medium is provided, having stored thereon a computer program, the computer program implementing the method steps in the above method for optimizing spectral data when executed by a processor.

[0149] In an implementable embodiment, a computer program product is provided, including a computer program, the computer program implementing the method steps in the above method for optimizing spectral data when executed by a processor.

[0150] ​Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0151] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0152] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method of optimizing spectral data, characterized by, An in-situ monitoring system for a semiconductor epitaxial growth device, the in-situ monitoring system comprising a spectrometer, the method comprising: acquiring raw spectral data output by the spectrometer; for each frame of raw spectral data, calculating a central wavelength value thereof; according to each central wavelength value, querying a corresponding instrumental line shape function from a preset database; wherein the instrumental line shape function comprises model parameters, and the model parameters are parameter functions that vary continuously with wavelength, and the model parameters comprise an amplitude, a Gaussian half-width parameter, a Lorentz half-width parameter, and a background offset; using a deconvolution algorithm, processing each frame of raw spectral data and its corresponding instrumental line shape function to obtain optimized real spectral data, the real spectral data being used as input into an optical model to obtain epitaxial growth parameters of a wafer; wherein the database is preset in the following manner: selecting a plurality of discrete calibration wavelength points within a measurement wavelength range of the spectrometer; measuring each calibration wavelength point using a standard reference light source, wherein the standard reference light source has an emission linewidth at each calibration wavelength point that is less than a resolution of the spectrometer; based on a Voigt function model, fitting the measurement data of each calibration wavelength point to determine a fitting parameter value of the model parameter of the Voigt function model at the calibration wavelength point; using an interpolation algorithm, continuously interpolating the fitting parameter values of all model parameters to generate a target parameter function that varies continuously with wavelength within the measurement wavelength range; based on each target parameter function, generating an instrumental line shape function that varies continuously with wavelength, and storing the instrumental line shape function as the database.

2. The method of claim 1, wherein, The fitting of the measurement data of each calibration wavelength point based on the Voigt function model to determine a fitting parameter value of the model parameter of the Voigt function model at the calibration wavelength point comprises: for each calibration wavelength point, taking the calibration wavelength point and an initial value of the model parameter as input to obtain an output result of the Voigt function model; using a non-linear least squares method, calculating a fitting difference between the output result and the measurement data; using an optimization algorithm, iteratively adjusting the model parameter based on the fitting difference until the fitting difference reaches a preset tolerance range, to obtain the fitting parameter value of the model parameter at the calibration wavelength point.

3. The method of claim 1, wherein, The use of a deconvolution algorithm to process each frame of raw spectral data and its corresponding instrumental line shape function to obtain optimized real spectral data comprises: using a deconvolution algorithm to solve a convolution equation to deconvolve and reconstruct each frame of raw spectral data to remove instrumental effects and random noise, to obtain optimized real spectral data, the convolution equation having an expression of: wherein represents raw spectral data; represents true spectral data; represents a convolution operation; represents an instrumental line function; represents random noise introduced during acquisition.

4. The method of claim 3, wherein, in the case where the deconvolution algorithm comprises a Richardson-Lucy algorithm, the use of a deconvolution algorithm to solve a convolution equation to deconvolve and reconstruct each frame of raw spectral data to remove instrumental effects and random noise, to obtain optimized real spectral data, comprises: based on a maximum likelihood estimation principle, using an iterative optimization method to repeatedly perform the following steps until a preset convergence condition is met: convolving the current real spectrum estimate with the instrument line shape function to obtain a simulated measured spectrum; dividing the original spectrum data by the simulated measured spectrum to obtain a correction factor; deconvolving the correction factor with the instrument line shape function and multiplying the result with the current real spectrum estimate to update the estimate; taking the estimate satisfying the convergence condition as the optimized real spectrum data.

5. The method of claim 3, wherein, In the case that the deconvolution algorithm comprises a Wiener filter algorithm, the deconvolution algorithm is used to deconvolve and reconstruct each frame of original spectrum data to remove instrument effects and random noise, and obtain the optimized real spectrum data, including: constructing a reference filter in the frequency domain based on the instrument line shape function and the signal-to-noise ratio prior information; performing Fourier transform on each frame of original spectrum data, and multiplying the transformed result with the reference filter to obtain a weighted frequency domain signal to suppress the influence of random noise through the weight; performing inverse Fourier transform on the frequency domain signal to obtain the optimized real spectrum data.

6. The method of claim 1, wherein, calculating the center wavelength value of each frame of original spectrum data, including: obtaining a characteristic spectrum line according to each frame of original spectrum data; the characteristic spectrum line is used to represent the corresponding relationship between light intensity and wavelength, and the characteristic spectrum line comprises at least one wave peak; assigning a weight to each wave peak according to the light intensity of the wave peak, and performing weighted average calculation on the wavelengths of all wave peaks according to the weights to obtain the center wavelength value.

7. An apparatus for optimizing spectral data, characterized by The device is arranged in an in-situ monitoring system of a semiconductor epitaxial growth device, and the in-situ monitoring system comprises a spectrometer, and the device comprises: an acquisition module configured to acquire original spectrum data output by the spectrometer; a query module configured to calculate, for each frame of original spectrum data, a center wavelength value thereof, and query a corresponding instrument line shape function from a preset database according to the center wavelength value; the instrument line shape function comprises model parameters, and the model parameters are parameter functions that vary continuously with wavelength; the model parameters comprise an amplitude, a Gaussian half-width parameter, a Lorentz half-width parameter, and a background offset; an optimization module configured to process each frame of original spectrum data and the corresponding instrument line shape function thereof by using a deconvolution algorithm to obtain optimized real spectrum data, which is used as input to an optical model to obtain epitaxial growth parameters of a wafer. The database setting module is configured to set the database in advance, and the setting manner comprises: selecting a plurality of discrete calibration wavelength points in a measurement wavelength range of the spectrometer; and measuring each calibration wavelength point by using a standard reference light source to obtain measurement data, wherein an emission line width of the standard reference light source at each calibration wavelength point is less than a resolution of the spectrometer; fitting the measurement data of each calibration wavelength point based on a Voigt function model to determine a fitting parameter value of a model parameter of the Voigt function model at the calibration wavelength point; continuously interpolating the fitting parameter values of all the model parameters by using an interpolation algorithm to generate a target parameter function continuously changing with wavelength in the measurement wavelength range; generating an instrument line shape function continuously changing with wavelength according to each target parameter function, and storing the instrument line shape function as the database.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1-6.

Images