A method and system for detecting dopant element distribution based on hyperspectral imaging technology

Abstract

This application relates to a method and system for detecting the distribution of doped elements based on hyperspectral imaging technology. The method for detecting the distribution of doped elements based on hyperspectral imaging technology includes the following steps: irradiating the ion implantation region on the back side of a wafer with light of different wavelengths; capturing the spectral characteristic data of the doped elements in the ion implantation region on the back side of the wafer using a hyperspectral imaging device, wherein the spectral characteristic data includes spatial-spectral cubic data; identifying the three-dimensional distribution of the doped elements based on the spectral characteristic data and using a spectral matching method; and matching the spectral characteristic data with a quantitative relationship model between spectral intensity and doping concentration to obtain the doping concentration of the doped elements.

Description

Technical Field

[0001] This application relates to the field of wafer manufacturing technology, and in particular to a method and system for detecting the distribution of doped elements based on hyperspectral imaging technology. Background Technology

[0002] In semiconductor manufacturing processes, ion implantation is a crucial step in achieving device doping, and the three-dimensional distribution of dopant elements directly affects device performance. Traditional detection methods primarily rely on sheet resistance measurement, which only obtains the average surface resistance and cannot acquire information on the depth distribution, concentration gradient, and lateral distribution of dopant elements. For back-side ion implantation processes on wafers, such as in radio frequency devices or 3D packaging applications, precise control of doping depth and distribution is particularly important. The limitations of traditional methods can easily lead to device performance parameters deviating from design targets.

[0003] While existing technologies such as secondary ion mass spectrometry and transmission electron microscopy can provide depth distribution information, these methods require destructive processing of samples, have long detection cycles, and are costly, failing to meet the real-time monitoring needs of production lines. Furthermore, these methods cannot provide two-dimensional distribution information of doped elements on the wafer surface, making it difficult to comprehensively assess the quality of the ion implantation process. Summary of the Invention

[0004] This application provides a method and system for detecting the distribution of doped elements based on hyperspectral imaging technology, which has the advantages of non-destructive and high-resolution acquisition of the three-dimensional distribution and concentration gradient of doped elements.

[0005] To achieve the above objectives, in a first aspect, this application provides a method for detecting the distribution of doped elements based on hyperspectral imaging technology, comprising:

[0006] The ion implantation area on the back side of the wafer is illuminated with light of different wavelengths;

[0007] The spectral characteristic data of doped elements in the ion-implanted region on the back side of the wafer are captured using a hyperspectral imaging device. This spectral characteristic data includes spatial-spectral cubic data.

[0008] The three-dimensional distribution of doping elements is identified based on the spectral feature data and by using a spectral matching method; and the spectral feature data is matched with a quantitative relationship model between spectral intensity and doping concentration to obtain the doping concentration of the doping elements.

[0009] In some embodiments of this application, after the steps of "identifying the three-dimensional distribution of doping elements based on the spectral feature data and using spectral matching; and matching the spectral feature data with a quantitative relationship model between spectral intensity and doping concentration to obtain the doping concentration of the doping elements", the method further includes the step of generating a two-dimensional doping concentration distribution map and a three-dimensional depth-concentration profile map and comparing them with the design value to achieve process monitoring.

[0010] In some embodiments of this application, the step of "identifying the three-dimensional distribution of doping elements based on the spectral feature data and using a spectral matching method; and matching the spectral feature data with a quantitative relationship model between spectral intensity and doping concentration to obtain the doping concentration of the doping element" includes: comparing the spectral feature data with a spectral feature library of doping elements to identify the type and location of the doping element; and inputting the spectral feature data into the quantitative relationship model between spectral intensity and doping concentration to calculate the doping concentration of the doping element.

[0011] In some embodiments of this application, in the step of "comparing the spectral feature data with the spectral feature library of doped elements to identify the type and location of doped elements", the depth distribution of doped elements is analyzed by multi-wavelength penetration depth difference analysis, with short wavelengths analyzing shallow layers and long wavelengths analyzing deep layers.

[0012] In some embodiments of this application, in the step of analyzing the doping depth distribution using multi-wavelength penetration depth differences, it is also necessary to combine the implantation energy and annealing process parameters during ion doping to analyze the depth distribution of the doped elements.

[0013] In some embodiments of this application, before the steps of "identifying the three-dimensional distribution of doping elements based on the spectral feature data and using spectral matching; and matching the spectral feature data with a quantitative relationship model between spectral intensity and doping concentration to obtain the doping concentration of the doping elements", the method further includes the step of preprocessing the acquired hyperspectral data; wherein the preprocessing includes at least one of denoising, background subtraction, and normalization.

[0014] In some embodiments of this application, the step of “capturing spectral characteristic data of specific doped elements in the ion implantation region on the back side of a wafer using a hyperspectral imaging device” includes: the incident angle of the light of different wavelengths is 30°-70°.

[0015] In some embodiments of this application, the establishment of the spectral feature library of doped elements includes: establishing characteristic absorption / reflection spectral curves of different doped elements in their activated states through experiments or simulations, wherein the characteristic absorption / reflection spectral curves include peak wavelength parameters and full width at half maximum (FWHM) parameters; and / or

[0016] The establishment of the quantitative relationship model between spectral intensity and doping concentration includes: using training data of samples calibrated with known concentrations of doping elements, and using machine learning or physical models to achieve concentration inversion.

[0017] Secondly, this application also provides a dopant distribution detection system based on hyperspectral imaging technology, comprising: a light source for emitting light of different wavelengths and illuminating the ion implantation region on the back side of a wafer; a hyperspectral camera for capturing spectral characteristic data of specific dopant elements in the ion implantation region on the back side of the wafer; an oblique incident optical module for adjusting the incident angle of the light emitted by the light source; and a data processing unit for storing a quantitative relationship model between spectral intensity and dopant concentration and a dopant element spectral characteristic library; comparing the spectral characteristic data with the dopant element spectral characteristic library to identify the type and location of the dopant element; inputting the spectral characteristic data into the quantitative relationship model between spectral intensity and dopant concentration to calculate the dopant concentration of the dopant element.

[0018] In some embodiments of this application, the data processing unit is further configured to generate a two-dimensional doping concentration distribution map and a three-dimensional depth-concentration profile map and compare them with the design values ​​to achieve process monitoring.

[0019] This application provides a method and system for detecting dopant distribution based on hyperspectral imaging technology. The detection method includes: irradiating the ion-implanted region on the back side of a wafer with light of different wavelengths; capturing spectral characteristic data of specific dopant elements in the ion-implanted region on the back side of the wafer using a hyperspectral imaging device, wherein the spectral characteristic data includes spatial-spectral cubic data; identifying the three-dimensional distribution of dopant elements based on the spectral characteristic data and using a spectral matching method; and matching the spectral characteristic data with a quantitative relationship model between spectral intensity and dopant concentration to obtain the dopant concentration of the dopant elements. By introducing a hyperspectral imaging device and combining it with a spectral matching method and a quantitative relationship model between spectral intensity and dopant concentration, this application can obtain information on the concentration gradient, implantation depth, and lateral distribution of specific dopant elements in the ion-implanted region on the back side of the wafer (three-dimensional distribution and concentration gradient of dopant elements) without destroying the wafer. Therefore, this application provides a method and system for detecting dopant distribution based on hyperspectral imaging technology with the advantages of non-destructive and high resolution.

[0020] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] To gain a more complete understanding of this application and its beneficial effects, the following description will be provided in conjunction with the accompanying drawings, wherein the same reference numerals in the following description denote the same parts.

[0023] Figure 1 This is a flowchart illustrating a method for detecting the distribution of doped elements based on hyperspectral imaging technology, provided for some exemplary embodiments of this application.

[0024] Figure 2 for Figure 1 The flowchart shown is a schematic diagram of step S2 in the flowchart of the method for detecting the distribution of doped elements based on hyperspectral imaging technology.

[0025] Figure 3 A schematic diagram of a dopant element distribution detection system based on hyperspectral imaging technology provided for some exemplary embodiments of this application.

[0026] Explanation of reference numerals in the attached figures:

[0027] 100. Detection system; 10. Light source; 20. Hyperspectral equipment; 30. Inclined incident optical module; 40. Data processing unit. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0029] In existing technologies, the semiconductor manufacturing field has long relied on the sheet resistance method to evaluate the doping effect after ion implantation. However, this method only reflects the average surface resistance and cannot reveal the depth distribution and concentration gradient of doped elements. For back-side implantation processes, precise control of doping depth and lateral distribution directly affects device performance. The two-dimensional detection characteristics of traditional methods result in a lack of effective data support for process parameter adjustments. While destructive testing methods can obtain depth information, they require wafer dicing and cannot meet the requirements for in-line inspection, leading to extended production cycles and increased costs.

[0030] To address the aforementioned issues, the inventors noted that hyperspectral imaging technology can simultaneously acquire spatial information of a target and continuous or quasi-continuous narrow-band spectral data. Its high-resolution spectra typically contain characteristic absorption or reflection peaks related to the material's chemical composition (such as the fingerprint characteristics of specific elements), but overlapping signals need to be analyzed using calibration data or algorithms. By studying the optical response characteristics of the active states of doped elements, it was discovered that different elements possess distinguishable absorption or reflection characteristics at specific wavelengths. Furthermore, it was found that differences in the penetration depth of light at different wavelengths can reflect depth distribution information. Combining this with the physical correlation between spectral intensity and concentration, a technical approach was formed to invert the three-dimensional distribution and concentration through optical signals. The following will provide a detailed description of the doped element distribution detection method based on hyperspectral imaging technology of this application, with reference to the accompanying drawings.

[0031] Please see Figure 1 This application provides a method for detecting the distribution of doped elements based on hyperspectral imaging technology, including the following steps:

[0032] S1, using light of different wavelengths to irradiate the ion implantation area on the back side of the wafer;

[0033] S3, capturing spectral characteristic data of doped elements in the ion-implanted region on the back side of the wafer using a hyperspectral imaging device, wherein the spectral characteristic data includes spatial-spectral cubic data; and

[0034] S4. Based on the spectral feature data, identify the three-dimensional distribution of the doping element using the spectral matching method; and match the spectral feature data with the quantitative relationship model between spectral intensity and doping concentration to obtain the doping concentration of the doping element.

[0035] Hyperspectral imaging equipment refers to an imaging device capable of simultaneously recording the spatial location and continuous or quasi-continuous spectral information of a target area. Specifically, it can be implemented using a combination of a beam-splitting prism and a planar detector. The beam-splitting element decomposes the incident light into hundreds of continuous bands, with each pixel corresponding to a complete spectral curve. This hyperspectral imaging equipment provides a data foundation for the analysis of the three-dimensional distribution of doped elements.

[0036] Among them, spectral feature data refers to a data cube that includes spatial location information and continuous or quasi-continuous band spectral response, acquired by a hyperspectral imaging device. Specifically, it can be achieved by scanning the back area of ​​the wafer line by line in the ultraviolet to near-infrared band range with a hyperspectral camera and recording the full-band reflectance or transmittance of each pixel. It is used to characterize the differences in spectral characteristics at different locations. Its spatial dimension information is used to locate the doped region, and its spectral dimension information is used to identify the type and concentration of elements.

[0037] Among them, the spectral matching method refers to the method of identifying the distribution of elements by comparing the similarity between the measured spectrum and the characteristic spectral library. Specifically, it can be implemented by using spectral angle mapping algorithm or least squares fitting algorithm, and the element type and spatial location are determined by calculating the angle or residual between the spectral curves.

[0038] Among them, the quantitative relationship model refers to the algorithm model that establishes a mathematical relationship between spectral data and doping concentration values. Specifically, it can be achieved by using machine learning methods to train the mapping relationship between the spectral features of known concentration samples and concentration labels, which is used to convert spectral intensity into quantitative concentration values.

[0039] Specifically, hyperspectral imaging equipment illuminates the back side of the wafer at a specific incident angle to reduce surface reflection interference and enhance the optical response signal of the implanted region. Each pixel in the acquired spatial-spectral cube data contains a spectral curve of continuous or quasi-continuous wavelengths. Spectral matching algorithms, such as spectral angle mapping and least-squares fitting, are used to compare the data with a pre-established dopant element feature library to identify the type and lateral distribution of dopant elements. Differences in penetration depth of different wavelengths are used to analyze depth distribution; short-wavelength light reflects shallow doping information, while long-wavelength light reflects deep distribution. After the spectral intensity data is input into a quantitative relationship model, the concentration value of each pixel is output based on the calibration curve or training model, ultimately forming concentration distribution data in three-dimensional spatial coordinates.

[0040] Compared to existing technologies, traditional sheet resistance methods only provide average surface resistance values ​​and cannot distinguish concentration differences at different depths and lateral positions. This application, however, utilizes the three-dimensional characteristics of spectral data to achieve depth distribution detection with micron-level spatial resolution. Compared to destructive detection methods, this application's dopant distribution detection method based on hyperspectral imaging technology can complete full-area scanning without damaging the wafer, and the detection process is compatible with production lines.

[0041] Through the above technical solution, this application solves the problem that traditional methods cannot obtain the depth distribution and concentration gradient of doped elements, realizing non-destructive three-dimensional distribution detection of doped elements. By combining spectral matching and quantitative models, concentration inversion is achieved while maintaining high spatial resolution, providing accurate data support for process parameter optimization. The longitudinal distribution is analyzed using differences in optical penetration depth, avoiding complex physical stripping steps and significantly improving detection efficiency and repeatability.

[0042] Please continue reading. Figure 1 In some embodiments of this application, after step S4, the doping element distribution detection method based on hyperspectral imaging technology further includes step S5: generating a two-dimensional doping concentration distribution map and a three-dimensional depth-concentration profile map and comparing them with the design value to achieve process monitoring.

[0043] The two-dimensional doping concentration distribution map is an image generated by mapping the horizontal spatial coordinates of hyperspectral data to the corresponding doping concentration values. Specifically, pseudo-color encoding techniques can be used to convert different concentration values ​​into color gradients for visualization, visually displaying the concentration gradient distribution of doped elements on the wafer surface. The three-dimensional depth-concentration profile map is a stereoscopic image generated by obtaining vertical depth information through multi-wavelength penetration depth difference analysis technology and reconstructing the concentration data corresponding to each depth layer. Specifically, tomographic imaging algorithms can be used to deconvolve different wavelength spectral data to reveal the concentration variation patterns of doped elements in the vertical direction. Design value comparison involves analyzing the differences between the measured two-dimensional concentration distribution data and the three-dimensional depth-concentration data and the preset distribution parameters in the process design document. Specifically, image registration techniques can be used to spatially align the measured data with the design template and calculate the deviation, providing real-time feedback on the process execution status.

[0044] Specifically, after preprocessing the spectral data from hundreds of bands acquired by hyperspectral imaging equipment, the characteristic spectral curve of each pixel is extracted. A spectral matching algorithm is used to identify the types of doping elements and determine their spatial distribution. Combined with a quantitative concentration model, the doping concentration value of each pixel is calculated, forming a two-dimensional concentration distribution map. Simultaneously, based on the differences in the penetration depth of different wavelengths of light in silicon material—short-wavelength data reflecting shallow concentration information and long-wavelength data reflecting deep concentration information—a three-dimensional depth-concentration profile is constructed through multi-wavelength tomography. The generated distribution map is then compared point-by-point with the target distribution range specified in the process design document. An alarm is triggered when the measured concentration distribution exceeds the design tolerance band, guiding real-time adjustments to process parameters.

[0045] Compared to existing technologies, traditional sheet resistance measurement can only obtain the average surface resistance value and cannot distinguish lateral distribution differences or depth-direction concentration gradients. Destructive detection methods such as secondary ion mass spectrometry require offline sampling and are time-consuming, taking several hours. This application achieves non-destructive detection through hyperspectral imaging technology. While maintaining wafer integrity, it can complete the two-dimensional concentration distribution and three-dimensional depth profile detection of the entire wafer within five minutes. The detection data is directly linked with the production line control system to achieve closed-loop process control.

[0046] Through the above technical solution, this application solves the technical bottleneck that traditional detection methods cannot obtain three-dimensional distribution information of doped elements online. It realizes the real-time identification of process deviations by visualizing the concentration distribution map, and feeds the detection results back to the ion implanter for dose compensation or scanning path optimization, effectively preventing device performance failure caused by abnormal doping distribution.

[0047] Please see Figure 2In some embodiments of this application, step S4 includes: step S41, comparing the spectral feature data with a dopant element spectral feature library to identify the type and location of the dopant element; and step S42, inputting the spectral feature data into the quantitative relationship model between spectral intensity and doping concentration to calculate the doping concentration of the dopant element.

[0048] Among them, the doped element spectral feature library refers to a pre-established database containing the characteristic absorption or reflection spectral curves of different doped elements in the activated state. Specifically, it can be realized by simulating the spectral response of different elements at a specific wavelength through experimental measurement or simulation, and is used to identify the type of element through spectral matching.

[0049] Specifically, after acquiring hyperspectral data, the similarity between the spectral curve of each pixel and the standard spectra in the feature library is first calculated. By matching the peak wavelength and half-width at half-maximum (WHM) parameter, the type of dopant element and its spatial distribution are determined. Then, the matched spectral data is input into a trained quantitative relationship model between spectral intensity and doping concentration. Based on the nonlinear relationship between spectral intensity and concentration established within the model, the concentration value of each pixel is output. Through layer-by-layer analysis of the spatial-spectral data, the three-dimensional distribution of the dopant element and its concentration gradient in the lateral and depth directions are finally obtained.

[0050] In this embodiment, in step S1, a light source is incident on the silicon wafer after backside ion implantation and annealing. The light source is a halogen lamp (400 nm-2500 nm), and the incident angle of the light source is 60°. In step S3, the resolution of the hyperspectral device is 5 μm / pixel. In step S41, characteristic bands (such as the 1050 nm reflection valley corresponding to phosphorus doping) are extracted, and an elemental distribution map is generated using spectral angle mapping. In step S42, the concentration distribution differences at different wavelengths (450 nm and 1000 nm) are compared to calculate the deviation between the doping depth and the design value.

[0051] Compared to existing technologies, traditional sheet resistance methods can only measure the average surface resistance and cannot distinguish element types or depth distributions. Destructive detection methods such as secondary ion mass spectrometry require wafer cutting and cannot be applied online. This solution utilizes a dual matching mechanism of spectral feature library and quantitative model to simultaneously achieve element identification, three-dimensional localization, and concentration inversion under non-destructive conditions, overcoming the limitations of single-dimensional detection in traditional technologies.

[0052] Through the above technical solution, this application can acquire the three-dimensional distribution and concentration gradient data of doped elements in the ion implantation region in real time during the semiconductor manufacturing process, avoid damaging the wafer structure, provide a quantitative basis with high spatial resolution for process parameter adjustment, and effectively improve the accuracy of doping depth and concentration control.

[0053] In some embodiments of this application, in the step of "comparing the spectral feature data with the spectral feature library of doped elements to identify the type and location of doped elements", the depth distribution of doped elements is analyzed by multi-wavelength penetration depth difference analysis, with short wavelengths analyzing shallow layers and long wavelengths analyzing deep layers.

[0054] The difference in penetration depth across wavelengths refers to the difference in the penetrating power of light of different wavelengths in semiconductor materials. This can be achieved using ultraviolet (UV) and near-infrared (NIIR) light; UV light has weaker penetrating power, while NIIR light has stronger penetrating power. Short wavelengths refer to light with a wavelength range of 300-450 nanometers, which can be achieved using UV light sources or blue-violet lasers. This type of light has a penetration depth of less than 100 nanometers in silicon. Long wavelengths refer to light with a wavelength range of 800-2500 nanometers, which can be achieved using halogen lamps or near-infrared lasers. This type of light can achieve a penetration depth in silicon that is on the micrometer scale.

[0055] Specifically, during the data acquisition phase, light of different wavelengths is sequentially projected onto the back of the wafer using a broadband light source, and a hyperspectral camera simultaneously acquires the reflection or absorption spectral data corresponding to each wavelength. During the data processing phase, for the same spatial location, spectral data corresponding to shorter wavelengths are extracted to analyze the distribution of shallow doped elements, while spectral data corresponding to longer wavelengths are extracted to analyze the distribution of deep doped elements. By superimposing the spectral matching results of different wavelength layers, a three-dimensional depth distribution model is constructed. For example, for a phosphorus-doped region, spectral data at a wavelength of 450 nm reflects the doping concentration within 50 nm of the surface, while spectral data at a wavelength of 1050 nm reflects the doping concentration gradient at a depth of 1 micrometer.

[0056] Compared to existing technologies, traditional spectral matching methods rely solely on a single wavelength or a fixed combination of bands for elemental identification, failing to distinguish doping distributions at different depths. This proposed method achieves tomographic analysis through differences in physical penetration depth, obtaining depth-dimensional information without damaging the sample. Existing spectral analysis techniques can only generate two-dimensional planar distribution maps, while this method directly constructs a three-dimensional depth-concentration profile through multi-wavelength data fusion.

[0057] Through the above technical solution, this application solves the problem that traditional detection methods cannot non-destructively analyze the depth distribution of ion-implanted dopant elements. By synergistically combining short and long wavelengths, information on shallow and deep dopant elements can be acquired simultaneously during the same detection process, achieving three-dimensional concentration gradient analysis from the surface to the interior. This method avoids destructive sampling and eliminates the information loss in the depth dimension inherent in traditional spectral matching, providing complete dopant distribution data for process monitoring.

[0058] In some embodiments of this application, in the step of analyzing the doping depth distribution using multi-wavelength penetration depth differences, it is also necessary to combine the implantation energy and annealing process parameters during ion doping to analyze the depth distribution of the doped elements.

[0059] The implantation energy refers to the initial implantation depth of ions determined by the accelerating voltage during ion implantation. This value can be obtained from process parameter records or equipment logs and is used to constrain the initial distribution position of doped elements.

[0060] Among them, the annealing process parameters refer to the temperature, time and atmosphere conditions during the annealing process. Specifically, temperature gradient data and time series data can be used to correct the influence of the diffusion behavior of doped elements caused by annealing on the depth distribution.

[0061] Specifically, in resolving the doping depth distribution, preliminary depth stratification information is first obtained through multi-wavelength spectral data. Then, the implantation energy parameter is used as a physical constraint on the initial depth distribution; for example, a higher implantation energy confines the initial position of the dopant element to a deeper region. Simultaneously, annealing process parameters are used to calculate the diffusion range of the dopant element during annealing. For instance, high-temperature annealing causes the dopant element to diffuse into shallower layers, thus adjusting the resolving results based on spectral penetration depth. By fusing process parameters with optical data, a depth resolving model incorporating physical laws is established, ensuring that the depth distribution results conform to both the spectral detection data and the physical mechanisms of ion implantation and annealing.

[0062] Compared to existing technologies, which rely solely on multi-wavelength transmission differences for depth analysis without considering the impact of actual process parameters on dopant element distribution, the analysis results may deviate from the actual physical process. For example, when annealing causes dopant element diffusion, simple spectral transmission depth analysis cannot accurately reflect the actual distribution after diffusion. This proposed solution effectively eliminates analysis errors introduced by process parameters as constraints.

[0063] Through the above technical solution, this application solves the problem of physical mechanism deviation that may exist when resolving depth distribution by relying solely on optical penetration difference, improves the resolution accuracy of doped element depth distribution, and makes the detection results more consistent with the element distribution state under actual process conditions.

[0064] In some embodiments of this application, before the steps of "identifying the three-dimensional distribution of doping elements based on the spectral feature data and using spectral matching; and matching the spectral feature data with a quantitative relationship model between spectral intensity and doping concentration to obtain the doping concentration of the doping elements", the method further includes the step of preprocessing the acquired hyperspectral data; wherein the preprocessing includes at least one of denoising, background subtraction, and normalization.

[0065] Hyperspectral data preprocessing is a crucial step in eliminating interference and enhancing effective signals, directly impacting the reliability of subsequent analyses. Reducing random noise (such as sensor noise and environmental interference) and improving the signal-to-noise ratio (SNR) are essential to prevent noise from interfering with subsequent quantitative analyses (such as concentration inversion). In this embodiment, Savitzky-Golay filtering can be used for denoising, a method that smooths the spectral curve through local polynomial fitting while preserving peak characteristics. Background subtraction eliminates the influence of non-target signals (such as light source fluctuations, dark current, and stray light), highlighting the material's inherent spectral characteristics. For example, dark current correction, whiteboard correction, baseline correction, and spectral differentiation can be used for background subtraction. Normalization eliminates dimensional differences, making data from different samples or bands comparable and reducing the influence of factors such as light source intensity and distance on the spectrum. For example, maximum value normalization, standard normal transformation, multivariate scattering correction, and band ratio methods can be used for normalization.

[0066] This application preprocesses the acquired hyperspectral data to eliminate interference and enhance the effective signal, thereby improving the reliability of subsequent three-dimensional distribution and concentration analysis of dopant elements.

[0067] In some embodiments of this application, the establishment of the spectral feature library of doped elements includes: establishing characteristic absorption / reflection spectral curves of different doped elements in their activated states through experiments or simulations, wherein the characteristic absorption / reflection spectral curves include peak wavelength parameters and full width at half maximum (FWHM) parameters; and / or

[0068] The establishment of the quantitative relationship model between spectral intensity and doping concentration includes: using training data of samples calibrated with known doping element concentrations, and employing machine learning (such as support vector regression, neural networks) or physical models (such as the Beer-Lambert law correction formula) to achieve concentration inversion.

[0069] For example, phosphorus doping exhibits characteristic reflectance changes in the near-infrared band (e.g., 900-1100 nm), while boron doping shows specific absorption in the visible light band (e.g., 500-600 nm).

[0070] Among them, the characteristic absorption or reflection spectrum curve refers to the spectral response characteristics exhibited by the dopant element under specific energy level transitions. Specifically, it can be achieved by experimental measurement or electromagnetic simulation software to simulate the optical characteristics of the activated dopant element, and is used to accurately identify the type and activation state of the element.

[0071] Among them, the peak wavelength parameter refers to the wavelength position of the characteristic peak of absorption or reflection in the spectral curve. Specifically, it can be achieved by measuring the spectral response of different doped elements and extracting the peak position using a spectrometer, which is used to distinguish the characteristic spectra of different elements.

[0072] The half-width at half-maximum (WHM) parameter refers to the wavelength width of the characteristic peak at half its maximum intensity. It can be calculated by measuring the WHM value of the spectral curve and is used to quantify the broadening of the spectral characteristic peak to improve the specificity of identification.

[0073] Among them, the quantitative relationship model between spectral intensity and doping concentration refers to the algorithm that establishes a mathematical correlation between spectral data and concentration values. Specifically, it can be achieved by using neural network training to calibrate the mapping relationship between the spectral intensity of the sample and the known concentration, or by establishing physical equations based on the modified Beer-Lambert law, which is used to convert spectral signals into concentration values.

[0074] Specifically, when establishing characteristic absorption or reflectance spectral curves, spectral data of the dopant element in its activated state are obtained through experimental measurement or simulation, and the peak wavelength and full width at half maximum (FWHM) are extracted as characteristic parameters and stored in a database. For example, after annealing a boron-doped sample, its absorption spectrum in the visible light band is acquired using a hyperspectral imaging system, determining that the characteristic peak is located at 550 nm and the FWHM is 30 nm. When establishing a quantitative relationship model, a calibration sample set with known concentrations is used, and the acquired hyperspectral data is input into a neural network for training to establish a nonlinear mapping relationship between spectral intensity and concentration. For example, a gradient descent algorithm is used to optimize the neural network weights, minimizing the error between the model's output concentration prediction and the actual measured value.

[0075] Compared with existing technologies, traditional methods rely on destructive detection methods and lack reliable spectral feature libraries, making it impossible to achieve non-destructive element identification and concentration inversion. Existing technologies have not established spectral feature parameters for active-state doped elements, resulting in low element identification accuracy; at the same time, they lack concentration inversion methods that combine machine learning and physical models, making it difficult to adapt to the detection requirements under complex process conditions.

[0076] Through the above technical solutions, this application achieves the technical effect of non-destructive detection of the type and concentration of doping elements. By establishing a spectral feature library containing peak wavelength and full width at half maximum (FWHM) parameters, the spectral fingerprint characteristics of different doping elements can be accurately identified; by fusing a concentration inversion model that integrates data-driven and physical mechanism approaches, hyperspectral data can be converted into quantitative concentration values, solving the technical problem that traditional methods cannot simultaneously achieve element identification and concentration measurement.

[0077] In some embodiments of this application, the step of “capturing spectral characteristic data of specific doped elements in the ion implantation region on the back side of a wafer using a hyperspectral imaging device” includes: the incident angle of the light of different wavelengths is 30°-70°.

[0078] Here, different wavelengths of light refer to light sources from at least two of the ultraviolet, visible, and near-infrared bands. Specifically, this can be achieved using halogen lamps combined with filters or tunable lasers. By switching between different wavelengths of light, the detection of elements at different depths of doping can be achieved. The incident angle of 30°-70° refers to the range of angles formed between the light ray and the normal to the wafer surface. This can be achieved through the mirror angle adjustment mechanism in the tilted incident optical module. By controlling the incident angle, interference from surface reflected light on the effective spectral signal can be suppressed.

[0079] Specifically, a broadband light source illuminates the back side of the wafer at an oblique incidence angle. Short-wavelength light (e.g., ultraviolet) is preferentially absorbed by shallow (<100 nm) doped regions, while long-wavelength light (e.g., near-infrared) penetrates to deeper regions (micrometer scale), creating a multi-layered detection effect. A hyperspectral imaging device acquires the reflectance spectrum line by line in a line-scan manner, constructing cubic data containing two-dimensional spatial coordinates and continuous spectral dimensions. When the incidence angle is controlled within the 30°-70° range, the specular reflection light from the surface deviates from the detector's receiving direction, effectively capturing the diffuse reflection signal generated by the doped regions while ensuring sufficient luminous flux to maintain the signal-to-noise ratio. Through coordinated optimization of wavelength selection and incidence angle, the characteristic absorption peaks of the doped elements are highlighted in the spectral data. Specifically, when short-wavelength light illuminates the back side of the wafer at an incidence angle of 30°-70° (e.g., 50°), its shallow penetration characteristics effectively capture the spectral characteristics of surface doped elements, while the oblique incidence angle deviates the reflected light from the detection path. For deeply doped elements, long-wavelength light is used to irradiate them at an incident angle of 30°-70° (e.g., 65°). Its deep penetration capability can excite the spectral response of the deeply doped elements, and effective signal acquisition is still maintained even with oblique incidence. Hyperspectral imaging equipment records the spectral intensity of each spatial location at different wavelengths through point-by-point scanning, forming a three-dimensional dataset containing lateral position, depth information, and spectral characteristics.

[0080] Compared to existing technologies, traditional methods employ a fixed wavelength and vertical incidence, resulting in aliasing between surface reflected light and effective signals, and failing to distinguish features at different depths. This application, through the coordinated control of wavelength and incident angle, suppresses surface reflection noise and achieves layered extraction of spectral signals at different depths.

[0081] Through the above technical solution, this application effectively eliminates spectral distortion caused by wafer surface reflection, and realizes full-depth spectral feature capture from submicron-level shallow doping to micron-level deep doping, providing a high-fidelity original data foundation for subsequent three-dimensional distribution reconstruction.

[0082] This invention can replace traditional destructive detection methods, improve the monitoring efficiency of ion implantation processes, and is suitable for the stringent requirements of doping distribution in advanced processes such as Fin Field-Effect Transistors (Fin-FETs) or Three-Dimensional Integrated Circuits (3D ICs), and has significant industrial value.

[0083] Please see Figure 3 Secondly, this application also provides a dopant distribution detection system 100 based on hyperspectral imaging technology. The detection system 100 includes a light source 10, a hyperspectral device 20, an oblique incident optical module 30, and a data processing unit 40. The light source 10 is used to emit light of different wavelengths and illuminate the ion implantation region on the back side of the wafer. The hyperspectral device 20 is used to capture the spectral characteristic data of specific dopant elements in the ion implantation region on the back side of the wafer. The oblique incident optical module 30 is used to adjust the incident angle of the light emitted by the light source. The data processing unit 40 is used to store a quantitative relationship model between spectral intensity and doping concentration and a dopant element spectral characteristic library; and compares the spectral characteristic data with the dopant element spectral characteristic library to identify the type and location of the dopant element, inputs the spectral characteristic data into the quantitative relationship model between spectral intensity and doping concentration, and calculates the doping concentration of the dopant element.

[0084] In some embodiments of this application, the data processing unit 40 is further configured to generate a two-dimensional doping concentration distribution map and a three-dimensional depth-concentration profile map and compare them with the design values ​​to achieve process monitoring.

[0085] Among them, the doping element distribution detection system 100 based on hyperspectral imaging technology can be integrated into the semiconductor production line to support online detection and process feedback control.

[0086] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0087] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0088] The embodiments, implementation methods, and related technical features of this application can be combined and substituted for each other without conflict.

[0089] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Although the descriptions of each embodiment in this application have different focuses, and the parts not described in detail in a certain embodiment can be referred to the relevant embodiments in other embodiments, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of this application without departing from the content of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for detecting the distribution of doped elements based on hyperspectral imaging technology, characterized in that, Including the following steps: The ion implantation area on the back side of the wafer is irradiated with light of different wavelengths; among which, the different wavelengths of light include short wavelength light with a wavelength of 300 nm to 450 nm and long wavelength light with a wavelength of 800 nm to 2500 nm. The spectral characteristic data of doped elements in the ion implantation region on the back side of the wafer are captured by a hyperspectral imaging device, and the spectral characteristic data includes spatial-spectral cubic data. The spectral feature data is compared with a dopant element spectral feature library to identify the type and location of the dopant element. The spectral feature data is input into a quantitative relationship model between spectral intensity and doping concentration to calculate the doping concentration of the doping element; In the step of "comparing the spectral feature data with a dopant element spectral feature library to identify the type and location of the dopant element", By utilizing the difference in penetration depth at different wavelengths, spectral data corresponding to the short wavelengths are extracted for the same spatial location to analyze the distribution of shallow doped elements, and spectral data corresponding to the long wavelengths are extracted to analyze the distribution of deep doped elements. Then, by superimposing the spectral matching results of different wavelength layers, a three-dimensional depth distribution model is constructed. In addition, the depth distribution of doped elements is analyzed by combining the implantation energy and annealing process parameters during ion doping.

2. The method for detecting dopant element distribution based on hyperspectral imaging technology as described in claim 1, characterized in that, Following the step of "inputting the spectral feature data into a quantitative relationship model between spectral intensity and doping concentration to calculate the doping concentration of the doping element", the following step is also included: Two-dimensional doping concentration distribution maps and three-dimensional depth-concentration profile maps are generated and compared with design values ​​to achieve process monitoring.

3. The method for detecting dopant element distribution based on hyperspectral imaging technology as described in claim 1, characterized in that, Before the steps of "comparing the spectral feature data with a dopant element spectral feature library to identify the type and location of the dopant element; inputting the spectral feature data into a quantitative relationship model between spectral intensity and doping concentration to calculate the doping concentration of the dopant element", the following steps are also included: Preprocess the acquired hyperspectral data; The preprocessing includes at least one of denoising, background subtraction, and normalization.

4. The method for detecting dopant element distribution based on hyperspectral imaging technology as described in claim 1, characterized in that, The incident angle of the light of different wavelengths is 30°-70°.

5. The method for detecting dopant element distribution based on hyperspectral imaging technology as described in claim 1, characterized in that, The establishment of the dopant element spectral feature library includes: For different doped elements, characteristic absorption / reflection spectra in their activated states are established through experiments or simulations. These characteristic absorption / reflection spectra include peak wavelength parameters and full width at half maximum (FWHM) parameters; and / or The establishment of the quantitative relationship model between spectral intensity and doping concentration includes: Training data of calibrated samples using known dopant element concentrations are used to achieve concentration inversion using machine learning or physical models.

6. A dopant element distribution detection system based on hyperspectral imaging technology, characterized in that, include: A light source is used to emit light of different wavelengths and illuminate the ion implantation area on the back side of the wafer; wherein, the different wavelengths of light include short wavelength light with a wavelength of 300 nm to 450 nm and long wavelength light with a wavelength of 800 nm to 2500 nm. Hyperspectral equipment is used to capture spectral characteristic data of specific doped elements in the ion-implanted region on the back side of a wafer; An inclined incident optical module is used to adjust the incident angle of the light emitted from the light source; and The data processing unit is used to store the quantitative relationship model between spectral intensity and doping concentration and the spectral feature library of doping elements; compare the spectral feature data with the spectral feature library of doping elements to identify the type and location of doping elements; input the spectral feature data into the quantitative relationship model between spectral intensity and doping concentration to calculate the doping concentration of doping elements. In the step of "comparing the spectral feature data with a dopant element spectral feature library to identify the type and location of the dopant element", By utilizing the difference in penetration depth at different wavelengths, spectral data corresponding to the short wavelengths are extracted for the same spatial location to analyze the distribution of shallow doped elements, and spectral data corresponding to the long wavelengths are extracted to analyze the distribution of deep doped elements. Then, by superimposing the spectral matching results of different wavelength layers, a three-dimensional depth distribution model is constructed. In addition, the depth distribution of doped elements is analyzed by combining the implantation energy and annealing process parameters during ion doping.

7. The dopant element distribution detection system based on hyperspectral imaging technology as described in claim 6, characterized in that, The data processing unit is also used to generate two-dimensional doping concentration distribution maps and three-dimensional depth-concentration profile maps and compare them with design values ​​to achieve process monitoring.

Images