Pr film thickness measurement method based on multi-modal data fusion and light absorption collaborative correction
By employing a multimodal data fusion and optical absorption co-correction method, the problem of large measurement error in PR adhesive film thickness was solved, achieving high-precision and rapid film thickness measurement, which is suitable for online inspection in semiconductor photolithography processes.
Patent Information
- Application Number
- CN202611195224.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-07
- Publication Date
- 2026-11-10
AI Technical Summary
In existing semiconductor lithography processes, the limitations of a single data source, fixed optical parameters, and lack of coupled correction for multiple interferences lead to large measurement errors in the PR resist layer thickness, especially under conditions of high absorption and high roughness. Furthermore, the lack of absorption constraints in the fitting process results in a high divergence rate.
A method combining multimodal data fusion and optical absorption co-correction is adopted. By measuring the initial film thickness and optical absorption coefficient, multimodal optical data is collected by combining broadband light source, coupled correction and synchronous correction are performed, a dedicated fitting function is configured and iterative solution is obtained, and temperature-absorption co-compensation is performed to achieve consistency of optical parameters and fitting accuracy.
It achieves high-precision PR resist film thickness measurement across the entire process stage and absorption range, with a measurement error ≤0.12nm and a single measurement time ≤50ms, significantly improving efficiency and providing strong anti-interference capabilities, thus meeting the requirements for online inspection of photolithography processes.
Smart Images

Figure CN123065359A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor manufacturing process testing technology. Background Technology
[0002] In semiconductor photolithography, the uniformity of the PR (photoresist) layer thickness directly determines the accuracy of the exposure pattern. Furthermore, the light absorption characteristics of the PR layer, especially thick layers, after exposure can significantly interfere with optical measurement signals. Existing technologies suffer from the following core defects: 1. Limitations of a single data source: The ellipsometer method relies solely on polarization parameters and does not incorporate the light absorption coefficient. Correction is needed because thick adhesive layers (>5μm) cause phase difference calculation deviations due to light absorption, resulting in measurement errors exceeding ±0.5nm; the spectral reflectance method only utilizes reflectance intensity, which is insufficient for high-absorption PR adhesive layers ( >0.1μm -1 The signal attenuation compensation is insufficient, and the detection accuracy of thin films below 500nm is less than 85%.
[0003] 2. Fixed optical parameters: using a fixed refractive index With extinction coefficient The PR adhesive layer was not considered during the application of the adhesive. 0.01 0.05μm -1 ),exposure( 0.05 0.2μm -1 ),development( 0.2 0.5μm -1 The absorption characteristics change during the ) stage, resulting in poor parameter adaptability.
[0004] 3. Multi-interference uncoupled correction: Ignore surface roughness Film thickness and light absorption coefficient Cross-interference, correcting a single parameter individually leads to error superposition, and high absorption + high roughness cause the measurement deviation of PR adhesive layer to exceed 1nm.
[0005] 4. No absorption constraint in fitting: A single fitting function (such as polynomial fitting) is used across the entire range, without optimization for different absorption intervals. The thick adhesive layer has a fitting divergence rate of over 5% due to fluctuations in the absorption coefficient. Summary of the Invention
[0006] To address the aforementioned technical problems, this invention provides a method for measuring the thickness of PR film using multimodal data fusion and coordinated optical absorption correction.
[0007] A method for measuring PR film thickness using multimodal data fusion and optical absorption co-correction includes the following steps: After the wafer to be tested is fixed, the preliminary film thickness and the preliminary light absorption coefficient at the characteristic wavelength are measured for light source wavelength adaptation. A broadband light source illuminates the wafer under test, and multimodal optical data parameters are synchronously acquired and then coupled and corrected. The multimodal optical data includes polarization parameters, reflection intensity spectrum, and light absorption coefficient spectrum. Simultaneous correction of refractive index, absorption coefficient, and extinction coefficient ensures consistency of optical parameters; Based on the initial film thickness, intervals are divided, and each interval is configured with a dedicated fitting function, initial value, and light absorption coefficient constraint threshold. The solution is obtained iteratively through a multi-weighted objective function containing an absorption term. Temperature-absorption synergistic compensation is performed to dynamically correct measurement results and output film thickness data and measured light absorption coefficient values that meet the standards.
[0008] Technical effects: This invention utilizes "reflection" polarization "Absorption" integrates multi-dimensional optical data to construct light absorption Refractive index synergistic correction is employed, extracting roughness through high-frequency fluctuations in the reflection spectrum and simultaneously eliminating cross-interference among the three factors. An absorption-constrained piecewise fitting strategy is designed, dividing the PR adhesive layer into three intervals based on its thickness, with each interval configured with a dedicated fitting function, initial values, and... The constraint threshold is solved iteratively using a multi-weighted objective function containing an absorption term to address the accuracy of fitting across the entire range. Speed is a contradiction. Achieving full-process PR adhesive coating (adhesive application) exposure Development), total absorption range ( 0.01 0.5μm -1 The high-precision measurement has an error of ≤0.12nm and a single measurement time of ≤50ms, significantly improving efficiency. Attached Figure Description
[0009] Figure 1 This is a flowchart illustrating the overall process of this invention.
[0010] Figure 2 This is a comparison chart showing the effects before and after coupling correction.
[0011] Figure 3 The graph shows the wavelength and corresponding fitted function values for the three film thickness intervals in S4.
[0012] Figure 4 The graph shows the relationship between the number of iterations and the film thickness error when performing piecewise fitting of the absorption constraint in S4. Detailed Implementation
[0013] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments; therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.
[0014] S1 and PR adhesive layer wafers are fixed to a six-degree-of-freedom fine-tuning platform by electrostatic adsorption, with a positioning accuracy of ±0.5μm. The spectral ellipsometry is set at 1mm. Preliminary film thickness measured at 1mm sampling interval For parameter adaptation; temperature sensor collects ambient temperature. Preliminary light absorption coefficient at the characteristic wavelength (633nm) was collected. Used for wavelength adaptation of light sources: i.e. <1000nm and <0.05μm -1 The wavelength of the light source is selected as 380nm. ≥1000nm or ≥0.05μm -1 The wavelength of the light source is selected as 650nm.
[0015] S2. Multimodal optical data acquisition and coupling correction processing: A broadband light source is used to illuminate the wafer under test for multimodal optical data parameter acquisition. The broadband light source is incident at a 15° angle with a power density of 5 mW / cm². 2 This avoids damage to the adhesive layer and allows for the simultaneous acquisition of polarization parameters. , 400 1000nm reflection intensity and measured values of light absorption coefficient Sampling interval 5nm; data acquisition time ≤20ms, synchronously stored as " matrix spectral vector Spectral vector.
[0016] The coupling correction specifically refers to: First, dark field correction eliminates dark current. 4-wavelet 3-layer noise reduction; A modified model trained using images of 1000 wafers with varying PR adhesive thicknesses calculates surface roughness. : , The extinction coefficient is set to 0.02. It is a variance operation function. The collected reflection intensity The high-frequency components, the original reflection intensity after High-frequency band signals extracted after wavelet 3-level decomposition are used to characterize surface scattering features. Reflection intensity coupling modification: , The reflection intensity is after coupling correction, in au; The original reflected intensity is the input signal directly acquired by the spectrometer, expressed in au. The roughness interference correction coefficient is set to 0.05 to compensate for the light intensity attenuation caused by surface scattering. This is the light absorption interference correction coefficient, set to 0.01, to compensate for the light intensity attenuation caused by light absorption in the adhesive layer; The wavelength of the incident light is in nm and corresponds one-to-one with the wavelength of the current sampling point. This is the preliminary light absorption coefficient, in μm. -1 ; For preliminary measurement of film thickness, the unit is μm.
[0017] Revised and , , The comparison of the effects before and after normalization alignment and coupling correction is shown in the figure below. Figure 2 As shown.
[0018] Table 1. Comparison of Coupling Correction Effects: Original Reflection Intensity Data Before Normalization
[0019] S3, Optical Absorption-Refractive Index Synergistic Correction: First, dynamic parameters are calculated, including the refractive index correction parameter, and the final corrected refractive index is output. , , in, This is a temperature correction term; This is a process stage correction item. For process weights: For absorption characteristic correction items; Coating = 0.3, Exposure = 0.5, Development = 0.2; The reference refractive index is obtained by interpolation from a wafer database. : Temperature-induced refractive index shift, dimensionless, compensates for refractive index changes caused by an ambient temperature deviation of 25℃; : Refractive index shift caused by process stage, dimensionless, compensated by coating exposure Differences in the physical properties of the gel layer at different stages of development; : The refractive index shift caused by light absorption characteristics, dimensionless, and the dispersion relationship between the correlation absorption coefficient and the refractive index; The final extinction coefficient synchronous correction formula is as follows, and the corrected final extinction coefficient is output. , , The baseline extinction coefficient is obtained by interpolation from the wafer database; The refractive index shift caused by absorption characteristics, and the refractive index correction formula. Same origin, ensuring consistency of optical parameters.
[0020] S4. Absorption constraint fitting: Configure specific fitting parameters based on the initial film thickness range. The wavelength and corresponding fitting function value curves within the three film thickness ranges are shown below. Figure 3 As shown, 1. When the film thickness is in the range of 0~1μm, a quadratic polynomial is used to fit the curve: Quadratic polynomial: , The incident wavelength is in nm. The values are the fitted function values; only the coefficients of the quadratic terms are marked in the formula, while the linear and constant terms are determined by the initial values of the interval.
[0021] 2. When the film thickness is in the range of 1~5μm, use Gaussian... Lorentz mixture function fitting curve: Gauss Lorentz mixing function: , The incident wavelength is in nm; the center wavelength is 600 nm, and the full width at half maximum (FWHM) is 100 nm; the Gaussian and Lorentz components each have a weight of 0.75. 3. When the film thickness is in the range of 5~10μm, a piecewise linear function is used to fit the curve: when At <600 nm: when When ≥ 600 nm: The incident wavelength is in nm; with 600 nm as the segment node, different linear slopes are used for the high and low bands.
[0022] Table 2. Correspondence between different film thickness ranges and fitting functions
[0023] Perform iterative analysis of the objective function with absorbing constraints, using... The algorithm's iteration termination condition is set to two iterations. If the difference is ≤0.005nm or the number of iterations is ≥50, the relationship between the number of iterations and the film thickness error is shown in the figure below. Figure 4 As shown.
[0024] Multi-weighted objective function: , in, To fit the objective function value, the smaller the value of the minimization objective is, the higher the fitting accuracy. The film thickness value to be solved is the core variable for iteration, and its unit is nm; The weight for the reflection intensity term is set to 0.5. This is the weight for the polarization phase term, with a fixed value of 0.3; This is the weight of the light absorption term, with a fixed value of 0.2; Corrected measured reflection intensity, and the measured signal corresponding to the film thickness; Theoretical calculation of reflection intensity, based on film thickness. Final absorption coefficient Calculated values from the optical thin film model; : Measured light absorption coefficient, absorption coefficient obtained from spectral acquisition; Theoretical calculation of polarization phase difference based on film thickness Final absorption coefficient Calculated values from the elliptic model; : Measured light absorption coefficient, absorption coefficient obtained from spectral acquisition; Theoretical calculation of light absorption coefficient, based on the current film thickness. The absorption model calculates the value and uses it to determine the absorption interval constraint. If during iteration... If the value exceeds the interval threshold, the step size is automatically reduced by 50% to ensure convergence of the fit.
[0025] S5, Temperature-Absorption Synergistic Compensation, specifically includes: 1. Thermal expansion compensation: , : Film thickness after temperature compensation, intermediate output, unit: nm; The optimal film thickness obtained through fitting and iteration is expressed in nm. 1.2×10 -5 ℃ is The coefficient of thermal expansion of the adhesive; 2. Absorption and compensation: , : Final film thickness after absorption compensation, basic output, in nm; : Film thickness after temperature compensation, calculated and output in the previous step, in nm; The corrected final light absorption coefficient, in μm. -1 ; This is the absorption compensation coefficient, used to compensate for the measurement deviation in film thickness caused by the absorption effect. 0.1 is the reference absorption coefficient, a baseline value, in μm. -1 ; 3. Thick adhesive stress additional compensation type: At that time, superimposed stress compensation term , This is the process stress coefficient, with a value ranging from 0 to 0.5.
[0026] Output parameters, including final film thickness , Accuracy ±0.12nm Measurement time, temperature compensation, and process compatibility score (0-10 points); data format conforms to... Standard, including wafers PR adhesive process stage, thickness-absorption thermal map (1mm×1mm resolution); when film thickness uniformity deviation >5% or Fluctuation > 0.1 μm -1 At that time, an audible and visual warning will be triggered.
[0027] Furthermore, it provides a self-calibration mechanism, allowing daily calibration using standard calibration sheets containing 0.5μm / 3μm / 8μm PR adhesive layers. =0.05 / 0.15 / 0.3μm -1 Calibration parameters are set to Measurement deviation ≤ 0.002 Measurement deviation ≤ 0.005 μm -1 Film thickness deviation ≤ 0.05nm; environmental compensation is set to adjust light source power (0.015 / ℃) for every 1℃ change in temperature; and to adjust humidity for every 5% change. Data (0.003 / 5%).
[0028] The following specific embodiment further clarifies the overall process of the present invention.
[0029] Using a 300mm silicon-based wafer with a 6μm thick positive PR resist layer, the absorption coefficient after exposure is... =0.25μm -1 Surface roughness Using 0.4nm as a detection example, testing was conducted using a 650nm light source at an ambient temperature of 25.5℃. Wafer fixation and initial parameter measurements were completed. =5.98μm, =0.24μm -1 And obtain through database interpolation =1.62、 =0.0015、 =0.23μm -1 Subsequently, within 18ms, Ψ=45.32° was synchronously acquired. =121.8° and the original light intensity in the range of 400~1000nm and Spectrum. Extracted using wavelet denoising. =0.38nm and after implementing coupling correction, The deviation elimination rate reached 92%.
[0030] Based on this, a collaborative correction is performed, and the result is calculated. =1.6222、 =0.2425μm -1 Then, it undergoes segmented fitting (range 5~10μm, initial thickness) =7μm, using a piecewise linear function Constraint ≥ 0.2μm -1 After 35 iterations, it converged to... =6.002nm, and finally obtained after temperature compensation (0.0036nm) and absorption compensation (0.006nm) to obtain =6.0116nm.
[0031] Verification results show that the measured value has an error of only 0.0016 nm compared to the standard film thickness of 6.01 nm, meeting the accuracy requirement of ±0.12 nm; when In 0.2-0.3μm -1The measurement error is ≤0.08nm when fluctuating within the range, demonstrating good absorption adaptability; the total time for a single measurement is 45ms, which is 40% more efficient than traditional non-absorption correction algorithms; and in With an accuracy of ≤0.5nm and an error of ≤0.1nm under temperature fluctuation of ±2℃, its anti-interference capability meets the requirements for online inspection of photolithography processes.
[0032] All content not described in detail in this specification belongs to the prior art known to those skilled in the art. Furthermore, for those skilled in the art, there will be changes in specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
Claims
1. A method for measuring the thickness of PR film using multimodal data fusion and coordinated optical absorption correction, characterized in that, Includes the following steps: After the wafer to be tested is fixed, the preliminary film thickness and the preliminary light absorption coefficient at the characteristic wavelength are measured for light source wavelength adaptation. A broadband light source illuminates the wafer under test, and multimodal optical data parameters are synchronously acquired and then coupled and corrected. The multimodal optical data includes polarization parameters, reflection intensity spectrum, and light absorption coefficient spectrum. The refractive index, absorption coefficient, and extinction coefficient are simultaneously corrected to ensure the consistency of optical parameters; Based on the initial film thickness, intervals are divided, and each interval is configured with a dedicated fitting function, initial value, and light absorption coefficient constraint threshold. The solution is obtained iteratively through a multi-weighted objective function containing an absorption term. Temperature-absorption synergistic compensation is performed to dynamically correct measurement results and output film thickness data and measured light absorption coefficient values that meet the standards.
2. The PR film thickness measurement method based on multimodal data fusion and optical absorption co-correction according to claim 1, characterized in that, The coupling correction specifically refers to: First, dark field correction eliminates dark current. 4-wavelet 3-layer noise reduction; Surface roughness is calculated by modifying the model. , The extinction coefficient is set to 0.
02. It is a variance operation function. The original reflection intensity collected The high-frequency components, the original reflection intensity after The high-frequency band signal extracted after wavelet three-level decomposition is used to characterize surface scattering features.
3. The PR film thickness measurement method based on multimodal data fusion and optical absorption co-correction according to claim 2, characterized in that, Reflection intensity coupling modification , The reflection intensity is after coupling correction, in au; The original reflected intensity is expressed in au. The roughness interference correction coefficient is set to 0.05 to compensate for the light intensity attenuation caused by surface scattering. This is the light absorption interference correction coefficient, set to 0.01, to compensate for the light intensity attenuation caused by light absorption in the adhesive layer; The wavelength of the incident light is in nm and corresponds one-to-one with the wavelength of the current sampling point. This is the preliminary light absorption coefficient, in μm. -1 ; For preliminary measurement of film thickness, the unit is μm.
4. The PR film thickness measurement method based on multimodal data fusion and optical absorption co-correction according to claim 1, characterized in that, The synchronization correction specifically refers to: First, dynamic parameters are calculated, including the refractive index correction parameter, and the final corrected refractive index is output. , , in, This is a temperature correction term; This is a process stage correction item. For process weights: For absorption characteristic correction items; Coating = 0.3, Exposure = 0.5, Development = 0.2; The reference refractive index is obtained by interpolation from a wafer database. : Temperature-induced refractive index shift, dimensionless, compensates for refractive index changes caused by an ambient temperature deviation of 25℃; : Refractive index shift caused by process stage, dimensionless, compensated by coating exposure Differences in the physical properties of the gel layer at different stages of development; : The refractive index shift caused by light absorption characteristics, dimensionless, and the dispersion relationship between the correlation absorption coefficient and the refractive index; Final output extinction coefficient after correction , , The baseline extinction coefficient is obtained by interpolation from the wafer database; The refractive index shift caused by absorption characteristics, and the refractive index correction formula. Same origin, ensuring consistency of optical parameters.
5. The PR film thickness measurement method based on multimodal data fusion and optical absorption co-correction according to claim 1, characterized in that, The specific fitting function is as follows: When the film thickness is in the range of 0~1μm, a quadratic polynomial is used to fit the curve: Quadratic polynomial: , The incident wavelength is in nm. The values are for the fitted function; only the coefficients of the quadratic term are marked in the formula, while the linear term and the constant term are determined by the initial value constraints of the interval. When the film thickness is in the range of 1~5μm, a Gaussian film is used. Lorentz mixture function fitting curve: , The incident wavelength is in nm; the center wavelength is 600 nm, and the full width at half maximum (FWHM) is 100 nm; the Gaussian component and the Lorentz component each have a weight of 0.
75. When the film thickness is in the range of 5~10 μm, a piecewise linear function is used to fit the curve: when When < 600 nm: , when When ≥ 600 nm: , The incident wavelength is in nm; with 600 nm as the segment node, different linear slopes are used for the high and low bands.
