Ultrasonic parameter adaptive adjustment and cleaning evaluation method in wafer cleaning process

Abstract

The invention provides an ultrasonic parameter adaptive adjustment and cleaning evaluation method in a wafer cleaning process, and relates to the technical field of semiconductor manufacturing, and the method comprises the steps: calculating sound field energy distribution through collecting wafer surface temperature distribution and cavity vibration response data; based on the temperature gradient and the sound intensity gradient, calculating the micro-flow field distribution of the cleaning fluid, and establishing a cavitation bubble collapse characteristic matrix; recognizing and cleaning a dead zone through singular value decomposition of the feature matrix, and controlling a piezoelectric transducer to generate focused ultrasonic waves; meanwhile, the plasma resonance sensor is used for monitoring the surface adsorption characteristic change and evaluating the cleanliness. According to the invention, real-time self-adaptive adjustment of ultrasonic cleaning parameters is realized, and the cleaning uniformity and efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing technology, and in particular to a method for adaptive adjustment of ultrasonic parameters and cleaning evaluation in wafer cleaning processes. Background Technology

[0002] Wafer cleaning is a critical process step in semiconductor manufacturing. Its purpose is to remove impurities such as particles, metal ions, organic contaminants, and natural oxides from the wafer surface, ensuring high yield rates for subsequent processes. Traditional wafer cleaning technologies mainly include chemical cleaning, physical cleaning, and ultrasonic cleaning. Among these, ultrasonic cleaning is widely used due to its high efficiency, non-destructive nature, and environmental friendliness. Ultrasonic cleaning utilizes the cavitation effect generated when sound waves propagate in a liquid. Energy is released through the formation and collapse of cavitation bubbles, forming microjets and shock waves, thereby removing contaminants from the wafer surface.

[0003] With the continuous development of integrated circuit manufacturing processes, wafer sizes are constantly increasing while feature sizes are constantly decreasing, placing higher demands on the precision and uniformity of cleaning processes. Currently, ultrasonic cleaning technology faces several technical challenges in practical applications.

[0004] Existing ultrasonic cleaning equipment typically uses fixed parameter settings, which cannot adaptively adjust according to the wafer surface structure characteristics and real-time status during the cleaning process. This results in uneven cleaning effects, especially for wafer surfaces with complex structures, which can easily create cleaning dead zones and affect cleaning quality.

[0005] Traditional ultrasonic cleaning processes lack real-time monitoring of the sound field distribution and liquid microflow field inside the cleaning chamber, making it impossible to accurately control the local sound energy distribution. This results in uneven energy distribution during the cleaning process, causing some areas to be under-cleaned or over-cleaned, affecting the integrity of the wafer surface and the yield of subsequent processes.

[0006] Existing technologies lack quantitative assessment methods for wafer surface cleanliness, typically relying on visual inspection or offline measurement. This approach cannot achieve real-time assessment and feedback adjustment during the cleaning process, making it difficult to guarantee the consistency and reliability of cleaning quality, nor can it provide accurate data support for optimizing the cleaning process. Summary of the Invention

[0007] This invention provides a method for adaptive adjustment of ultrasonic parameters and cleaning evaluation in wafer cleaning processes, which can solve the problems in the prior art.

[0008] A first aspect of this invention provides a method for adaptive adjustment of ultrasonic parameters and cleaning evaluation in a wafer cleaning process, comprising: During the cleaning process, wafer surface temperature distribution and cavity vibration response data are collected. The vibration principal frequency and vibration mode are obtained by analyzing the cavity vibration response data. The cavity resonant frequency is calculated by combining the vibration principal frequency with the wafer surface structural characteristics. The sound field energy distribution is calculated based on the cavity resonant frequency and the vibration mode. Based on the temperature distribution and the sound field energy distribution, the temperature gradient and sound intensity gradient of the wafer surface are calculated. The driving force boundary conditions are obtained based on the temperature gradient and the sound intensity gradient. The microfluidic field distribution of the cleaning fluid is obtained by combining fluid dynamics simulation calculations. A cavitation bubble collapse feature matrix is ​​established based on the microfluidic field distribution of the cleaning fluid. The collapse mode is extracted by singular value decomposition of the cavitation bubble collapse feature matrix. Local cleaning dead zones are identified in real time based on the spatiotemporal distribution of the collapse mode. The piezoelectric transducer and ultrasonic reflector are controlled to generate phase-adjustable focused ultrasonic waves to focus the sound energy on the local cleaning dead zones. During the cleaning process, the changes in the molecular layer adsorption characteristics of the wafer surface are monitored in real time by a plasma resonance sensor, and the changes in the resonance peak displacement of the local surface plasma resonance angle over time are measured. Based on the changes in the molecular layer adsorption characteristics and the changes in the resonance peak displacement, the thickness of the surface adsorption layer and the concentration of adsorbed molecules are calculated, and combined with the surface wettability data obtained by the contact angle measuring device, a quantitative evaluation curve of surface cleanliness is generated.

[0009] Based on the analysis of the cavity vibration response data, the dominant vibration frequency and vibration mode are obtained. The cavity resonant frequency is calculated by combining the dominant vibration frequency with the wafer surface structure characteristics. The sound field energy distribution is calculated based on the cavity resonant frequency and the vibration mode, including: The cavity vibration response data is subjected to Fast Fourier Transform to obtain a frequency domain signal, and the vibration main frequency is extracted from the frequency domain signal. Wavelet multi-resolution analysis is performed on the cavity vibration response data to obtain the energy distribution of different frequency bands. Based on the energy distribution, energy vibration frequency bands with an energy ratio exceeding a preset frequency band energy threshold are selected. The energy vibration frequency bands are subjected to Hilbert-Huang transform to extract vibration mode feature vectors and eigenvalue decomposition to obtain vibration modes. The height distribution data and size characteristic data of the wafer surface structure are obtained as the boundary conditions for acoustic wave propagation. Based on the acoustic wave propagation boundary conditions and the dominant vibration frequency, the eigenvalues ​​are solved by the modal iteration method. A correction coefficient is introduced in combination with the wafer surface structure characteristics. The cavity resonant frequency is obtained by multiplying the eigenvalues ​​by the correction coefficient. Using the resonant frequency of the cavity as excitation, acoustic-structural coupling iterative calculations are performed based on the vibration modes to obtain three-dimensional sound pressure distribution data; the density parameter of the cleaning fluid inside the cavity is obtained, and the sound field propagation impedance matrix is ​​calculated based on the cleaning fluid density parameter. The three-dimensional sound pressure distribution data and the sound field propagation impedance matrix are convolved to obtain the local sound field impedance. The local sound field impedance is multiplied by the three-dimensional sound pressure distribution data and then divided by the square of the cleaning fluid density to obtain the sound field energy density distribution.

[0010] Based on the temperature gradient and the acoustic intensity gradient, the driving force boundary conditions are obtained. Combined with fluid dynamics simulation calculations, the microfluidic field distribution of the cleaning fluid is obtained. Based on the microfluidic field distribution of the cleaning fluid, a cavitation bubble collapse characteristic matrix is ​​established, including: The thermocapillary force is obtained by taking the negative value of the product of the temperature gradient and the surface tension coefficient of the temperature; the acoustic radiation force is obtained by multiplying the acoustic intensity gradient by the acoustic field attenuation coefficient of the cleaning fluid; the total driving force is obtained by adding the thermocapillary force and the acoustic radiation force; the total driving force is applied to the boundary of the fluid computational domain to establish the fluid boundary driving force distribution; the fluid boundary driving force distribution is used as the driving force boundary condition. The pressure correction amount is obtained by combining the driving force boundary conditions with the initial pressure distribution of the fluid. The pressure field is updated according to the pressure correction amount. The calculation is iterated until the difference between two adjacent pressure fields is less than the convergence accuracy threshold, and the microfluidic field distribution of the cleaning fluid is obtained. The fluid shear stress distribution is extracted from the microfluidic field distribution of the cleaning fluid, and dimensionless parameters are calculated based on the fluid shear stress distribution. The internal pressure and bubble wall acceleration of the bubbles in the cleaning fluid are calculated based on the dimensionless parameters. The bubble wall acceleration is integrated to obtain the bubble wall velocity. The bubble wall velocity is multiplied by the internal pressure of the bubble to obtain the collapse power density. The collapse power density is combined with the bubble collapse time to form a cavitation bubble collapse feature matrix.

[0011] Collapse modes are extracted through singular value decomposition of the cavitation bubble collapse feature matrix, and local cleaning dead zones are identified in real time based on the spatiotemporal distribution of the collapse modes, including: The cavitation bubble collapse feature matrix is ​​decomposed by singular value to obtain a spatiotemporal feature vector matrix and a singular value diagonal matrix. The cumulative contribution rate is obtained by calculating the ratio of the sum of squares of the singular values ​​in the singular value diagonal matrix to the total sum of squares. The spatiotemporal feature vector matrices corresponding to the singular values ​​whose cumulative contribution rate exceeds the preset contribution ratio are selected to form the collapse mode. The local collapse intensity is obtained by summing the products of each column vector of the spatiotemporal feature vector matrix with the corresponding singular values. The local collapse intensity is normalized to obtain the collapse energy weight. The collapse intensity gradient field is obtained by multiplying the collapse energy weight with the local collapse intensity. The local cleaning dead zone is determined based on the local minimum point of the collapse intensity gradient field.

[0012] Controlling the piezoelectric transducer and ultrasonic reflector to generate phase-tunable focused ultrasonic waves, and focusing the acoustic energy onto the localized cleaning dead zone includes: The collapse period is obtained by performing spectral analysis on the spatiotemporal feature vector matrix, and the product of the collapse period and the preset dwell coefficient is used as the sound field switching period. The phase delay distribution of the piezoelectric transducer is calculated and used as the excitation signal. Simultaneously, the radius of curvature of the ultrasonic reflector is adjusted according to the spatial coordinates of the local cleaning dead zone. A phase-adjustable focused ultrasonic wave is emitted according to the excitation signal and the radius of curvature, and the acoustic energy is focused according to the spatial coordinates of the local cleaning dead zone. When the focusing time reaches the sound field switching cycle, the focused ultrasonic wave is switched to the next local cleaning dead zone.

[0013] The phase delay distribution of the piezoelectric transducer is calculated, and the phase delay distribution is used as the excitation signal. Simultaneously, the radius of curvature of the ultrasonic reflector is adjusted according to the spatial coordinates of the local cleaning dead zone. A phase-adjustable focused ultrasonic wave is emitted according to the excitation signal and the radius of curvature, and the focused acoustic energy includes the following based on the spatial coordinates of the local cleaning dead zone: The initial phase delay distribution of the piezoelectric transducer is calculated based on the relationship between the coordinates of any point on the surface of the piezoelectric transducer and the focal length. The phase delay correction amount is determined in combination with the characteristic parameters of the piezoelectric material. The initial phase delay distribution and the phase delay correction amount are added together to obtain the target phase delay distribution. A coordinate mapping matrix is ​​established based on the target phase delay distribution, and the spatial coordinates of the local cleaning dead zone are converted into mapped coordinates through the coordinate mapping matrix. The initial radius of curvature of the ultrasonic reflector is calculated based on the mapped coordinates. The sound field pressure distribution at the mapped coordinates is obtained. The first and second derivatives of the sound field pressure distribution are multiplied by the corresponding adjustment coefficients and summed to obtain the radius of curvature adjustment amount. The initial radius of curvature and the radius of curvature adjustment amount are superimposed to obtain the target radius of curvature. The target phase delay distribution is converted into a phase delay term, the target radius of curvature is converted into a phase modulation term, and the phase delay term and the phase modulation term are combined to generate an excitation signal; The ultrasonic wave is excited at the mapped coordinates according to the excitation signal. The ratio of the sound intensity of the target area at the mapped coordinates to the sound intensity of the entire sound field area is calculated to obtain the sound energy focusing degree. The phase component is iteratively optimized based on the sound energy focusing degree to obtain the optimized ultrasonic excitation signal. The optimized ultrasonic excitation signal is used to achieve sound energy focusing cleaning of the local cleaning dead zone.

[0014] A second aspect of the present invention provides a system for adaptive adjustment of ultrasonic parameters and cleaning evaluation in a wafer cleaning process, comprising: The first unit is used to collect wafer surface temperature distribution and cavity vibration response data during the cleaning process, analyze the cavity vibration response data to obtain the main vibration frequency and vibration mode, combine the main vibration frequency with the wafer surface structural characteristics to calculate the cavity resonant frequency, and calculate the sound field energy distribution based on the cavity resonant frequency and the vibration mode. The second unit is used to calculate the temperature gradient and sound intensity gradient of the wafer surface based on the temperature distribution and the sound field energy distribution, obtain the driving force boundary conditions based on the temperature gradient and the sound intensity gradient, obtain the micro-flow field distribution of the cleaning fluid by combining fluid dynamics simulation calculations, establish a cavitation bubble collapse feature matrix based on the micro-flow field distribution of the cleaning fluid, extract the collapse mode through the singular value decomposition of the cavitation bubble collapse feature matrix, identify local cleaning dead zones in real time according to the spatiotemporal distribution of the collapse mode, and control the piezoelectric transducer and ultrasonic reflector to generate phase-adjustable focused ultrasonic waves to focus the sound energy on the local cleaning dead zones. The third unit is used to monitor the changes in molecular layer adsorption characteristics on the wafer surface in real time during the cleaning process using a plasma resonance sensor, and to measure the changes in the resonance peak displacement of the local surface plasma resonance angle over time; based on the changes in molecular layer adsorption characteristics and the changes in resonance peak displacement, the thickness of the surface adsorption layer and the concentration of adsorbed molecules are calculated, and combined with the surface wettability data obtained by the contact angle measuring device, a quantitative evaluation curve of surface cleanliness is generated.

[0015] A third aspect of the present invention, An electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0016] Fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0017] The beneficial effects of this application are as follows: This invention acquires real-time data on wafer surface temperature distribution and cavity vibration response, analyzes vibration characteristics, and calculates sound field energy distribution. It can accurately identify local dead zones during the cleaning process, achieve adaptive adjustment of ultrasonic parameters, and significantly improve the uniformity and efficiency of wafer cleaning.

[0018] This invention utilizes a piezoelectric transducer and an ultrasonic reflector to generate phase-adjustable focused ultrasonic waves, precisely focusing acoustic energy onto the cleaning dead zone. This solves the cleaning problem of areas that are difficult to reach in traditional cleaning technologies, significantly reduces cleaning defects and residual contaminants, and improves wafer manufacturing yield.

[0019] This invention innovatively combines plasma resonance sensing technology and contact angle measurement method to achieve real-time monitoring of the thickness and molecular concentration of the adsorption layer on the wafer surface, establishes a quantitative evaluation system for surface cleanliness, provides a scientific basis for precise control and quality assurance of the cleaning process, and reduces the risk of subsequent process failures. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the adaptive adjustment of ultrasonic parameters and cleaning evaluation method in the wafer cleaning process according to an embodiment of the present invention. Figure 2 Flowchart for vibration analysis and sound field calculation of acoustic cleaning cavity; Figure 3 This diagram illustrates the performance comparison of different cleaning algorithms. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0023] Figure 1 This is a flowchart illustrating the adaptive adjustment of ultrasonic parameters and cleaning evaluation method in the wafer cleaning process according to an embodiment of the present invention. Figure 1 As shown, the method includes: During the cleaning process, wafer surface temperature distribution and cavity vibration response data are collected. The vibration principal frequency and vibration mode are obtained by analyzing the cavity vibration response data. The cavity resonant frequency is calculated by combining the vibration principal frequency with the wafer surface structural characteristics. The sound field energy distribution is calculated based on the cavity resonant frequency and the vibration mode. Based on the temperature distribution and the sound field energy distribution, the temperature gradient and sound intensity gradient of the wafer surface are calculated. The driving force boundary conditions are obtained based on the temperature gradient and the sound intensity gradient. The microfluidic field distribution of the cleaning fluid is obtained by combining fluid dynamics simulation calculations. A cavitation bubble collapse feature matrix is ​​established based on the microfluidic field distribution of the cleaning fluid. The collapse mode is extracted by singular value decomposition of the cavitation bubble collapse feature matrix. Local cleaning dead zones are identified in real time based on the spatiotemporal distribution of the collapse mode. The piezoelectric transducer and ultrasonic reflector are controlled to generate phase-adjustable focused ultrasonic waves to focus the sound energy on the local cleaning dead zones. During the cleaning process, the changes in the molecular layer adsorption characteristics of the wafer surface are monitored in real time by a plasma resonance sensor, and the changes in the resonance peak displacement of the local surface plasma resonance angle over time are measured. Based on the changes in the molecular layer adsorption characteristics and the changes in the resonance peak displacement, the thickness of the surface adsorption layer and the concentration of adsorbed molecules are calculated, and combined with the surface wettability data obtained by the contact angle measuring device, a quantitative evaluation curve of surface cleanliness is generated.

[0024] In one optional implementation, the dominant vibration frequency and vibration mode are obtained based on the cavity vibration response data. The cavity resonant frequency is calculated by combining the dominant vibration frequency with the wafer surface structure characteristics. The sound field energy distribution is calculated based on the cavity resonant frequency and the vibration mode, including: The cavity vibration response data is subjected to Fast Fourier Transform to obtain a frequency domain signal, and the vibration main frequency is extracted from the frequency domain signal. Wavelet multi-resolution analysis is performed on the cavity vibration response data to obtain the energy distribution of different frequency bands. Based on the energy distribution, energy vibration frequency bands with an energy ratio exceeding a preset frequency band energy threshold are selected. The energy vibration frequency bands are subjected to Hilbert-Huang transform to extract vibration mode feature vectors and eigenvalue decomposition to obtain vibration modes. The height distribution data and size characteristic data of the wafer surface structure are obtained as the boundary conditions for acoustic wave propagation. Based on the acoustic wave propagation boundary conditions and the dominant vibration frequency, the eigenvalues ​​are solved by the modal iteration method. A correction coefficient is introduced in combination with the wafer surface structure characteristics. The cavity resonant frequency is obtained by multiplying the eigenvalues ​​by the correction coefficient. Using the resonant frequency of the cavity as excitation, acoustic-structural coupling iterative calculations are performed based on the vibration modes to obtain three-dimensional sound pressure distribution data; the density parameter of the cleaning fluid inside the cavity is obtained, and the sound field propagation impedance matrix is ​​calculated based on the cleaning fluid density parameter. The three-dimensional sound pressure distribution data and the sound field propagation impedance matrix are convolved to obtain the local sound field impedance. The local sound field impedance is multiplied by the three-dimensional sound pressure distribution data and then divided by the square of the cleaning fluid density to obtain the sound field energy density distribution.

[0025] like Figure 2 As shown, the method includes: The acquired cavity vibration response data were processed using Fast Fourier Transform (FFT) to convert the time-domain vibration signal into a frequency-domain signal. During processing, a Hanning window function was used to window the signal to reduce spectral leakage. Taking a specific data acquisition as an example, after FFT, three distinct frequency peaks—25kHz, 28kHz, and 40kHz—were clearly observed in the frequency domain signal. The amplitude at 25kHz was the largest, therefore 25kHz was determined as the dominant vibration frequency.

[0026] For the cavity vibration response data, a db4 wavelet basis was selected, and a 5-level wavelet decomposition was performed to obtain the energy distribution in different frequency bands. The energy proportion of each frequency band was calculated, and a preset energy threshold of 5% was used to select frequency bands with an energy proportion exceeding 5% as energy vibration bands. In the example, the selected energy vibration bands were the 20-30kHz and 38-42kHz bands.

[0027] Hilbert-Huang transform was applied to the selected energy vibration frequency band data. First, the original signal was processed through empirical mode decomposition (EMD). After identifying local extrema, cubic spline interpolation was used to fit the upper and lower envelopes. The mean envelope was calculated and subtracted from the original signal to obtain candidate intrinsic mode functions (EMFs). This process was repeated until the EMF conditions were met, and the residual signal was further decomposed until it became a monotonic function. Then, a Hilbert transform was performed on each EMF, the Hilbert transform value was calculated, and an analytic signal was constructed to obtain the instantaneous amplitude, instantaneous phase, and instantaneous frequency. For the 20-30kHz frequency band, three main EMFs were extracted; for the 38-42kHz frequency band, two main EMFs were extracted. Eigenvalue decomposition was performed on these EMFs to obtain vibration mode eigenvectors. In the example, the first vibration mode corresponds to the vertical vibration of the bottom of the cavity, with an eigenvalue of 1.24; the second vibration mode corresponds to the radial vibration of the sidewall of the cavity, with an eigenvalue of 0.89; and the third vibration mode corresponds to the torsional vibration of the entire cavity, with an eigenvalue of 0.63.

[0028] The surface structural features of the wafer were obtained through measurements using a high-precision 3D profilometer. Taking a typical etched wafer as an example, the measured surface groove depth distribution was between 0.5 and 2 μm, the lateral dimension was 10 to 50 μm, and the pattern coverage was approximately 60%. These data were used as boundary conditions for acoustic wave propagation and input into the computational model.

[0029] Based on the obtained sound wave propagation boundary conditions and the dominant vibration frequency of 25 kHz, a modal iteration method was adopted, with a convergence threshold of 0.001 and a maximum number of iterations of 100. After 73 iterations, the eigenvalue was obtained as 1.047. Considering the influence of the wafer surface structure on the sound field propagation, a correction factor was introduced. According to the characteristics of the wafer surface structure, when the pattern coverage is 60% and the average groove depth is 1.2 μm, the correction factor is set to 0.942. Multiplying the eigenvalue of 1.047 by the correction factor of 0.942 yields a cavity resonant frequency of 24.7 kHz.

[0030] Using the cavity resonant frequency of 24.7 kHz as the excitation frequency and combining it with the previously obtained vibration modes, an acoustic-structural coupling model was established. In this model, the vibration of the cavity wall serves as the boundary condition for the sound field, while the sound field exerts a feedback effect on the cavity structure. Through iterative calculations, convergence was considered achieved when the relative error between two adjacent iterations was less than 0.5%. After 12 iterations, the three-dimensional sound pressure distribution data within the cavity was obtained.

[0031] The density of the cleaning solution used was measured to be 998 kg / m³. 3 The speed of sound is 1480 m / s. The sound field propagation impedance matrix is ​​calculated based on these parameters. Specifically, the sound field propagation impedance Z is equal to the product of the cleaning fluid density and the speed of sound, i.e., Z = 1476040 kg / (m²). 2 The local sound field impedance distribution is obtained by convolving the three-dimensional sound pressure distribution data with the sound field propagation impedance matrix.

[0032] Multiplying the local acoustic field impedance by the three-dimensional sound pressure distribution data and then dividing by the square of the cleaning fluid density yields the acoustic field energy density distribution. Calculations show that the maximum acoustic field energy density, located 10 mm above the wafer surface, occurs in the wafer's central region, with an energy density of 0.324 J / m². 3 In the wafer edge region, the energy density decreases to 0.118 J / m². 3 It exhibits a clear center-edge gradient distribution characteristic.

[0033] In practical applications, adjusting the frequency of the ultrasonic generator to close to the calculated cavity resonant frequency of 24.7 kHz can effectively improve the cleaning effect. Compared with the cleaning effect without frequency optimization, the particle removal rate is increased by 23%, and the cleaning uniformity of the wafer surface is significantly improved.

[0034] In one optional implementation, the driving force boundary conditions are obtained based on the temperature gradient and the sound intensity gradient, and the microfluidic field distribution of the cleaning fluid is obtained by combining fluid dynamics simulation calculations. The cavitation bubble collapse characteristic matrix is ​​then established based on the microfluidic field distribution of the cleaning fluid, including: The thermocapillary force is obtained by taking the negative value of the product of the temperature gradient and the surface tension coefficient of the temperature; the acoustic radiation force is obtained by multiplying the acoustic intensity gradient by the acoustic field attenuation coefficient of the cleaning fluid; the total driving force is obtained by adding the thermocapillary force and the acoustic radiation force; the total driving force is applied to the boundary of the fluid computational domain to establish the fluid boundary driving force distribution; the fluid boundary driving force distribution is used as the driving force boundary condition. The pressure correction amount is obtained by combining the driving force boundary conditions with the initial pressure distribution of the fluid. The pressure field is updated according to the pressure correction amount. The calculation is iterated until the difference between two adjacent pressure fields is less than the convergence accuracy threshold, and the microfluidic field distribution of the cleaning fluid is obtained. The fluid shear stress distribution is extracted from the microfluidic field distribution of the cleaning fluid, and dimensionless parameters are calculated based on the fluid shear stress distribution. The internal pressure and bubble wall acceleration of the bubbles in the cleaning fluid are calculated based on the dimensionless parameters. The bubble wall acceleration is integrated to obtain the bubble wall velocity. The bubble wall velocity is multiplied by the internal pressure of the bubble to obtain the collapse power density. The collapse power density is combined with the bubble collapse time to form a cavitation bubble collapse feature matrix.

[0035] To establish the driving force boundary conditions under the combined effect of temperature and acoustic intensity gradients, the cleaning area is first meshed, dividing the cleaning fluid surface into m×n micro-elements. The temperature distribution T(x,y) of each micro-element is measured, and the temperature difference between adjacent micro-elements is calculated to obtain the temperature gradient ▽T. In practical applications, an infrared thermal imager can be used to measure the temperature field distribution. For example, during the cleaning of microelectronic devices, the measured surface temperature distribution range is 293K to 323K, and the calculated maximum temperature gradient is approximately 0.5K / μm.

[0036] Simultaneously, the acoustic field distribution I(x,y) of the cleaning fluid is measured using a sound intensity measuring device, and the acoustic intensity difference between adjacent regions is calculated to obtain the acoustic intensity gradient ▽I. In a typical ultrasonic cleaning device, the acoustic intensity distribution starts from 25 W / cm² at the center. 2 Gradually reduce to 5 W / cm at the edge 2 The corresponding maximum sound intensity gradient is approximately 0.4 W / cm². 3 .

[0037] When calculating thermocapillary force based on the temperature gradient, the temperature gradient ▽T is multiplied by the surface tension coefficient σT and the negative value is taken, resulting in the thermocapillary force Ft = -σT×▽T. For water-based cleaning fluids, the surface tension coefficient σT is approximately -0.15×10⁻⁶. -3 N / (m·K). Therefore, in the region with a temperature gradient of 0.5 K / μm, the amplitude of the thermocapillary force is approximately 7.5 × 10⁻⁶ N / (m·K). -5 N / m.

[0038] The acoustic radiation force is calculated by multiplying the sound intensity gradient ▽I by the sound field attenuation coefficient α of the cleaning fluid, i.e., Fs = α × ▽I. For ultrasonic waves with a frequency of 40 kHz propagating in water, the sound field attenuation coefficient α is approximately 1.3 × 10⁻⁶. -5 m -1 At a sound intensity gradient of 0.4 W / cm 3 In that area, the acoustic radiation force is approximately 5.2 × 10⁻⁶. -6 N / m 3 .

[0039] The total driving force F is obtained by the vector superposition of the thermocapillary force Ft and the acoustic radiation force Fs, F = Ft + Fs. At each grid element on the surface of the cleaning fluid, the total driving force at that location is calculated, forming the fluid boundary driving force distribution matrix F(i,j), where i=1,2,...,m and j=1,2,...,n, representing the grid position. This driving force distribution matrix serves as the boundary condition for the fluid dynamics simulation.

[0040] The initial fluid pressure field P0(x,y,z) is set to hydrostatic pressure distribution, i.e., P0(x,y,z) = P0 - ρgz, where P0 is atmospheric pressure, ρ is the density of the cleaning fluid, g is the acceleration due to gravity, and z is the depth coordinate. Driving force boundary conditions are applied to the computational domain boundary, and the pressure correction ΔP is calculated based on the momentum conservation equation. The pressure field P = P0 + ΔP is updated through an iterative solution process. When the difference in pressure field calculated in two adjacent iterations is less than a preset convergence accuracy threshold (e.g., 10), the convergence is considered complete. -6 When the pressure is P(x,y,z), the calculation is considered to have converged, and the final microfluidic field distribution of the cleaning fluid is obtained, including the velocity field V(x,y,z) and the pressure field P(x,y,z).

[0041] In typical semiconductor device cleaning applications, the calculated microfluidic velocity field distribution ranges from 0 to 15 cm / s, with the maximum velocity occurring near the center of the cleaning area, forming a ring-shaped flow region. The pressure field distribution shows a local pressure difference of approximately 200 Pa in the central region; this microfluidic field distribution helps enhance the removal efficiency of contaminants.

[0042] The fluid shear stress distribution τ(x,y,z) is extracted from the calculated microfluidic field distribution of the cleaning fluid. The shear stress is calculated using the velocity gradient; for a Newtonian fluid, the shear stress τ = μ × ( ), where μ is the dynamic viscosity of the cleaning fluid, and u and v are the velocity components in the x and y directions, respectively. In practical applications, the shear stress near the surface of the cleaning fluid can reach 0.05-0.2 Pa.

[0043] Based on the fluid shear stress distribution, the dimensionless parameters Re (Reynolds number), We (Weber number), and Ca (capillary number) are calculated. Where Re = ρVL / μ, and We = ρV2 L / σ, capillary number Ca=μV / σ, where V is the characteristic flow rate, L is the characteristic length, and σ is the surface tension. Typical parameter values ​​in microelectronic device cleaning processes are Re=50-200, We=0.01-0.1, and Ca=0.001-0.01.

[0044] Using these dimensionless parameters and a simplified form of the Rayleigh-Plesset equation, the internal pressure Pg and bubble wall acceleration a of the bubble are calculated. The internal pressure Pg is related to the surrounding liquid pressure P, the bubble radius R, and the surface tension σ. For a bubble with an initial radius of 5 μm, the internal pressure can reach 10 during collapse. 6 The Pa level. The bubble wall acceleration a can reach 10. 9 -10 10 m / s 2 .

[0045] The bubble wall velocity *v* is obtained by integrating the bubble wall acceleration *a* over time. At the end of collapse, the bubble wall velocity can reach 100-300 m / s. The collapse power density *Pd* is calculated as the product of the bubble wall velocity *v* and the internal pressure *Pg*, i.e., *Pd* = *v* × *Pg*. Under typical ultrasonic cleaning conditions, the collapse power density can reach 10... 13 -10 14 W / m 3 The duration is approximately 10 -6 -10 -7 Second.

[0046] By combining the collapse power density Pd with the bubble collapse time t, a cavitation bubble collapse feature matrix M is constructed. The matrix element M(i,j,k) represents the collapse characteristic of the k-th bubble at position (i,j). This feature matrix reflects the spatiotemporal distribution characteristics of bubble collapse in the cleaning fluid, providing a basis for optimizing cleaning parameters. In practical applications, by analyzing the collapse feature matrix, the optimal combination of process parameters for cleaning effect can be determined, such as an ultrasonic power of 200W, a frequency of 40kHz, and a temperature gradient controlled within the range of 0.3-0.6K / μm.

[0047] Figure 3This diagram illustrates the performance comparison of different cleaning algorithms, showcasing a comprehensive comparative analysis of the key performance indicators of the present invention and four traditional algorithms in wafer cleaning processes. From the perspective of cleaning efficiency, the present invention achieves a cleaning efficiency of 95.8%, significantly surpassing the traditional SIMPLE algorithm's 78.2%, PISO algorithm's 82.6%, SIMPLEC algorithm's 80.4%, and PCG algorithm's 76.9%, demonstrating the significant advantage of optimizing cavitation bubble collapse characteristics under the synergistic driving of temperature gradient and acoustic intensity gradient. Regarding convergence speed, the present invention requires only 27 iterations to achieve the required convergence accuracy, achieving a computational efficiency improvement of over 50% compared to the 55-58 iterations required by other algorithms, verifying the rapid convergence characteristics of the adaptive driving force boundary condition establishment method. In terms of energy consumption reduction, the present invention achieves a 42.3% energy saving effect, avoiding energy waste through precise control of the phase-adjustable focused ultrasonic waves. The dead zone recognition rate reached 96.2%. The collapse mode extraction method based on the singular value decomposition of the cavitation bubble collapse feature matrix achieved accurate local cleaning dead zone positioning. The overall leading performance fully demonstrates the innovation and practicality of the multi-physics coupling modeling and real-time adaptive adjustment technology route of this invention.

[0048] In one optional implementation, the collapse mode is extracted by singular value decomposition of the cavitation bubble collapse feature matrix, and local cleaning dead zones are identified in real time based on the spatiotemporal distribution of the collapse mode, including: The cavitation bubble collapse feature matrix is ​​decomposed by singular value to obtain a spatiotemporal feature vector matrix and a singular value diagonal matrix. The cumulative contribution rate is obtained by calculating the ratio of the sum of squares of the singular values ​​in the singular value diagonal matrix to the total sum of squares. The spatiotemporal feature vector matrices corresponding to the singular values ​​whose cumulative contribution rate exceeds the preset contribution ratio are selected to form the collapse mode. The local collapse intensity is obtained by summing the products of each column vector of the spatiotemporal feature vector matrix with the corresponding singular values. The local collapse intensity is normalized to obtain the collapse energy weight. The collapse intensity gradient field is obtained by multiplying the collapse energy weight with the local collapse intensity. The local cleaning dead zone is determined based on the local minimum point of the collapse intensity gradient field.

[0049] The obtained cavitation bubble collapse feature matrix is ​​decomposed using singular value decomposition (SVD), which results in the product of three matrices: a left singular vector matrix, a singular value diagonal matrix, and the transpose of a right singular vector matrix. The left and right singular vector matrices form the spatiotemporal feature vector matrix, while the singular value diagonal matrix contains the singular values ​​arranged in descending order. In practical applications, for example, a 10000×10000 feature matrix can yield 10000 singular values ​​and their corresponding eigenvectors.

[0050] The cumulative contribution rate of singular values ​​can be obtained by calculating the ratio of the square of each singular value in the singular value diagonal matrix to the sum of the squares of all singular values. In practical applications, a preset contribution rate of 95% can be set, that is, singular values ​​with a cumulative contribution rate exceeding 95% and their corresponding eigenvectors can be selected. For example, in an experimental case, the cumulative contribution rate of the first 50 singular values ​​reached 96.8%, so these 50 singular values ​​and their corresponding eigenvectors can be selected to constitute the collapse mode.

[0051] The local collapse intensity is obtained by summing the products of each column vector of the spatiotemporal eigenvector matrix with its corresponding singular value. In practical applications, a 100×100 pixel spatial region can yield a 10000-dimensional local collapse intensity vector. For example, in one experiment, the local collapse intensity ranged from 0 to 5000, with the collapse intensity in most areas between 2000 and 3000.

[0052] The maximum-minimum normalization method is used, which involves subtracting the minimum value from each local collapse intensity and then dividing by the difference between the maximum and minimum values. After normalization, the collapse energy weight ranges from 0 to 1. For example, in the experiment mentioned above, the area with a normalized collapse energy weight between 0.4 and 0.6 accounts for approximately 70%.

[0053] Multiplying the collapse energy weight by the local collapse intensity yields the collapse intensity gradient field. This field reflects the spatial distribution of the collapse energy of the cavitation bubble. In practical applications, the collapse intensity gradient field can be represented as a two-dimensional matrix, where each element corresponds to the collapse intensity gradient value at a spatial location. For example, within a 100×100 pixel region, the collapse intensity gradient field can be represented as a 100×100 matrix.

[0054] A local minimum is a location where the collapse intensity gradient value is less than that of all its neighboring points. In practical applications, a threshold can be set to identify local minimums where the collapse intensity gradient value is below the threshold as cleaning dead zones. For example, setting the threshold to 30% of the average collapse intensity gradient field value can filter out areas with particularly low collapse intensity. In one experimental case, cleaning dead zones identified by this method accounted for approximately 8% of the entire cleaning area; these areas were mainly distributed at the corners of the container and behind obstructions.

[0055] In practical applications, this method can process the cavitation bubble collapse feature matrix in real time, for example, processing 10 frames of image data per second, calculating the collapse intensity gradient field, and identifying cleaning dead zones. Based on the identification results, the parameters of the ultrasonic cleaning equipment, such as frequency, power, or transducer position, can be automatically adjusted to reduce or eliminate cleaning dead zones and improve cleaning efficiency. Experimental data shows that after optimizing cleaning parameters using this method, the cleaning dead zone area can be reduced by approximately 60%, and the cleaning efficiency can be improved by approximately 40%.

[0056] This method is not only applicable to conventional ultrasonic cleaning equipment, but can also be applied to fields such as medical device cleaning, semiconductor wafer cleaning, and precision component cleaning. By identifying cleaning dead zones in real time and adjusting parameters, it can significantly improve cleaning quality and efficiency, reduce cleaning time and energy consumption, and achieve intelligent and precise control of the cleaning process.

[0057] In one alternative implementation, controlling the piezoelectric transducer and ultrasonic reflector to generate phase-tunable focused ultrasonic waves, focusing the acoustic energy onto the localized cleaning dead zone, includes: The collapse period is obtained by performing spectral analysis on the spatiotemporal feature vector matrix, and the product of the collapse period and the preset dwell coefficient is used as the sound field switching period. The phase delay distribution of the piezoelectric transducer is calculated and used as the excitation signal. Simultaneously, the radius of curvature of the ultrasonic reflector is adjusted according to the spatial coordinates of the local cleaning dead zone. A phase-adjustable focused ultrasonic wave is emitted according to the excitation signal and the radius of curvature, and the acoustic energy is focused according to the spatial coordinates of the local cleaning dead zone. When the focusing time reaches the sound field switching cycle, the focused ultrasonic wave is switched to the next local cleaning dead zone.

[0058] The acquired spatiotemporal feature vector matrix of the cleaning dead zone was subjected to spectral analysis using the Fast Fourier Transform (FFT) method to transform the time-domain signal into the frequency domain, thereby identifying the main frequency characteristics of bubble collapse. By analyzing the distribution pattern of spectral peaks, the periodic characteristics of bubble collapse, i.e., the collapse period, were determined. In a typical cleaning system, the collapse period usually varies between 0.5 ms and 2 ms, depending on the characteristics of the cleaning fluid and the ultrasonic power. For example, when using deionized water as the cleaning medium and the ultrasonic power is set to 500 W, the typical collapse period obtained through spectral analysis is approximately 1.2 ms.

[0059] The sound field switching cycle is obtained by multiplying the identified collapse cycle by a preset residence coefficient. The residence coefficient is an empirical parameter used to control the duration of acoustic energy in a single cleaning dead zone, and its value is usually between 2.5 and 5. A larger residence coefficient means that the acoustic energy stays in a single dead zone for a longer time, which is beneficial for removing hard-to-clean dirt, but will reduce the overall cleaning efficiency; a smaller residence coefficient can improve cleaning efficiency, but is not effective in removing stubborn dirt. In practical applications, this coefficient can be dynamically adjusted according to the type of dirt and cleaning requirements. Taking the aforementioned collapse cycle of 1.2ms as an example, when the residence coefficient is set to 3, the calculated sound field switching cycle is 3.6ms, that is, the focusing time of the ultrasonic wave in each cleaning dead zone is 3.6ms, after which it switches to the next cleaning dead zone.

[0060] A three-dimensional coordinate system is established, with the center of the bottom of the cleaning tank as the origin, and the coordinate information of each cleaning dead zone is recorded. For a cleaning dead zone located at coordinates (x, y, z), the distance difference from each piezoelectric transducer unit to that point is calculated and converted into the corresponding phase delay. In a system equipped with 64 piezoelectric transducer units, the phase delay of each unit can be controlled independently with an accuracy of 0.1 degrees. For example, when the cleaning dead zone is located at coordinates (50mm, 30mm, 20mm), the phase delay of the first transducer unit is 0 degrees, while the phase delay of the 64th unit needs to be adjusted to 285.7 degrees to ensure coherent superposition of sound waves at the target location.

[0061] The calculated phase delay distribution is used as the excitation signal for the piezoelectric transducers. This excitation signal is generated by a digital signal processor and drives each piezoelectric transducer unit through a power amplifier. The frequency of the excitation signal is typically set between 20 kHz and 40 kHz, which is the commonly used ultrasonic cleaning frequency range. In specific applications, 28 kHz can be selected as the operating frequency, in which case the amplitude of the excitation signal is set to 80 V to 120 V to generate sufficient ultrasonic energy for cleaning.

[0062] The radius of curvature of the ultrasonic reflector is adjusted based on the spatial coordinates of the localized cleaning dead zone. The ultrasonic reflector employs a deformable reflective structure made of shape memory alloy, allowing its curvature to be dynamically adjusted under the control of an electronic system. When the cleaning dead zone is located at a greater distance, the radius of curvature of the reflector needs to be increased to obtain a longer focal length; when the cleaning dead zone is located at a closer distance, the radius of curvature should be decreased. In practical applications, the adjustment range of the radius of curvature is typically from 150mm to 350mm. For example, when the cleaning dead zone is located 200mm from the reflector, the optimal radius of curvature for the reflector is approximately 240mm.

[0063] Based on the calculated excitation signal and the adjusted radius of curvature, a phase-adjustable focused ultrasonic wave is emitted. The ultrasonic wave propagates through the medium and focuses in the target cleaning dead zone, generating a high-intensity sound field. At the focal point, the sound intensity can reach 8 to 12 times that of the non-focused area, sufficient to produce a strong cavitation effect and effectively remove dirt from the dead zone. The focusing diameter of the sound field is typically between 3 mm and 5 mm, a precision sufficient for targeted treatment of small cleaning dead zones.

[0064] The focusing time of each cleaning dead zone is monitored by a timer. When the focusing time reaches the previously calculated acoustic field switching cycle, the focused ultrasonic wave is switched to the next local cleaning dead zone according to a pre-planned path. During the switching process, the phase delay distribution required for the new target position is recalculated, and the radius of curvature of the reflector is adjusted. The switching response time is typically within 0.5ms, ensuring the continuity and efficiency of the cleaning process. In this way, multiple cleaning dead zones can be processed sequentially according to a predetermined order, significantly improving the overall cleaning effect.

[0065] In one optional implementation, the phase delay distribution of the piezoelectric transducer is calculated, and the phase delay distribution is used as an excitation signal. Simultaneously, the radius of curvature of the ultrasonic reflector is adjusted according to the spatial coordinates of the local cleaning dead zone. A phase-adjustable focused ultrasonic wave is emitted according to the excitation signal and the radius of curvature, and the focused acoustic energy includes the following based on the spatial coordinates of the local cleaning dead zone: The initial phase delay distribution of the piezoelectric transducer is calculated based on the relationship between the coordinates of any point on the surface of the piezoelectric transducer and the focal length. The phase delay correction amount is determined in combination with the characteristic parameters of the piezoelectric material. The initial phase delay distribution and the phase delay correction amount are added together to obtain the target phase delay distribution. A coordinate mapping matrix is ​​established based on the target phase delay distribution, and the spatial coordinates of the local cleaning dead zone are converted into mapped coordinates through the coordinate mapping matrix. The initial radius of curvature of the ultrasonic reflector is calculated based on the mapped coordinates. The sound field pressure distribution at the mapped coordinates is obtained. The first and second derivatives of the sound field pressure distribution are multiplied by the corresponding adjustment coefficients and summed to obtain the radius of curvature adjustment amount. The initial radius of curvature and the radius of curvature adjustment amount are superimposed to obtain the target radius of curvature. The target phase delay distribution is converted into a phase delay term, the target radius of curvature is converted into a phase modulation term, and the phase delay term and the phase modulation term are combined to generate an excitation signal; The ultrasonic wave is excited at the mapped coordinates according to the excitation signal. The ratio of the sound intensity of the target area at the mapped coordinates to the sound intensity of the entire sound field area is calculated to obtain the sound energy focusing degree. The phase component is iteratively optimized based on the sound energy focusing degree to obtain the optimized ultrasonic excitation signal. The optimized ultrasonic excitation signal is used to achieve sound energy focusing cleaning of the local cleaning dead zone.

[0066] For any point P(x, y, z) on the surface of a piezoelectric transducer, the initial phase delay can be calculated by the distance difference from that point to the focal point F(xf, yf, zf). For example, if the reference point is the transducer center O(0, 0, 0), the initial phase delay of point P can be expressed as the difference between the distance from point P to the focal point F and the distance from point O to the focal point F, multiplied by the wavenumber. For a circular transducer with a diameter of 50 mm, when the focal length is set to 100 mm, the phase delay difference between the edge point and the center point is approximately π / 2.

[0067] The phase delay correction is determined by considering the piezoelectric constant, dielectric constant, and mechanical mass factors of the material. For example, for the commonly used PZT-5H material, when operating at a frequency of 1 MHz and a temperature of 25 °C, a correction of approximately 0.05π is required due to the material's inherent nonlinear response. Adding the initial phase delay distribution to the phase delay correction yields the target phase delay distribution, which more accurately describes the phase characteristics of the actual transducer.

[0068] Based on the target phase delay distribution, an interpolation algorithm is used to discretize the continuous phase distribution into a matrix form to establish a coordinate mapping matrix. For a matrix resolution of 50×50, each matrix element corresponds to a small region on the transducer surface. This coordinate mapping matrix can convert the spatial coordinates (xd, yd, zd) of the local cleaning dead zone into mapped coordinates (xm, ym, zm). For example, when the cleaning dead zone is located at (30mm, 25mm, 120mm) in actual space, it can be converted to coordinates (0.6, 0.5, 1.2) in the normalized mapped space using the mapping matrix.

[0069] The initial radius of curvature of the ultrasonic reflector is calculated using geometric formulas based on mapped coordinates. For a single focal point, the initial radius of curvature R0 is typically set to twice the distance from the mapped coordinates to the center of the reflector. For example, for mapped coordinates (0.6, 0.5, 1.2), if the center of the reflector is located at (0, 0, 0), the initial radius of curvature is approximately 2.62 times the normalized unit, corresponding to an actual physical size of 262 mm.

[0070] The sound pressure distribution at the mapped coordinates was obtained through sound field simulation or actual measurement. For operation at 1 MHz and a sound power of 50 W, the sound pressure in the central region reached 0.5 MPa, while the edge region was approximately 0.1 MPa. The first and second derivatives of the sound pressure distribution were calculated, representing the sound pressure gradient and gradient rate of change, respectively. For the above sound field, the first derivative in the central region was close to zero, while the first derivative in the transition region was approximately ±0.02 MPa / mm, and the second derivative was approximately ±0.001 MPa / mm. 2 .

[0071] The first derivative coefficient α of the sound field pressure distribution is set to 10 mm / MPa, and the second derivative coefficient β is set to 100 mm / MPa. 2 / MPa. An adjustment coefficient is set empirically. The first and second derivatives are multiplied by the adjustment coefficient and summed to obtain the radius of curvature adjustment. For the above sound field, the calculated radius of curvature adjustment is approximately ±20mm. The initial radius of curvature is then superimposed with the radius of curvature adjustment to obtain the target radius of curvature, which is approximately 282 mm in this example.

[0072] The continuous phase distribution is discretized into a digital control signal, and the target phase delay distribution is converted into a phase delay term. For example, the phase values ​​in the range [-π, π] are quantized into 256 levels, corresponding to 8-bit digital control precision. When converting the target radius of curvature into a phase modulation term, the calculation is performed based on the relationship between the reflector curvature and the acoustic wave phase modulation. The phase delay term and the phase modulation term are combined to generate an excitation signal, which can be a set of time series or frequency components.

[0073] After the ultrasonic excitation signal generates ultrasonic waves at the mapped coordinates, the ratio of the sound intensity in the target area at the mapped coordinates to the sound intensity of the entire sound field needs to be calculated to obtain the acoustic energy focusing degree. The acoustic energy focusing degree is an important indicator for evaluating the focusing effect, and ideally should be close to 1. In practical applications, the acoustic energy focusing degree is usually between 0.7 and 0.9. For example, for a cleaning dead zone with a target area radius of 5 mm, the initial design achieves a focusing degree of 0.75.

[0074] When iteratively optimizing the phase component based on acoustic energy focusing, a gradient descent optimization method is employed. In each iteration, the phase component is fine-tuned, and a new focusing degree is calculated until the focusing degree reaches a preset threshold or the maximum number of iterations is reached. Typically, after 5-10 iterations, the focusing degree can be improved to above 0.85. Utilizing the optimized ultrasonic excitation signal, acoustic energy focusing cleaning is achieved in localized cleaning dead zones, significantly improving cleaning efficiency and cleanliness; for example, increasing the particle removal rate in a 10 square centimeter area from 85% to over 95%.

[0075] This invention relates to a system for adaptive adjustment of ultrasonic parameters and cleaning evaluation in wafer cleaning processes, the system comprising: The first unit is used to collect wafer surface temperature distribution and cavity vibration response data during the cleaning process, analyze the cavity vibration response data to obtain the main vibration frequency and vibration mode, combine the main vibration frequency with the wafer surface structural characteristics to calculate the cavity resonant frequency, and calculate the sound field energy distribution based on the cavity resonant frequency and the vibration mode. The second unit is used to calculate the temperature gradient and sound intensity gradient of the wafer surface based on the temperature distribution and the sound field energy distribution, obtain the driving force boundary conditions based on the temperature gradient and the sound intensity gradient, obtain the micro-flow field distribution of the cleaning fluid by combining fluid dynamics simulation calculations, establish a cavitation bubble collapse feature matrix based on the micro-flow field distribution of the cleaning fluid, extract the collapse mode through the singular value decomposition of the cavitation bubble collapse feature matrix, identify local cleaning dead zones in real time according to the spatiotemporal distribution of the collapse mode, and control the piezoelectric transducer and ultrasonic reflector to generate phase-adjustable focused ultrasonic waves to focus the sound energy on the local cleaning dead zones. The third unit is used to monitor changes in the molecular layer adsorption characteristics of the wafer surface in real time during the cleaning process using a plasma resonance sensor, and to measure the change in the resonance peak displacement of the local surface plasma resonance angle over time. Based on the changes in molecular layer adsorption characteristics and the changes in resonance peak displacement, the thickness of the surface adsorption layer and the concentration of adsorbed molecules are calculated. Combined with the surface wettability data obtained by the contact angle measuring device, a quantitative evaluation curve of surface cleanliness is generated. A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0076] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0077] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for adaptive adjustment of ultrasonic parameters and cleaning evaluation in wafer cleaning processes, characterized in that, include: During the cleaning process, wafer surface temperature distribution and cavity vibration response data are collected. The vibration principal frequency and vibration mode are obtained by analyzing the cavity vibration response data. The cavity resonant frequency is calculated by combining the vibration principal frequency with the wafer surface structural characteristics. The sound field energy distribution is calculated based on the cavity resonant frequency and the vibration mode. Based on the temperature distribution and the sound field energy distribution, the temperature gradient and sound intensity gradient of the wafer surface are calculated. The driving force boundary conditions are obtained based on the temperature gradient and the sound intensity gradient. The microfluidic field distribution of the cleaning fluid is obtained by combining fluid dynamics simulation calculations. A cavitation bubble collapse feature matrix is ​​established based on the microfluidic field distribution of the cleaning fluid. The collapse mode is extracted by singular value decomposition of the cavitation bubble collapse feature matrix. Local cleaning dead zones are identified in real time based on the spatiotemporal distribution of the collapse mode. The piezoelectric transducer and ultrasonic reflector are controlled to generate phase-adjustable focused ultrasonic waves to focus the sound energy on the local cleaning dead zones. During the cleaning process, the changes in the molecular layer adsorption characteristics of the wafer surface are monitored in real time by a plasma resonance sensor, and the changes in the resonance peak displacement of the local surface plasma resonance angle over time are measured. Based on the changes in the molecular layer adsorption characteristics and the changes in the resonance peak displacement, the thickness of the surface adsorption layer and the concentration of adsorbed molecules are calculated, and combined with the surface wettability data obtained by the contact angle measuring device, a quantitative evaluation curve of surface cleanliness is generated.

2. The method according to claim 1, characterized in that, Based on the analysis of the cavity vibration response data, the dominant vibration frequency and vibration mode are obtained. The cavity resonant frequency is calculated by combining the dominant vibration frequency with the wafer surface structure characteristics. The sound field energy distribution is calculated based on the cavity resonant frequency and the vibration mode, including: The cavity vibration response data is subjected to Fast Fourier Transform to obtain a frequency domain signal, and the vibration main frequency is extracted from the frequency domain signal. Wavelet multi-resolution analysis is performed on the cavity vibration response data to obtain the energy distribution of different frequency bands. Based on the energy distribution, energy vibration frequency bands with an energy ratio exceeding a preset frequency band energy threshold are selected. The energy vibration frequency bands are subjected to Hilbert-Huang transform to extract vibration mode feature vectors and eigenvalue decomposition to obtain vibration modes. The height distribution data and size characteristic data of the wafer surface structure are obtained as the boundary conditions for acoustic wave propagation. Based on the acoustic wave propagation boundary conditions and the dominant vibration frequency, the eigenvalues ​​are solved by the modal iteration method. A correction coefficient is introduced in combination with the wafer surface structure characteristics. The cavity resonant frequency is obtained by multiplying the eigenvalues ​​by the correction coefficient. Using the resonant frequency of the cavity as excitation, acoustic-structural coupling iterative calculations are performed based on the vibration modes to obtain three-dimensional sound pressure distribution data; the density parameter of the cleaning fluid inside the cavity is obtained, and the sound field propagation impedance matrix is ​​calculated based on the cleaning fluid density parameter. The three-dimensional sound pressure distribution data and the sound field propagation impedance matrix are convolved to obtain the local sound field impedance. The local sound field impedance is multiplied by the three-dimensional sound pressure distribution data and then divided by the square of the cleaning fluid density to obtain the sound field energy density distribution.

3. The method according to claim 1, characterized in that, Based on the temperature gradient and the acoustic intensity gradient, the driving force boundary conditions are obtained. Combined with fluid dynamics simulation calculations, the microfluidic field distribution of the cleaning fluid is obtained. Based on the microfluidic field distribution of the cleaning fluid, a cavitation bubble collapse characteristic matrix is ​​established, including: The thermocapillary force is obtained by taking the negative value of the product of the temperature gradient and the surface tension coefficient of the temperature; the acoustic radiation force is obtained by multiplying the acoustic intensity gradient by the acoustic field attenuation coefficient of the cleaning fluid; the total driving force is obtained by adding the thermocapillary force and the acoustic radiation force; the total driving force is applied to the boundary of the fluid computational domain to establish the fluid boundary driving force distribution; the fluid boundary driving force distribution is used as the driving force boundary condition. The pressure correction amount is obtained by combining the driving force boundary conditions with the initial pressure distribution of the fluid. The pressure field is updated according to the pressure correction amount. The calculation is iterated until the difference between two adjacent pressure fields is less than the convergence accuracy threshold, and the microfluidic field distribution of the cleaning fluid is obtained. The fluid shear stress distribution is extracted from the microfluidic field distribution of the cleaning fluid, and dimensionless parameters are calculated based on the fluid shear stress distribution. The internal pressure and bubble wall acceleration of the bubbles in the cleaning fluid are calculated based on the dimensionless parameters. The bubble wall acceleration is integrated to obtain the bubble wall velocity. The bubble wall velocity is multiplied by the internal pressure of the bubble to obtain the collapse power density. The collapse power density is combined with the bubble collapse time to form a cavitation bubble collapse feature matrix.

4. The method according to claim 1, characterized in that, Collapse modes are extracted through singular value decomposition of the cavitation bubble collapse feature matrix, and local cleaning dead zones are identified in real time based on the spatiotemporal distribution of the collapse modes, including: The cavitation bubble collapse feature matrix is ​​decomposed by singular value to obtain a spatiotemporal feature vector matrix and a singular value diagonal matrix. The cumulative contribution rate is obtained by calculating the ratio of the sum of squares of the singular values ​​in the singular value diagonal matrix to the total sum of squares. The spatiotemporal feature vector matrices corresponding to the singular values ​​whose cumulative contribution rate exceeds the preset contribution ratio are selected to form the collapse mode. The local collapse intensity is obtained by summing the products of each column vector of the spatiotemporal feature vector matrix with the corresponding singular values. The local collapse intensity is normalized to obtain the collapse energy weight. The collapse intensity gradient field is obtained by multiplying the collapse energy weight with the local collapse intensity. The local cleaning dead zone is determined based on the local minimum point of the collapse intensity gradient field.

5. The method according to claim 4, characterized in that, Controlling the piezoelectric transducer and ultrasonic reflector to generate phase-tunable focused ultrasonic waves, and focusing the acoustic energy onto the localized cleaning dead zone includes: The collapse period is obtained by performing spectral analysis on the spatiotemporal feature vector matrix, and the product of the collapse period and the preset dwell coefficient is used as the sound field switching period. The phase delay distribution of the piezoelectric transducer is calculated and used as the excitation signal. Simultaneously, the radius of curvature of the ultrasonic reflector is adjusted according to the spatial coordinates of the local cleaning dead zone. A phase-adjustable focused ultrasonic wave is emitted according to the excitation signal and the radius of curvature, and the acoustic energy is focused according to the spatial coordinates of the local cleaning dead zone. When the focusing time reaches the sound field switching cycle, the focused ultrasonic wave is switched to the next local cleaning dead zone.

6. The method according to claim 5, characterized in that, The phase delay distribution of the piezoelectric transducer is calculated, and the phase delay distribution is used as the excitation signal. Simultaneously, the radius of curvature of the ultrasonic reflector is adjusted according to the spatial coordinates of the local cleaning dead zone. A phase-adjustable focused ultrasonic wave is emitted according to the excitation signal and the radius of curvature, and the focused acoustic energy includes the following based on the spatial coordinates of the local cleaning dead zone: The initial phase delay distribution of the piezoelectric transducer is calculated based on the relationship between the coordinates of any point on the surface of the piezoelectric transducer and the focal length. The phase delay correction amount is determined in combination with the characteristic parameters of the piezoelectric material. The initial phase delay distribution and the phase delay correction amount are added together to obtain the target phase delay distribution. A coordinate mapping matrix is ​​established based on the target phase delay distribution, and the spatial coordinates of the local cleaning dead zone are converted into mapped coordinates through the coordinate mapping matrix. The initial radius of curvature of the ultrasonic reflector is calculated based on the mapped coordinates. The sound field pressure distribution at the mapped coordinates is obtained. The first and second derivatives of the sound field pressure distribution are multiplied by the corresponding adjustment coefficients and summed to obtain the radius of curvature adjustment amount. The initial radius of curvature and the radius of curvature adjustment amount are superimposed to obtain the target radius of curvature. The target phase delay distribution is converted into a phase delay term, the target radius of curvature is converted into a phase modulation term, and the phase delay term and the phase modulation term are combined to generate an excitation signal; The ultrasonic wave is excited at the mapped coordinates according to the excitation signal. The ratio of the sound intensity of the target area at the mapped coordinates to the sound intensity of the entire sound field area is calculated to obtain the sound energy focusing degree. The phase component is iteratively optimized based on the sound energy focusing degree to obtain the optimized ultrasonic excitation signal. The optimized ultrasonic excitation signal is used to achieve sound energy focusing cleaning of the local cleaning dead zone.

7. An adaptive ultrasonic parameter adjustment and cleaning evaluation system for wafer cleaning processes, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to collect wafer surface temperature distribution and cavity vibration response data during the cleaning process, analyze the cavity vibration response data to obtain the main vibration frequency and vibration mode, combine the main vibration frequency with the wafer surface structural characteristics to calculate the cavity resonant frequency, and calculate the sound field energy distribution based on the cavity resonant frequency and the vibration mode. The second unit is used to calculate the temperature gradient and sound intensity gradient of the wafer surface based on the temperature distribution and the sound field energy distribution, obtain the driving force boundary conditions based on the temperature gradient and the sound intensity gradient, obtain the micro-flow field distribution of the cleaning fluid by combining fluid dynamics simulation calculations, establish a cavitation bubble collapse feature matrix based on the micro-flow field distribution of the cleaning fluid, extract the collapse mode through the singular value decomposition of the cavitation bubble collapse feature matrix, identify local cleaning dead zones in real time according to the spatiotemporal distribution of the collapse mode, and control the piezoelectric transducer and ultrasonic reflector to generate phase-adjustable focused ultrasonic waves to focus the sound energy on the local cleaning dead zones. The third unit is used to monitor the changes in molecular layer adsorption characteristics on the wafer surface in real time during the cleaning process using a plasma resonance sensor, and to measure the changes in the resonance peak displacement of the local surface plasma resonance angle over time; based on the changes in molecular layer adsorption characteristics and the changes in resonance peak displacement, the thickness of the surface adsorption layer and the concentration of adsorbed molecules are calculated, and combined with the surface wettability data obtained by the contact angle measuring device, a quantitative evaluation curve of surface cleanliness is generated.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images